Professional Documents
Culture Documents
Platform Economics Essays On Multi Sided Businesses
Platform Economics Essays On Multi Sided Businesses
David S. Evans
David S. Evans
www.competitionpolicyinternational.com
iii
Preface
G
iven the subject of this collection, there is some irony in how I’ve chosen to bring these
essays to you. Publishing has traditionally been a two-sided model. Publishers get authors
and readers together. They typically make their money by charging the reader and giving
some fraction of the earnings to the author as royalties.
This 20th century model of publishing doesn’t serve authors of academic books well. Often,
publishers set the price of academic books relatively high, expecting to earn the greatest profits
from libraries and a handful of aficionados. For most books that aren’t aimed towards a popular
audience, including most academic books, royalties are quite small. Optimistically, I might have
been able to buy a pretty good new bicycle if I had published these essays in the traditional fashion,
but I’d rather have more people read my work than collect the chump change from royalties.
Therefore, the two-sided publishing model fails in two ways: the author doesn’t make much
money, and the author doesn’t get read by very many people. Moreover, most publishers in my ex-
perience are still using 20th century technology to produce and distribute books. It can take many
months—if not years—from a book’s conception to its appearance in a reader’s hands.
And therein lies the paradox. In order to bring my work into the 21st century, I have de-
cided to publish my collection of essays about two-sided markets in a one-sided way. I ditched
the intermediary and chose to connect directly with likely readers. I’m sure some of you would
prefer the feel of paper and leather but hopefully the price is right. It was easy for me to decide
to make this volume free (a bit more on Amazon) because it cost almost nothing to produce and
distribute it.
An earlier version of this book appeared in 2010. It consisted of a series of urls (website ad-
dresses) that took readers to the original papers which they could then download. I promised
a real e-book in the early part of 2011. At least I got the year right which for an economist is
pretty good.
David S. Evans
iv
Acknowledgements
I
would like to thank Richard Schmalensee for his collaboration on several of the articles and
books on multi-sided platforms we have done over the last decade as well as Jean-Charles
Rochet and Jean Tirole who have been generous with their comments and time. I also grate-
fully acknowledge research support from Microsoft and Visa for some of the articles. None of
them necessarily agrees with anything I say of course. I also want to extend my appreciation to
Almira Joy Bautisa and Dean Whitney of Aericon who led the effort to design and assemble
this book.
v
Contents
Preface ................................................................................................................................................. iii
Acknowledgements ............................................................................................................................ iv
Introduction.......................................................................................................................................... vi
Bibliography...................................................................................................................................... 430
vi
Introduction
T
his volume collects a series of essays that I have written, sometimes with colleagues, over
the last decade on businesses that create value by providing products that enable two or
more different types of customers to get together, find each other, and exchange value.
These businesses were called “two-sided markets” in the seminal paper by Jean-Charles Rochet and
Jean Tirole that was first circulated in 2001.1 I typically avoid this term since it tends to obscure
the fact that we are talking about businesses rather than markets. I prefer the term “multi-sided
platforms” because these businesses provide a place for customers to meet and interact and often
support more than two interdependent types of customers. In writing for business audiences I use
the term “catalyst” to denote the fact that these businesses create value that couldn’t be had without
bringing these customer types together.2 The term “two-sided markets” has stuck, though, and I
will use it here.
The Rochet and Tirole paper ignited work on two-sided markets in economics, law, and busi-
ness. Several other papers, in circulation around 2000, touched on some of the interesting aspects
of intermediaries3 or on the increasingly widespread phenomenon of giving one product away for
free to attract revenue from another product.4 The Toulousians’ contribution was fundamental be-
cause it recognized for the first time that a very diverse set of businesses were two-sided, presented
an elegant economic model of them, and derived several robust aspects of these businesses includ-
ing the critical importance of the price structure (the relative prices charged to the various types of
customers) in their making money.
Economic theorists and empiricists were quickly attracted to this topic. The mill of articles
and dissertations has flourished ever since.5 It soon became apparent that this new area had
important implications for antitrust. Many competition authorities took notice and law review
1
Jean-Charles Rochet and Jean Tirole, “Platform Competition in Two-Sided Markets,” Journal of the European
Economic Association 1, no. 4 (2003): 990-1209.
2
David S. Evans and Richard Schmalensee, Catalyst Code: The Strategies behind the World’s Most Dynamic
Companies (Cambridge, MA: Harvard Business School Press, 2007).
3
Bernard Caillaud and Bruno Jullien, “ Chicken & Egg: Competition among Intermediation Service Providers,”
RAND Journal of Economics, The RAND Corporation, vol. 34(2), pages 309-28, Summer.
4
Geoffrey Parker and Marshall Van Alstyne (2005). “Two-Sided Network Effects: A Theory of Information Product
Design.” Management Science, Vol. 51, No. 10.
5
With the recent important work of Glen Weyl we have moved on to Two-Sided Markets 2.0. See Glen Weyl, “A Price
Theory of Multi-Sided Platforms,” American Economic Review, 2010, 100(4).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Introduction vii
articles addressed various aspects of these newly-recognized business forms.6 The corporate world
seized on this new field. Strategy articles and courses focused on these platforms began appearing.7
Companies found the subject eye opening, including firms that were multi-sided platforms but
had not quite understood the ramifications.8
My contributions to this literature are presented in this volume. Part I presents background
pieces on the economics of multi-sided platforms and industries in which these platforms are
common. Part II examines the antitrust economics of two-sided markets including the difficult
problem of defining the boundaries of competition. Part III comprises several papers that apply
two-sided market analysis to web-based businesses. Part IV does the same for payment cards which
is the industry that attracted much of the early two-sided analysis— in part because this framework
was helpful for understanding the hotly debated issue of interchange fees. Part V collects several ar-
ticle and book chapters on software platforms. These platforms have become especially important
in the last several years because they are now the basis for revolutionary developments with mobile
devices (e.g. the iPhone and Android), social networking (Facebook in particular), and payments
(PayPalX). The essays are published as originally written (usually, in fact, whatever version I could
make freely available).
When the theory of two-sided markets was first introduced it was common to hear at least
two complaints. The first was that there was nothing new—from economists who suggested that
it was just the indirect network effects wine in new bottles or antitrust analysts who commented
that it had all been considered before in advertising cases. The second was that it was a theory of
everything, and therefore nothing, since everything seems to be two-sided.
There’s some truth, of course to both arguments. Indirect network effects are usually essential
to understanding two-sided markets. But the network effects literature missed the importance of
these effects for a very diverse group of industries among other things; the literature spent a lot of
time on fax machines and video standards but not so much on more general business issues such as
pricing or industries such as shopping malls that did not obviously have indirect network effects.
Two-sided market analysis is at least a richer and more subtle wine.
One of the problems with two-sided market analysis is that it is hard to find formal limiting prin-
ciples. But that isn’t uncommon in economics. Sometimes a two-sided market perspective is highly
informative while other times it isn’t. It matters when it matters. What’s now very clear, with the
benefit of a decade of work, is that the study of multi-sided platforms has provided valuable insights
to economists, policymakers and business people, as I hope the chapters in this volume demonstrate.
6
OECD Organization for Economic Co-operation and Development (“OECD”), Two-Sided Markets, DAF/COMP
(2009) 20, 28.
7
See, for example, Thomas Eisenmann “Managing Networked Businesses: Course Overview.” Harvard Business
Online , 2006;Thomas Eisenmann, Geoffrey Parker, and Marshall Van Alstyne, “Strategies for Two-Sided Markets.”
Harvard Business Review, 2006.
8
Companies in the payment card industry, for example, have fundamentally changed how they think about that
business as a result of the two-sided market concepts.
Part One
ECONOMICS OF MULTI-SIDED
PLATFORM BUSINESSES
CHAPTER
The Industrial Organization of
1 Markets with Two-Sided Platforms
David S. Evans and Richard Schmalensee
ABSTRACT
Many diverse industries are populated by businesses that operate “two-sided platforms.” These
businesses serve distinct groups of customers who need each other in some way, and the core
business of the two-sided platform is to provide a common (real or virtual) meeting place and to
facilitate interactions between members of the two distinct customer groups. Platforms play an
important role throughout the economy by minimizing transactions costs between entities that
can benefit from getting together. In these businesses, pricing and other strategies are strongly
affected by the indirect network effects between the two sides of the platform. As a matter of the-
ory, for example, profit-maximizing prices may entail below-cost pricing to one set of customers
over the long run and, as a matter of fact, many two-sided platforms charge one side prices that
are below marginal cost and are in some cases negative. These and other aspects of two-sided
platforms affect almost all aspects of antitrust a
nalysis—from market definition, to the analysis
of cartels, single-firm conduct, and efficiencies. This chapter provides a brief introduction to the
economics of two-sided platforms and the implications for antitrust analysis.
I. INTRODUCTION
Many diverse industries are populated by businesses that operate “two-sided platforms.” These
businesses serve distinct groups of customers who need each other in some way, and the core
business of the two-sided platform is to provide a common (real or virtual) meeting place and to
facilitate interactions between members of the two distinct customer groups. Two-sided platforms
are common in old-economy industries such as those based on advertising-supported media and
new-economy industries such as those based on software platforms and web portals. They play an
important role throughout the economy by minimizing transactions costs between entities that
can benefit from getting together.
In these businesses, pricing and other strategies are strongly affected by the indirect network
effects between the two sides of the platform. As a matter of theory, for example, profit-maxi-
mizing prices may entail below-cost pricing to one set of customers over the long run and, as
2
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 3
a matter of fact, many two-sided platforms charge one side prices that are below marginal cost
and are in some cases negative. These and other aspects of two-sided platforms affect almost all
aspects of antitrust analysis—from market definition, to the analysis of cartels, single-firm con-
duct, and efficiencies.1
This paper provides a brief introduction to the economics of two-sided platforms and the im-
plications for antitrust analysis.
Two-sided platforms were first identified clearly in pioneering work by Jean Charles Rochet
and Jean Tirole, which began circulating in 2001.2 A significant theoretical and empirical litera-
ture quickly emerged, and the subject has become a very active area of research in economics.3 For
the purposes of this paper, it is helpful to clarify some terminology that is used in the economics
literature and which sometimes causes confusion. Rochet and Tirole used the term “two-sided
markets” to refer to situations in which businesses cater to two interdependent groups of custom-
ers. The term “market” was meant loosely and does not refer to how that term is often used in
antitrust. This paper refers to “two-sided platforms” but it is synonymous with “two-sided markets”
as used in much of the economics literature. How to determine what market a two-sided platform
competes in, from an antitrust perspective, is one of the questions considered here.4 Two-sided
platforms often compete with ordinary (single-sided) firms and sometimes compete on one side
with two-sided platforms that serve a different second side.
1
See David S. Evans, The Antitrust Economics of Multi-Sided Platform Markets, 20 YALE J. ON REG. 325 (2003) and
Julian Wright, One-Sided Logic in Two-Sided Markets, 3 Rev. of Network Econ. 44 (2004).
2
Jean-Charles Rochet & Jean Tirole, Platform Competition in Two-Sided Markets, 1 J. Eur. Econ. ASS’N 990 (2003).
Some of the key issues were identified in the context of payment cards in an important contribution Wlliam F.
Baxter, Bank Exchange of Transactional Paper: Legal and Economic Perspectives, 26 J.L. & Econ. 541 (1983). There are
also literatures for particular industries that also provide precursors.
3
See Conference on Competition Policy in Two-Sided Markets (Institute d’Economie Industrielle, U. Toulouse)
(Jun. 29 - Jul. 1, 2006), available at http://idei.fr/doc/conf/tsm/programme.pdf.
4
Although, for the most part, we will use the term two-sided platform the reader should note that some platforms
have more than two distinct groups of customers. Digital media platforms, for example, often have four: users, de-
velopers, hardware makers, and content providers.
PLATFORM ECONOMICS:
4 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
one way to get the balance right. The club might want to offer women a break if they are in short
supply (through a lower price or free drinks). Or it might want to ration the spots to ensure the
appropriate number of women; popular clubs typically have queues waiting outside, and women
are picked out of line disproportionately.
The dating club example motivates the informal definition of a two-sided platform that we
introduced in the beginning paragraph. There are two groups of customers—men and women.
Members of each group value members interacting with members of the other group. And the
platform provides a place for them to get together and interact. By doing so it enables members of
these two groups to capture various benefits from having access to each other.
In their 2006 paper, Rochet and Tirole have proposed a formal definition:
“A market5 is two-sided if the platform can affect the volume of transactions by charging
more to one side of the market and reducing the price paid by the other side by an equal
amount; in other words, the price structure matters, and platforms must design it so as to
bring both sides on board.”6
To satisfy this definition, “the relationship between end-users must be fraught with residual
externalities” that customers cannot sort out for themselves.7 That is clear in the case of the dating
environment. In contrast, in the textbook wheat market there are no externalities connecting buy-
ers and sellers, and the price structure doesn’t matter: a tax on wheat levied on buyers has the same
effect on quantity as the same tax levied on sellers.
In addition, it must not be possible for the two sides to arbi- Think of two-sided
trage their way around the price structure chosen by the platform. platforms as arising in
Men and women, for example, want to be able to search for dates situations in which there
among a large number of opposites. It is hard to conceive of a prac- are externalities and in
tical mechanism for women to reward men who come to a singles which transactions costs,
club but who they reject. Likewise, for the other two-sided platform broadly considered,
industries we consider it is difficult, if not impossible, for customers prevent the two sides
on one side to make side payments to customers on the other side. from solving this
As a result the platform owner can institute a pricing structure to externality directly.
harness indirect network effects, and it is not feasible for customers
5
Note that the word market below is being used in the loose manner that is the custom among economists and not
in the antitrust sense. The Rochet-Tirole definition would be more precise if it said “A two-sided platform business
exists if ....”
6
Jean-Charles Rochet & Jean Tirole, Two-Sided Markets: A Progress Report, RAND J. Econ. (Autumn 2006).
7
As a result a necessary condition for a market to be two-sided is that the Coase theorem does not apply to the transac-
tion between the two sides. For more details, see Rochet & Tirole (2006), id.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 5
to defeat this pricing structure through arbitrage. Generally, one can think of two-sided platforms
as arising in situations in which there are externalities and in which transactions costs, broadly
considered, prevent the two sides from solving this externality directly. The platform can be
thought of as providing a technology for solving the externality in a way that minimizes transac-
tions costs.
It is helpful to review four different types of two-sided platforms: exchanges, advertiser-
supported media, transaction devices, and software platforms.8
A. Exchanges
Exchanges have two groups of customers, who can generally be considered “buyers” and “ sellers.”
The exchange helps buyers and sellers search for feasible contracts—that is where the buyer and
seller could enter into a mutually advantageous trade—and for the best prices—that is where
the buyer is paying as little as possible and the seller receiving as much as possible. (In organized
exchanges, As a result a necessary condition for a market to be two-sided is that the Coase theo-
rem does not apply to the transaction between the two sides. For more details, see Rochet &
Tirole (2006), id. such as the New York Stock Exchange, it is often more useful to think of the
two sides as liquidity providers—specialists or market-makers who quote prices to both buyers
and sellers and thus bring liquidity to the market—and liquidity consumers—ordinary custom-
ers who accept liquidity providers’ offers.9) We use the term buyers and sellers here loosely. The
term, “ exchanges,” covers various match-making activities such as dating services and employment
agencies. It also covers traditional exchanges such as auction houses, internet sites for business-to-
business, p erson-to-business, and person-to-person transactions, various kinds of brokers (insur-
ance and real estate) and financial exchanges for securities and futures contracts. Finally, exchanges
include a variety of businesses that provide brokerage services. These include publishers (readers
and authors), literary agents (authors and publishers), travel services (travelers and travel-related
businesses), and ticket services (people who go to events, and people who sponsor events).
Exchanges provide participants with the ability to search over participants on the other side
and the opportunity to consummate matches. Having large numbers of participants on both sides
increases the probability that participants will find a match. Depending on the type of exchange,
8
For discussion, see David S. Evans, Andrei Hagiu, & Richard Schmalensee, Invisible Engines: How Software
Platforms Drive Innovation and Transform Industries, ch. 3 (MIT Press 2006). We refer there to software
platforms more generally as shared input facilities. Armstrong uses the term “competitive bottle-necks” to refer to
certain shared-input facilities. Although his discussion is analytically sound, his term is pejorative and has a meaning
in competition law that differs from the one he assigns to it. See Mark Armstrong, Competition in Two-Sided
Markets (EconWPA, working paper, 2005).
9
Bernhard Friess & Sean Greenaway, Competition in EU Trading and Post-Trading Service Markets, 2 Competition
Pol’y Int’l (2006).
PLATFORM ECONOMICS:
6 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
however, a larger number of participants can lead to congestion. That is the case with physical
platforms such as singles clubs or trading floors. Moreover, participants may derive some value
from having the exchange prescreen participants to increase the likelihood and quality of matches.
Some exchanges charge only one side. For example, only sellers pay directly for the services
provided by eBay. This is also true for real-estate sales in the United States. Other exchanges charge
both sides, although the prices may bear little relation to side-specific marginal costs. Internet
matchmaking services charge everyone the same, for instance, while, as we mentioned, physical
dating environments sometimes charge men more than women. Auction houses charge commis-
sions to buyers and sellers. Insurance brokers historically charged both insurance customers and
insurance providers in some types of transactions (some have agreed not to charge both as a result
of settlements of lawsuits brought by the New York State Attorney General).
10
See, e.g., James M. Ferguson, Daily Newspaper Advertising Rates, Local Media Cross-Ownership, Newspaper Chains,
and Media Competition, 26 J.L. & Econ. 637 (1983) (“Readership studies show that advertising, especially retail
advertising, is considered as important as, or more important than, editorial content.”) and R.D. Blair & R.E.
Romano, Pricing Decisions of the Newspaper Monopolist, 59 Southern Econ. J. 731 (1993) (“circulation demand
rises with increases in the quantity of advertising”).
Other studies have shown that, unlike Americans, readers in certain European countries are averse to advertising.
See, e.g., Nathalie Sonnac, Readers’ Attitudes Toward Press Advertising: Are They AdLovers or Ad-Averse?, 13 J. Media
Econ. 249 (2000). On the other hand, TiVo and other related products that permit ad avoidance and deletion are
very popular currently, with one study citing that TiVo viewers skip about 60 percent of commercials. See A Farewell
to Ads?, The Economist, Apr. 15, 2004.
11
In a two-sided platform there is no rigorous way to define the profit “earned” by one side or the other. Not only are
there typically costs that are common to both sides (the floor of the New York Stock Exchange, for instance), outlays
that build business on one side of the market (via product enhancement, say) will also tend, via the externality, to
build business on the other side. By “gross margin” we mean the difference between revenue and the variable costs,
if any, that depend entirely on the volume on only one side of the market. The cleanest examples of such a cost
would be the manufacturing costs of videogame consoles or the marginal printing costs of newspapers or yellow
page directories.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 7
to or below the marginal cost of printing and distribution.12 In some cases—such as yellow page
directories and some newspapers—they are provided for free. Free television is just that. And most
web portals—Google and Yahoo for example—receive revenue only from advertisers.
12
Blair & Romano, supra note 10.
PLATFORM ECONOMICS:
8 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
v ariable transaction prices. Card associations such as MasterCard and Visa have been examples of
cooperative two-sided platforms. For a transaction to be consummated there has to be an agree-
ment on the division of profits and the allocation of various risks between the entity that services
the merchant and the entity that services the cardholder. Most card associations set this centrally
as, in effect, a standard contract between the businesses that service the two sides. Typically, they
agree that the entity that services the merchant pays a percentage of the transaction—the “inter-
change fee”—to the entity that services the cardholder. This fee ultimately determines the relative
prices for cardholders (issuers obtain a revenue stream which they compete for) and merchants
(acquirers pass the cost of the interchange fee onto merchants). This centrally set fee has been the
subject of litigation and regulatory scrutiny, as we discuss below.13
13
David S. Evans & Richard Schmalensee, The Economics of Interchange Fees and Their Regulation:
An Overview (MIT Sloan, Working Paper, 2005), in Interchange Fees in Credit and Debit Card Industries 73-120
(Federal Reserve Bank of Kansas City, 2005).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 9
programs, and reduces learning costs.14 An important consequence of this reduction in cost is an
increase in the supply of applications for the platform, an increase in the value of the software
platform to end users, and positive feedback effects to application developers.
14
See Evans, Hagiu, & Schmalensee, supra note 8.
15
See David S. Evans & Richard Schmalensee, Catalyst Code: The Strategies Behind the World’s Most
Successful Companies (Harvard Business School Press 2007).
16
That is, the models are based on assuming particular functional forms—e.g. linear—for relationships.
PLATFORM ECONOMICS:
10 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
A. Pricing
To see the intuition behind pricing consider a platform that serves two customer groups A and B.
It has already established prices to both groups and is considering changing them.17 If it raises the
price to members of group A fewer As will join. If nothing else changed the relationship between
price and the number of As would depend on the price elasticity of demand for As. Since members
of group B value the platform more if there are more As fewer Bs will join the platform at the
current price for Bs. That drop-off depends on the indirect network externality which is measured
by the value that Bs place on As. But with fewer Bs on the platform, As also value the platform
less leading to a further drop in their demand. There is a feedback loop between the two sides.
Once this effect is taken into account, the effect of an increase in price on one side is a decrease in
demand on the first side because of the direct effect of the price elasticity of demand and on both
sides as a result of the indirect effects from the externalities.
A few equations will make this point more sharply for readers familiar with the concept
of elasticity. The situation described just above can be summarized by two demand functions:
Q A 5 D A(P A, Q B) and Q B 5 D B(P B, Q A). The first of these gives participation by members of
group A as a function of the price charged to group A and participation by group B, and the second
gives participation by members of B similarly. Let e I 5 2(D I/P I )(P I/Q I ), for I 5 A,B. These are
the own-price elasticities for each group, holding constant participation by the other (i.e., ignoring
the externalities linking the two groups). Let uIJ 5 (D I/Q J )(Q J/Q I ) for I, J 5 A, B and I J.
These elasticities measure the strengths of the externalities connecting the two groups. In the nor-
mal two-sided case, both would be expected to be positive. Finally, let E I 5 2(dQ I/dP I )(P I/Q I )
for I 5 A, B. These are the ordinary own price elasticities, computed assuming other prices remain
constant but allowing participations (quantities) to vary. Differentiating both demand functions
totally with respect to either price, and solving, yields:
Even if the As are not particularly price-sensitive, and as long as the externalities between the
groups are strong (in either direction!), participation by group A may be highly sensitive to the
price its members are charged, and similarly for group B. Even a small response by group A to a
price change will trigger a response by group B, which in turn will produce a response by A, and
so on. (The equation above assumes that these response sequences converge.)
The platform of course would like to find the prices that maximize its profits by taking these
same sorts of considerations into account. For a single-sided business that would occur by selecting
17
To keep matters simple we consider the case where each side is charged a membership fee as in Mark Armstrong,
Competition In Two-Sided Markets (EconWPA, Working Paper, 2005). More generally, platforms are natural
businesses for two-part tariffs involving an access fee and a usage fee.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 11
the output at which marginal revenue equals marginal cost and then charging the corresponding
price for this quantity from the demand curve. (This equilibrium is often described by the Lerner
formula that says that the price marginal-cost margin equals the inverse of the own-price elasticity
of demand.) For two-sided platforms three results appear to be robust:
1. The optimal prices depend in a complex way on the price sensitivity of demand on both
sides, the nature and intensity of the indirect network effects between the two sides, and the
marginal costs that result from changing output of each side.
2. The profit-maximizing, non-predatory price for either side may be below the marginal cost
of supply for that side or even negative.
3. The relationship between price and cost is complex, and the simple formulas that have been
derived for single-sided markets do not apply.
For many platforms it is possible to charge two different kinds of prices: an access fee for join-
ing the platform and a usage fee for using the platform. Although these are interdependent, one
can think of the access fee as mainly affecting how many customers join the platform and the usage
fee as mainly affecting the volume of interactions between members of the platform. Most software
platforms charge access fees to users—they have to license the software platform but then can use
it as much as they want—and do not charge access or usage fees to developers. Videogame console
vendors, though, charge a usage fee to game developers—a royalty based on the numbers of games
that are sold; users pay this usage fee indirectly through their purchase of games for the console.
Payment card systems generally charge merchants a usage fee but no access fee. Cardholders may
pay an access fee (the annual card fee); they often pay either no usage fee or a negative one (to the
extent they receive rewards based on transactions volume).
The profit-maximizing reliance on access versus usage fees depends on many factors includ-
ing the difficulty of monitoring usage and the nature of the externality between the two sides.
Cardholders care about card acceptance, for instance, while merchants care about usage. It thus
seems sensible not to charge merchants for access and not to charge consumers for usage.
The empirical evidence suggests that prices that are at or below marginal cost are common for
two-sided platforms. Table 1 summarizes some relevant evidence.
18
This table shows pricing structures that are common in these industries. In many cases, fees will differ from these
pricing structures. For example, some clubs offer free entry to women, some magazines offer free subscriptions, some
videogame players pay fees for on-line play, and some payment cardholders do not pay fees for their cards and/or
get usage based rewards. For dating clubs, usage fees for men and women refer to fees for drinks in the club. For real
estate, the usage fee for sellers refers to the fee for selling a house; there is typically no fee for using the system to list
or show a house. For shopping malls, the negative usage fee for shoppers refers to the free parking that is commonly
available. For videogame consoles, players do not pay a fee for using the console, although they do pay for video
games to the game developer (which in some cases is the same firm that makes the console and in other cases pays
a royalty to the console manufacturer). For payment cards, cardholders are also subject to penalty fees, such as for
exceeding credit limits or for late payments; we have not included these fees in the table.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 13
fore the number of potential shoppers. Shopping malls are sometimes designed to encourage shop-
pers to pass by many stores (e.g., by putting the up and down escalators at different ends of the mall).
Advertising-supported media are another obvious example. Viewers would like to gain access
to the content—and perhaps even the advertisements of their choice—in the most convenient way.
Some magazines are laid out to make it difficult even to find the table of contents or to find the
continuation of an article without thumbing through many advertisements. Television watchers
might benefit from having advertisements clustered at the beginning or the end of each program,
but television providers (in the United States, at least) typically intersperse the advertisements and
precede them perhaps with a cliffhanger to discourage viewers from taking a long break.
Two-sided platforms may also bundle features that directly benefit side A but harm side B (put-
ting aside the indirect externalities from increasing the participation of side A).19 All software plat-
forms include features for example that do not benefit most users. However, some developers value
each of these features and in particular value knowing that any user of the software will have that
feature and therefore be able to run its applications. All payment card systems require merchants
that take their cards for payment to take any of their cards for payment, regardless of who presents
it or which entity issued it. Some merchants would benefit from being selective—taking cards only
from people who lack cash, for example. But this would reduce the confidence that cardholders
have that their cards will be taken at stores that display the acceptance mark. (We will see later that
special cases of these requirements, linking acceptances of credit and debit cards, have given rise to
tying claims. This paragraph is not meant to suggest that tying could not be used in an anticom-
petitive way by two-sided platforms but rather to point out that there is an additional efficiency
explanation for at least one aspect of this practice that does not arise in one-sided businesses.)
19
See Rochet & Tirole (2006), supra note 6.
PLATFORM ECONOMICS:
14 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
ample, has rules that govern the appearance of cards issued by mem- Cooperative two-sided
bers, to provide some uniformity for the common brand, as well platforms have further
as to prevent members from using the brand inappropriately. The need for rules and
system also has rules that address disputed transactions. Acquirers regulations because
would have an incentive to favor their customers (merchants) in a the behavior of their
dispute while issuers would favor their customers (cardholders). The members can affect the
system’s rules attempt to find a balance between these competing value of the two-sided
interests, to increase the attractiveness of the system as a whole. platform as a whole.
with more customers on both sides more valuable to both sets of customers. To take another ex-
ample, a payment card system whose cards are taken at more merchants is more valuable to card
users—that is why we see card systems touting their acceptance (“MasterCard: No card is more
accepted.”) in consumer advertisements.
If there were no countervailing factors, we would expect that indirect network effects would
lead two-sided platforms to compete for the market. First movers would have an advantage, all else
being equal. We would have the familiar story that the firm that obtains a lead tends to widen that
lead as a result of positive feedback effects and therefore wins the race for the market.20 Other firms
could compete with this advantage only if they offered consumers on either side something that
offset the first mover’s size advantage.
Indirect network effects may decline with the size of the platform. For example, the probability
of finding a match increases at a diminishing rate with the number of individuals on either side
(buyers or sellers, men or women).21 At some point positive externalities from more participants
may turn into negative externalities in the form of congestion as discussed below.
20
See, e.g., David S. Evans & Richard Schmalensee, A Guide to the Antitrust Economics of Networks, 10 Antitrust
Mag. 36 (1996) and Carl Shapiro & Hal R. Varian, Information Rules: A Strategic Guide to the Network
Economy (Harvard Business School Press 1999).
21
See Evans, supra note 1.
PLATFORM ECONOMICS:
16 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
complex, it has become more expensive and time consuming to add features and functionality.
The most recent version of the Apple OS took four months longer to develop than the previous
version.22 Microsoft’s Vista operating system has also been plagued with very long delays.
22
For Apple OS release dates, see Jason Snell, Jaguar unleashed: Mac OS X 10.2 Arrives, Macworld, Sept. 1, 2002;
Sarah Stokely, Apple Sets Panther Release Date, IDG Data, Oct. 10, 2003.; and, Steven Musil, This Week in Tiger:
Apple releases Mac OS X 10.4, CNET NEWS, Apr. 29, 2005.
23
For a general discussion on matching, search, and congestion see, for example, Robert Shimer & Lones Smith,
Matching, Search, and Heterogeneity, 1 Advances In Macroeconomics (2001) and Mark Rysman, Competition
Between Networks: A Study of the Market for Yellow Pages, 71 Rev. Econ. Studies 483 (2004b).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 17
example of multi-homing on both sides. Most merchants accept credit and debit cards from several
systems, including ones that have relatively small shares of cardholders. Many cardholders carry
multiple cards, although they may tend to use a favorite one most often.24 Advertising supported
media also has multi-homing on both sides—advertisers and viewers rely on many differentiated
platforms. Other two-sided platforms have multi-homing only on one side. Most end-users rely on
a single software platform for their personal computers, for instance, while many developers write
for several platforms.
24
Mark Rysman, An Empirical Analysis of Payment Card Usage (Boston University Department of Economics,
Working Paper, 2004).
PLATFORM ECONOMICS:
18 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
Residential Buyer Uncommon: Multi-homing may be unnecessary, Fifty largest firms have a 23%
Property Seller since a multiple listing service allows the listed share. (2002)
Brokerage property to be seen by all member agencies’
customers and agents.
Securities Buyer Common: The average securities brokerage client Four largest firms accounted
Brokerage Seller has accounts at three firms. Note that clients can for 37% of in securities bro-
be either buyers or sellers or both. kerage and 16% in financial
portfolio management. (2002)
Newspapers Reader Common: In 1996, the average number of maga- Wall Street Journal had a
and Magazines Advertiser zine issues read per person per month was 12.3. 28% share of the five largest
Also common for advertisers: for example, AT&T newspapers. (2001)
Wireless advertised in the New York Times, The
Wall Street Journal, and Chicago Tribune, among
many other newspapers, on Aug. 26, 2003.
Network Viewer Common: For example, viewers in Boston, U.S. law forbids broadcast-
Television Advertiser Chicago, Los Angeles, and Houston, among other ers from owning TV stations
major metropolitan areas, have access to at least reaching more than 35% of
four main network television channels: ABC, CBS, the nation’s television audi-
FOX, and NBC. Also common for advertisers: for ence.
example, Sprint places television advertisements
on ABC, CBS, FOX, and NBC.
Operating End User Uncommon for users: Individuals typically use Microsoft has a 96% share of
System Application only one operating system. Common for devel- revenue of client operating
Developer opers: As noted earlier, the number of develop- systems. (2004)
ers that develop for various operating systems
indicates that developers engage in significant
multi-homing.
Video Game Game Varies for players: The average household (that Sony PS1 and PS2 had a 63%
Console Player owns at least one console) owns 1.4 consoles. share of console shipments in
Game Common for developers: For example, in 2003, North America. (2003)
Developer Electronic Arts, a game developer, developed for
the Nintendo, Microsoft, and Sony platforms.
Payment Card Cardholder Common: Most American Express cardholders The Visa system had a 45%
Merchant also carry at least one Visa or MasterCard. In share of all credit, charge,
addition, American Express cardholders can use and debit purchase volume.
Visa and MasterCard at almost all places that (2004)
take American Express.
Source: Adapted from David S. Evans, The Antitrust Economics of Multi-Sided Platform Markets, 20 Yale J. ON Reg. 325 (2003). Industry
share data from United States Census Bureau, 2002 Economic Census, available at http://www.census.gov/econ/census02/guide/
INDSUMM.HTM; “Top 20 U.S. Daily Newspapers by Circulation,” Newspaper Association of America (2001), at http://www.naa.org/info/
facts01/18_top20circ/index.html (accessed Feb. 21, 2007); Stephen Labaton, U.S. Backs Off Rules for Big Media, NY TIMES, Jan. 28, 2005;
A. Gillen & D. Kusnetzky, Worldwide Client and Server Operating Environments 2004-2008 Forecast, IDC Market Analysis, No. 32452 (Dec.
2004); Schelley Olhava, Worldwide Videogame Hardware and Software 2004-2008 Forecast and Analysis, IDC Market Analysis, No. 31260
(May 2004); The Nilson Report, No. 828 (Feb. 2005); The Nilson Report, No. 833 (May 2005).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 19
Transaction
Times Picayune Monopolization
Systems
Media
Platforms
Software
Tying
Exchanges
ally familiar.26 The cases span all of the major categories of two-sided platforms and involve the
spectrum of competition policy issues. This section summarizes some key issues that arose in
several of these cases.
A. NaBanco
In NaBanco v. Visa, the federal district court and the U.S. Court of Appeals for the Eleventh
Circuit recognized several of the key features of what have become known as two-sided platforms.
Visa was (and is) a cooperative of banks that issued cards and acquired those card transactions from
merchants. It established a rule for governing the situation in which an individual whose card was
25
United States v. Times-Picayune Publishing Co., 345 U.S. 594 (1953); Joined Cases C-241/91 P and C-242/91
P, RTE, BBC, and ITP v. Commission of the European Communities (Magill), 1995 E.C.R. I-00743 (Apr. 6,
1995); U.K. Competition Commission, Classified Directory Advertising Services (1996); U.K. Office
of Fair Trading, Classified Directory Advertising Services: Review of Undertakings Given by BT to
the Secretary of State in July 1996 (2001); United States v. Lorain Journal Co., 342 U.S. 143 (1951); United
States v. Taubman, 297 F.3d 161 (2d Cir. 2002); State of New York v. Marsh & McLennan Companies, Inc., et al.,
Complaint filed October 14, 2004, Index No. 04-403342; U.K. Competition Commission, A Report on the
Proposed Acquisition of London Stock Exchange PLC by Deutsche Börse AG or Euronext NV (2005);
U.S. Department of Justice, Department of Justice Antitrust Division Statement on the Closing of
its Two Stock Exchange Investigations (Nov. 16, 2005); U.K. Office of Communications, Wholesale
Mobile Voice Call Termination (Jun. 1, 2004); National Bancard Corp. v. Visa U.S.A., Inc., 779 F.2d 592,
602 (11th Cir. 1986); In re Visa Check/MasterMoney Antitrust Litigation, 192 F.R.D. 68 (E.D.N.Y 2000); United
States v. Microsoft, 87 F. Supp. 2d 30 (D.D.C. 2000); Commission of the European Communities v. Microsoft,
Case COMP/C-3/37.792/Microsoft; Atari Games Corp. v. Nintendo, 975 F.2d 832 (Fed. Cir. 1992).
26
See J. Wotton’s article in this issue (John Wotton, Are Media Markets Analyzed as Two-Sided Markets?, (3)1
Competition Pol’y Int’l 237–47 (2007)).
PLATFORM ECONOMICS:
20 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
issued by bank A paid with that card at a merchant acquired by bank B, where A and B are differ-
ent banks. Although those banks could have a bilateral agreement, Visa established a default rule
that among other things determined the allocation of the profits and risks of the transaction. This
rule provided that given the various allocations of risks and costs that the bank that acquired the
transaction (B) had to pay the bank (A) that issued the card a percent of the transaction amount;
this percent is known as the interchange fee, and it was initially set at 1.95 percent.
NaBanco argued that the interchange fee violated Section 1 of the Sherman Act because it was
a price set collectively by competitors. Visa argued that unlike classic price-fixing, the ability to
set an interchange fee was a mechanism to allocate costs between the issuing and acquiring sides
of the business and enhanced output by, among other things, limiting opportunistic behavior by
individual members and avoiding the chaos of bilateral negotiations among thousands of member
banks. The Eleventh Circuit concluded:
“Another justification for evaluating the [interchange fee] under the rule of reason is because it is a
potentially efficiency creating agreement among members of a joint enterprise. There are two possible
sources of revenue in the VISA system: the cardholders and the merchants. As a practical matter, the
card-issuing and merchant-signing members have a mutually dependent relationship. If the revenue
produced by the cardholders is insufficient to cover the card-issuers’ costs, the service will be cut back or
eliminated. The result would be a decline in card use and a concomitant reduction in merchant-signing
banks’ revenues. In short, the cardholder cannot use his card unless the merchant accepts it and the mer-
chant cannot accept the card unless the cardholder uses one. Hence, the [interchange fee] accompanies
“the coordination of other productive or distributive efforts of the parties” that is ‘capable
of increasing the integration’s efficiency and no broader than required for that purpose.’”27
Professor William Baxter worked for Visa on this matter. His 1983 article in the Journal of Law
and Economics presented many of the key concepts of two-sided markets within the context of the
determination of interchange fees.28 The modern literature now recognizes that the interchange fee
is at least partly a device for determining the pricing structure for the card system.29 Some regula-
tors and antitrust authorities, while recognizing the two-sided nature of the business, have argued
in recent years that the interchange fee is set at a level that encourages the overuse of cards.
27
National Bancard Corp. v. Visa U.S.A., Inc., 779 F.2d 592, 602 (11th Cir. 1986).
28
Baxter, supra note 2.
29
See, e.g., Richard Schmalensee, Payment Systems and Interchange Fees, 50 J. Indus. Econ. 103 (2002);
Jean-Charles Rochet & Jean Tirole, Cooperation among Competitors: Some Economics of Credit Card Associations,
33 Rand J. Econ. 549 (2002); See Rochet & Tirole, supra note 2; See Wright, supra note 1; David S. Evans &
Richard Schmalensee, Paying with Plastic: The Digital Revolution in Buying and Borrowing (MIT Press
2005); and Evans & Schmalensee (2005), supra note 13.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 21
30
See Friess & Greenaway, supra note 9.
PLATFORM ECONOMICS:
22 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
hover below 40 percent. Stock exchanges have been ordered to provide such linkage; this is expected
to happen in the first half of 2007 and may have a major effect on the competitive landscape.
In Europe, on the other hand, there has thus far been very little direct competition between
the London Stock Exchange and other European exchanges, such as Euronext and Deutsche Börse.
One key question in mergers between stock exchanges is whether network effects will continue to
limit the scope for competition or whether falling communications costs and the computerization
of the securities business will make global competition—of one sort or another—inevitable.
D. Magill
Magill is a leading EC case involving the compulsory licensing of intellectual property. What
makes it interesting from a two-sided standpoint is that it involved several interlinked two-sided
platforms. The defendants in the case were three television networks (RTE, BBC, and ITV) whose
broadcasts were received in Ireland. RTE and ITV were two-sided platforms, receiving revenues
from advertisers. RTE was also supported by licenses paid by consumers for having television sets.
The BBC received similar revenues from licenses for television sets in the United Kingdom (but not
31
For contrary views on this case, see Maurits Dolmans & Thomas Graf, Analysis of Tying Under Article 82 EC: The European
Commission’s Microsoft Decision in Perspective, 27 World Competition 225 (2004). See also David S. Evans
& A. Jorge Padilla, Tying Under Article 82 EC and the Microsoft Decision: A Comment on Dolmans and Graf, 27
World Competition 503 (2004).
32
CFI Order of Dec. 22, 2004, Case T-201/04 R 2, Microsoft Corporation v. Commission, at paras. 365 and 388,
available at http://curia.eu.int/en/content/juris/index_form.htm.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 23
Ireland). The BBC did not allow advertising and was not a two-sided platform. All three networks
published an advertising-supported television guide that contained their own weekly listings; these
were two-sided platforms. In addition they each provided their daily listings to newspapers—other
two-sided platforms—that combined the listings.
Magill TV Guide (Magill) wanted to publish a weekly advertising-supported guide that con-
tained the listings of the three networks. The networks complained that this violated their copy-
rights. The Commission and ultimately the EC courts concluded that there would be a market—in
the antitrust sense—for a weekly television guide and that the refusal to supply the copyrighted
information prevented the emergence of the weekly guide product. As it turns out, the weekly
newspapers were the main beneficiaries of this decision since they started weekly television supple-
ments included in the Sunday newspapers. Magill never made a successful go of it.
We will return to these issues when we discuss the analysis of market definition and market
power. The key point is that the analysis by all the parties (including the television networks)
ignores a key side of the two-sided industry here—the advertisers who were the likely source of
much of the revenue and profits—as well as the link between the guides and the television business.
or are unlikely to acquire it are often presumed benign (except of course for naked price-fixing
and related cartel practices).
The economics of two-sided platforms provides several insights into analysis of market power.
1. The link between the customers on the two-sides affects the price elasticity of demand
and thus the extent to which a price increase on either side is profitable. It therefore
necessarily limits market power all else equal. Consider two sides A and B. An increase
in the price to side A reduces the number of customers on side A and therefore reduces
the value that customers on side B receive from the platform. That in turn reduces the
price that side B will pay and the number of customers on side B. The reduction in the
number of customers on side B in turn reduces the demand on side A and thus the price
that customers on side A will pay. These positive feedback effects may take some time to
work themselves out, but, as we demonstrated above, even if, say, customers on side A are
not very sensitive to price, all else (including the behavior of those in side B) equal, de-
mand from side A may nonetheless end up being very price-sensitive indeed when these
feedback effects work themselves out.
2. For two-sided platforms it can be important to recognize that competition on both sides of a
transaction can limit profits. Suppose in a market without multi-homing that there is limited
competition on side A because customers cannot easily switch between vendors of that side,
but there is intense competition on side B because customers can and do switch between
vendors based on price and quality. Then if competitors on side B cannot differentiate their
products and otherwise compete on an equal footing, the ability to raise prices on side A will
not lead to an increase in profits. Any additional profits on side A will be competed away
on side B. This is different from a simple multi-product setting, since the platform cannot
stop serving side B without leaving the business entirely. This point is especially relevant for
assessing incentives and recoupment. It is also worth noting that the possibility of multi-
homing on side B will permit positive profits, since it reduces the intensity of competition.
3. Price equals marginal cost (or average variable cost) on a particular side is not a relevant
economic benchmark for two-sided platforms for evaluating either market power, claims of
predatory pricing, or excessive pricing under EC law. As we saw above, the non-predatory,
profit-maximizing price on each side is a complex function of the elasticities of demand on
both sides, indirect network effects, and marginal costs on both sides. Thus it is incorrect to
conclude, as a matter of economics, that deviations between price and marginal cost on one
side provide any indication of pricing to exploit market power or to drive out competition.33
33
For the two-sided platform as a whole, a formula similar to the standard Lerner index emerges in the Rochet-Tirole
model. This is not a general result, and it thus suggests that the overall price-cost margin is somewhat less relevant
than in single-sided businesses for evaluating overall market power.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 25
SIDE 1
B
A
Coincident
C
D
One-sided
Intersecting
E
SIDE 2
F
SIDE 3
The constraints on market power that result from interlinked demand also affect market defi-
nition. Market definition assists in understanding constraints on business behavior and assessing
the contours of competition that are relevant for evaluating a practice. In some cases, the fact that
a business can be thought of as two-sided may be irrelevant. That could happen either because the
indirect network effects though present are small or because nothing in the analysis of the practices
really hinges on the linkages between the demands of participating groups. In other cases, the fact
that a business is two-sided will prove important both by identifying the real dimensions of com-
petition and focusing on sources of constraints.34
X X X X
Figure 1 shows potential sources of competitive constraints for a two-sided platform denoted
X X X X X
by A. It faces competition of some degree from other differentiated two-sided platforms that serve
the same customer groups (e.g., the newspapers in a city).X It also faces competition from single-
sided businesses that provide competitive services to one side only (e.g., billboards). And it faces
34
See David S. Evans & Michael Noel, Defining Antitrust Markets When Firms Operate Multi-Sided Platforms, 3
Colum. Bus. L. Rev. 667 (2005).
PLATFORM ECONOMICS:
26 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
competition from other two-sided platforms that provide a product that competes mainly with
one side but not the other (e.g., advertising-supported television). Again, the existence of these
constraints does not mean they are important, only that they need to be looked at.
its price on one side so low as to deny other platforms access to this side of the market. It is also
possible for a two-sided platform to engage in two-sided predatory pricing, charging below cost
overall on both sides with the purpose of foreclosing competitors. Cost-based tests make some
sense in the latter case, but it is hard to see how they could be used to analyze an allegation of one-
sided predation.
Under Article 82 of the EC Treaty a dominant firm can be found to have made an abuse by
charging “unfair purchase or selling prices.” Just as a below-cost price on one side can emerge in
long-run market equilibrium so can an above-cost price on the other side. Indeed, such below-cost/
above-cost prices will come together. This issue has come up in a series of cases in Europe in which
regulatory authorities have found mobile telephone operators to have charged fixed-line carriers
excessive prices for terminating calls on their networks; the authorities recognize that the profits
from these excessive prices are competed away in part through low prices for handsets and call orig-
ination. Indeed, the U.K.’s Office of Communication (OfCom) recognized that mobile telephone
platforms were highly competitive (on the mobile subscriber side at least) and did not overall earn
supracompetitive returns.35 Although they did not accept that this was a two-sided business, and
did not apply two-sided analysis, OfCom did provide an “indirect network e xternality” kicker to
the regulated price it imposed on the mobile termination side.36
2. Tying
Under a rule of reason analysis37 the economics of two-sided platforms can provide an explanation
for certain tying practices that seem to reduce consumer choice and harm consumers. As we dis-
cussed above, the platform provider designs the platform—including the constellation of services
and features—to harness internalized externalities, minimize transactions costs between the custom-
ers and both sides, and maximize the overall value of the platform. As part of harnessing externalities
this platform provider wants to increase positive indirect network effects while limiting negative
35
See, e.g., U.K. Office of Telecommunications, Discontinuing Regulation: Mobile Access and Call
Origination Market §1.2 (2003), available at http://ofcom.org.uk/static/archive/oftel/publications/eu_
directives/2003/discon1103.pdf (“no mobile network operator, either individually or in combination with one or
more other mobile network operators, has [significant market power] in that market.”). No provider has a share
exceeding 28 percent. See, e.g., Economist Intelligence Unit, United Kingdom: Telecoms and Technology
Background (2005).
36
U.K. Office Of Communications, Wholesale Mobile Voice Call Termination 163-72 (2004), available
at http://www.ofcom.org.uk/consult/condocs/mobile_call_termination/wmvct/wmvct.pdf. See Armstrong, supra
note 8.
37
Economists and legal scholars generally agree that tying should be considered under a rule of reason analysis rather
than a per se test. That is not the state of the law in the United States or the European Community, both of whose
highest courts have adopted something closer to a per se test of liability. However, both courts admit that efficien-
cies can at least play a limited role in the analysis (in the United States through the separate product test and in the
European Union through the possibility of “objective justification” of the practice).
PLATFORM ECONOMICS:
28 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
indirect network effects. As a consequence, the two-sided platform Under a rule of reason
may impose requirements on side A that do not benefit them directly analysis the economics of
and which customers on that side might even reject after comparing two-sided platforms can
private benefits and costs. But such requirements may benefit side provide an explanation
B. And if the demand increases on side B, these requirements may for certain tying
increase the value placed on the platform on side A—and in fact practices that seem to
could increase value so much that the feature provides a net benefit reduce consumer choice
to side A.38 and harm consumers.
The honor-all-cards rule for payment cards is a possible example. Card systems generally require
that merchants that agree to take the system’s branded cards agree to take all branded cards that
are presented by shoppers. Thus, merchants that have a contract to take American Express (Amex)
cards cannot decide to take payment by Amex corporate cards but not Amex personal cards, or to
take payment from visibly wealthy travelers but not from locals. For at least some merchants the
private cost of this requirement outweighs its benefits (generally we would expect that merchants
would privately want a choice to take whatever card they wanted).39 However, this rule makes the
system’s branded card more valuable to its cardholders, who have the assurance that their card will
be accepted for payment at merchants that display the system’s acceptance mark. By increasing the
number of cardholders it makes the card a more valuable payment device for merchants to accept.40
38
See Rochet & Tirole (2005), supra note 6.
39
For a discussion of this issue, see Robert E. Litan & Alex J. Pollock, The Future of Charge Card Networks
(AEI-Brookings Joint Center for Regulatory Studies, Working Paper, 2006).
40
A class of merchants claimed that Visa and MasterCard had illegally tied by requiring merchants that accepted their
credit cards to also accept their debit cards. The card associations agreed to end this practice after a federal district
court judge applied the per se tying test and ruled that the associates failed several prongs of this test as a matter
of law. In re Visa Check/MasterMoney Antitrust Litigation, 192 F.R.D. 68 (E.D.N.Y 2000). American Express
has been sued by a class of merchants for illegally tying its corporate and personal cards. See Lavonne Kuykendall,
Merchants Suing Amex Add Citi, MBNA as Defendants, 170 Am. Banker (2005).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Industrial Organization of Markets with Two-Sided Platforms 29
that it is at least theoretically possible for a two-sided platform to use exclusive contracts to exclude
competitors, although the welfare consequences of these contracts are not clearly harmful.41
As with exclusivity in one-sided markets, however, this can only be a concern if one firm has
exclusivity over most or all of the market and if the exclusivity is persistent and durable. For ex-
ample, consumers on the nonexclusive side could respond by moving to a competing platform,
thus exerting pressure on consumers on the exclusive side to end exclusivity. Moreover, in markets
with significant buyer concentration, the buyers would be reluctant to agree to exclusivity if there
is some expectation that it will lead to dominance by that platform, as that will likely result in
higher prices in the future for all sides. As with one-sided markets, one needs to consider whether
the efficiencies from exclusive contracts—for example, in helping to create a platform that might
not otherwise exist for the benefit of consumers—offset possible costs from reducing competition.
41
See Mark Armstrong & Julian Wright, Two-Sided Markets, Competitive Bottlenecks and Exclusive Contracts, Econ.
Theory (forthcoming 2006).
CHAPTER
Some Empirical Aspects of
2 Multi-Sided Platform Industries
David S. Evans
ABSTRACT
Multi-sided platform markets have two or more different groups of customers that businesses
have to get and keep on board to succeed. These industries range from dating clubs (men and
women), to video game consoles (game developers and users), to payment cards (cardholders
and merchants), to operating system software (application developers and users). They include
some of the most important industries in the economy. A survey of businesses in these industries
shows that multi-sided platform businesses devise entry strategies to get multiple sides of the
market on board and devise pricing, product, and other competitive strategies to keep multi-
ple customer groups on a common platform that internalizes externalities across members of
these groups.
I. INTRODUCTION
Multi-sided platforms coordinate the demand of distinct groups of customers who need each other
in some way. Dating clubs, for example, enable men and women to meet each other; yellow pages
provide a way for buyers and sellers to find each other; and computer operating system vendors
provide software that applications users, applications developers, and hardware providers can use
together. When devising pricing and investment strategies, multi-sided platforms must account for
interactions between the demands of multiple groups of customers. In theory, the optimal price
to customers on one side of the platform is not based on a markup formula such as that given
by the Lerner condition, and price does not track marginal cost. Competition among platforms
takes place when seemingly distinct customer groups are connected through interdependent de-
mand and a platform that, acting as an intermediary, internalizes the resulting indirect network
externalities (Rochet and Tirole, 2003). Platforms are central to many key industries including
computer games, information technology, many internet-based industries, media, mobile telepho-
ny and other telecommunications industries, and payment systems.
This chapter provides an empirical survey of entry, pricing and other strategies in platform
industries. It provides background for the emerging theoretical (Rochet and Tirole, 2003; Julien,
30
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Some Empirical Aspects of Multi-Sided Platform Industries 31
2001; Armstrong, 2002; and Parker and Van Alstyne, 2002); empirical (Rysman, 2002), and policy
(Evans, 2003) literature on multi-sided platform markets. Although it does not provide empirical
tests of either the key assumptions of this literature or its implications, it confirms that multi-sided
platforms are an important, and until recently unrecognized, part of the industrial organization
landscape, and that this new area of economic research has potentially rich empirical implications
and relevance. This chapter is based, and reports early results, of a series of detailed case studies of
multi-sided platform industries I have been conducting.
Section 2 summarizes the conditions under which a multi-sided platform may emerge and
the main theoretical findings of the literature to date. It also provides a brief overview of the three
major kinds of platforms. Section 3 reviews some common business practices followed in multi-
sided platform industries studied thus far. I then turn to two more detailed case studies. Section
4 reviews the entry of Diners Club in the payment card industry and summarizes the pricing
strategies that continue to this day. Section 5 examines the entry of the Palm operating system for
personal digital assistants. Section 6 makes some brief concluding remarks.
1. There are two or more distinct groups of customers. These customers may be quite different
from each other, such as the men and women for a dating platform or retailers and custom-
ers for a shopping mall. Alternatively, these customers may be different only for the purpose
of the transaction at hand – eBay users are sometimes buyers, sometimes sellers; mobile
phone users are sometimes callers, sometimes receivers.
2. There are externalities associated with customers A and B becoming connected or coordi-
nated in some fashion. A cardholder benefits when a merchant takes his card for payment;
a merchant benefits when a cardholder has a form of payment he takes. The presence of in-
direct network effects seems to be an empirically important explanation for the emergence
1
See Rochet and Tirole (2002), Rochet and Tirole (2003), Rochet (2003), Armstrong (2002), and Parker and Van
Alstyne (2002).
PLATFORM ECONOMICS:
32 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
of a platform although not necessary as a matter of theory. Sellers of PEZ dispensers value
exchanges that have more people who would like to buy PEZ dispensers. (See discussion of
eBay below.)
3. An intermediary can internalize the externalities created by one group for the other group.
Obviously, if the members of group A and group B could enter into bilateral transactions
they would be able to internalize the indirect externalities under the second condition (2).
However, in practice information and transaction costs and free-riding problems make it
difficult for members of distinct customer groups to internalize the externalities on their
own.2 You can look for your new sweetheart by strolling around the Boston Public Garden;
the Yahoo! Personals are less romantic but perhaps more efficient. The intermediary does
not have to be a business in the usual sense. Cooperatives have emerged in payment cards
(Visa International) and software (Linux). Governments sometimes act as the intermediary
– currency is an example.
Several articles have examined the economics of price determination in multi-sided platform
markets. A key finding is that optimal prices for the multiple customer groups must align – or
balance – the demand among these groups – and indeed the emergence of a pricing structure as
well as a pricing level is the defining characteristic of such industries (Rochet and Tirole, 2002).
Optimal prices are not proportional to marginal costs as is the case with the familiar Lerner condi-
tions3 or its multi-product variants.4 Indeed, it is possible that the optimal price for one side will
be less than the marginal cost for that side (Parker and Van Alstyne, 2002). (The assignment of
costs to one side or another may not be well defined either. When it is necessary to get both sides
together for a platform product to exist – that is for either customer to have anything to purchase
– one may not be able to say that one side or another “caused” a cost.) Platform businesses may
tend to skew prices towards one side or another depending upon the magnitude of the indirect
network externalities resulting from that side. If side A generates a much greater degree of exter-
nalities for side B than side B does for side A, side A may tend to get a lower price (Parker and Van
Alstyne, 2002). As in Ramsey-type models of multi-product pricing, in which firms are pricing in
part to recover common costs of production, one side may end up contributing more to common
costs than another side. However, the economic reasons for this are different in multi-sided than
in single-sided markets.5 Note that the theoretical literature is based on quite rarefied assumptions
and has thus far focused on static pricing issues.
2
In unpublished work Jean Tirole points out that a necessary condition for multi-sided platforms to arise is that the
Coase Theorem does not apply.
3
The Lerner condition was first stated in Lerner (1934).
4
See generally Baumol, Panzar and Willig (1982).
5
See generally Rochet and Tirole (2003), and Parker and Van Alstyne (2002).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Some Empirical Aspects of Multi-Sided Platform Industries 33
The remainder of this chapter examines several multi-sided platform industries. It is helpful to
divide multi-sided platforms into three categories: (1) market-makers; (2) audience-makers; and
(3) demand coordinators. Table 1 provides further examples of multi-sided platform markets and
businesses that participate in these markets. While by no means exhaustive, it illustrates the variety
of multi-sided platform industries.
Market-makers enable members of distinct groups to transact with each other. Each member
of a group values the service more highly if there are more members of the other group – because
that increases the likelihood of a match and reduces the time it takes to find an acceptable match.
Examples include exchanges such as NASDAQ and eBay, shopping malls such as those that dot
the New Jersey Turnpike, and dating services such as Yahoo! Personals.6
Audience-makers match advertisers to audiences. Advertisers value the service more if there
are more members of an audience who will react positively to their messages; audiences value the
service more if there are more useful messages (Goettler, 1999). Advertising-supported media
such as magazines, newspapers, free television, yellow pages, and many Internet portals are audi-
ence makers.7
Demand-coordinators make goods and services that generate indirect network effects across
two or more groups. They are a residual category but economically the most interesting and the
least studied. These platforms do not strictly sell “transactions” like a market maker or “messages”
like an audience-maker. Software platforms such as Windows and the Palm OS, payment systems
such as debit cards, and mobile telephones are examples (Rochet and Tirole, 2003).8
6
See NASDAQ (2003) “Market Characteristics” http://www.nasdaq.com/about/market_characteristics.pdf, eBay
Inc. (2003) “Company Overview,” http://pages.ebay.com/community/aboutebay/overview/index.html, Pashigian
and Gould (1998), and Yahoo! Inc. (2003) “Yahoo! Personals,” http://personals.yahoo.com/.
7
“In a fundamental sense, what advertisers demand, and what the various advertising media outlets supply, are units
of audience for advertising messages. Thus advertiser demand for space in the print media and time in the broadcast
media is a derived demand stemming from a demand for audience, and is a positive function of the size and quality
of audience” (Ferguson, 1983).
8
Mobile telephone services are not a one-sided market because most users both make and receive calls. A high termi-
nation charge raises the marginal cost of calls and lowers the marginal cost of call receptions. For a given call, end
users are on a single side and their consumption behaviors depend on their own price (calling price for the caller and
receiving price for the receiver). As a consequence, the choice of termination charge is not neutral. See Jeon, et al.
(2001) for more detail.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Some Empirical Aspects of Multi-Sided Platform Industries 35
when Palm provides free tools and support to PDA applications software developers, it encour-
ages those developers to write programs that work on the Palm OS platform, but it also induces
those developers to spend less time writing programs for other operating systems (See discussion
of Palm below.)
9
If finance charge revenues are included, American Express earned 62 percent of its revenues from merchants in
2001 (American Express Company (2001) “Annual Report,” http://www.onlineproxy.com/amex/2002/ar/pdf/axp_
ar_2001.pdf.).
10
From 1988 through 2000, Microsoft earned at least 67 percent of its revenues from licensing packaged software
(such as Windows and Office) to end-users (either directly at retail or through manufacturer pre-installation on
PCs). See IDC (1994) 1994 Worldwide Software Review and Forecast, Report #9358, November, IDC (1995)
1995 Worldwide Software Review and Forecast, Report # 10460, November, IDC (1996) 1996 Worldwide Software
Review and Forecast, Report #12408, November, IDC (1997) 1997 Worldwide Software Review and Forecast, Report
#14327, October, IDC (1999) 1999 Worldwide Software Review and Forecast, Report #20161, October, IDC (2001)
Worldwide Software Market Forecast Summary, 2001-2005, Report #25569, September. Note that the 67 percent
figure underestimates the amount of revenue Microsoft earns from end-users because the other third of revenue
coming from “Applications Development and Deployment” includes some end-user revenues as well. For example,
database products used by business IT departments are included in the Applications Development category.
11
See British Retail Consortium (2002) “Retail Disappointment as Visa Exempted from EU Competition Rules,”
Press Release, July 25, http://www.brc.org.uk/Archive.asp, Guerrera (2002), MacKintosh (2001), and Pacelle,
Sapsford and Scannell (2003).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Some Empirical Aspects of Multi-Sided Platform Industries 37
IBM, Palm and other operating system companies could have charged higher fees to applications
developers and lower fees to end-users. They all discovered that it made sense to charge developers
relatively modest fees for developer kits and, especially in the case of Microsoft, to give a lot away
for free. Nevertheless, Microsoft is known for putting far more effort into the developer side of the
business than the other operating system companies (Gawer and Cusumano, 2002).
The debit card is an example in which different platforms made different pricing choices. In
the late 1980s, the ATM networks had a base of cardholders who used their cards to withdraw
cash or obtain other services at ATMs. They had no merchants that took these cards. To add debit
services to existing ATM cards, the ATM networks charged a small interchange fee than the card
associations charged (8 cents per transaction on a typical $30 transaction compared) to encourage
merchants to install PIN pads that could read the ATM cards that cardholders already had and
accept the pins they used to access the ATM machines (Evans and Schmalensee, 1999). Many
merchants invested in the PIN pads – the number of PIN pads increased from 53,000 in 1990 to
about 3.6 million in 2001.12 The credit-card associations had a base of merchants who took their
cards but it did not have cards that, like the ATM cards, accessed consumers’ checking accounts.
The credit-card systems imposed a much higher interchange fee than the ATM networks, about 37
cents versus 8 cents on a typical $30 transaction.13 They did this to persuade banks to issue debit
cards and cardholders to take these cards.14 The number of Visa debit cards in circulation increased
from 7.6 million in 1990 to about 117 million in 2001.15
Two other factors influence the pricing structure. There may be certain customers on one side
of the market – Rochet and Tirole refer to them as “marquee buyers” – that are extremely valuable
to customers on the other side of the market. The existence of marquee buyers tends to reduce the
price to all buyers and increases it to sellers. A similar phenomenon occurs when certain customers
are extremely loyal to the two-sided firm – perhaps because of long-term contracts or sunk-cost
investments. For example, American Express has been able to charge a relatively high merchant
discount as compared to other card brands, especially for their corporate card, because merchants
viewed the American Express business clientele as extremely attractive. Corporate expense clients
were “marquee” customers that allowed American Express to raise its prices to the other side of
the market, merchants. In contrast, when the ATM systems entered into debit, they had “cap-
tive” cardholders – ATM cards could be used for debit transactions, so consumers did not need
to be courted to accept the new payment form. Therefore, it has been the merchants – who must
12
See Evans and Schmalensee (1999) and The Nilson Report (2002) No. 759, March.
13
The ATM systems typically charged a flat interchange fee per transaction, while the interchange fee set by Visa and
MasterCard varied with the size of the transaction. The reported interchange fee comparison is from 1998, around
the time of substantial growth in debit for the ATM and credit-card systems (Evans and Schmalensee, 1999).
14
Visa attracted consumers through an effective advertising campaign and attracted issuers through heavy investment
in a debit processing facility, among other strategies (Evans and Schmalensee, 1999).
15
See The Nilson Report (2002) No. 760, March and The Nilson Report (1991) No. 500, May.
PLATFORM ECONOMICS:
38 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
purchase and install expensive machinery in order to process online debit transactions – who have
been courted, as we saw above.
16
IDC (2001) Worldwide Software Market Forecast Summary, 2001-2005, Report #25569, September.
17
The percentages total to 205, indicating substantial multihoming on the part of developers. See Lerner (2002) and
Corporate Technology Directory (1990-2000).
18
See Lerner (2002) and Corporate Technology Directory (1990-2000).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Some Empirical Aspects of Multi-Sided Platform Industries 39
Multihoming and intersecting platforms affect both the price level and the pricing structure.
Theory and empirics are not far enough advanced to say much more.
19
See eBay Inc. (2001) “Annual Report.”
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Some Empirical Aspects of Multi-Sided Platform Industries 41
Palm also did not invest large amounts into developer support until it achieved sufficient
liquidity on the user side of the market. While small investments were made in releasing a software
development kit (SDK) with the Palm Pilot and making the Palm OS and architecture accessible
to outside developers, Palm made few efforts to sign up key development partners or to support the
Palm development community through classes, conferences, and other community support activi-
ties until a critical mass of users was assembled in 1998. (See discussion of Palm below.)
Many successful multi-sided firms have tested and modified their platforms with minimal
investment and then scaled up according to what works best. An example comes from Yahoo!, a
firm that has, over time, experimented with providing pages geared to different audiences such as
Yahoo!Personals, Yahoo!Finance, and Yahoo!Travel. In 1996, Yahoo! developed Yahooligans!, a direc-
tory site that linked to content that appealed to children.20 Yahoo! invested only one programmer and
one business developer in their early efforts on Yahooligans!, both working half time for three months
before the launch. As it turned out, the bandwidth available to most children at the time – 28 kbps –
was too slow to allow for content that kept kids’ attention. Yahooligans! was a relative failure among
Yahoo! pages at the time, but that failure came at a minimal loss to Yahoo! (Bergin, 1998).
A final observation on business models. Contrary to the traditional economic theory of net-
work effects (Arthur, 1989) and the business advice based on that theory (Shapiro and Varian,
1999), there is no evidence that building up market share quickly is a recipe for market domina-
tion in platform industries most, if not all of which, are precisely those industries that economists
have cited as having strong network effects. Many of the early entrants in these industries ultimate-
ly did not retain the leadership position: Diners Club in cards, Apple in personal computer, Apple
in hand-held devices, and OnSale in on-line exchanges. Also, as noted above, despite network
effects many platform industries have several overlapping competing platforms and most have
multi-homing on at least one side.
20
See http://www.yahooligans.com
21
“Dining on the Cuff” (1951) Newsweek, January 29.
PLATFORM ECONOMICS:
42 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
McNamara and an associate, Ralph Schneider, started small. Beginning with less than $1.6
million (in 2002 dollars)22 of start-up capital they signed up 14 New York restaurants, charging
them 7 percent of the tab, and initially gave cards away to people.23 (The observant reader will note
that for most of us owning a credit card would not help our predicament when we leave our wallets
at home. The proverbial apple falling on the head is a mysterious force.)
By its first anniversary there were 42,000 cardholders who were each paying $18 year for
membership in the “club.” There were 330 U.S. restaurants, hotels and nightclubs that accepted
these cards; they paid an average of 7 percent of the cardholder’s bill to Diners Club. In March
1951, Diners Club handled about $3 million of exchanges between cardholders and merchants (it
reportedly made almost $60,000 in profit before taxes).24 In 1951 Frank McNamara, founder of
Diners Club, predicted that monthly transaction volume would increase to about $7 million by
the end of the year.25 Diners Club was therefore getting more than three quarters of its revenues
from merchants.26 Moreover, the margin on cardholders was low since they were getting free float
for an average of two weeks.
By 1958, it had raised the cardholder fee to about $26 but left the merchant fee at 7 percent.
Nevertheless, it continued to earn most of its revenues, about 70 percent, from merchants.27 The
same year saw the birth of two of the major competitors to Diners Club. American Express, which
had long been in the travel and entertainment business with its travelers checks and travel of-
fices, decided to enter the industry. It recognized need to get both sides on board. Even before the
first transaction, it had already acquired a cardholder base of 150,000 from the American Hotel
Association’s card program, as well as the program’s 4,500 hotels for its merchant base. It bought
another 40,000 cardholders from Gourmet magazine’s dining card program (Grossman, 1987). It
soon had a cardholder base to sell to T&E merchants, and sent out representatives to sign them up.
It also advertised in newspapers for additional cardholders (Grossman, 1987). American Express’s
positive reputation in the T&E industry was also a benefit in attracting both sides of the business
(Grossman, 1987). By the time of its entry on October 1958, it already had 17,500 merchant loca-
tions and 250,000 cardholders (Grossman, 1987).
American Express adopted a slightly different pricing policy than Diners Club. It initially set its
annual fee at $31, $5 higher than Diners Club, thereby suggesting that it was the more “ exclusive”
card (Friedman and Meehan, 1992). But it set the initial merchant discount slightly lower: 5 to 7
percent for restaurants; and 3 to 5 percent for the recalcitrant hotel industry.28 Within a year, its
22
Except where noted, all dollar figures in this paper have been adjusted to constant 2002 dollars.
23
See Mandell (1990) and Grossman (1987).
24
“Charge It, Please” (1951) Time, April 9.
25
“Dining on the Cuff” (1951) Newsweek, January 29.
26
Cardholder fees totaled about $745,000 while merchant fees totaled about $2.5 million.
27
“On-the-Cuff Travel Speeds Up” (1958) Business Week, August 16.
28
“On-the-Cuff Travel Speeds Up” (1958) Business Week, August 16.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Some Empirical Aspects of Multi-Sided Platform Industries 43
cardholder base had grown to 700,000 and its merchant base to 37,000.29 With its slightly higher
cardholder fee and slightly lower merchant fee, American Express received just under 55 percent of
its revenue from merchants, compared to over 65 percent for Diners Club in 1959.30 This would
grow over time, however, as spending per card increased faster than annual fees.
American Express had been successful getting both sides on board but initially struggled to
make a profit. It responded by putting more pressure on cardholders to pay promptly. It also man-
aged to raise cardholder fees without suffering significant attrition. There were 900,000 cardhold-
ers that could use their cards at 82,000 merchant locations by the end of 1962, the first year that
the card operation posted a small profit (Hammer, 1962).
The other significant entrant in 1958 was Bank of America (this card evolved into Visa).
Unlike many other banks, it had always focused on lending money to the middle class. By 1958,
Bank of America had extensive experience in making small loans on consumer durables such as
refrigerators and automobiles. It had also become the largest bank in the country (Mandell, 1990).
One of its small competitors, the First National Bank of San Jose, started a charge card in 1953.
At the time, Bank of America considered introducing its own card but decided that there was not
a good enough business case. After studying the emerging industry over the next few years they de-
cided to introduce a credit card. Credit-worthy customers would receive cards with limits of either
$1,600 or $2,600; prior authorization would be required for purchases over $130; and a revolving
credit option was available for some cardholders (Wolters, 2000). Revolving credit was the feature
that distinguished this card from existing charge cards. The merchant fee was initially set at 5 per-
cent and lowered soon thereafter (Wolters, 2000). Cardholders did not pay an annual fee but paid
finance charges on any revolved balances at an annual interest rate of 18 percent (Wolters, 2000).
This was a more merchant-friendly balance than that set by Diners Club or American Express,
especially considering that Bank of America’s card was also extending credit on a revolving basis.
Bank of America did a market test in Fresno in the fall of 1958. Three hundred retailers signed
up initially, and every Bank of America customer in the Fresno area received a card. According to
one study, “This mass mailing of 60,000 cards had been William’s (the Bank of America leader
of the effort) solution to the problem of how to convince retailers that enough individuals would
possess a card to make their participation in the program worthwhile. His solution worked, for
during the next five months another eight hundred Fresno-area retailers joined the newly named
‘BankAmericard’ program” (Wolters, 2000). They expanded throughout the state during the fol-
lowing year. By the end of 1959, 25,000 merchants accepted the card and almost two million
California households had one.
Things did not go well. Fraud was rampant. The number of delinquent accounts was five times
higher than expected. Large retailers resisted joining. The program lost $45 million in 1960. The
bank worked on collection problems and lowered the merchant fee to as low as three percent to
29
“Towards an Ever-Fuller Life on Credit Cards” (1959) Newsweek, September 28.
30
Figures taken from Newsweek 1959 article, and assuming average of 5 percent for Amex per Newsweek 1958 article.
PLATFORM ECONOMICS:
44 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
entice retailers. Delinquencies declined and the merchant base increased to 35,000 in 1962. It was
profitable by the early to mid 1960s.
One notable common theme across the experiences of the three startups is the importance of
getting both sides on board. Diners Club, starting from scratch and trying to sell a new product,
had to build slowly. It set a relatively cardholder-friendly balance, collecting most of its revenues
from merchants. This made it possible to sign up cardholders even before a significant merchant
base had been signed. The cardholder base could then be used, of course, to sign up merchants.
By the time of American Express’s entry, the idea of a charge card had already been established by
Diners Club and others. American Express was able to capitalize on that by entering at a larger
scale. Indeed, it probably did not have the option of entering as gradually as Diners Club did,
since what cardholders or merchants would sign up with a fledgling American Express card given
the option of an established alternative. American Express therefore bought up customer bases on
both sides before it entered. It was also able to use its established position in the T&E business to
set a slightly higher cardholder fee, to cultivate the upscale image that served it well for a long time.
Bank of America also developed a large cardholder base in entering the business. It used its base
of depository customers, the largest in California, to develop an instant cardholder base, which it
then used to sign up merchants. Even with its prominent position among consumers in California,
however, Bank of America had to strike a more merchant-friendly balance, lowering its merchant
fees until it could get enough merchants on board.
Perhaps as a consequence of entering at a larger scale, both American Express and Bank of
America, unlike Diners Club, had a hard time making a profit early on. Diners Club made a profit
in its first year, while the other two firms did not see a profit until their fourth or fifth years. It
took time to weed out delinquent cardholders. Having to make an upfront investment in build-
ing a profitable cardholder base is a lesson that holds even for new card issuers today (although
they can join Visa or MasterCard and do not have to build an entire system). There were also the
operational problems that would be expected in starting up any new business. Bank of America
had such (misplaced) confidence its customers would pay their bills that it did not even set up a
collections department (Nocera, 1994).
Apple had made the first attempt with its Apple Newton in 1993. Other entrants into the PDA mar-
ket from 1993 to 1996 were the Psion Series 3, Hewlett-Packard LX, Hewlett-Packard OmniGo,
Sharp Zaurus, Sharp Wizard, and Microsoft’s Windows CE31 PDAs. With the exception of the
Windows CE PDA, all followed an integrated strategy where the PDA manufacturer produced the
hardware, operating system, and applications. Windows CE PDAs had a non-integrated strategy,
where the hardware, operating system, applications were all developed by different firms (although
Microsoft did end up writing the operating system as well as many applications for Windows CE).
The integration of application and operating system development within one firm was common in
the software industry, for example Microsoft produced some of the most popular applications for
the Windows OS (Evans, 2003).
Palm re-entered the handheld market in 1996 with the Palm Pilot. This integrated hardware,
an operating system, and some applications. Palm overcame the handwriting recognition hurdle
with the invention the Graffiti text entry method, providing users a simple method to enter text.
Industry experts described Graffiti as the “killer app” for the Palm Pilot (Feldstein and Flanagan,
2001). The pricing of the Palm Pilot also reflected Palm’s desire to have deep market penetration.
While 3Com, Palm’s parent company from June 1997 to March 2000, wanted Palm to price the
Palm Pilot high to extract profits, Palm insisted that prices be kept low to expand the user base
(Gawer and Cusumano, 2002). The Palm Pilot’s combination of low price and robust features were
a hit with consumers and provided the “most bang for the buck” (Atluru and Wasserstein, 1998).
The Palm Pilot met immediate success with over 360,000 units sold in 1996, representing 51 per-
cent share of the market (Atluru and Wasserstein, 1998).32
Palm has gone through three stages in developing what is now a three-sided platform busi-
ness. In the first phase from about 1996 to 1998, it entered as an integrated platform. It made the
hardware, operating system, and applications for the Palm Pilot and integrated them together. It
planned to court developers, however, only after it had a significant user base. According to Donna
Dubinsky, CEO and one of its founders, “We are a highly integrated product that delivers end
user results. We are not having a developer conference until we sell a million units” (Atluru and
Wasserstein, 1998). A developer conference is one of the major methods used by software plat-
forms to stimulate the writing of applications. Nevertheless, even during this initial phase Palm
laid the groundwork for getting a developer community on board.33 In early 1996, Palm released
its first software development kit (SDK), which included the source code for the Palm Pilot’s bun-
dled applications. This source code served as a model that outside developers could reference to
build other applications (Gawer and Cusumano, 2002).
31
Windows CE was renamed Pocket PC at version 6.0.
32
This represents market share of Palm hardware PDA and does not include licensed Palm OS devices.
33
In fact, in the first year of release, the Palm attracted over 2,000 third-party developers to its platform without any
developer support efforts (Atluru and Wasserstein, 1998).
PLATFORM ECONOMICS:
46 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
In 1998, Palm sold over 2 million PDAs and launched the second phase of market develop-
ment. Palm invested resources in persuading developers to write applications for the Palm OS. It
offered business development resources to developers, including joint development, marketing,
and bundling. Palm’s most important business development resource was its software development
forums (SDF), which helped external developers start and grow their own businesses. The SDF
offered advisory meetings, business seminars, workshops, and networking events. It also started a
$54 million venture capital unit called Palm Ventures to support businesses focusing on Palm OS
applications. The company also offered Palm OS development classes regularly, and encouraged
other activities among its community of users through developer portals (Gawer and Cusumano,
2002). The strategy to actively get developers on board was successful. The number of registered
developers grew from 7,500 in 1998 to 220,000 in 2002.34 Palm continued to secure its market
leadership, with over 7 million units shipped by 2000.
The third phase, from about 2001 to the present, has moved Palm closer to being a pure
operating system company. Palm has created alliances with Sony, Nokia, Handspring, IBM,
Qualcomm, Supra, Symbol Technologies, Fossil35, and TRG Products to create PDAs and other
computer appliances based on the Palm OS (Gawer and Cusumano, 2002). Palm has reported
that this greatly increases the market share of Palm OS based PDAs over their potential market
share with an integrated strategy.36 In addition, Palm OS licensing has expanded the scope of
the Palm OS beyond traditional PDAs to such devices as wrist watches (Fossil Wrist PDA), cel-
lular phones (Handspring Treo), and multimedia players (Sony Cliè). While Palm manufactured
PDAs claimed 50 percent of the market in 2000, devices powered by the Palm OS operating
system had over 75 percent share of the worldwide personal companion device market according
to a report by International Data Corporation.37 This market share increased to over 82 percent
by the second half of 2002.38 In January 2002 as Palm powered devices hit 20 million sold, Palm
formed a Palm OS platform subsidiary, PalmSource, further separating the Palm OS from Palm’s
hardware solutions.39
Palm claims that its operating system was designed from the start to provide a platform for
these three sides. According to Palm’s web site, “From the start, the flexible, extensible Palm OS has
been designed to grow and evolve in response to user needs. The open, modular architecture allows
34
Palm, Inc. (1998) “Key Strategic Relationships And Growth To 7,500 Developers Propel 3Com Palm Computing®
Platform Into Corporate America,” June 15, http://www.palm.com/pr/enterprise.html; Palm, Inc. (2002) “Annual
Report.”
35
Palm, Inc. (2002) “Fossil Signs License Agreement with Palm Source For First Ever Palm Powered Wrist Devices,”
November 18, http://www.palmsource.com/press/2002/111802.html.
36
Palm, Inc. (2001) “Annual Report.”
38
Palm, Inc. (2003) “Palm Powered Share Increases in Second Half of 2002,” January 27, http://www.palmsource.
com/press/2003/012703.html.
39
Palm, Inc. (2002) “Palm Completes Formation of Palm OS Subsidiary as Palm Powered Devices Hit 20 Million
Sold,” January 21, http://www.palmsource.com/press/2002/012102.html.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Some Empirical Aspects of Multi-Sided Platform Industries 47
our more than 50,000 developers, licensees, alliance, and OEM partners to develop innovative new
products and applications for a rapidly expanding global market” (Gawer and Cusumano, 2002).
Palm has subsidized both the application developer side of the market and the user side of the
market by licensing their operating system at a loss and giving their development tools away for
free. It is able to do this because of its profits in hardware sales. Palm provides an SDK, Palm OS
Emulator, Palm OS Simulator, and sample source code free for download on its web site. Palm
also provides links to third-party integrated development environments, both free and not, for
download on their web site.40 Palm also offers an online developer support program to offer a full
range of development services. The Basic Level Membership is offered at no cost. The Advanced
Level Membership is available for just $500 per year and includes direct technical, marketing and
training support.41 Only a small percentage of Palm’s revenues were from licensing of the Palm OS.
PalmSource revenues totaled $67, $88, and $50 million, while operating loss totaled $17, $8, and
$8 million in fiscal years 2002, 2001 and 2000, respectively.42 This continued loss of PalmSource
represents Palm’s continued efforts in trying to expand the user base of the Palm platform.
VI. SUMMARY
Multi-sided platform markets are becoming an increasingly important part of the economy. They
range from relatively small emerging companies like eBay, Yahoo!, and Palm, to relatively large and
mature companies like American Express. These markets also have had a large impact on the recent
information technology boom, and undoubtedly, they will continue to be important, as internet
based commerce expands its scope to include both new and old economy firms. In addition to in-
ternet commerce, other increasingly important industries, such as credit cards, operating systems,
shopping centers, and mass media, are all governed by the economics of multi-sided platforms.
While it is not my intention to generalize from a small sample, and many issues need to be
investigated in more depth, a number of business models appear common across both the cases
studied here. First, differential pricing is used to get both sides on board. Second, once both sides
are on board, pricing continues to play a key role in maintaining both sides of the platform. Third,
multi homing often occurs in multi-sided platforms. And, fourth, one solution to the pricing com-
plexity that has been used successfully is to start with a small but scalable platform.
40
Palm, Inc. (2003) “PalmOS: Desktop Development,” http://www.palmos.com/dev/tools.
41
Palm, Inc. (2003) “PalmOS.com: Palm OS Developer Program,” http://www.palmos.com/dev/programs/pdp.
42
Palm, Inc. (2002) “Annual Report.”
CHAPTER
How Catalysts Ignite:
The Economics of Platform-Based
3 Start-Ups
David S. Evans
ABSTRACT
Entrepreneurs who start multi-sided platforms must secure enough customers on both sides, and
in the right proportions, to provide enough value to either group of customers and to achieve
sustainable growth. In particular, these entrepreneurs must secure “critical mass” to ignite the
growth of their platforms; the failure to achieve “critical mass” quickly results in the implosion of
the platform. There are a number of strategies available to entrepreneurs to reach critical mass.
For example, the “zigzag” strategy involves successive accretions of customers on both sides to
build up the value to both. The relevant strategies depend in large part on whether the nature of
the platform requires securing participation by both platform sides at launch (e.g. dating ven-
ues), whether it is possible to acquire one side before approaching the other side (e.g. search
engines), and whether it is necessary to make precommitments to one side to induce them to
make investments (e.g. video games).
I. INTRODUCTION
Starting a business and getting it to the point where it is economically viable is the most difficult
problem for all entrepreneurs. Most new businesses fail. In the United States, 61 percent of new
businesses that were started in second quarter of 1998 had ceased business within five years.1
Venture capital firms that invested in new firms do not get any of their money back in 43.7 percent
of the first-round investments they make and get less than their initial investment back in 66.7
percent of the first-round investments.2
The start-up problem is particularly difficult for firms that are based on multi-sided platforms.
In addition to the usual problems faced by new firms they often must contend with the well-known
chicken-and-egg problem. Their firm can deliver value to one side of the platform only if there are
1
Amy Knaup and Merissa C. Piazza, “Business employment dynamics data: survival and Longevity, II,” Monthly
Labor Review Online 130, no.9 (September 2007), http://www.bls.gov/opub/mlr/2007/09/art1full.pdf.
2
Andrew Metrik, Venture Capital and the Finance of Innovation (New York: Wiley, 2006), Exhibit 7-4.
48
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 49
articipants on the other side of the platform. They have to figure out how to get both sides on board
p
their platform. That problem, which is the subject of this chapter, is very different from that faced by a
one-sided startup whose main challenge is getting just one set of customers to buy its product or s ervice.3
The chicken-and-egg problem is central to the study of multi-sided platforms. Yet most of the
theoretical and empirical research on two-sided businesses has focused on mature platforms and
examined their pricing structures and other properties. Little attention has been given to critical
issues that entrepreneurs must solve to create a viable platform business.4 These include strategies
for getting both sides on board, the role of critical mass in establishing the foundations for success,
and the particularly thorny issues that arise when both sides must arrive simultaneously.
This introduction provides an overview of multi-sided platforms and the start-up problem.
Section II then presents some building blocks for solving this problem. A few of these building blocks
are similar to those for one-sided businesses while others are unique to multi-sided ones. Section III
describes several complementary strategies for solving the chicken-and egg problem. We then turn to
two case studies. Section IV examines a modern classic in a failed strategy for starting up multi-sided
platforms: the en masse demise of the B2B exchanges whose assured success was extolled by many
academics and practitioners.5 Section V considers how social networking sites have started up and
considers a successful and unsuccessful one. Section VI presents some concluding remarks.
3
These comments also pertain to communication platforms in which it is important to distinguish between senders
and receivers—even though they are the same people—for both economic analysis and business strategy.
4
There are several notable exceptions to this statement. The role of critical mass in launching exchanges is analyzed in a dy-
namic model in Gabor Fath and Miklos Sarvaryi, “Adoption Dynamics in Buyer-Side Exchanges,” Quantitative Marketing
and Economics, Vol. 1, No. 3 (September), pp. 305-335. Also see Bernard Caillaud, and Bruno Jullien, (2003), “Chicken
& Egg: Competition Among Intermediation Service Providers,” Rand Journal of Economics, Vol. 34, No. 2 (Summer), pp.
309-328 which considers platform launch but in a single-period setting. The role of commitment is analyzed in a two-
period game in Andrei Hagiu, “Pricing and Commitment by Two-Sided Platforms”, The RAND Journal of Economics 37,
no. 3 (fall 2006). A managerial treatment of the issues is presented in Annabelle Gawer and Michael Cusumano : How
Companies Become Platform Leaders (Sloan Management Review, Winter 2008, Vol. 49, No. 2, pp.28-35.
5
See David Lucking-Reiley and Daniel F. Spulber, “Business-to-Business Electronic Commerce,” Journal of Economic
Perspectives 15, no.1 (2001):55-68; Nicholas Carr, The Digital Enterprise: How to Reshape Your Business for a Connected
World, (Massachusetts: Harvard Business Press, 2001); and Arthur Sculley and William Wood, B2B Exchanges: The
Killer Application in the Business-to-Business Internet Revolution, (ISI Publications, 1999).
6
See David S. Evans & Richard Schmalensee, Catalyst Code: The Strategies Behind the World’s Most Dynamic Companies
(Massachusetts: Harvard Business School Press, 2007).
PLATFORM ECONOMICS:
50 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
cilitate matching, and make it easier for the two groups of economic agents to exchange value
between each other. In the traditional literature, a catalyst is referred to as a “two-sided mar-
ket” or as a “multi-sided platform.”7 In this chapter we use catalyst and multi-sided platform
interchangeably.
The economic value created by the catalytic reaction is essential for understanding the feasible set of
business strategies that a multi-sided platform can use. That value must be significant enough to warrant
the cost and risk of investment in developing the platform. The value also provides the “pie” that can be
split among the distinct groups of economic agents to provide incentives for them to join and interact on
the platform with a portion of the pie going to the platform for performing its role. Some of the pie can
be used to subsidize certain groups of economic agents, or members of those groups, to join the platform.
Catalyst innovators are ones who discover that it is possible to create economic value by getting
two or more groups of economic agents together on a shared platform or develop a more efficient
platform for starting and accelerating a catalytic reaction.
7
I prefer not to use the “two-sided market” term because the term really refers to a platform business that provide a
matchmaking service in competition with other platform businesses in a market for that service.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 51
In the literature on market microstructure, for example, for financial exchanges a “liquid”market
is one that has enough buy and sell-orders to facilitate transactions.8 Markets that are too “thin”, or too
illiquid, collapse. There is a critical mass of buy and sell orders that allows markets to sustain themselves.
We will see below that the B2Bs failed because they did not achieve a critical mass of buyers and seller;
in particular they did not create enough value to suppliers to entice them to participate in the exchanges.
8
In financial exchanges liquidity providers (such as market makers) provide buy and sell orders. They attract liquidity
takers (such as investors who buy and sell through broker-dealers). In other exchange markets there is no institution
that plays the role of liquidity providers. However, in these other exchange markets it is essential that the platform
attract enough buyers and sellers to make the market liquid. See Larry Harris, Trading and Exchanges: Market
Microstructure for Practitioners (Oxford University Press, 2002).
9
For the technical development of this framework see David S. Evans and Richard Schmalensee, “Failure to Launch:
Critical Mass in Platform Businesses,”. We show that in certain plausible cases there is a critical level of liquidity
that platforms need to get to in order to evolve towards a stable equilibrium with largest platform size. This frame-
work does not cover the process of getting to critical mass. This chapter uses the framework to motivate an informal
analysis of how catalysts can achieve critical liquidity.
10
See Steve Johnson, “Google Video isn’t ready for Prime Time,” Chicago Tribune, January 18 , 2006, and David
Pogue, “Google Video: Trash Mixed with Treasure,” The New York Times, January 19th , 2006, http://www.nytimes.
com/2006/01/19/technology/circuits/19pogue.html
11
See Evans and Schmalensee, “Failure to Launch,” op. cit.
PLATFORM ECONOMICS:
52 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
Ignition
D*- Long-run equilibrium
C``
O
Side B
friends, if the platform does not grow quickly enough. The entrepreneur must increase the number of
customers on each side to a point where there are enough to reach critical mass. That involves hori-
zontal and vertical movements within the cone-shaped area in Figure 1. A transaction platform, for
example, needs to take actions that increase buyers and sellers.12 The entrepreneur must also maintain
the right proportions of customers on each side so that there are enough participants on each side to
interest the participants on the other side. That involves diagonal movements to the northeast in Figure
1. An exchange platforms needs to make sure there are enough buyers to interest sellers and vice versa.
12
For most financial exchanges the two sides are liquidity providers who announce prices at which they stand ready to
buy or sell and liquidity takers who provide order flow that interacts with those standing orders. See Larry Harris,
Trading and Exchanges: Market Microstructure for Practitioners (Oxford University Press, 2002). For non-financial
exchanges such as e-commerce the two sides are simply buyers and sellers.
13
See, e.g., Susan Hart and Nikolaos Tzokas, “New Product Launch “Mix” in Growth and Mature Product Markets,”
Benchmarking: An International Journal, Vol. 7, No. 5, (2000): 389-406 and Biren Prasad, “Analysis of Pricing
Strategies for New Product Introduction,” Pricing Strategies and Practice, Vol. 5, No. 4, (1997): 132-141. For litera-
ture summary see, V. Krishnan and Karl T. Ulrich, “Product Development Decisions: A Review of the Literature,”
Management Science 47, no. 1, (Jan., 2001): 1-21.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 53
and distribution system for getting products and services to consumers. They need to provide sell
consumers on the merits of trying a product through the dissemination of information as well as
persuasion. Finally, they need to set prices recognizing both the competition and the fact that con-
sumers may need an incentive to try to a new and unproven product. These traditional strategies
facilitate obtaining consumers on both sides of a two-sided platform, that is, in making horizontal
and vertical movements.
This section focuses on several aspects of product launch and product design that are likely to
be particularly applicable to the ignition problem faced by catalysts. Before we begin it is useful
to clarify some issues concerning the relationships among different platform members which will
affect the discussion below.
For some platforms, customer groups are very distinct. Companies such as Electronic Arts that
develop and sell videogames for the Sony PlayStation platform are distinct from people who buy
and use Sony PlayStations.
For other platforms, the customers appear so similar that the platform may not appear
multi-sided at all. A telephone platform connects people who talk to each other. People are
people. Closer inspection, though, often reveals that people fall into one of two distinct po-
sitions on the platform at any point in time, that they need people in the other positions to
connect to, and that it is possible to manipulate the pricing structure to mediate the externali-
ties between these two groups.14 At any point in time members of a phone network are either
calling someone or being called and carriers can adopt pricing structures to alter the incentives
to make or take calls. The same is true for social networking sites that involve inviting and
accepting friends and then involve participating in other interactions that entail initiating or
receiving messages communications.15
The fact that customers on the different sides may be the same economic agents clearly facili-
tates platform ignition. While the catalyst entrepreneur might have to provide incentives for these
economic agents to engage in different types of behavior it can focus on securing the participation
of one well-identified group of agents.
14
The fact that the pricing structure—the relative charges to the two groups matters—makes these a two-sided plat-
form under the definition proposed by J.C. Rochet and J. Tirole, “Platform Competition in Two-Sided Markets,”
J. European Economic Association 1, no.4 (2003): 990-1029. See e.g, Jerry Hausman, and Julian Wright, “Two Sided
Markets with Substitution: Mobile Termination Revisited,” 2006, mimeo. Using data from Australia they explore
the equilibrium prices of fixed-to-mobile calls by taking into account that fixed-line callers might be cell-phone
users themselves, and thus be from either side of the market. They also point out that it is important to consider
whether the mobile receiver pays for the call (as in US) or does not (as in most EU countries).
15
The social networking literature distinguishes between senders and receivers and highlights the important role that
gregarious members of a network (i.e. individuals who initiate many interactions) and prestigious members of a net-
work (i.e. individuals who are the recipients of many messages) play. See Wasserman, S and Katherine Faust, Social
Network Analysis: Methods and Applications (Cambridge University Press, 1994).
PLATFORM ECONOMICS:
54 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
Adoptions
Due to
Internal Influence
pm
Adoptions Due to
External Influence
Time
16
See Vijay Mahajan, Eitan Muller, and Frank M. Bass, “New Product Diffusion Models in Marketing: A Review and
Directions for Research,” Journal of Marketing 54, No. 1, (Jan., 1990): 1-26
17
See Vijay Mahajan, Eitan Muller, and Frank M. Bass, “New Product Diffusion Models in Marketing: A Review and
Directions for Research,” Journal of Marketing 54, No. 1, (Jan., 1990): 1-26.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 55
The literature on social networks provides insights into the process of word-of-mouth com-
munication.18 The social graph describes the relationships among members of a network. It
consists of nodes which reflect the agents and lines that show the connection between the
agents. The connections can be unidirectional (x communicates with y) or bidirectional (x and
y communicate with each other). The nature of the relationships among members of the net-
work can provide insights into the organic workings of the network. There are three key rami-
fications for product diffusion. Word of mouth will spread more quickly: (1) the more con-
nections innovators have; (2) the more connections friends of the innovators have; and (3) the
denser the network is in the sense of there being fewer degrees of separation among members
of the network.
Two types of agents facilitate the diffusion information within the network. “Influencers”—or
gregarious members—account for a disproportionate share of communications. They send a lot of
messages out to a broad range of connections. “Centers” have connections with many people who
are not connected to other agents.19 They are important because they are the only way to reach
isolated agents within the network.
Product diffusion may provide double duty for two-sided businesses in which economic agents
have shifting roles in the platform. Individuals who upload photos on the photo-sharing network
Flickr may view photos from their friends. People who use eBay’s auto exchange may use it for both
buying and selling, though some users may specialize in selling, and some may never sell.
18
See Jacqueline Brown and Peter H. Reingen, “Social Ties and Word-of-Mouth Referral Behavior,” The Journal of
Consumer Research,Vol.14, No. 3, (Dec. 1987): 350–362, and Jo Brown, A. Broderick, and Nick Lee, “Word of
Mouth Communication Within Online Communities: Conceptualizing the Online Social Network,” Journal of
Interactive Marketing 21, no. 3, (2007): 2–20.
19
The centers are the nodes for “stars” in which the center is connected to other nodes and the points of the star are
connecter to the center node.
20
See Carl Shapiro and Hal. Varian, Information Rules, A Strategic Guide to Network Economics, (Harvard Business
School Press, 1998).
PLATFORM ECONOMICS:
56 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
Points of Inflection
Time
equal direct network effects increase the likelihood that each subsequent agent that is contacted by
an earlier adopter will adopt the product also. Figure 3 shows how direct network effects modify
the S-curve of diffusion.
Social networking theory provides some guidance on how to use direct network effects strategi-
cally. I would conjecture that economic agents that are more densely connected in a network have
stronger direct network effects among them. Thus using influencers to connect to more densely
connected portions of networks will tend to have higher payoffs.
21
See Shapiro and Varian, id.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 57
video games) and the maker of those products want to be focus its efforts on platforms that have
users who will demand its products. There are also negative indirect network effects: one type of
economic agent on the network harms another type of agent. The leading case of this for platform
businesses is advertising-supported media. Consumers may dislike the advertisements. The plat-
form solves the externality problem between advertisers and consumers by using content to bribe
people into viewing ads.
Indirect network effects are the key aspect of multi-sided platforms. They are the source of the
catalytic reaction—and much of the value—created by the platform. A key practical aspect of these
indirect network effects is that they require that the platform “balance” the two sides to maximize
the value of the platform to either side. The platform has zero value to either side if the other side
is not on board. For many platforms the optimal balance is likely well into the interior and away
from the axes as was shown in Figure 1.22
22
See Evans and Schmalensee, “Failure to Launch,” op. cit.
23
See J.C. Rochet and J. Tirole, “Platform Competition in Two-Sided Markets,” J. European Economic Association 1,
no.4 (2003): 990-1029.
PLATFORM ECONOMICS:
58 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
24
This is the case treated in Andrei Hagiu, “Pricing and Commitment by Two-Sided Platforms,” The RAND Journal
of Economics 37, no. 3 (fall 2006): 720-737.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 59
Side A
C`
C*
C``
Side B
network effects, the platform is more valuable to each successive group of prospective customers.
Figure 4 shows that basic zig-zag approach to growth.
eBilleMe provides an example of this strategy. Consumers who click on the eBillMe sign at the
checkout for an e-tailer can pay with their online banking account. They then send an email which
contains details for paying from their online banking account. After they enter the information
into their online banking account they receive a receipt and the product is shipped. This payment
alternative is attractive to people who are either concerned about the security of paying with cards
online or do not happen to have a card conveniently available. A small but significant fraction of
consumers, as it turns out, like paying this way.
To get started, eBilleMe persuaded ToolKing to offer eBillMe at checkout.25 A small percent-
age of customers used this payment alternative. eBillMe then went to other online retailers. Each
led to eBillMe having more people who were accustomed to using its service. For each subsequent
merchant it went to it offered an increasingly valuable offer since it had more users who were pre-
disposed to use this payment alternative. At the same time it let its users know that they could pay
at more places thereby increasing the value to the merchants. eBilleMe grew from 1 merchant and
hundreds of users during its first year of entry in 2005 to hundreds of online stores taking 2% to
10% of the merchant’s transaction volume by 2008.
25
Interview with Marwan Forzley, founder and chairman of eBillMe.
PLATFORM ECONOMICS:
60 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
26
or description and discussion of this example, see David S. Evans & Richard Schmalensee, Catalyst Code: The
F
Strategies Behind the World’s Most Dynamic Companies (Massachusetts: Harvard Business School Press, 2007), p. 1.
27
At this point in time credit cards were mainly used in local markets so Diners Club had to secure critical mass in
a number of different cities to ignite the platform. The 330 restaurants and 42,000 cardholders refers to the total
across these separate markets.
28
Id., p. 89.
29
Hagiu, id.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 61
30
See Evans & Schmalensee, Catalyst Code, id.
31
Google Milestones: http://www.google.com/corporate/history.html.
32
See YouTube videos: “The History of YouTube” and “The Real History of YouTube in 3 minutes.”
PLATFORM ECONOMICS:
62 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
F. Summary
We have introduced several concepts.
• First, multi-sided platforms often must attain critical mass to ignite a catalytic reaction that
leads to organic growth. Platforms that do not reach this critical mass implode.
• Second, to reach critical mass platforms can engage in a number of strategies to get “enough”
customers on either side, and in the right proportions, on board. These include the zig-zag
and the two-step.
• Third, these strategies can usefully employ many of the tactics used for new product intro-
ductions by non-platform businesses.
We have taken some liberties in explaining the dynamics of platform ignition. As a formal
mathematical matter one would have to reach critical mass instantaneously to ignite the reaction.
In practice it appears that platforms have some limited time to get to critical mass. Early adopters
use a platform. If they come back and if later adopters also find value then it is possible to reach
critical mass. If the platform does not grow quickly enough to critical mass early adopters lose in-
terest, fewer later adopters come, and word-of-mouth referrals stop or turn negative.
Second, a new platform must allocate that value between the two sides to provide both with a
sufficient incentive to join the platform. The B2Bs were not able to offer enough of a value proposi-
tion to sellers to get enough of them on board. The sellers saw the B2B exchanges as methods for
driving down their prices through auctions. That did not leave them with enough of a long-term
return and they therefore stayed away.
Third, a new platform must get enough of both sides on board and in the right proportions to
achieve the critical mass that provides a minimum amount of value to both sides. Sellers came on
board too slowly. Buyers lost interest. That provided even less motivation for sellers. Platforms that
do not achieve critical mass implode and never have the chance to grow.
36
Arthur Sculley and William Woods, B2B Exchanges: The Killer application in the Business-to-Business Internet
Revolution (ISI Publications, 1999).
37
David Lucking-Reiley and Daniel F. Spulber,”Business-to-Business Electronic Commerce,” Journal of Economic
Perspectives 15, no. 1, (2001): 55-68.
38
According to a research study, Shakeouts in Digital Markets: Lessons from B2B Exchanges, conducted by Wharton
School and Pembroke Consulting in November 2002, of the 1,500 independent business exchanges that existed in
2001, less than 200 remained at the end of 2003.
PLATFORM ECONOMICS:
64 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
A number of observers of the demise of the B2Bs observed that a major problem was that sup-
pliers were scared, as Kabir puts its, “of comparison shopping and brand dilution.”40
Once we go beyond the perfectly competitive model it is easy to see why exchanges that were
built around efficient procurement auctions were not attractive to suppliers. Most businesses have
fixed costs that they have to recover. They therefore cannot survive if their prices are competed
down to short-run variable costs. Department stores can have sales periodically where merchandise
is sold at low margins; they could not survive if they were selling at those prices all the time. B2B
auctions will tend to attract suppliers that can offer low prices because they have excess inventories
as a result of lack of success.
The B2Bs were not necessarily advantageous to buyers either. Company procurement officers
typically learn about the quality of their suppliers. Through interactions over time and discussions
with people in the industry they discover the quality of supplier goods and services as well as their
reliability for delivery. They can take this information into account in deciding who to put on the
bid list thereby selecting the lowest price from pre-qualified bidders, or they can take non-price
39
See, Mark Odell, “B2B Struggle to Achieve Take-off,” Financial Times, June 18, 2001.
40
See Nowshade Kabir, “E-marketplace; Facts and Fictions,” WebProNews, posted on August 20th, 2003, http://www.
webpronews.com/topnews/2003/08/20/emarketplace-facts-and-fictions.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 65
features into account in selecting the winning bid. Many of the B2B exchanges focused mainly
on price and did not provide companies with significant services for assessing quality. Indeed, the
importance of quality and experience was learned the hard way by American toy manufactur-
ers who found suppliers through some of the successful B2Bs that have focused on connecting
Chinese suppliers to multinational buyers. Many of these suppliers used lead in their products
which led to significant health risks and product recalls when this was discovered.41
41
David Barbosa, “Mattel recalls toys made in China,” Herald Tribune, August 2, 2007.
42
See the case studies in Anne Engstrom and Esmail Salehi-Sangari, “Assessment of Business-to-Business e-Market-
places’ Performance,” (PhD diss., Luleå University of Technology, 2007). Of them the 6 B2Bs they examined, five
started trying to earn revenues from both buyers and suppliers and one of these moved towards a model in which
suppliers obtain services largely for free.
43
See M. Meyer, Neil de Crescenzo, and Bruce Russell, “Case Study: Chemdex: In Search of a Viable Business Model,”
International Journal for Entrepreneurship Education 2, no.2. (2004), http://web.cba.neu.edu/~mmeyer/cases/
Ventro-010105.pdf
PLATFORM ECONOMICS:
66 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
buyers and s uppliers already on—at that point suppliers might lack access to a significant market
if they refused to pay the fees. But at the start of these exchanges buyers and sellers would have to
be persuaded to switch from the widely used bilateral procurement methods. In any event, it ap-
pears that many of the exchanges ended up well off the path to critical mass because they did not
offer suppliers with enough of an incentive to join. That could have come in the form of subsidies
in cash or kind to suppliers to join the exchange. If the B2Bs could have gotten enough suppliers
through what is known as a “divide and conquer” strategy they might have gotten enough buyers
and then more suppliers.44 Of course, such subsidies could have been large risky investments in the
success of the B2Bs and granting them would have been foolish if, as suggested above, there was
no feasible profitable model for the B2B.
44
See Jack Nickelson and Hideo Owen, “A Theory of B2B Exchanges Formation,” 2001, manuscript.
45
http://www.easy-viral-traffic.com/blog/index.php/top-list-of-social-networking-sites/
46
http://www.comscore.com/press/release.asp?press=2396
47
ComScore Data
48
See D. Boyd and N. Ellison, “Social Networking Sites: Definition, History, and Scholarship,” Journal of Computer-
Mediated Communication, 13:1, 2007, available at http://jcmc.indiana.edu/vol13/issue1/boyd.ellison.html, last
visited on September 27, 2008. In addition to these “friending” sites there are also professional social networking
sites such as Linked In which in theory help people form professional relationships and in practice help people find
jobs and recruiters to find candidates.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 67
B. Friendster
Friendster was the first social networking site to gain wide popularity.49 It attracted users initially
by providing a dating venue where people could find romantic partners among friends of friends.
That was a different approach than existing online data venues that matched up strangers based on
predictions of their likely compatibility. People posted their profile and were able to communicate
with other users who were no more than four-degrees of separation away. People interested in
finding romantic partners liked the fact that there were social connections which not only helped
screen people but also helped police bad behavior.
Friendster started with a prototype in 2002. Its founder invited 20 close friends to join.50
The site launched formally in March 2003. It grew virally from the initial seed of 20 friends
49
See Boyd, D. M., & Ellison, N. B., “Social network sites: Definition, history, and scholarship” Journal of Computer-
Mediated Communication, 13(1), article 11 (2007).
50
http://www.inc.com/magazine/20070601/features-how-to-kill-a-great-idea_pagen_2.html
PLATFORM ECONOMICS:
68 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
FRIENDSTER MEMBERSHIP
25
40.00%
28.10% 30.00%
20 24.00%
22.20%
In Hundred Thousand
19.90%
20.00%
14.30%
9.60%
15 7.60% 6.60% 10.00%
4.50% 3.20%
0.00%
-1.10%
10 -3.80%
-10.00%
-9.80%
-20.00%
5
-24.40%
-30.00%
-29.30%
0 -40.00%
3
7
2
7
3
04
05
06
07
08
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
b-
b-
b-
b-
b-
ay
ay
ay
ay
ay
ar
pr
ep
ug
ug
ug
ug
ov
ov
ov
ov
Fe
Fe
Fe
Fe
Fe
M
M
N
N
S
A
d
e
ch
un
La
FACEBOOK MEMBERSHIP
350.00
146.10%
150.00%
300.00
In Hundred Thousand
250.00
100.00%
200.00
59.20%
50.00%
150.00 33.90% 33.80%
26.60%
18.70%
12.90%
100.00 5.10%
0.30%
0.00%
-0.30% -3.60%
50.00 -5.00%
0.00 -50.00%
5
7
4
7
4
8
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
ay
ay
ay
eb
ug
ug
ug
ec
eb
ov
eb
ov
eb
ov
eb
M
M
D
N
F
A
F
F
d:
e
ch
un
La
Note: Based on ComScore data for the period Feb 04 –Feb 08 and “Facebook Case Study”, Harvard Business School, 2006, for the period
Apr 03-Feb-03.
and had 835,000 registered users by June 2003 and 1.5 million by September. There is no
evidence that it focused on influencers or gregarious people—instead it just grew naturally.
The growth was not driven entirely by the users looking for people to date. Most people used
it to communicate with and expand their network of friends. Many people used it as a source
of entertainment from looking the profiles of people that they would not otherwise have
contact with.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 69
Although it reached critical mass and grew explosively Friendster encountered two major prob-
lems.51 The first was technological. It was not prepared for the upsurge in users and users started
finding that the site was not reliable. The four- degree of separation rule was a major source of the
problem. Every time someone signed on to the site the software had to determine who was within
the individual’s community. As the size of the overall network grew it took longer and longer to
do that calculation.
The second involved changes in the community and Friendster’s response to those changes. As
the community grew the four-degree of separation rule made people’s profiles accessible to people
that they did not necessarily want in their social network—for example their bosses. At the same
time the four-degree of separation rule limited the size of the community that one could acquire.
To counter this, people started gaming the system to create more friends. One of these efforts
involved creating Fakesters that were fake profiles that were designed to attract a lot of attention
and therefore many friends. Friendster banned fake profiles and also dropped the “most popular”
attribute that had encouraged people to engage in some of the bad behavior. As it turns out, the
Fakesters were a popular source of entertainment and way to create groups such as alumni of par-
ticular universities. Friendster also policed the profiles and kicked users off who did not comply.
The user community became dissatisfied with Friendster. Growth slowed.52 In the U.S. market
which is the most significant source of advertising revenue Friendster declined dramatically after
the third quarter of 2003 and remained roughly flat for the next three years while other social net-
working sites grew explosively. Nevertheless, it achieved a critical mass of users in other countries,
and remains the 3rd largest social networking site worldwide;53 in August 2008, announced that
they received a $20 million of venture capital.54
C. Facebook
Facebook’s founders also started with the idea that they would provide a convenient method for
dating. As students at Harvard they thought it would be helpful to have a place where students
who saw each other in class and other settings could connect. Within the first month of starting
in February 2004, more than fifty percent of Harvard college students had put their profiles on
51
See Gary Rivlin, “Wallflower at the Web Party,” New York Times, October 15, 2006.
52
MySpace picked up a number of Friendster’s users. It provided a platform in which people could see anyone’s profile
and in which people had a great deal of freedom in designing their profiles. An important group that it attracted
were indie rock groups, some of whom had been ejected from Friendster, and then musicians more broadly who
used the site to connect with their fans.
53
http://www.prnewswire.com/cgi-bin/stories.pl?ACCT=109&STORY=/www/story/08-18-2008/0004869579
&EDATE=
54
See Tom Taulli, “Friendster is still alive and well ... and gets $20 million,” Bloggingstocks, posted on Aug 5, 2008,
http:// ww.bloggingstocks.com/2008/08/05/friendster-is-still-alive-and-well-and-gets-20-million/ (accessed September
24, 2008).
PLATFORM ECONOMICS:
70 Part One : Economics of Multi-sided Platform Businesses Essays on Multi-Sided Businesses
the site.55 Since Facebook was designed to facilitate communication within a closed community
it was able to obtain critical mass—enough people found it valuable enough to use the site—very
quickly and then captured most of the population of potential users.
Shortly thereafter Facebook extended its model to other universities beginning with Stanford,
Columbia, and Yale. In each case a significant portion of the undergraduate student body set up
profiles. By the end of the year Facebook had almost 1 million active users. All users had to have
an “.edu” email extension to verify that they were part of an academic community. Facebook con-
tinued to add local networks of students. It had more than 800 college networks by May 2005. It
then added international schools and work environments. By August 2006 it had 14,782 million
registered users.
Facebook opened its doors to everyone in September 2006. However, it offered a much dif-
ferent level of privacy than either the imploding Friendster or the exploding MySpace. When a
person creates a profile on Facebook you cannot interact with anyone else at the beginning. That
person can build a community of friends either by sending out friend requests, or by being invited
by others to be a friend. Users can also “de-friend” people and have various ways to restrict access
to portions of their pages.
Facebook did not have any significant revenue stream for its first 36 months and focused in-
stead on building its user base. In February 2007 when it had 16,737 million unique users world-
wide56 it introduced Facebook Gifts. These are icons such as flowers, hearts, and balloons that users
can purchase for $1 can send to other friends. In May 2007 they launched Marketplace for classi-
fied listings for sales, housing, and jobs. In November 2007 the company launched Facebook Ads,
an ad system that allows businesses to target their advertising to the precise audience they seek to
reach57. As of September 2008, it is estimated that Facebook earns $350 million58 from advertising
and $35 million59 from the sale of gifts.
Facebook’s successful ignition was based on a form of the two-step. The first step involved de-
veloping the user base. There was a chicken-and-egg problem to be solved there since the network
could grow only if there was a reciprocal relationship between enough people. However, by provid-
ing an effective vehicle for communication social networking sites can grow very quickly virally as
friends invite friends who invite more friends.
The second step involved earning revenue from that user base. An important aspect of that was
selling access to the user base to advertisers. Facebook and other social networking sites have been
disappointed at their ability to secure advertising revenue. Online advertising is sold on the basis
55
http://www.guardian.co.uk/technology/2007/jul/25/media.newmedia
56
http://www.comscore.com/press/release.asp?press=1519
57
http://www.facebook.com/press/info.php?timeline
58
http://www.emarketer.com/Article.aspx?id=1005257
59
http://lsvp.wordpress.com/2008/01/23/facebook-digital-gifts-worth-around-15myear/
http://facereviews.com/2008/09/02/facebook-virtual-gifts-make-big-bucks/
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses How Catalysts Ignite: The Economics of Platform-Based Start-Ups 71
of “cost per thousand” viewers of an advertisement. The average CPM for online advertising is in
the range of $10-15. The average for Facebook is reportedly below $0.50. (Of course although
Facebook may not prove as valuable as its founders and investors might have liked it is likely to be
enormously profitable.)
Part of the difficulty which appears not to have been anticipated is that advertising is a bit alien
to a social networking community. It is natural to see advertising on publisher web sites. It is less
natural to see advertising on your own profile page or the profile pages of friends—especially if you
have been using Facebook for a while without such advertising. Facebook alienated its customers
when as part of its Beacon advertising strategy it started tracking users` purchases from third-party
websites. Facebook and other social networking sites are looking for innovative ways to advertise
that satisfy the needs of advertisers without reducing the value of the social networking aspects of
the sites.
The social networking sites illustrate the power of diffusion in generating web traffic. The suc-
cessful sites grew very quickly because they provided existing physical networks of friends much
more powerful tools for interacting with each other more efficiently virtually. That, however, is
only the first step in a sequential entry strategy. The later strategies have required getting advertisers
on board the platform.60
60
At this point in time the social networking sites have, as noted above, found it difficult to obtain much interest from
advertisers and are charging vastly lower prices for viewers than are other content-oriented sites.
Part Two
TWO-SIDED ANTITRUST
ECONOMICS
CHAPTER
Antitrust Economics
4 of Two-Sided Markets
David S. Evans
ABSTRACT
“Two-sided” markets have two different groups of customers that businesses have to get on
board to succeed—there is a “chicken-and-egg” problem that needs to be solved. These indus-
tries range from dating clubs (men and women), to video game consoles (game developers and
users), to credit cards (cardholders and merchants), and to operating system software (applica-
tion developers and users). They include some of the most important industries in the economy.
Two-sided firms behave in ways that seem surprising from the vantage point of traditional
industries, but in ways that seem like plain common sense once one understands the business
problems they must solve. Prices do not and prices cannot follow marginal costs in each side
of the market. Price levels, price structures, and investment strategies must optimize output
by harvesting the indirect network effects available on both sides. By doing so, businesses in
two-sided industries get both sides on board and solve the chicken-and-egg problem. There
is no basis for asking regulators or antitrust enforcers to steer clear of these industries or to
spend extra effort on them. The antitrust analysis of these industries, however should heed the
economic principles that govern pricing and investment decisions in these industries.
I. INTRODUCTION
Dating clubs—typically bars or cafes—are an innovative way for men and women to meet each
other in Japan.1 At one club, men and women sit on opposite sides of a glass divide. If a man sees
a woman he likes, he can ask a waiter to carry a “love note” to her. Dating clubs sell patrons the
prospect of making a match. Their business works only if they attract enough members of the op-
posite sex to their club to make a match likely. Enough men must participate to attract women,
and enough women to attract the men. The club must figure out how much to charge men and
1
Howard W. French, Osaka Journal; Japanese Date Clubs Take the Muss Out of Mating, N.Y. TIMES, February 13,
2000.
73
PLATFORM ECONOMICS:
74 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
women so that the club gets the right number and mix of patrons, while at the same time making
money since most of these clubs are in business to turn a profit. One bar does this by charging men
$100 for membership, plus $20 a visit, and letting women in for free.2 The bar presumably believes
that at equal prices it would attract too many men, or too few women. That pricing structure—one
that obtains a disproportionate share of the revenues from the men—seems common, based on an
unscientific survey, in singles bars, discotheques and other legitimate businesses around that world
that help men and women find companionship.3
Dating clubs are an example of a “two-sided market.”4 In such a market there are two sets
of customers who, in effect, need each other. Each type of customer values the service more if
the other type of customer also buys the service. Businesses service such markets by acting as
“matchmakers.” To do so, they must get both types of customers on board to have a product to
sell. Indeed, in such markets the product or service is consumed jointly by two customers and, in
a sense, only exists at all if a “transaction” takes place between them.5 For example, the “product”
created by the dating club is, in the end, a date from which both parties expect to benefit, and
which occurs only if both of them participate. A fundamental economic characteristic of two-sided
markets is the presence of positive externalities from having the other side on board (lots of guys to
meet) and the inability of the parties to the transaction to internalize these externalities themselves
2
Id.
3
Here are some examples based on recent (Web-site) visits: C2K, a dance club in Las Vegas, is free for local women
while the cover charge is $10 for out-of-state women and $15 for men (visited Aug. 15, 2002) <http://www.lasvegas.
com/localmusic/danceclubs.html>; the Buddha Lounge in Chicago charges $5-$15 less to women, depending on
the day of the week, than to men (visited Aug. 15, 2002) <http://centerstage.net/dance/clubs/buddha-lounge.html>;
and on Saturday nights, The Wave Nightclub in Atlantic City lets women in for free while men are assessed a cover
charge of $10 (visited Aug. 15, 2002) http://www.poolspanews.com/2002/011/ac_nightlife.html>. A recently devel-
oped online matching service has chosen equal prices—it specializes in matching identical twins. Twins Seek Twins
in Online Matchmaking First, REUTERS NEWS, Apr. 11, 2002; and Twins Realm website (visited Aug. 15, 2002)
<http://www.twinsrealm.com/>.
4
The general economics of two-sided markets are discussed in a seminal paper by Jean-Charles Rochet & Jean Tirole.
See Jean-Charles Rochet & Jean Tirole, Platform Competition in Two-Sided Markets, Financial Markets Group
Discussion Paper #dp0409, (Nov. 26, 2001) <http://www.idei.asso.fr/Commun/Articles/Tirole/PlatformNov26.
pdf>. See also Bernard Caillaud & Bruno Jullien, Chicken & Egg: Competing Matchmakers, Cepr Working Paper
#2885 (Apr. 24, 2001), at 5; and Bruno Jullien, Competing in Network Industries: Divide and Conquer, Idei Working
Paper, (Jul. 2001). Many of the notions discussed in this article were first introduced in papers that analyzed the
payment card industry as a two-sided market. See, e.g., Richard Schmalensee, Payment Systems and Interchange Fees,
L(2) J. Indus. Econ. 103 (Jun. 2002); and Jean-Charles Rochet & Jean Tirole, Cooperation Among Competitors: Some
Economics of Payment Card Associations, Cepr Working Paper #DP2101 (Mar. 1999). This work is based in part
on notions that were first recognized in W. F. Baxter, Bank Interchange of Transactional Paper: Legal and Economic
Perspectives, 23(3) J.L. & Econ. 541 (1983).
5
Although we shall see shortly that in many two-sided markets the parties do not even know one another. There is an
intermediary who matches them and each only deals with that intermediary.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 75
(only one or a few of the guys get your attention).6 Firms profit themselves and society by figuring
out ways to internalize these externalities.
Many important industries, including some central to what is often called the “new” e conomy,7
are based on business models that are similar to dating clubs. Computer operating systems pro-
vide features that software developers can use in writing applications, and a platform on which
computer users can run applications that use those features. Both developers and users must use
the operating system for it to be a viable product. The PalmOS for handheld devices, Microsoft
Windows for the desktop, and Sun Solaris for servers all depend for their success on obtaining
application users and application developers. Videogames are similar. People who buy video game
consoles, such as the Sony PlayStation, want games to play.8
Developers of games want to write for consoles that have many players.9 Console manufactur-
ers must attract both developers and users. Payment cards—credit, debit and charge cards—are
yet another example.10 They are used by consumers to make payments and by merchants to take
payments. Merchants are more willing to accept cards that are more widely held by shoppers,
and shoppers are more willing to carry cards that are accepted by more merchants. Industries
that arrange for buyers and sellers to meet each other also are two-sided markets: Internet-based
business-to-business exchanges (B2B), apartment rentals, and corporate bond trading are but a few
examples. Dating clubs, computer-operating system makers, video-game manufacturers, payment-
card systems, and B2B exchanges produce “platforms” that make “matches” between two or more
distinct groups of consumers.
6
A positive externality is something that party A generates for party B but for which party A has no practical way
to demand compensation. Planting flowers in your front yard is an example. A negative externality like auto-
mobile exhaust fumes causes harm. See Paul Samuelson, The Pure Theory of Public Expenditure, 36 REV. ECON.
STAT. 387-89 (1954). Dennis W. Carlton & Jeffrey M. Perloff, Modern Industrial Organization 82-
83 (3d ed. 2000).
7
For discussion of the empirical importance of the new economy, see David S. Evans & Richard Schmalensee,
Some Economic Aspects of Antitrust Analysis in Dynamically Competitive Industries, in Innovation Policy and The
Economy Vol. 2 (Adam B. Jaffe et al. eds., 2002).
8
See, e.g., Chris Morris, Console Wars: Round Two, CNN.COM (visited Aug. 7, 2002) <http://money.cnn.
com/2002/05/22/technology/e3_consolewards>. Morris states: “Console hardware sales help establish a user base,
but it’s ultimately the games that make or break a system.”
9
See, e.g., Jim Davis, Microsoft Woos Game Developers, CNET.COM (visited Aug. 5, 2002) <http://news.com.
com/2100-1040-237819.html>. In connection with the launch of Microsoft’s video game console, Davis states:
“Microsoft will have to convince people (i.e., developers) that there will be an installed base that makes development
efforts worthwhile (…). In other words, there have to be enough potential customers for people to sell games at a
profit.”
10
David S. Evans & Richard Schmalensee, Paying With Plastic: The Digital Revolution in Buying and
Borrowing 150 (1999).
PLATFORM ECONOMICS:
76 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
A key aspect of the business model for most of these industries involves the optimal pricing
structure: the division of revenues between the two sides of the market that gets both sides on
board.11 Most computer operating system vendors do not seek significant revenues from software
developers, choosing instead to collect mainly from users (Windows) or from the sale of com-
plementary hardware (such as Palm and Sun). Sellers of video game consoles, however, do earn
significant revenues from the game developers. Charge-card companies like American Express earn
a disproportionate share of their revenue from merchants. And to take an example not discussed
above, media sites tend to give readers content for free and collect their revenue from advertis-
ers. The need for a pricing structure as well as a pricing level distinguishes industries based on a
two-sided market from the industries ordinarily studied by economists. In two-sided markets, the
product may not exist at all if the business does not get the pricing structure right.12
It turns out that most, if not all, industries characterized by network effects—a subject of
considerable economic theorizing since the mid 1980s—are two-sided markets.13 A network
effect arises when the value that one user receives from a product increases with the number of
other users of that product. A modern, but already almost quaint example, is the fax machine.
You value a fax machine if there are many people to whom you can send faxes and who can send
faxes to you. Most network effects arise because the product tends to be two sided. That is clear
when there are two distinct types of customers (such as men and women in the dating club ex-
ample). In other cases the two-sided nature it is subtler. Consider the fax machine. Most people
want to both send and receive faxes. There is just one kind of customer who wants to be on a
network with other customers. The same is true for telephones, instant messaging, and e-mail
services. However, at any given instant, these markets can be divided into senders and receivers
and the pricing strategies adopted by businesses in these markets are driven in part by these two
kinds of customers.14
11
We focus on division of the revenue since many of the costs of providing two-sided markets are joint. For docu-
mentation of the assertions in this paragraph, see the discussion of observed business models in Section II and the
discussion of the theory of two-sided market pricing strategies in Section III.
12
The district court in NaBanco noted that at an interchange fee (which, as discussed later, is used by the card systems
to balance demands on the cardholder and merchant sides) of zero, the Visa system would have likely been substan-
tially less successful. See National Bancard Corp. (NaBanco) v. VISA U.S.A., 596 F. Supp. 1231, 1262 (S.D. Fla.
1984).
13
For a discussion of network effects, see generally Michael L. Katz & Carl Shapiro, Network Externalities, Competition,
and Compatibility, 75 Am. Econ. Rev. 424 (1985); Michael L. Katz & Carl Shapiro, Technology Adoption in the
Presence of Network Externalities, 94 J. Pol. Econ. 822 (1986); Chien-fu Chou & Oz Shy, Network Effects without
Network Externalities, 8 Int’l J. Indus. ORG. 259 (1990); S.J. Leibowitz & S.E. Margolis, Network Externality: An
Uncommon Tragedy, 8(2) J. Econ. Persp. 133 (1994).
14
See Rochet & Tirole (2001), supra note 4, at 36 n.26. For economic models that treat the two-sided nature of com-
munication markets, see Jean-Jacques Laffont et al., Internet Interconnection and the Off-Net-Cost Pricing Principle,
Rand J. Econ. (forthcoming).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 77
Industries with network effects have been under close examination by U.S. and foreign antitrust
enforcement agencies.15 Several significant cases were brought by these agencies during the 1990s and
early 2000s and businesses that competed in two-sided markets were at the center of most of these
cases: the AOL-Time Warner merger (U.S. and European authorities investigated two-sided markets
including Internet portals, magazines, and free television);16 the credit-card association investigations
(Australian and European authorities investigated a two-sided market involving merchants and card
users);17 U.S., European and private antitrust cases against Intel (which competes in a two-sided
hardware platform market);18 the Microsoft cases (U.S. and European authorities investigated two-
sided markets involving operating systems and other possible computer platforms);19 and probes into
15
For instance, in his speech Robert Pitofsky, a former FTC Commissioner, noted, “The importance of these industries
[high-tech industries] to the economy, and limited antitrust experience and precedent with regard to some uses of
intellectual property, suggests that there ought to be careful antitrust attention to be certain that critical economic
growth is not compromised by the abuse of private market power.” See Robert Pitofsky Speech before the American
Bar Association, Antitrust Analysis in High-Tech Industries: A 19th Century Discipline Addresses 21st Century Problems,
Feb. 25-26, 1999, (visited Aug. 16, 2002) <http://www.ftc.gov/speeches/pit1.htm>. Joel Klein, a former Assistant
Attorney General of Antitrust Division of the Department of Justice, noted, “Civil non-merger enforcement has
become especially important in this era of rapid technological change and the growth of the network industries, and
we have also been very active in this area to ensure that antitrust enforcement keeps up with these changes…” See Joel
Klein Statement before the Committee of the Judiciary of the United States House of Representatives, Washington
D.C., Apr. 12, 2000 (visited Aug. 16, 2002) <http://www.usdoj.gov/atr/public/testimony/4536.pdf>. The EC
Commissioner for Competition Policy Mario Monti also noted that “competition rules are all the more necessary in
the era of the Internet. The Internet is a wonderful enabling technology, which will in principle, increase competition
in many markets. Nevertheless, that does not mean that it is immune from competition problems.” See Mario Monti
Speech, Competition and Information Technologies, Sept. 18, 2000 (visited Aug. 16, 2002) <http://europa.eu.int/
rapid/start/cgi/guesten.ksh?p_action.gettxt=gt&doc=SPEECH/00/315|0|RAPID&lg=EN>. For a discussion of four
U.S. antitrust cases during the Clinton Administration that involved network industries, see David S. Evans, All the
Facts That Fit: Square Pegs and Round Holes in U.S. v. Microsoft, 22 REG. 54, (1999).
16
See Complaint, FTC Docket #C-3989 (downloaded Jul. 25, 2002) <http://www.ftc.gov/os/2000/12/aolcomplaint.
pdf>. Other relevant documents may be found at <http://www.ftc.gov/os/caselist/c3989.htm> (visited Jul. 25, 2002).
17
Commission Plans to Clear Certain Visa Provisions, Challenge Others, European Commission Press Release IP/00/1164,
Oct. 16, 2000 (visited Aug. 14, 2002) <http://europa.eu.int/comm/competition/antitrust/cases/29373/studies/>;
Designation of Credit Card Schemes in Australia, Reserve Bank of Australia Media Release, Apr. 12, 2001 (visited
Aug. 20, 2002) <http://www.rba.gov.au/MediaReleases/2001/mr_01_09.html>
18
See Matt Loney, EC to Drop Intel Antitrust Investigation, ZDNET, Feb. 4, 2002 (visited Aug. 20, 2002); <http://
news.zdnet.co.uk/story/0,,t269-s2103680,00.html James Niccolai>; James Niccolai, FTC Closes Intel Investigation,
PC WORLD, Sept. 26, 2000; Michael Kanellos, Court Lifts Injunction in Intel-Integraph Case, CNET.COM, Nov.
5, 1999 (visited Aug. 20, 2002) <http://news.com.com/2100-1040-232538.html?legacy=cnet>.
19
See United States v. Microsoft Corp., Civil Action No. 98-1232(CKK), Stipulation, Nov. 2, 2001 (downloaded
Jul. 25, 2002) <http://www.usdoj.gov/atr/cases/f9400/9462.htm>; and Commission Initiates Additional Proceedings
against Microsoft, European Commission Press Release IP/01/1232, Aug. 30, 2001 (visited Aug. 20, 2001) <http://
europa.eu.int/comm/competition/index_en.html>. Other relevant documents may be found at <http://www.usdoj.
gov/atr/cases/ms_index.htm> (visited Jul. 25, 2002).
PLATFORM ECONOMICS:
78 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
online broker-dealers (six separate U.S. regulatory investigations and one European investigation
looking into anticompetitive behavior in two-sided e-dealer markets).20
In some cases, the two-sided nature of the market was central to the allegations in the anti-
trust case. For example, the credit-card investigations involved the pricing structure used to bal-
ance the two-sided demand21 and the U.S. Microsoft case included the claim that one side of the
market (applications) was the source of a barrier to entry.22 In other cases, the two-sided nature
of the market provided an important backdrop for understanding the workings of the business.
For example, current investigations into online bond and currency exchanges are examining how
dealers encourage the use of their trading platforms among buyers and sellers.23 As another exam-
ple, the European Commission was concerned that the AOL Time Warner merger would create a
dominant platform in a two-sided market. The concern was that the merged company could use
its allegedly dominant position in on-line music content: AOL, through its contractual agreements
with Bertelsmann, a German media group, and Time Warner would have had a combined share of
30-40 percent of music content in Europe according to the Commission.24 Before approving the
merger, the Commission required the companies to take steps to eliminate AOL’s contractual links
to Bertelsmann, so that the Commission was satisfied that the merged firm “would not have the
critical mass in terms of music publishing rights to dominate the market.”25 In the late 1980s and
early 1990s, Nintendo’s videogame business was also the subject of FTC and state investigations
that looked at Nintendo’s licensing practices and the technology it used to restrict unauthorized
access to the platform.26
20
See Chris Sanders, Update 1 – BrokerTec Says Profitable since Q4 of 2001, Reuters News, Jun. 6, 2002, in forbes.
com, (visited Jun. 10, 2002) <http://www.forbes.com/newswire/2002/06/10/rtr627112.html>. The article names
several online trading companies as being investigated, including the bond trader BrokerTec and currency trader
FXall. See also Chris Sanders, BrokerTec Confirms Probe by US Antitrust Official, Reuters News, May 16, 2002;
and Online Trading Draws Greater Scrutiny, Reuters News, in Cnet.com, May 17, 2002 (visited Aug. 21, 2001)
<http://news.com.com/2100-1017-916334.html?legacy=cnet&tag=lh>.
21
See Christian Ahlborn et al., The Problem of Interchange Fee Analysis: Case Without a Cause? 22(8) Eur. Competition
L. Rev. 304 (2001).
22
See United States v. Microsoft Corp., 253 F.3d 34, 52 (D.C. Cir. 2001). For other relevant documents see Doj
website (visited Aug. 30, 2002) <http://www.usdoj.gov/atr/cases/ms_index.htm>.
23
See Chris Sanders, Analysis—Investigators Sniff Out Online Trading, Again, Reuters News, May 16, 2002; and
Sanders, supra note 20.
24
Commission Opens Full Investigation into AOL/Time Warner Merger, European Commission Press Release IP/00/634,
Oct. 19, 2000 (visited Aug. 20, 2002) <http://europa.eu.int/comm/competition/index_en.html>. See also EEC
Regulation No. 4064/89, Merger Procedure, Art. 8(2) ¶ 46 (Nov. 10, 2000).
25
See EEC Regulation No. 4064/89, Merger Procedure, Art. 8(2) (Nov. 10, 2000); Commission Gives Conditional
Approval to AOL/Time Warner Merger, European Commission Press Release IP/00/1145, Oct. 11, 2000 (visited Aug.
30, <http://europa.eu.int/rapid/start/cgi/guesten.ksh?p_action.gettxt=gt&doc=IP/00/1145|0|RAPID&lg=EN>.
26
See D. Sheff, Game Over: How Nintendo Zapped An American Industry, Captured Your Dollars, and
Enslaved Your Children ch. 12 (1993).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 79
Despite their economic importance, however, two-sided markets have only recently received
attention from economists and, with the exception of some recent work on payment cards, have
received virtually no attention in the scholarly literature on antitrust.27 This chapter explains the
economics of two-sided markets, considers several important examples, and discusses the implica-
tions of this kind of market structure for antitrust analysis.
The remainder of this chapter is divided into four parts. Part II describes the features of two-
sided market industries and explores the business models that firms servicing them tend to adopt. It
finds that those firms search for a pricing structure that will solve the well-known chicken-and-egg
problem common to firms in two-sided markets. It also shows that firms usually invest resources in
developing both sides—sometimes by subsidizing supply on one side of the market and sometimes
by producing that supply themselves.28 Part III then delves into the economic reasons behind these
business practices. It shows that businesses serving two-sided markets have to set pricing structures
that balance the demand on the two sides of the market. The combined price charged to those
on both sides of the market will have a relationship to the matchmaking firm’s marginal costs of
making a match. However, the prices charged on a particular side of a market will ordinarily not
have a relationship either to the marginal cost of making the match or to costs specific to that side.
Indeed, a key feature of these markets is that, because the product jointly benefits two parties, there
is no basis for separating benefits or costs.29 It also explains why investment in one side of the mar-
ket or other, possibly through self-production, is necessary. Part IV discusses the implications of
these features of two-sided markets for antitrust analysis. It shows how standard market definition,
predatory pricing, vertical restraints, and coordinated effects analyses must be modified to take
into account the two-sided nature of these markets. Part V presents brief conclusions.
27
See Schmalensee (2002), supra note 4; and Rochet & Tirole (1999), supra note 4.
28
For a related discussion, see Annabelle Gawer & Michael A. Cusumano, Platform Leadership: How Intel,
Microsoft, and Cisco Drive Industry Innovation (2002).
29
It is well known in economics that the allocation of joint costs is arbitrary. See David M. Kreps, A Course In
Microeconomic Theory 255 n.i (1990). See also Carlton & Perloff, supra note 6, at app. 3B.
PLATFORM ECONOMICS:
80 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Part A examines match-making services similar to the dating club. The residential real-estate
industry operates much like a computerized dating service while investment bankers are more like
the village matchmakers. Part B considers advertising-supported media. Here the intermediaries
match groups of sellers (advertisers) with groups of buyers (consumers). Part C looks at computer
operating systems. Here the intermediary provides a platform—the software—in which those who
write application software and those who use application software can “meet.” Part D examines
video games—a market that is similar to computer operating systems but which has developed a
somewhat different business model. Part E considers payment-card systems. The intermediaries—
American Express for example—have something in common with the individual match makers
(real estate) and the group match makers (media). Individual matches are important—I can use
my American Express card to pay for something at a merchant that takes the American Express
card. Group matches are important too. American Express delivers a particular class of merchants
to a particular class of consumers—and vice versa.
In all these cases we shall see that the intermediary—like the dating club—helps customers
complete a “transaction” by providing a “platform.” The transaction occurs when members of both
sides get together. Looked at from one side, a transaction is buying a house, watching a television
show, buying a software application that runs on a computer, buying a video game that runs on a
game console, and using a credit card at a store. The platform is the product (or set of institutional
arrangements) upon which this transaction occurs. A free television show is a platform where adver-
tisers meet consumers who may be receptive to their advertisements. A video game console is a place
where people who make games can transact with people who play games. A payment-card system is
a place where merchants who sell goods can meet customers who want to buy those goods.
The intermediaries succeed in the businesses discussed below by figuring out a pricing struc-
ture that internalizes the externalities between the two sides. The residential real estate industry
does this by charging potential buyers nothing to look, even though each buyer imposes a cost on
the system; and by charging sellers only if a sale results, even though costs are incurred when a sale
does not take place. This pricing structure tends to increase the number of potential buyers and
sellers just like the dating club pricing structure tends to increase the number of men and women
looking for dates. Software operating system vendors internalize these externalities by reaping most
of their revenue from end-users, while video game console manufacturers earn significant revenues
from royalties from software game developers.
portionate share of revenues from one side of the market, perhaps even taking a loss on one side
of the market.30
Residential real estate sales in the United States is an example of a market maker that typically
charges one side (the seller) while providing free (or low cost) services for the other (the buyer).
Most residential property sales result from the services of a real estate agent or broker.31 Real estate
agents place the property on a common database called a multiple listing service (MLS),32 where
sellers are able to show their homes to a large audience of buyers, thus potentially increasing the
number of offers that they would otherwise receive without the help of a broker.33 Full service
agents typically list the property on a local MLS as part of the overall service package without ex-
plicitly charging for the listing.34 Real estate agents then charge the property seller a commission
when the property is actually sold.35 The traditional sales commission in the United States is six
30
Because sometimes both the costs and the product are joint, the concept of “loss” on “one” side of the market
is not precise. Here I refer to loss loosely as situations in which one side of the market generates a de minimis
(and sometimes zero) share of the revenues and has substantial investments made in it. This will suffice for our
purposes.
31
Approximately 92 percent of the sales analyzed in studies reviewed by Zorn and Larsen were accomplished through
brokers. See Thomas S. Zorn & James E. Larsen, The Incentive Effects of Flat-Fee and Percentage Commissions for Real
Estate Brokers, 14(1) Areuea J. 24, 27 n.3 (1986).
32
Multiple listing services are offered through local non-profit real estate boards or associations. Agents pay a one-time
membership plus a monthly flat fee in order to list properties on an MLS. MLS fees are not tied to the number of
properties listed and can be viewed as a fixed cost for agents. The only condition dictated by MLS membership is
that agents representing sellers split the commission with agents representing buyers––the specific terms of com-
mission-sharing arrangements, as well as the commission itself, are determined by the individual agents. See Metro
MLS website (visited Aug. 30, 2002) <http://www.metromls.com/>; see also Regional MLS of Minnesota website
(visited Aug. 30, 2002) <http://www.rmls-mn.com/rules.html>.
33
See Metro MLS website (visited Aug. 30, 2002) <http://www.metromls.com/>. Property owners can always at-
tempt to sell their own homes. A seller will only list with a broker if the broker has lower transaction costs. That
is, if the broker can show the home to more potential buyers in a shorter span of time at a lower cost than the
seller. James R. Frew & G. Donald Jud, Who Pays the Real Estate Broker’s Commission? in Researchin Law and
Economics: The Economics of Urban Property Rights 177, 178 (Austin J. Jaffe ed. 1987).
34
However, discount real estate brokers (who simply facilitate the transaction without providing advisory services
and can have commissions as low as one percent of the selling price) may add an extra 3 percent of the selling
price to their fee for listing the property on the MLS. See Robert Erwin, Are Discount Brokers Too Good to be True?
Wall St. J. Online (visited Aug. 22, 2002) <http://homes.wsj.com/columnists/qa/20010914-irwin.html>. The
typical (non-discount) broker or agent’s fee will include such services as advising the seller on a reasonable
listing price and showing the property to prospective buyers, without charging extra for the MLS listing. Id.
at 177.
35
The individual agents set commission percentages. While all licensing is conducted through local non-profit real es-
tate boards or associations, implying that each licensed real estate agent is a member of the local board, these boards
do not dictate pricing terms to their members. See Regional MLS of Minnesota website (visited Aug. 30, 2002)
<http://www.rmls-mn.com/rules.html>.
PLATFORM ECONOMICS:
82 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
percent of the property’s selling price.36 In contrast, the buyer has access to all of the properties
listed and can usually buy a home through an agent without paying that agent.37 Subsidies of this
sort get participation from a key side of the market—in this case, the buyer.
Residential real estate agencies have an apparently simple method for pricing—actual sellers
pay a fixed percentage of the sales price. But there is a more complex method in the background.
Potential buyers and sellers are not charged for access to the MLS. They also are not charged for
house showings. Potential sellers pay a commission only if a sale is consummated. The prices to
potential buyers and sellers do not bear any obvious relationship to the any costs that are specific
to serving each side.
Three points are worth noting. First, potential buyers and sellers of real estate benefit from
having someone organize a “bazaar” for them. They all have greater match prospects. Second, po-
tential buyers and sellers have no practical way to internalize externalities from this bazaar. Each
seller has benefited from all the buyers who came through his house; and each buyer has benefited
from all the houses he has seen. Third, the market intermediary (the real estate agencies and the
local associations that provide the MLS) appears to internalize these externalities by adjusting
prices between the potential buyer side and the potential seller side. They do not charge buyers
or sellers any search fees; and they impose the transaction fee entirely on the seller. Whether they
have found the best possible price structure from the standpoint of operating the residential real
estate industry efficiently is beyond our purview. Indeed, as we consider different industries, what
is interesting is that searching for the “right” price structure is a very important aspect of building
a successful business.38
A similar business model applies to an otherwise very different business. Companies that want
to be acquired (sellers) usually retain an investment banker to seek out potential acquirers (buyers).
The services of an investment banker with good contacts in key industry segments is important
for making matches, especially among firms that are privately held.39 If the firm is purchased, the
36
Motoko Rich, Sellers Trade Brokers for Internet Discounters, Wall St. J. Online (visited Aug. 2, 2002) <http://homes.
wsj.com/buysell/agentsandbrokers/20001221-rich.html>; Patrick Barta, Real Estate Agents Face a Threat from the
Web, WALL ST. J. ONLINE (visited Aug. 2, 2002) <http://homes.wsj.com/buysell/agentsandbrokers/20020108-
barta.html>; Danielle Reed, Buyers and Sellers Decide to Go It Alone, Wall St. J. Online (visited Aug. 2, 2002)
<http://homes.wsj.com/buysell/agentsandbrokers/20010330-reed.html>.
37
That is not true, for example, in France where residential real estate is generally not centrally listed. In the United States,
buyers may end up not purchasing a house through a broker, in which case the broker does not even get paid by the seller.
38
Apartment rental agents and location services operate in a somewhat different fashion: typically the rental property
owner pays fees to the rental agent in exchange for the agent channeling renters to the property. However, in tight hous-
ing markets, such as those in New York City or Boston, the renters are required to pay the real estate broker’s commis-
sion instead of the landlord. See Courtney Ronan, Apartment Locaters: How Do They Make Their Money? Realty Times
website, Jun. 30, 1998 (visited Aug. 22, 2002) <http://realtytimes.com/rtnews/rtcpages/19980630_aptlocator.htm>.
39
See Gabor Garai, How to Use Intermediaries to Sell Your Business, in Handbook Of Business Strategy 285, 288
(Pamela Goett ed., 2001).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 83
banker is paid a transaction fee (similar to the commission fee above). Most often, the transaction
fee is a portion of the purchase price,40 although the investment banker may also set a minimum
charge.41 The seller also typically pays a retainer (an up-front fee) to the banker to cover the cost
of researching potential matches based on the seller’s demands.42 Potential buyers pay nothing.43
Similar fee structures prevail for other investment banking services, such as leveraged buyouts and
private placements.44
Some matchmaker markets have not adopted pricing structures that are clearly biased
towards buyers or sellers. Consider electronic business-to-business (B2B) exchanges, which
bring buyers and sellers together usually over the Internet or through a privately operated net-
work.45 B2Bs usually follow either a transaction-fee-only model or a model that includes any
combination of registration fees, transaction fees, and listing (or hosting) fees.46 Registration
fees may be charged to buyers, sellers, or both, and typically involve either a one-time pay-
ment or annual fees in exchange for access to the products or services of the B2B47—including
40
One common formula for determining the broker fee in a large transaction is the Lehman formula, or 5-4-
3-2-1 formula. Under this arrangement, 5 percent is paid on the first $1 million of the sale price, 4 percent on
the next million, 3 percent on the next million, 2 percent on the next million, and 1 percent on the remaining
excess. However, in most small transactions as well as many large transactions, a fixed percentage, equal to
approximately 5 to 10 percent of the entire purchase, is generally used instead. See John W. Herz at al., Broker
and Finder Agreements, in The Mergers and Acquisitions Handbook, 135, 137-8 (Milton L. Rock et al.
eds., 2d ed. 1994).
41
For example, the investment banker may set a minimum transaction fee of $500,000 for a sale price (“considera-
tion”) of up to $10 million, 1.5 percent of consideration for the first $100 million over the minimum threshold and
an additional 1 percent of consideration for amounts over the first $100 million. For a $125 million acquisition,
then, the fees would break out as follows: $500,000 minimum 1 (1.5 percent 3 $90 million 5 $1,350,000) 1 (1.0
percent 3 $25 million) 5 $2,100,000. If the client had paid a retainer of $50,000, this amount would be applied
to the transaction fee, leaving $2,050,000 due upon deal completion. For a more general discussion, including the
varying fee structures in broker-arranged M&A deals, see id. at 135-147.
42
Andrew J. Sherman, Mergers and Acquisitions From A To Z: Strategic and Practical Guidance For
Small- and Middle-Market Buyers and Sellers 28 (1998).
43
Pricing is rarely this simple. Depending on the relative bargaining power of the two firms, the buyer may end up
paying all or some of the transaction fee. See Silverstein, supra note 40, at 140.
44
See Shelley Winston, Producer Price Indexes for Stockbrokers, Dealers, and Investment Banking Companies, Producer
Price Index Program, U.S. Bureau of Labor Statistics (visited Aug. 14, 2002) <http://www.bea.doc.gov/bea/papers/
stockbrokers.pdf >, at 10.
45
See generally Steven Kaplan & Mohanbir Sawhney, E-Hubs: The New B2B Marketplaces, Harv. Bus. Rev. 97 (May-
June 2000).
46
Entering the 21st Century: Competition Policy in the World of B2B Electronic Marketplaces, Part 1: Overview of B2B
Electronic Marketplaces, Federal Trade Commission Staff Report (Oct. 2000) (downloaded Aug. 14, 2002)
<http://www.ftc.gov/os/2000/10/index.htm#26>, at 14-16.
47
Id. at 14-16.
PLATFORM ECONOMICS:
84 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
reduced costs of searching for an audience of buyers/sellers.48 Transaction fees are traditionally
based on either the monetary value of the transaction (and can be assessed to either buyers or
sellers) or on savings realized by the buyer as a result of conducting the transaction through
the particular online B2B.49 Listing or hosting fees are generally paid by the seller in exchange
for permission to market products or services over the online B2B’s website; the B2B operates
as a “catalog” for the seller to market its products to the B2B’s audience.
The B2B’s are like the dating clubs. Successful exchanges need many buyers and sellers. The
potential buyers and sellers cannot internalize the benefits of having many potential parties to
transact with on the other side. However, it does not seem to have been necessary—at least at this
point in their development—to bias the pricing structure to one side or the other. The B2Bs seem
to be able to get both sides on board without much ingenuity in either pricing or investment.
Many stock-trading systems also function without offering prices that obviously benefit one side
over the other. For example, most U.S. stocks trade on auction-style systems, such as the New York
Stock Exchange, in which participants offer to buy and sell at posted prices.50 As with all exchanges,
this only works if there are buyers and sellers looking to trade frequently.51 Otherwise, an interested
buyer may look to trade on the exchange but find no sellers, or vice-versa, within a reasonable time-
frame. Such an exchange has little “liquidity” and would have minimal appeal to many buyers or sell-
ers. There is enough interest in buying and selling U.S. stocks to make auction-style exchanges viable.
In contrast to stocks, corporate bonds are infrequently traded and are, at least for now, traded
through dealers rather than exchanges.52 The lack of liquidity is due to a lack of trading activity gen-
erally rather than any overall imbalance between buyers and sellers. The corporate bond dealer helps
address this liquidity problem by holding an inventory of bonds. It is willing to buy a bond from a
customer, and hold it in inventory to be sold later, even if a buyer cannot be located right away.53 The
dealer attempts to make money by selling at a higher price than it bought. A dealer also provides li-
quidity by actively trying to find buyers and sellers for different bonds.54 In the language of two-sided
48
Entering the 21st Century: Competition Policy in the World of B2B Electronic Marketplaces, Part 2: Efficiencies of B2B
Electronic Marketplaces, Federal Trade Commission Staff Report (Oct. 2000) (downloaded Aug. 14, 2002) <http://
www.ftc.gov/os/2000/10/index.htm#26>, at 4-7. In addition, Internet technology makes it possible for buyers and sellers
in an electronic B2B marketplace to obtain a great deal of information about the B2B customer base, including number
of transactions, transaction volumes, the number of other products the customer viewed before making the transaction,
and customer’s identity, among other things. Entering the 21st Century, Part 1, supra note 46, at 4-5.
49
Entering the 21st Century, Part 1, supra note 46, at 14.
50
Antony Santomero & David Babbel, Financial Markets, Instruments & Institutions 440-441, 445-449 (2d ed.
2001).
51
Id. at 440.
52
Id. at 439.
53
Even on auction-style exchanges, firms known as “specialists” in specific stocks perform a similar role. Id. at 443-444.
54
In this context, the dealer is also acting partly as a “broker” between buyers and sellers. For more on this, see Id. at
437-440.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 85
markets, the dealer attempts to resolve temporary imbalances between the two sides of the market
by either standing in for, or investing efforts to locate, the missing side. In essence, it is willing to
provide, or find, the “chicken” or the “egg,” as needed, to ensure the viability of its platform. This is
in contrast to other markets we have discussed where there is a need to systematically develop one side
of the market. Here, the side that is missing varies from transaction to transaction. Relative prices to
buyers and sellers in dealer markets are difficult to assess, in part because prices are not generally pub-
licly reported. There is no obvious bias in favor of buyers or sellers, which is what would be expected
given the lack of a need to systematically develop a particular side of the market.
By the late 1990s, a number of firms had started developing auction-style electronic exchanges
for trading corporate bonds.55 The idea was that advances in technology (many of these exchanges
operated over the Internet) might allow buyers and sellers to join a virtual exchange, on which
enough buy and sell orders might be posted to generate sufficient liquidity to make the exchange
viable. Some participants hoped that such an exchange, if it were successful, might offer better
prices than available from dealers.56 However, while technological advances might have allowed a
larger exchange (with more liquidity) than before, the basic liquidity problem remained difficult
to surmount. Customers, which are primarily large institutional investors, complained that elec-
tronic exchanges without dealer participation did not solve the liquidity problem.57 The other side
that this two-sided market needed was the dealer community, which had the liquidity that the
institutional investors required. Dealers, of course, had little interest in providing their liquidity to
auction-style exchanges that were designed to ultimately replace them, but many corporate bond
dealers did develop their own systems. Such electronic exchanges owned or run by dealers, were re-
garded by some as promising,58 although it is unclear whether they would eventually replace dealer
markets with auction markets, or simply complement dealer markets.
55
The 1999 Review of Electronic Transaction Systems: eCommerce in the U.S. Fixed Income Market, The Research Department
of the Bond Market Association Report (Nov. 1999) (visited Aug. 29, 2002) <http://www.bondmarkets.com/re-
search/ecommerce_survey99.pdf>; see also The 2001 Review of Electronic Transaction Systems: eCommerce in the U.S. Fixed
Income Market, The Research Department of the Bond Market Association Report (Dec. 2001) (visited Aug. 29,
2002) <http://www.bondmarkets.com/research/ecommerce/ETSRprt1201.pdf>.
56
See Chuck Epstein, Electronic Bond Trading Systems Pass Critical Mass, Wall St. Tech. Online, Feb. 1, 1999 (visited
Aug. 29, 2002) < http://www.wallstreetandtech.com/story/WST20000619S0015>.
57
Bonds Get Wired, The Forrester Report (Sep. 2000), at 5.
58
Id. at 13.
PLATFORM ECONOMICS:
86 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
b uyers with a group of sellers. The sellers in this case are the advertisers looking for a platform to
pitch their products to a receptive group of readers or viewers. The newspapers, television channels,
and Web sites that form media platforms are “audience makers” rather than “market makers.”59
That is, advertising-supported media providers are interested in attracting advertisers on one side
and subscribers on the other to form an audience for the advertisers.
Like many market makers, most audience makers earn a disproportionate share of their revenues
from one side of the market and, like some market makers, may lose money on one side of the mar-
ket. Newspapers, magazines, and television networks charge advertisers a placement fee dependent
on the size or length of the ad and on the circulation or subscriber base of the media platform.60 For
example, full page ads cost more than quarter page ones and television ads that run during hit shows
or special programs, like the Super Bowl, cost considerably more than those that run during non-
prime-time slots. For instance, during the 2002 Super Bowl, which ran on FOX and attracted 86.8
million television viewers, the average price for a thirty-second commercial was $1.9 million.61 In
contrast, the average price for a thirty-second commercial on FOX is slightly over $150,000.62
The fees that media platforms collect from advertisers pay for the content that the media pre-
sents to the audience. That is, FOX buys the rights to shows such as Malcolm in the Middle with
proceeds from advertising, and in exchange, advertisers are allotted a specific amount of time dur-
ing the show to market to the show’s audience (which may include a somewhat specialized demo-
graphic group).63 Contracts between the advertiser and television media platform generally include
59
“In a fundamental sense, what advertisers demand, and what the various advertising media outlets supply, are units
of audience for advertising messages. Thus advertiser demand for space in the print media and time in the broadcast
media is a derived demand stemming from a demand for audience, and is a positive function of the size and qual-
ity of audience.” James M. Ferguson, Daily Newspaper Advertising Rates, Local Media Cross-Ownership, Newspaper
Chains, and Media Competition, 26 J.L. Econ. 635, 637 (1983).
60
Newspapers also have different rates for different classes of advertising, such as display versus classified, and differ-
ent ad types, such as black-and-white versus colored. Fees can also vary depending on the day of the week (Sunday
charges may be higher) and time of day (evening edition versus morning edition). A typical newspaper charge is
determined by the “milinch,” the column inch per thousand circulation. Id. at 653. For television, approximately
70 to 80 percent of network television advertising time is sold prior to the start of the television season, through
contracts specifying price and guaranteed minimum rating of the television shows. Ronald Goettler, Advertising
Rates, Audience Composition, and Competition in the Network Television Industry, Carnegie Mellon University
Gsia Working Paper #1999-E28 (Aug. 8, 1999), at 4.
61
TV, Internet and Advertising Trends Surrounding the 2002 Super Bowl, Nielson Media Research (Feb. 19, 2002)
(downloaded Aug. 15, 2002) <www.nielsenmedia.com/newsreleases/2002/2002%20Super%20Bowl_WriteUp-
POST.doc>.
62
Prime-Time Pricing Survey, Advertising Age (2002).
63
A recent survey of by the Magazine Publishers of America (MPA) found that 45.1 percent of magazine content con-
sisted of advertisements, while 54.9 percent comprised the editorial content. Providing advertising space generates rev-
enue for the magazine and providing editorial content helps retain a reader base. See Historical Advertising /Editorial
Ratios, MPA website (visited Aug. 20, 2002) <http://www.magazine.org/resources/fact_sheets/ed2_9_02.html>.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 87
the price paid by the advertiser for commercial time and a minimum rating for the television show
that is guaranteed by the media platform.64 Advertising prices vary with the television show’s audi-
ence size, age, and gender make-up.65 For example, higher advertising prices are generally associ-
ated with a large and primarily homogenous viewing audience, along with a higher proportion of
viewers between 35 to 49 years of age.66
On the other side, free television viewers pay only an implicit price: the cost of having to
watch commercials or waiting for the show to resume. Other platforms, such as newspapers and
magazines, charge the audience an explicit price. But even here readers are heavily subsidized as
advertisers pay the bulk of the costs of obtaining the content that attracts their desired subscriber
demographics.67
Audience-maker businesses have the basic characteristics of two-sided markets.68 Their pric-
ing structure does not have a direct relationship to the marginal cost of providing the media or
its content to either side of the market. The economist’s usual “marginal revenue equals marginal
cost” condition does not help us understand why television viewers pay nothing for watching
content that was expensive to create or why many newspapers and magazines are distributed for
less than the marginal cost of production and distribution. The audience-makers are helping
buyers and sellers internalize externalities. The advertisers benefit more the larger the number
of potential buyers they can reach. The potential buyers realize a negative externality from ad-
vertising messages—most people would pay to avoid them; the more messages they receive the
bigger the externality. These externalities seem to be addressed by a pricing structure in which
advertisers pay for the audience, and the audience is paid with content for putting up with the
64
Television show ratings are measured by Nielson Media Research. Goettler, supra note 60, at 4.
65
For instance, television shows like Roseanne and Melrose Place that had desirable viewer demographics (large audi-
ence with a high proportion being female and between 18 to 49 years of age) also had substantial gains in ad revenue
for each additional million viewers. This is compared with less considerable increases in ad revenue per one million
additional viewers for Hat Squad, a show with less desirable viewer demographics (audience with a smaller propor-
tion being female compared to Melrose Place and a large proportion being 50 years of age and older). Id. at 3.
66
Id. at 16.
67
Approximately 80 percent of newspaper revenue comes from advertisers. Lisa George & Joel Waldfogel, Whom
Benefits Whom in Daily Newspaper Markets? Nber Working Paper #7944 (Oct. 2000), at 11.
68
The two-sided nature of advertising is well-understood in the advertising literature. “[M]edia firms, such as news-
papers, magazines and commercial television channels, operate simultaneously in two sub-markets. Not only they
sell their products to readers, viewers or listeners, they also sell advertising space to firms. Moreover, these markets
are generally interrelated on the demand side. For example, the value of placing an ad in a local newspaper depends
on the paper’s circulation, and the subscribers’ valuation of the newspaper is, at least to some extent, affected by
the type and amount of advertising.” See Jonas Hackner & Sten Nyberg, Price Competition, Advertising and Media
Market Concentration, Stockholm University Research Papers in Economics #2000:3 (2000), at 1. See also
Kyle Bagwell, ed., The Economics of Advertising Pt. VII-VIII (2001). This paper shows that advertising is part
of a much broader class of two-sided markets. Also, the theoretical results discussed below do not appear to have
been developed in the literature on the economics of advertising.
PLATFORM ECONOMICS:
88 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
ads. Again, whether this is the most efficient method is beyond the scope here, although its
long-term survival in many media suggests that it is. An interesting issue is whether this business
model will work for Web portals. Advertisers seem reluctant to pay much for the occasional click
while browsers seem reluctant to pay for content.69
69
“[M]any traditional advertisers have yet to be convinced of the cost effectiveness of spending their marketing budget
on the web.” See Can Dot.Coms Make Money from Advertising? BBC News, Mar. 8, 2001 (visited Aug. 27, 2002)
<http://news.bbc.co.uk/1/hi/business/1208861.stm>. Many Internet media companies have already been looking
for other sources of income apart from online advertising. For instance, in 1997 Microsoft’s online magazine, Slate,
planned to begin asking its readers to pay $19.95 a year for the magazine they used to receive for free. Eventually,
Slate abandoned this idea because it could lock out too many potential readers. See Slate Clean of Subscription Fees,
CNET.com, Jan. 10, 1997 (visited Aug. 27, 2002) <http://news.com.com/2100-1023-261158.html?tag=bplst>.
In general, consumers are reluctant to pay for most information on the Web including Webzines and music. See
Pay for Music Online? No Way! ZDNet May 17, 2000 (visited Aug. 27, 2002) <http://comment.zdnet.co.uk/
story/0,,t479-s2113872-p2,00.html>.
70
For definitions, see NetLingo The Internet Dictionary (visited Aug. 21, 2002) <http://www.netlingo.com/inframes.cfm>.
71
Id.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 89
There is nothing preventing operating system sellers from collecting revenues from both sides
of the market. Users can be charged license fees for the software.72 Software developers can be
charged for information, tools, and other services necessary for accessing the code they want to
rely on in their programs.73 In fact, all operating system vendors have chosen to get most of their
revenues from the user-side of the market; vendors differ in the extent to which they invest in the
applications-side of the market.74
Consider Microsoft. It licenses its Windows operating system to computer manufacturers
(and also sells some copies at retail and through other channels). This is the source of most of
the considerable revenue that it realizes from Windows.75 Microsoft does not charge software
developers for information on using the features of the operating system. Developers, in fact,
have gotten numerous development tools and considerable support for free from Microsoft,
including the free Microsoft Developer Network (MSDN) Web site and free software develop-
ment kits (SDKs) for Windows.76 The company also shares early beta versions of new operating
system releases with developers.77 The company holds annual development conferences to dem-
onstrate how the Windows programming interfaces are evolving.78 It does earn some revenues,
however, from developers. It offers five different development packages ranging from $99 to
72
For instance, Microsoft Corp. charges end-users a $199 fee for Microsoft Windows XP Home Edition. See Microsoft
Corp. website (visited Aug. 21, 2002) <http://www.microsoft.com/windowsxp/order/>. Apple’s charge for Mac OS
10 X v10.2 is $129. See Apple website (visited Aug. 21, 2002) <http://www.apple.com/macosx/>.
73
For example, Xtras.net supplies certain development tools for .NET at discounted prices. See Xtras website (visited
Aug. 21, 2002) <http://www.xtras.net/>. Also, companies that make development tools, such as Oracle, Borland,
IBM, and Rational Software offer to rent web-based tools to developers. See Wylie Wong, For Rent: Software
Development Tools, Cnet.com, Jan. 30, 2001 (visited Aug. 21, 2002) <http://news.com.com/2100-1001-251751.
html?legacy=cnet>.
74
For general commentary on the importance of investment in third party software developers, see FM Interview
with Linus Torvalds, What motivates free software developers? FIRST MONDAY, Mar. 2, 1998 (visited Jul. 31,
2002) <http://www.firstmonday.dk/issues/issue3_3/torvalds/index.html>; Jerry Pournelle, Jerry’s Take on the
Microsoft Decision: Wrong! BYTE.COM, Nov. 8, 1999, (visited Aug. 9, 2002) <http://www.byte.com/documents/
s=200/byt19991108s0001/index.htm>; and Richard Shim & Sandeep Junnarkar, Palm Reaches Out to
Developers, CNET.COM, Oct. 23, 2001 (visited Aug. 13, 2002) <http://news.com.com/2100-1040-274797.
html?legacy=cnet>.
75
See Microsoft Corp. 10-K (FY2001) (downloaded Aug. 30, 2002) <http://www.sec.gov/Archives/edgar/data/789019/
000103221001501099/0001032210-01-501099.txt>.
76
MSDN website provides online support for developers including technical programming information (sample code,
documentation, technical articles, and reference guides), troubleshooting, access to Microsoft MSDN magazine and
other information. See Microsoft Corp. website (visited August 30, 2002) http://msdn.microsoft.com/ >.
77
See Gawer & Cusumano, supra note 28, at 150-51.
78
For example, the Windows .NET Server Developer Conference in Seattle, Washington from September 3–6, 2002
was focused on providing various technical content to architects, enterprise developers, ISV developers, and techni-
cal decision makers. The cost to register for the entire conference was $1,295. See Windows .NET Server DevCon,
Microsoft Corp. website (visited Aug. 22, 2002) <http://microsoft.com/misc/external/serverdevcon/>.
PLATFORM ECONOMICS:
90 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
$2,799 per individual license.79 The packages include varying levels of access to and discounts
on technical resources and support, code samples, documentation, development tools, software,
and hardware.80
Microsoft has also encouraged end-user demand by developing applications programs that run
on its operating system. In many cases, Microsoft has done this for the same reason any applica-
tions developer does this—to make money from the sales of a successful application.81 In other cas-
es, Microsoft has produced applications to encourage users to license its operating system (which
in turn encourages other applications developers to write applications). In fact, during the early
years of Windows Microsoft had a difficult time persuading other developers to write applications
for it. In 1992 there were only 1,438 applications available specifically for Windows compared
with 22,328 for DOS and 4,213 for Apple’s MacOS.82 As Bill Gates puts it, “In 1989, I person-
ally went to all the applications developers and asked them to write applications for Microsoft
Windows. They wouldn’t do it. So I went to the Microsoft Applications Group, and they didn’t
have that option.”83
Apple, which makes both the Macintosh computer and its operating system MacOS, has
followed a similar approach.84 Apple provides interfaces (the APIs, or application program inter-
faces) for its operating system to developers at no charge.85 They also provide free SDKs that de-
velopers can download from their website.86 However, Apple, like Microsoft, does impose some
charges for tools that help software developers use the APIs. The Apple Developer Connection
(ADC) program provides software and hardware developers access to development tools, soft-
ware, kits, and reference materials, with membership fees ranging from the $99 per year Student
Program to the $3,500 per year Premier Program.87 Based on various membership levels, in
addition to development materials and support, members of the ADC may also have access to
discounted Apple hardware, discounted technical and business services, and Apple’s Worldwide
79
Discounts are offered on volume licenses and academic licenses orders. See MSDN Subscriptions Pricing, Microsoft
Corp. website (visited Aug. 13, 2002) <http://msdn.microsoft.com/subscriptions/prodinfo/pricing.asp#univ>.
80
See MSDN Subscriptions Level & Feature Overview, Microsoft Corp. website (visited Aug. 13, 2002) <http://
msdn.microsoft.com/subscriptions/prodinfo/levels.asp>.
81
For example, the current retail price for the 2002 version of Microsoft Word is $339 (the upgrade from an older version
costs $79.95). See Look Up Prices, Microsoft Corp. website (visited Aug. 21, 2002) <http://www.microsoft.com/office/
howtobuy/pricing.asp>.
82
Software Product Specifications from Computer Select Cd (Jan. 1992).
83
See Pournelle, supra note 74.
84
This and the following sentences in this paragraph are based on information available in the developer section of
Apple’s website (visited Jul. 31, 2002) <http://developer.apple.com/membership/descriptions.html>.
85
See Development Kit, Apple website (visited Aug. 21, 2002) <http://developer.apple.com/sdk/>.
86
Id.
87
See Apple Developer Connection Programs, Apple website (visited Aug. 21, 2002) <http://developer.apple.com/
membership/descriptions.html>.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 91
Developers Conference. Like Microsoft, Apple earns relatively little of its revenues from the
developer side of the business despite the large costs of writing operating system software code
that these developers use.88 Apple, however, has been less assiduous at courting developers than
Microsoft.89
Palm, which is the leader in the market for handheld personal digital assistants (PDAs), cur-
rently follows a similar business model although it took a different approach to solve the chicken-
and-egg problem in the beginning.90 In 1996 Palm introduced the PalmPilot PDA, developed a
new operating system to run on it, and also invested in designing application programs.91 After
obtaining critical mass on one side of the market (users), Palm reduced its investments in applica-
tions, focused on the hardware and operating system, and began to attract independent developers
for applications.92 Palm, in effect, solved the chicken-and-egg problem by initially providing its
own chicken.93 By supplying one side of the market itself, Palm could focus all of its efforts on at-
tracting end-users. Now, however, it expends considerable effort on wooing software developers by
hosting developer forums, providing technical training, and even supplying marketing for comple-
mentary software developers.94 In September 2000, Palm launched Palm Ventures, a $50 million
venture unit to aid firms developing complementary products.95 For individual developers, Palm
provides free development kits, product images, limited access to source code, and early access to
tools and information.96 Palm also offers an advanced plan for a $495 annual fee, which includes
direct technical support, marketing, and quarterly updates on development tools and technology.97
88
In 2001, Apple derived 82 percent of its total revenue from sales of Power Macintosh, PowerBook, iMac and iBook
to end-users, and 18 percent from third-party developers, services, and sales of iPad, various peripheral products
and software. See Apple Computer, Inc. 10-K (FY2001) (visited Aug. 27, 2002) <http://www.corporate-ir.net/ireye/
ir_site.zhtml?ticker=aapl&script=800&layout=8>.
89
See Pournelle, supra note 74.
90
See Shim & Junnarkar, supra note 74; Media Backgrounder, Palm website (visited Aug. 13, 2002) <http://www.
palm.com/about/pr/background.html>.
91
See Gawer & Cusumano, supra note 28, at 190-91, 194.
92
Id.
93
Palm’s approach was, at least in part, dictated by earlier PDA failures by Apple with the Newton and by Palm with
the Zoomer. As Palm cofounder Donna Dubinsky explained, they were determined not to rely on outside develop-
ers until they had sold at least a million devices. “We are going to prove it to developers,” she stated, “We are not
going to ask them to trust us.” See D.B. Yoffie & M. Kwak, Judo Strategy Ch. 5 (2001). Gawer and Cusumano
argue that a do-everything strategy, while helpful initially, is not a viable long term strategy. “If the product turns
out to be successful with customers, specialist firms are likely to jump into the market. Often, they become better
at specific tasks (like hardware design or applications design) than firms that try to do everything.” See Gawer &
Cusumano, supra note 28, at 249.
94
See Related Developer Programs, Palm website (visited Aug. 15, 2002) <http://www.palmos.com/dev/programs/>.
95
Palm Inc. Lauches VC Unit to Aid Complementary Firms, Boston Globe, Sep. 7, 2000.
96
Shim & Junnarkar, supra note 74.
97
Id.
PLATFORM ECONOMICS:
92 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
In addition, developers can pay to attend brief courses, such as the two-day “Advanced Palm OS
Development and Debugging” course for $1,000.98,99
Palm’s efforts appear to have paid off: in March 2002, Palm boasted that it had nearly
200,000 registered software developers and more than 13,000 software titles already available.100
Applications range from electronic book-reading with thousands of e-books, to document-editing
programs that work with Microsoft Excel and Word files, to enterprise software tools that let busi-
ness users fill out forms and check inventories through a wireless connection.101 For consumers,
Palm OS powered devices typically start at $99.102 But like Microsoft and Apple, Palm earns the
preponderance of its revenues from the user-side of the business.103
Computer operating system characteristics are similar to those we have seen in our review
of other two-sided markets. The business is about the search for the right price structure that
will get both sides on board. The price structure is one that is not readily explained by the
approach economists apply to one-sided markets. Here one side seems to get extraordinary
benefits from the operating systems but pay little for it. Once again there is an externality
problem that only an intermediary between the two sides can solve. Users benefit when they
have many applications available to run on their computers and when the similar portions of
those applications are common. They have no way to pay software developers for the benefits
of this coordination. Likewise, developers benefit when they have a larger group of users for
their software and when they can avoid duplicating the code used by other packages. They
have no way to pay users for the benefits of this coordination. The intermediary can solve the
problem—internalize these externalities between the two groups of customers—by adopting a
98
See Developer Training, Palm website (visited Aug. 13, 2002) <http://www.palmos.com/dev/training/instructor/
advanced.html>.
99
As another means to encourage independent developers to write applications for Palm OS, Palm licenses its oper-
ating system to other hardware manufacturers. For example, Handspring, which makes PDAs that compete with
Palm’s, is one of many licensees of the Palm OS. See PalmSource website (visited Aug. 13) <http://www.palmos.
com/licensees/>. Currently, both Palm and Handspring make the interface specifications for their computing plat-
forms available at no charge to developers. Thus, developers profit from selling Palm- and Handspring-compatible
software, and in exchange, Palm and Handspring profit from having a platform that becomes increasingly popu-
lar to consumers as more compatible software is developed. See Douglas Lichtman, Property Rights in Emerging
Platform Technologies, XXIX J. Legal Stud. 615, 616 (2000).
100
See Media Backgrounder, supra note 90. In recent years, Palm has faced increasing competition from the Pocket PC
platform. See Mark Walser, The (M)empire Strikes Back, M Commerce Times, Jul. 9, 2001 (visited Aug. 29, 2002)
<http://mcommercetimes.com/Technology/147>.
101
See Product Overview, Palm website (visited Aug. 13, 2002) <http://www.palm.com/about/corporate/productoverview.
html>.
102
See Developer Training: Class Topics, Palm website (visited Aug. 13, 2002) <http://www.palmos.com/platform/
os5/topten.html>.
103
See Palm, Inc. 10-K (FY2002) (visited Aug. 27, 2002) <http://media.corporate-ir.net/media_files/nsd/palm/
reports/fy0210k.pdf>.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 93
pricing structure that recognizes the mutual benefits. As before, whether the pricing structure
the operating system vendors have settled on is the best one for solving this problem is not
readily known.
104
It should be emphasized that we focus on the history of home video games in the U.S. Some of the platform
launches we describe for the U.S. were preceded by releases in Japan.
105
See S. Kent, The Ultimate History of Video Games Ch.1-8 (2001). The Odyssey, manufactured by Magnavox
and introduced in 1972, was the first home video game console. It was the only multi-game console of this early
period. While Home Pong played only one game, Odyssey could play twelve. See KENT supra, at 80-81.
106
The Video Computer System was also known as the Atari 2600. Id. at xiii.
107
Id. at 94-98.
108
Id. at 107.
109
On the significance of game availability for the success of a console, see, among others, Morris, supra note 8. See
also David Becker, Who Will Be the King of the Game Consoles, CNET.Com, Nov. 21, 2000 (visited Aug. 5, 2002)
<http://news.com.com/2100-1040-248875.html>.
PLATFORM ECONOMICS:
94 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
in the home video game market.110 The company’s purpose was to create games for the VCS.
Activision soon released its first games and was an overnight success. Again like Palm, the VCS
game platform reached such a level of success with customers on one side of the market (end-users
of games) that it generated demand from customers on the other side (independent game develop-
ers). Activision represented a turning point in the video game industry: from this point on console
manufacturers entering the market had to attract both gamers and developers to the platform to
ensure its success. A two-sided market emerged.
Although it is unclear whether Atari or any other company in the early 1980s charged licens-
ing fees to independent developers, from the mid-1980s onward (and more specifically from the
time Nintendo entered the home video game market on), licensing fees from software developers
became a major source of revenues for console manufacturers.111 In fact, the business model that
developed at least from the mid-1980s on was based on selling the console to end-users near or less
than its marginal cost of production while relying on revenue from in-house-produced software
and from license fees charged to independent developers to recover fixed development costs and
earn a profit.112
Starting in the mid-1980s, Nintendo and its 8-bit Nintendo Entertainment System (NES) dis-
placed Atari as the video game leader. Nintendo was already a big arcade company prior to entering
the home segment of the market, and it attracted end-users by converting many popular arcade
games into home video games.113 As far as courting independent developers, Nintendo sought
quality over quantity. Nintendo protected the NES with a security chip that locked out unauthor-
ized, unlicensed cartridges. The licensing agreements that it signed with independent developers
imposed hefty royalty rates, established that they would only publish five games per year and that
those games would be exclusive to the NES for a two-year period.114
By the late 1980s a new competitor, Sega, emerged. Sega’s strategy to compete with Nintendo
was manifold. In the first place, it beat Nintendo to market with the release of a 16-bit console that
110
Still other Atari programmers left the company to form Imagic. Like Activision, Imagic enjoyed success very quick-
ly. See KENT supra note 105, at 192-193, 197.
111
See, e.g., Software Can Raise Platform Values, Video Games & PC Entertainment Software – Marketers, The Investext
Group, Jun. 1, 1998, available in NEXIS. See also Morris, supra note 8.
112
See, for example, David Becker, Revenue from Game Consoles will Plunge, Report Predicts, CNET.Com, Jul. 31,
2000 (visited Aug. 5, 2002) <http://techrepublic-cnet.com.com/2100-1040-243841.html?legacy=cnet>. See also
David Becker & Joe Wilcox, Will Xbox Drain Microsoft? CNET.Com, Mar. 6, 2001 (visited Aug. 5, 2002) <http://
techrepublic-cnet.com.com/2100-1040-253654.html>.
113
See Kent, supra note 105, at 278, 296-297, 299-300.
114
Id. at 308-309, 351. Nintendo did grant special favors to some of its most successful licensees. It gave a second
license to both Konami and Acclaim, two of the top independent game developers for the NES, so that each com-
pany would be able to produce 5 additional games per year. Id. at 422. On the relationships between Nintendo and
its licensees, see also SHEFF, supra note 26, at chapter 10.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 95
was considerably more powerful than the NES.115 Sega’s Genesis was launched in America in 1989,
almost two years earlier than Nintendo’s 16-bit console, the Super NES.116 Second, in order to at-
tract end-users Sega soon cut the price of the console from $189 to $149.117 Third, since Nintendo
dominated the arcade translation business, Sega signed contracts with athletes and celebrities in or-
der to create a large library of games based on their names and images.118 Fourth, Sega launched an
aggressive marketing campaign to highlight the technological superiority of the Genesis over the
NES.119 Finally, even though in general the Sega licensing deals with independent game developers
were similar to Nintendo’s,120 Sega cut deals with some key developers (e.g., Electronic Arts) by
offering them special licensing terms.121 Sega’s strategy paid off and it outsold Nintendo in 1991.122
Even after Nintendo launched its 16-bit Super NES, Sega remained competitive for some time on
the basis of its larger library of game and its faster pace of game publishing.123
In 1995 yet another competitor joined the fray: Sony’s 32-bit PlayStation. Sony took advan-
tage of a window of opportunity that arose in the mid 1980s. At this time Sega was struggling in an
attempt to support several incompatible platforms and Nintendo was starting to develop a 64-bit
console (Nintendo 64), which it would not release until 1996.124
The Sony model was different from both Sega’s and Nintendo’s in that Sony made a much
stronger effort to attract independent developers. For example, Sony offered excellent develop-
ment tools to third-party publishers and PlayStation developed a reputation for being very easy to
program.125 Combined with a liberal $10 per game licensing fee and Sony’s aggressive marketing
115
Sega had unsuccessfully released an 8-bit console, the Master System, in 1986. See Kent, supra note 105, at 401,
404-405.
116
Id. at 404, 431-434.
117
Id. at 404, 427, 433.
118
Id. at 404-406, 426. Sega did have its own portfolio of arcade games, but Nintendo had exclusivity on some of the
most popular arcade products.
119
Id. at 426-427.
120
Id. at 381, 409-410. Independent developers paid Sega between $10 and $15 per cartridge on top of the real hard-
ware manufacturing costs. Like Nintendo, Sega created security systems in its consoles to guard against unlicensed
publishers. Id. at 381-382.
121
In 1989 Electronic Arts’ technicians successfully reverse-engineered both the Nintendo NES and the Sega Genesis.
Electronic Arts rejected Sega’s licensing terms and decided to go ahead and start publishing their own games on
Genesis. Before proceeding, however, they offered Sega the possibility of entering into an agreement with special
licensing terms. Sega accepted and granted Electronic Arts a licensing contract with increased flexibility and lower
royalty rates. Id. at 408-410.
122
Id. at 434.
123
Id. at 447-448.
124
Id. at ch. 27.
125
Sega also launched its 32-bit platform, the Saturn, in 1995, but at launch time this platform had already developed
a reputation for being very difficult to program. Id. at 509, 516. As a result of Saturn’s failure, Sega stopped manu-
facturing consoles in early 2001. Next Generation of Gaming: Forecasts and Analysis, 2000-2005, IDC REPORT
#24432 (Apr. 2001), at 26-28.
PLATFORM ECONOMICS:
96 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
plans, PlayStation became a very attractive platform for game designers.126 By the time it launched
in the U.S., about 100 game developing companies had signed agreements with Sony––and more
than 300 projects were either planned or underway.127 Game developers were now writing similar
titles for multiple game platforms (Nintendo, Sega, and Sony).128 As far as the end-user side of the
market is concerned, Sony attracted consumers by offering a large library of games almost from the
beginning and selling the console at a very competitive price.129
Currently, video game business models resemble the operating system approach in that com-
panies entering the market make significant efforts to attract independent game developers. In
the process of promoting the Xbox console, for example, Microsoft announced two programs, the
Independent Developer Program and the Incubator Program, to encourage smaller developers by
providing free software tools and waiving normal pre-publishing requirements.130 Furthermore,
Microsoft had extensive meetings with developers before the hardware specs for the console were
set and incorporated many of their suggestions into the final design.131 Microsoft also made it
easier for developers with PC experience to develop games for the Xbox by relying on DirectX (a
collection of APIs that serves as the foundation for most PC games)132 in the design of the console.
The video game history illustrates how seemingly similar two-sided markets can arrive at differ-
ent pricing structures. It is possible that slight differences in technology and demand lead to differ-
ent pricing structures for internalizing externalities. There is another possibility though. Coming
up with the right pricing structure is a difficult problem that requires more information than set-
ting the price of toothpaste. It may take time and experimentation for industries to converge on
the optimal structure.
126
Sony’s licensing structure was based on a $10-per-game arrangement that included manufacturing disks, manuals,
and packaging. Nintendo’s game consoles were based on cartridges, which were much more expensive for game
developers than CDs. At the time, it cost more than $20 to manufacture an 8-megabyte cartridge, but it cost less
than $2 to press a 640-megabyte CD. Id. at 511.
127
Id. at 504.
128
See The Investext Group, supra note 111. See also Morris, supra note 8.
129
See Kent, supra note 105, at 516-520. Sega’s Saturn was launched with a $399 price tag. Sony’s PlayStation was
launched at $299.
130
Microsoft Embraces the Worldwide Independent Video Game Developer Community, Microsoft PressPass, (visited Aug.
2001, 2002) <http://www.microsoft.com/presspass/press/2000/Nov00/XPKPR.asp>.
131
See Becker, supra note 109.
132
See Davis, supra note 9. See also Becker, supra note 109.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 97
more valuable to sellers if more buyers take the tender and more valuable to buyers if more sellers
take the tender. Governments have helped solve the chicken-and-egg problem in the case of money
(of the sort that has no intrinsic value) by passing laws that require businesses and people to accept
money for discharging debts.133 More interesting from our standpoint are payment mechanisms in
which businesses have had to figure out ways to get both sides of the market on board.134
Diners Club was the first charge card that many people used to pay at many businesses. It
was introduced in 1950.135 The card (then a slip of cardboard) was distributed for free to well-off
Manhattan residents for the purpose of paying at restaurants. There was no fee. There were also
no charges if the cardholder paid within a certain period of time; that meant the cardholders paid
nothing for the float they received from the time of their meal to the time they paid. Using its
attractive clientele as a lure, Diners Club then persuaded restaurants to accept payment with the
card, charging the restaurants a cut of the tab—about 7 percent.136 In its fourth year of operation,
Diners Club added a membership fee of $34 per year.137
This business model has persisted for the last half century for the pure charge card (no credit,
pay in full within 30 days). Consider the American Express charge card. In exchange for the
rights to accept the American Express card, merchants are typically assessed a fee (known oddly
enough as the “merchant discount”) that is about 2.7 percent of the purchase amount on aver-
age.138 Cardholders pay, on average, a $34 annual fee for an American Express charge and credit
cards.139 The cardholders with credit cards that have no annual fees, such as Blue and Optima,
133
Until 1971, the gold standard system constrained the amount of money that government could issue to its gold
reserves. See Michael D. Bordo, Gold Standard, The Concise Encyclopedia of Economics Online (visited Aug.
19, 2002) <http://www.econlib.org/library/Enc/GoldStandard.html>. Under this system, paper money could be
exchanged for gold, which had intrinsic value. A few years before the gold standard was finally and completely
abandoned, Congress passed the Coinage Act of 1965, stating that, “All coins and currencies of the United States…
shall be legal-tender for all debts, public and private, public charges, taxes, duties and dues.” See Legal Tender:
A Definition, Bureau of Engraving and Printing (visited Aug. 16, 2002) <http://www.bep.treas.gov/document.
cfm/18/110>. In essence, before officially removing the value from the currency, government passed a law requiring
people to accept it for payment of debt. However, this does not mean that merchants are under legal obligation
to accept cash for payment. For example, some businesses do not take pennies and certain merchants do not ac-
cept cash and only allow credit card transactions. Thus, the laws encourage the use of cash generally but did not
specifically mandate merchant acceptance, leaving businesses free to form their own payment guidelines. See FAQs:
Currency, U.S. Treasury (visited Aug. 21, 2002) <http://www.ustreas.gov/education/faq/currency/legal-tender.html>.
134
See generally Evans & Schmalensee, supra note 10.
135
Id. at 62.
136
Id. at 181.
137
A $34 membership fee is reported in 2002 dollars and is equal to $5 charged by Diners Club in 1954. See Peter
Z. Grossman, American Express: The Unofficial History of the People Who Built The Great Financial
Empire 263 (1987).
138
See American Express Company Annual Report 2001 (visited Aug. 15, 2002), <http://www.onlineproxy.com/
amex/2002/ar/pdf/axp_ar_2001.pdf>, at 37.
139
Id.
PLATFORM ECONOMICS:
98 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
benefit from the float.140 Overall, excluding finance charge revenue, American Express earned 82
percent of its revenues from the merchant side of the business in 2001.141
The institutions and pricing mechanisms for getting both sides of the market on board are more
complicated in the case of “cooperative” card systems.142 MasterCard and Visa are associations whose
members consist of banks that provide payment services to individuals (“issuing”) or merchants
(“acquiring”) or both. In the case of individuals, the services include providing a card that can be
used to make payment at merchants and that may provide some long-term credit—the issuing bank
makes arrangements to pay the merchant and then bills the cardholder and may offer to finance the
purchase. In the case of merchants, the services include providing technology for processing card
transactions and paying the merchant. Frequently, the issuing and acquiring banks differ. The associa-
tions provide coordination: they operate networks and accounting systems that authorize and process
transactions and provide the appropriate credits and debits to member accounts.
Getting both sides of the market on board requires that these associations provide members
with the proper incentives to service both sides of the market and that there is a pricing structure
that provides the proper incentives for individuals and merchants to use the cards. To see why,
it is useful to go back to the case of American Express. As a company that both issues cards and
acquires from merchants it can devise a pricing structure consisting of a price to cardholders (pri-
marily an annual card fee) and a price to merchants (merchant discount). In contrast, the issuing
and acquiring members of an association can only determine their own prices (card fees and inter-
est charges in the case of issuing members and merchant discount and related fees in the case of
acquiring members). They cannot determine the relative price of card services to individuals versus
merchants; that must be done centrally.
The associations have done this through setting an “interchange fee.” This is a fee that the
bank that acquires a card transaction from a merchant charges the bank that issued the card to
the individual who made that transaction. Issuing and acquiring banks had to have an agreement
for consummating transactions. Both could take the position that the transaction could not take
place without what they brought to the table (and both would be right). It was not practical for
the thousands of members to negotiate individual agreements.143
The associations could have adopted a rule that said all transactions were exchanged at par—
issuing banks had to fully reimburse acquiring banks. However, that in effect amounts to setting
140
The benefit American Express cardholders receive from the float is estimated at approximately $32. This amount
is calculated using average basic card member spending in 2001, a 28-day repayment grace period and the 2001
average 1-year constant maturity Treasury bill interest rate. Id.
141
Including finance charge revenue in 2001, American Express earned 62 percent of its revenues from merchants. Id, at 35.
142
For a short introduction, see David S. Evans, Payment Card Business: Chickens and Eggs, and Other Conundrums,
Mastering Management Online (2001), in Fin. Times Online (visited Jul. 26, 2002) <http://www.ftmastering.
com/mmo/mmo03_2.htm>.
143
Section V reviews the origin of the interchange fee in the Visa association.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 99
an interchange fee equal to zero and adopting a particular pricing structure. Estimates place the
average Visa and MasterCard interchange fee in 2001 at 1.53 percent.144 Acquiring banks usually
pass that fee on to the merchant as either an explicit or implicit part of the merchant discount.
The resulting pricing structure is one in which the average merchant discount for the card associa-
tions is substantially less than the merchant discount charged on similar cards issued by American
Express and somewhat more than the merchant discount charged by Discover which is another
integrated payment system.145
Payment cards are a complex match-maker market. The payment systems have a two-stage
problem to solve. They have to persuade individuals and merchants to use the same platform—by
getting individuals to carry the card brand and merchants to take the card brand. Many consumers
do not carry American Express cards and many merchants do not accept those cards. They then
need to persuade individuals to use the card to make purchases rather than competing payment
devices and they need to persuade merchant to take the card for those purchases rather than steer-
ing consumers to other payment devices. Many consumers carry American Express cards but pay
with a Visa cards or with cash for numerous transactions. Some merchants take American Express
cards but discourage their use when the consumer has a payment alternative that is cheaper for the
merchant. In effect, the systems want a lot of people to join the club and to go out on a lot of dates.
The pricing structure, and investment strategy, seems to have arisen to solve these two problems:
getting people on board and getting those on board together a lot.
Despite these differences with other two-sided markets there are many similarities. The pay-
ment systems have two groups of customers. The customers engage in transactions (the use of cards
as a means of exchange) on a platform (the card systems). There are externalities—merchants value
more cardholders; cardholders value more merchants. And there is a pricing structure that seems
to skew prices towards one side—in this case the merchants.
144
The Nilson Report No. 758 (Feb. 2002), at 4.
145
See United States v. Visa, 163 F. Supp. 2d 322, 333 (S.D.N.Y. 2001).
PLATFORM ECONOMICS:
100 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
markets usually analyzed by economists. Neither the usual price equals marginal cost condition
found in perfectly competitive markets nor the usual marginal revenue equals marginal cost con-
dition found in imperfectly competitive markets applies. Fundamentally, simple pricing and cost
relationships cannot apply to either side because the transaction that takes place on the platform
jointly benefits both parties and often entails joint costs.
Part C examines how the unique economics of two-sided markets affects the business models
used by firms in these markets. It examines strategies for getting both sides on board, balancing the
competing interests of both sides, and how these strategies are affected by the extent of competi-
tion among two-sided firms operating in the same market.
Part D concludes with a brief discussion of how the pricing strategies used in two-sided mar-
kets deviate from what a benevolent social planner, trying to maximize social welfare, would like
to see. That will set up the discussion for the following section that explores the antitrust analysis
of two-sided markets.
146
One can take a communications network and think of there being two sides (senders and receivers) in which case
one can also think of the network effects as indirect. What distinguishes two-sided from one-sided markets, and
indirect from direct network effects, is whether one can identify distinct groups of users with different preferences.
There is an argument that almost all network effects in reality are indirect effects and that almost all network mar-
kets are two-sided.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 101
more if more customers have the card. Applications developers value an operating system more if
more prospective applications buyers use it. Purchasers of video game consoles value them more
if there are more users of these consoles because that will translate into more games. In all these
cases, each side values having the other side on board and benefits the more customers there are
on the other side.
147
Jeff Rohlfs, A Theory of Interdependent Demand for a Communications Service, 5 Bell J. Econ. 16 (1974); William
J. Baumol et al., Contestable Markets and The Theory Of Industry Structure (1982); John C. Panzar,
Technological Determinants of Firms and Industry Structure, in Handbook of Industrial Organization Vol. I 3
(Richard Schmalensee & Robert D. Willig eds., 1989).
148
Augustin A. Cournot, Researches Into The Mathematical Principles of the Theory of Wealth, (Nathaniel
T. Bacon trans., Macmillan 1897).
149
The classic example of this involves monopoly production of A and B. A merger of two monopolists in these com-
plementary products would result in the merged firm charging lower prices for at least one of the products com-
pared to the separate firm. See generally Thibaud Verge, Multiproduct Monopolist and Full-Line Forcing: The Efficiency
Argument Revisited, 12 Econ. Bull. 1 (2001).
PLATFORM ECONOMICS:
102 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
These standard explanations for why firms produce multiple products probably apply to many
of the firms discussed in the previous section. But there is a further reason why firms sell multiple
products in two-sided markets. Doing so helps them increase the indirect network effects discussed
above. Most firms in one-sided markets do not have to offer product A in order to offer product
B. GM could sell cars without trucks and Gillette could sell razor blades without selling razors.
Many firms in two-sided markets, however, have to produce multiple products in order to sell any
product at all. Diners Club could not have come into existence without providing different services
to both cardholders and merchants. Selling operating systems and video game consoles requires
firms in these industries to produce products for both users and developers. For others, producing
multiple products is important but perhaps not essential. There are companies that specialize in
sending advertising messages (direct mail, bill boards) and companies that specialize in just selling
content to consumers (the leading German newspaper Frankfurter Allgemaine Zeitung has no ads,
and Consumer Reports declines advertiser support).
150
I discuss what I mean by “balance the interests” below.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 103
quantity resulting from a one percent change in price (times minus one so it is always positive).151
If it were possible to identify separate groups of consumers with different demand the firm could
engage in price discrimination in which case it would charge different prices to these different
groups (where each price would be at the level at which the marginal revenue for that group equals
marginal cost).
Rochet and Tirole examine the pricing and production strategy of a firm in a two-sided market
that is motivated by payment cards but can be easily generalized to many two-sided match-maker
markets.152 They consider the case in which both sides of the market are buying a “transaction”
and in which the seller incurs a marginal cost of cT for consummating that transaction. The prices
charged to buyers and sellers respectively are two prices, pTB and pTS . The buyer’s demand DTB ( pTB )
depends only on the price faced by the buyer and the seller’s demand DTS ( pTS ) depends only on
the price faced by the seller. The demands can be thought of, roughly speaking, as the number of
buyers and sellers using the system. The transactions that a seller engages in, and its benefits from
those transactions, increase proportionally with the number of buyers on the system. The same
holds for an individual buyer.153
Total demand equals the product of the two demands: qT 5 DTB ( pTB ) DTS ( pTS ). Thus, if there
were 500 sellers and 100 buyers, there would be 50,000 transactions. This assumption mimics
the situation in credit cards where the number of card transactions depends on the interactions
between those who hold cards and those who take cards. It also captures the situation in which the
number of video games that are played depends on the interactions between those who play games
on a particular type of video game console and those who make games for that console. The two-
sided monopoly’s profits are: pT 5 ( pTB 1 pTS 2 cT ) 3 qT .
Although this model is special, the results described below are likely to hold generally: the firm
has to choose a pricing structure in addition to a pricing level and profit-maximization does not
result in the equilibration of marginal revenue and marginal cost in either market taken by itself.
The assumption of multiplicative demand between the two sides understates the importance of the
indirect network effects. It ignores the fact that the value each side obtains from the other side in-
creases with the number of customers on the other side. The results below would likely be stronger
if this feature were taken into account.154
151
See Carlton & Perloff, supra note 6, at 91–92. (The elasticity is sometimes defined without multiplying by mi-
nus one.)
152
See Rochet & Tirole, supra note 4.
153
Rochet and Tirole assume, for simplicity, that there are no fixed costs of joining the system. If there were fixed costs,
a buyer would have to weigh the overall gain from joining, which would depend on the number of sellers, against
the fixed costs, so that the number of buyers joining would depend on the number of sellers. As a result, the decision
of an individual seller to join the system does not depend on the number of buyers.
154
The particular model is used because it leads to tractable results that can be used to analyze the welfare consequences
of two-sided pricing.
PLATFORM ECONOMICS:
104 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
A key difference between the one-sided and two-sided firms is that, as Rochet and Tirole show,
the two-sided monopolist must choose a pricing level (what total price to charge to buyers and sell-
ers) and a pricing structure (how to divide the total price between buyers and sellers). The p ricing
( pTB 1 pTS ) 2 cT 1
level condition is: T 5 B . This condition is a variant of the Lerner condition
( pTB 1 pTS ) 1 S
above. Here however, analog of the price term in the one-sided Lerner condition is the sum of the
prices to both sides and the analog of the demand elasticity term is the sum of the demand elastici-
ties on both sides.
The optimal pricing structure depends on the following condition where the D′ refers
to the change in demand with respect to a change in the price on that side of the market:
( DTB )′DTS 5 DTB′( DTS ). Consider the left-hand side of this equality. A slight increase in price on the
seller’s side will decrease the seller side demand slightly, by ( DTB )′. The impact on total demand
(and profits) is the product of the seller side effect times the buyer side demand. The right hand
side is the analogous effect of increasing the buyer side price slightly. In equilibrium, the effect
on profits must be the same from increasing the seller side price versus the buyer side price. That
pB p S
implies TB 5 TS , where the ’s are the elasticities of demand for each side of the market.
This is very different from the Lerner condition, shown above, in one-sided markets. Moreover,
none of the conditions for determining the price level or the price structure in two-sided markets
corresponds to marginal revenue equaling marginal cost on either side of the market. In fact, such
conditions have no meaning in two-sided markets because there is no way to allocate the increases
in revenues from changes in prices to one side or the other. Changes in prices result in more “trans-
actions” from which each side jointly benefits. Nor is there any way to allocate the costs. Often
costs are jointly incurred for both parties to a transaction and we have the usual issue that any al-
location of cost is arbitrary.155
To try to understand the pricing structure condition intuitively, consider the following.
Suppose, for example, the seller side demand ( DTS ) is five times larger in equilibrium than the
buyer side ( DTB )—say, 500 to 100, with total output of 50,000. Since total demand is the product
of the demands on the two sides, a change in the buyer side demand has five times the effect on to-
tal demand as the same sized change in seller side demand. If the buyer side decreases by one unit,
to 99, then total output falls by 500 units. But if the seller side decreases by one unit, total output
falls by only 100 units. At equilibrium, the change in profits coming from a one-cent increase in
the seller price must be the same as from the same increase in the buyer price. The monopolist
therefore wants to set prices so that the change in buyer side demand from a marginal price change
is only 1/5 of the change in seller side demand from a marginal price change because the effect of
a buyer side change is multiplied against the much larger demand on the seller side.
155
See supra note 29.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 105
1. Getting both sides on board. There are many references in the literature on the firms
discussed earlier about solving the chicken-and-egg problem.157 For example, there would
be no demand by households for payment cards if they could not use them anywhere and
no demand by retailers for payment cards if no one had them. Which comes first—the
cardholder or the retailer?158 Investment and pricing strategies are key to getting both sides
on board.
2. Balancing interests. Even with both sides on board, businesses have to carefully balance
their two demands. They always have to consider how changing prices on one side of the
market will impact the other side of the market. Businesses sometimes differ over this. For
example, in the battle between Microsoft and Netscape over Internet browsers, Microsoft
156
“Multihomed” was originally an Internet term. According to Webopedia, an online technical dictionary, it is “an
adjective used to describe a host that is connected to two or more networks or has two or more network addresses.
For example, a network server may be connected to a serial line and a LAN or to multiple LANs.” For definition, see
Webopedia Online Dictionary (visited Aug. 13, 2002) <http://www.pcwebopaedia.com/TERM/m/multihomed.
html>. Rochet and Tirole adapt the term to describe two-sided networks where a fraction of end-users on one or
more sides connect to multiple platforms. See Rochet & Tirole, supra note 4, at 5.
157
See, e.g., Gawer & Cusumano, supra note 28, at 150-151, 60-61.
158
See Evans & Schmalensee, supra note 10, at 137-168.
PLATFORM ECONOMICS:
106 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
gave away developer kits to Internet portals while Netscape charged for them.159 Political
tensions can also manifest themselves—looking out for their narrow interests customers
on each side of the market would like the other side to pay more. This is a familiar prob-
lem in the payment-card industry—in Europe a retailers association asked the European
Commission to force the card associations to eliminate the interchange fee.160
3. Multihoming. Firms sometimes compete to become the dominant two-sided provider. A
monopoly provider can emerge if consumers are sufficiently homogeneous (so that one firm
can provide them all), the indirect network effects are strong enough, and there are suffi-
ciently large scale economies in production. But even in these circumstances, at the start of
an industry it is common to see several firms in a race to become the dominant firm—they
are likely to engage in multihoming strategies. If some cases it is possible that several two-
sided firms could co-exist and compete with each other—this has clearly been the case in
payment cards where there have been multiple providers for almost half a century. In this
case, multihoming has implications for how they compete with each other.
159
Cameron D. Myhrvold, Microsoft’s Efforts to Create Alliances with ISPs, Microsoft PressPass, Jan. 6, 1999, (visited
Aug. 21, 2002) <http://www.microsoft.com/presspass/trial/mswitness/myhrvold/myhrvold_pt2.asp>.
160
See Retail Disappointment as Visa Exempted from EU Competition Rules, British Retail Consortium Gruerrera,
Brussels Set to Let Visa off the Hook, Fin. Times, May 13, 2002; James MacKintosh, Watchdog Rules Credit Card Fees
to Retailers Excessive, Fin. Times, Sept. 26, 2001.
161
Wylie Wong, Netscape Applauds Microsoft Suit, Techweb, May 20, 1998 (visited Aug. 21, 2002) <http://www.
techweb.com/wire/story/msftdoj/TWB19980519S0007>.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 107
suit.161 Microsoft is reportedly subsidizing the sales of its X-box hardware to consumers to get
them on board.162
Another way to solve the chicken-and-egg problem is to invest in one side of the market to
lower the costs to consumers on that side of participating in the market. Microsoft provides a
good example of this. As we saw earlier it invests in applications developers by developing tools
that help them write applications and providing other assistance that makes it easier for devel-
opers to write applications using Microsoft operating systems. To take another example, bond
dealers take positions in their personal accounts for certain bonds they trade. They do this when
the bond is thinly traded and the long time delays between buys and sells would hinder the mar-
ket’s pricing and/or liquidity. By investing in this manner, two-sided intermediaries are able to
cultivate (or even initially supply) one side, or both sides, of their market in order to boost the
overall success of the platform.
Providing low prices or transfers to one side of the market helps the platform solve the
chicken-and-egg problem by encouraging the benefited group’s participation––which in turn,
due to network effects, encourages the non-benefited group’s participation. Bernard Caillaud
and Bruno Jullien refer to this strategy as “divide-and-conquer.”163 Another effect of providing
benefits to one side is that this assistance can discourage use of competing two-sided firms.
For example, when Palm provides free tools and support to PDA applications software devel-
opers, it encourages those developers to write programs that work on the Palm OS platform,
but it also induces those developers to spend less time writing programs for other operating
systems.164
162
David Becker, Xbox Drags on Microsoft Profit, Cnet.Com, Jan. 18, 2002 (visited Aug. 21, 2002) <http://news.com.
com/2100-1040-818798.html>.
163
See Caillaud & Jullien, supra note 4, at 16. See also Jullien, supra note 4, at 1.
164
See Rochet & Tirole, supra note 4, at 6.
165
If finance charge revenues are included, American Express earned 62 percent of its revenues from merchants in
2001. See American Express Company Annual Report 2001, supra note 138, at 35.
PLATFORM ECONOMICS:
108 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Windows to computer manufacturers or end-users.166 Real estate brokers usually earn most or all
of their revenues from the sellers.
Discerning the optimal pricing structure is one of the challenges of competing in a two-sided
market. Sometimes all the platforms converge on the same pricing strategy. Microsoft, Apple,
IBM, Palm and other operating system companies could have charged higher fees to applications
developers and lower fees to end-users. They all discovered that it made sense to charge developers
relatively modest fees for developer kits and, especially in the case of Microsoft, to give a lot away
for free. Nevertheless, Microsoft is known for putting far more effort into the developer side of the
business than the other operating system companies.167
The debit card is an example in which different platforms made different pricing choices. In
the late 1980s, the ATM networks had a base of cardholders who used their cards to withdraw
cash or obtain other services at ATMs. They had no merchants that took these cards. To add debit
services to existing ATM cards, the ATM networks charged a small interchange fee (8 cents per
transaction on a typical $30 transaction) to encourage merchants to install pin-pads that could
read the ATM cards that cardholders already had and accept the pins they used to access the ATM
machines.168 Many merchants invested in the pin-pads—the number of pin-pads increased from
53,000 in 1990 to about 3.6 million in 2001.169 The credit-card systems had a base of merchants
who took their cards but it did not have cards that, like the ATM cards, accessed consumers’
checking accounts. The credit-card systems imposed a much higher interchange fee than the ATM
networks, about 38 cents versus 8 cents on a typical $30 transaction.170 They did this to persuade
banks to issue debit cards and cardholders to take these cards.171 The number of Visa debit cards in
circulation increased from 7.6 million in 1990 to about 117 million in 2001.172
166
From 1988 through 2000, Microsoft earned at least 67 percent of its revenues from licensing packaged software
(such as Windows and Office) to end-users (either directly at retail or through manufacturer preinstallation on
PCs). See 1994 Worldwide Software Review and Forecast, IDC REPORT #9358 (Nov. 1994); 1995 Worldwide
Software Review and Forecast, IDC REPORT #10460 (Nov. 1995); 1996 Worldwide Software Review and Forecast,
IDC REPORT #12408 (Nov. 1996); 1997 Worldwide Software Review and Forecast, Idc Report #14327 (Oct.
1997); 1999 Worldwide Software Review and Forecast, Idc Report #20161 (Oct. 1999); Worldwide Software Market
Forecast Summary, 2001-2005, Idc Report #25569 (Sept. 2001). Note that the 67 percent figure underestimates
the amount of revenue Microsoft earns from end-users because the other third of revenue coming from “Applications
Development and Deployment” includes some end-user revenues as well. For example, database products used by
business IT departments are included in the Applications Development category.
167
See Gawer & Cusumano, supra note 28, at 150-151.
168
See Evans & Schmalensee, supra note 10, at 300.
169
Id. at 308-309; and The Nilson Report No. 759 (March 2002), at 6.
170
The ATM systems typically charged a flat interchange fee per transaction, while the interchange fee set by Visa and
MasterCard varied with the size of the transaction. The reported interchange fee comparison is from 1998, around
the time of substantial growth in debit for the ATM and credit-card systems. Id. at 300.
171
Visa attracted consumers through an effective advertising campaign and attracted issuers through heavy investment
in a debit processing facility, among other strategies. Id. at chapter 12.
172
See The Nilson Report No. 760 (March 2002), at 7; The Nilson Report No. 500 (May 1991), at 6.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 109
Two other factors influence the pricing structure. There may be certain customers on one side
of the market—Rochet and Tirole refer to them as “marquee buyers”—that are extremely valuable
to customers on the other side of the market. The existence of marquee buyers tends to reduce the
price to all buyers and increases it to sellers. A similar phenomenon occurs when certain customers
are extremely loyal to the two-sided firm––perhaps because of long-term contracts or sunk-cost
investments. For example, American Express has been able to charge a relatively high merchant
discount as compared to other card brands, especially for their corporate card, because merchants
viewed the American Express business clientele as extremely attractive. Corporate expense clients
were “marquee” customers that allowed American Express to raise its prices to the other side of the
market, merchants. In the online debit-card market, however, card issuers faced “captive” custom-
ers––ATM cards could be used as online debit cards, so consumers did not need to be courted
to accept the new payment form. Therefore, it has been the merchants––who must purchase and
install expensive machinery in order to process online debit transactions––who have been courted,
as we saw above.
3. Multihoming
Most two-sided markets we observe in the real world appear to have several competing two-sided
firms and at least one side appears to multihome. Table 2 in Appendix A presents a summary.
Consider, for example, personal computers. One could consider the two sides as consisting of
personal computer end-users and as developers of applications. The end-users do not multihome.
They almost always use a single operating system and by far the preponderance of them use a
Microsoft operating system.173 The developers do multihome. According to Josh Lerner, in 2000,
68 percent of software firms developed software for Windows operating systems, 19 percent for
Apple computers operating systems, 48 percent for Unix operating systems including Linux, and
36 percent and 34 percent for proprietary non-Unix operating systems that run on minicomputers
and proprietary operating systems that run on mainframes respectively.174 In fact, in recent years
the percentage of software firms developing for non-Microsoft operating systems has increased.
The fastest-growing category has been software firms developing for Unix operating systems in-
cluding Linux. The percentage of developers in this category increased from 29 percent in 1998 to
48 percent in 2000.175
173
Worldwide Operating Environments Market Forecast and Analysis Summary, 2001-2005, IDC Report #25119 (Aug.
2001), at 12.
174
See Josh Lerner, Did Microsoft Deter Software Innovation? Working Paper (Jan. 2002) (downloaded Aug. 15, 2002)
<http://gsbwww.uchicago.edu/research/workshops/elo/lerner2.pdf>; and Corporate Technology Directory,
Editions 1990-2000. The percentages total to 205, indicating substantial multihoming on the part of developers.
175
Id.
PLATFORM ECONOMICS:
110 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Multihoming affects both the price level and the pricing structure. Not surprisingly the price
level tends to be lower with multihoming—the availability of substitutes tends to put pressure on
the two-sided firms to lower their prices. The seller has more options dealing with a multihomed
buyer on the other side and can steer toward its preferred platform. As buyer multihoming becomes
more prevalent, prices to sellers will tend to decrease since they have more substitution options.
Even when multihoming is not prevalent on one side of a two-sided market the possibility
of multihoming may have significant consequences for pricing. The possibility of multihoming
may encourage firms to lower their prices on the side of the market in which multihoming could
occur. By lowering their prices they discourage customers on that side from affiliating with other
two-sided firms. This is not entirely a free lunch for consumers. The firm can then charge more to
customers on the other side, for whom fewer substitutes are available.176
1. The extent to which firms are competing to become established in a two-sided market.
Firms tend to compete to establish customer bases on both sides of the market. This results
in investments to court customers, to provide them with subsidies in the form of equip-
ment, and to offer them low or negative prices.177 If the competition is sufficiently intense
176
In Jullien’s model, when multiple platforms compete and price discrimination between the two customer types is
possible, then prices are lower overall: “This forces the established firm to set on average prices at a much lower level
than it would do with uniform prices. It turns out that it is impossible for a network to capture in equilibrium the
surplus generated by the inter-group network externalities.” Jullien assumes the incumbent initially offers uniform
prices because in his model the two customer types have identical valuations for the network goods and both receive
the same extra value if they both join the same network. See Jullien, supra note 4, at 4.
177
Losing money initially to buy penetration can also be an important phenomenon in one-sided networks.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 111
the losses incurred during the “getting both sides on board” stage may significantly offset
the profits earned during the mature phase of the industry. For example, firms that have
entered the payment-card industry have all incurred sizeable losses during their startup
phases.178
2. The extent to which there are first-mover advantages in getting either side of the market on
board. If any firm can make investments to get a side of the market on board then competi-
tion to make these investments should reduce the opportunities to earn significant supra-
competitive returns. Supra-competitive profits could arise if one firm has an advantage that
other firms cannot replicate. For example, the network economics literature argues that the
first mover in an industry always has an advantage over the subsequent movers.179 Since
customers value other customers whoever gets customers first naturally wins.180 It could
also be that a firm has some asset that gives it a hard start. For example, a firm may have
already developed a product that gives it a large customer base on one side of a market.
When the demand for a two-sided version of this (or a related) product appears it will have
a significant head start over rivals. Online debit is an example. Banks issued ATM cards to
customers for use at its ATM machines. The ATM networks they belonged to eventually
realized that these same ATM cards could be used to pay at merchants. The fact that they
had the cardholder side on board gave them a significant advantage.
3. Even markets that appear to be dominated by a single player may in reality be contestable.
Jullien’s “model suggests that it may be easier than expected for a superior technology to
enter, provided that the quality improvement is large enough.”181 Because many of the
two-sided markets are fast moving, current leaders often face considerable competition in
the form of potential entrants––other platforms striving to displace today’s leader. Caillaud
and Jullien argue that the Internet represents one such environment: “Too many ways of
stealing the competitors’ business appear. Unsurprisingly, the strategic situation is very un-
stable and the only equilibrium situation that is tenable is for a firm to exert dominance on
the intermediation market, i.e. to be the sole supplier of intermediation services, without
enjoying any market power as potential entrants create a strong disciplinary device for the
dominant firm. In some sense, this market is extremely contestable.”182
178
See Evans & Schmalensee, supra note 10, at ch. 4.
179
See, e.g., Carl Shapiro & Hal R. Varian, Information Rules 29-32 (1999).
180
There is little empirical support, however, for this view. See Evans & Schmalensee, supra note 7. See also Stan J.
Liebowitz & Stephen E. Margolis, Winners, Losers & Microsoft: Competition and Antitrust In High
Technology (1999).
181
See Jullien, supra note 27, at 34.
182
See Caillaud & Jullien, supra note 4, at 40. The authors are speaking of Internet intermediaries, but the point holds
for other fast-moving dynamic markets.
PLATFORM ECONOMICS:
112 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
4. Two-sided markets in which non-profit associations determine the pricing structure are not
likely to permit the participants to earn supracompetive profits. Payment-card associations
have put an effectively non-profit institution in charge of managing a physical network for
members and for determining pricing policies. Pricing levels are determined by competi-
tion among members of the association. For example, in the United States payment-card
industry, thousands of banks compete for cardholders; although a small number of firms
compete for merchant services this business is considered to be highly competitive also.
Interest rates, card fees, merchant charges, and so forth are determined through intense
competition. Pricing structures are affected by the non-profit institution. In the case of
payment cards, the interchange fee affects the relative price to cardholders and merchants.
The consequences of having relatively few competitors in two-sided markets, and the exist-
ence of network effects, raise familiar issues concerning the efficacy of competitive markets and
the possible roles for government intervention. The pricing and investment strategies that firms in
two-sided markets use to “get both sides on board” and “balance the interests of both sides” raise
novel ones. These pricing and other business strategies are needed to solve a fundamental economic
problem arising from the interdependency of demand on both sides of the market. In some cases,
the product could not even exist without efforts to subsidize one side of the market or the other.
Rochet and Tirole, in an admittedly simplified setting, have compared the pricing structure
adopted by firms in two-sided markets to the pricing structure that would maximize social welfare.
Interestingly, they find that a monopoly firm, a firm with competition, and a benevolent social
planner would adopt similar pricing structures. The precise relative prices would differ some-
what.183 However, Rochet and Tirole find that the pricing structure adopted by the market (mo-
nopoly or two-firm oligopoly) is not biased towards one side of the market or the other side of the
market compared to the pricing structure that would be adopted by a benevolent social planner.184
(Schmalensee finds similar results for interchange fees.) 185 Therefore, there is no reason to believe
that the preferences that one side or the other of the market have gotten, as summarized in Table
1, are systematically different from what a social planner would seek to achieve.
sides of the market means that there is no meaningful economic relationship between benefits and
costs on either side of the market considered by itself. It takes two to tango. Second, one cannot
talk about the individual prices in isolation. Any change in demand or cost on either side of the
market will necessarily affect both prices along with the sum of those prices. Third, products in
two-sided markets cannot come into existence and cannot remain in existence unless firms in those
markets get “both sides on board.” This gives rise to pricing and investment strategies that differ
from those taken in one-sided markets and seem odd unless considered in the context of competi-
tion in a two-sided market. Fourth, any analysis of social welfare must account for the pricing level,
the pricing structure, and the feasible alternatives for getting both sides on board. It must also ac-
count for the extent to which not-for-profit institutions manage those aspects of the network that
could give rise to supra-competitive profits.
These differences matter for antitrust analysis. To see the principles consider the following hypo-
thetical merger. There are two chains of dating clubs in Tokyo—AAA Mates and Best Match (clubs
A and B, respectively). They cater to somewhat different clienteles. Club A charges men $20 for
admission and women $0; Club B charges men $30 for admission and gives women a $5 credit (in
the form of free drinks). Club B has a 40 percent market share while Club A has a 10 percent mar-
ket share. Club B has been more successful because it attracts more women and as a result of that it
attracts more men. In fact, it is so successful that—like an “in” discotheque—it typically has a line
and can select the men and women to admit. It tries to weed out “undesirable” men and women.
In analyzing whether this merger would be bad for consumers, a competition regulator would
have to take several factors into account. Let us start by considering market definition. A prelimi-
nary issue is whether dating clubs is the relevant antitrust market in which this merger should be
evaluated. Under the U.S. merger guidelines, it would be a market if a merger of all the dating
clubs in Tokyo (assume Tokyo is the relevant geographic market) would enable the combined firm
to raise price by a small but significant non-transitory amount—let us say 10 percent. For analyz-
ing that question one should look at the “total price” charged by the club—although the average
price across all patrons of the club would provide the same information. There is no reason to focus
on the individual prices for men and women (except perhaps as they say something about product
differentiation among the clubs). Different competitors may choose different pricing structures
depending on the mix of men and women they want to attract. They could easily adjust these pric-
ing structures in response to a change in the market.
If dating clubs is a relevant antitrust market, the next issue is whether a merger of clubs A and B
would result in prices increasing. One cannot answer that question by looking just at the demand
for patrons overall—e.g. by estimating the demand for admission against the average price. The
mix of men and women is critically important. One would have to estimate the demand for men
and the demand for women simultaneously. Then, using the theory of pricing in two-sided mar-
kets considered earlier together with information on cost, one could predict whether the merger
would lead the combined firms to increase their total price.
PLATFORM ECONOMICS:
114 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Let us suppose that the competition authority has completed this exercise and found that the
merged dating clubs would charge $32 for men and give women a credit of $6. Assuming equal
numbers of men and women, the average price charged at Club A would rise from $10 to $13 and
the average price charged at Club B would rise from $12.50 to $13. Should the competition au-
thority ban the merger based on this price increase? From the standpoint of consumers the answer
is unclear. At this point in a merger investigation the companies could come back and argue that
there are efficiencies from the merger and that the reductions in cost will pull prices back down.
There are no such savings here. But there is something else. Club B may have efficiencies that
enable it to provide a higher value to consumers on the two sides of the market. On average the
customers pay more but the men may have a better selection of women to choose from and the
women may have a better selection of men to choose from.186 In effect, there are demand-side effi-
ciencies that result from increasing the indirect network effects realized by both sides of the market.
The remainder of this section examines how the economics of two-sided markets affects several
key areas of antitrust analysis. Part A considers market definition. It focuses on how the merger
guidelines approach should be modified for mergers of firms in two-sided markets. Part B examines
the analysis of market power—the ability of firms to hold prices significantly above marginal cost.
It considers the implications of the breakdown of the usual price equals marginal cost condition
in two-sided markets. Part C considers the analysis of barriers to entry in two-sided markets—this
is an important topic that comes up in both market definition and market power. Part D looks at
efficiency considerations for two-sided markets; as part of this exercise it examines how the two-
sided nature of markets should affect the classification of practices as per se versus rule of reason.
Part E examines the implication of getting and keeping both sides on board in two-sided markets
for the analysis of predation. Part F looks at a variety of other issues.
186
Merger analysis in one-sided markets faces similar problems. Sometimes mergers permit the parties to create new
products. The value of these new products should be considered as part of the efficiency analysis.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 115
graphic or product dimension) when the collection of firms could, acting as a monopolist, raise
price by a small but significant non-transitory amount (often taken to be 5-10 percent). If the col-
lection of firms could do so, then presumably the firms “outside of the market” do not constrain
the firms “inside the market” much.
For two-sided markets, this analysis must pay attention to both sides. It should consider firms
that currently supply both sets of customers. Here the main controversy is likely to be over whether
the two-sided firms offer sufficiently differentiated products that they should not be considered in
the same market. Dating Club B might be the uptown/yuppie club and Dating Club A might be
the downtown/blue collar club. To take another example, one of the issues in litigation concerning
credit cards is whether other two-sided products such debit cards, checks, and cash compete with
credit cards. The market analysis should also consider the possibility that other firms could enter.
Dating Club A may be able to enter the uptown/yuppie market rather easily. During the early years
of Palm PDAs it would have been reasonable to argue that Microsoft could easily enter this oper-
ating system category—in fact it eventually did so by producing and licensing a small operating
system to a PDA manufactures.187 Of course, in considering the possibility of entry it is necessary
to examine whether and to what extent there are barriers to entry, a topic we consider below.
The right question to ask, for the merger guidelines price test, is whether the firms under con-
sideration, if they merged today and priced as a monopolist, could raise the total price by 10 per-
cent or more. There is no particular reason to focus on price effects on one side or the other—both
groups of consumers matter. That is especially true in matching markets in which both sides must
buy for there even to be a product. Moreover, competition authorities should not consider a price
increase of 10 percent to one side of a two-sided market (in which the other side has no increase)
as seriously as a price increase of 10 percent price to a one-sided market. Likewise, they should not
ignore any countervailing price decreases that occur on the other side of the market.
Generally, it is not possible to examine price effects on one side of a market without consider-
ing the effect on the other side and the feedback effects between them.188 For example, holding the
female side of the market constant (price and quantity), perhaps the merger of dating clubs would
result in an increase in the price to the male side of the market by 10 percent. But in reality it is
not possible to hold the female side of the market constant. A 10 percent price increase to men
would reduce the number of male patrons and therefore reduce the value of the service to women.
Once these interdependencies are taken into account it could be that the merged firms could raise
combined prices by only a trivial amount.
187
See Microsoft Corp. website (visited Aug. 20, 2002) <http://www.microsoft.com/windows/embedded/>.
188
Ignoring interdependencies between the two sides is a common mistake in the antitrust analysis of two-sided markets.
For example, this mistake was made in the analysis of class certification issues in Visa Check/MasterMoney. See In re Visa
Check/MasterMoney Antitrust Litig.; Wal-Mart, et al. v. Visa U.S.A. Inc. and MasterCard International, Inc., 280
F.3d 124 (2d Cir. 2001). The plaintiffs’ expert claimed that it was possible to change a significant price on one side
of the market (debit-card interchange fees) without any other changes in the two-sided market. For more details, see
David S. Evans, Class Certification, the Merits, and Expert Evidence, Geo. Mason L. Rev. (forthcoming).
PLATFORM ECONOMICS:
116 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
The Justice Department’s approach in U.S. v. Visa U.S.A. et al.189 illustrates the mistake.
MasterCard and Visa service cardholders and merchants. The DOJ’s economic expert asked the ques-
tion of whether a hypothetical merger of all credit and charge card issuers could profitably raise prices
to cardholders, looking only at profits on the issuer/cardholder side.190 He thus failed to consider two
important factors. First, any decrease in cardholder volume would necessarily and directly lead to a
decrease in merchant volume. And if merchant volume decreases, then any profits on the merchant
side also decrease. DOJ’s economist did not consider effects on profits on the merchant side. And
second, a decrease in the cardholder base makes the system less attractive for merchants, thus poten-
tially leading to a decrease in merchant demand for the system. (Which could then lead to a decrease
in cardholder demand, and so on.) These changes, not accounted for by the government’s analysis,
would affect profits on both the issuing and acquiring sides. By focusing only on the cardholder side,
the analysis put forward by the government’s economist neglected at least half of the story.
189
See United States v. Visa, 163 F. Supp. 2d 322 (S.D.N.Y. 2001).
190
Id. at 336.
191
See Franklin M. Fisher & John J. McGowan, On the Misuse of Accounting Rates of Return to Infer Monopoly Profits,
73 Am. Econ. REV. 82 (1983) for the seminal treatment of this issue which turns on the fact that it is difficult to
measure ex post the expected return ex ante.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 117
ing one more CD copy of an existing game is trivial, considerable investment is required to initially
develop that game. Platforms, such as Sony PlayStation, need to price all of the components––the
console, the internally developed games, and the licensing fees for externally developed games––such
that it recovers its R&D expenses at the same time as it maximizes the popularity of the platform.
192
United States v. E.I. du Pont, 351 U.S. 377, 391 (1956).
193
Phillip E. Areeda et al., Antitrust Law, Volume Iia: An Analysis of Antitrust Principles and Their
Application 55-56 (1994). “A barrier to entry is any factor that permits firms already in the market to earn returns
above the competitive level while deterring outsiders from entering” (footnote omitted). This definition follows J.S.
Bain, Barriers To New Competition: Their Character and Consequences in Manufacturing Industries
(1962). For a list of court cases relying on this definition of entry barriers, see Phillip E. Areeda et al., Volume IIA:
An Analysis of Antitrust Principles and Their Application 123-124 (Supp. 1999). In practice, the condition
that the factor must result in the ability of the firm “to earn returns above the competitive level” is dropped in one of
two ways. First, the condition is just ignored. The correct economic concept is whether the expected rate of return to
the firm based on the information available to the market at the time it made the investment exceeds the competitive
level after adjusting for risk. See Fisher & McGowan, supra note 191, at 90-91. It is almost impossible to measure
this ex ante return with ex post information. Second, the competitive rates of return are calculated from ex post data
with no adjustment for the risk perceived ex ante. For example, it is common for analysts to calculate the returns based
on survivors in an industry without adjusting for the losses incurred by failures. The result is that almost all indus-
tries involving substantial investments in research and development or other fixed costs are identified, incorrectly, as
earning a supra-competitive return. See Fisher & McGowan, supra note 191, at 91; and Franklin M. Fisher et al,
Folded, Spindled and Mutilated: Economic Analysis And U.S. V. Ibm ch. 7 (1983). Not surprisingly, Areeda,
Hovenkamp and Solow identify almost any advantage of incumbency as a barrier to entry. They include “economies
of scale, high initial investment, capital market imperfections, risk, low prices, scarce inputs or customers, product
reputation and promotion, and government constraints” as barriers to entry. Areeda et al. (1994), supra, at 63.
PLATFORM ECONOMICS:
118 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Many economists have adopted a different definition of barriers to entry, at least for the pur-
pose of antitrust analysis. They argue that a barrier to entry should refer to an advantage that the
incumbent firm has that an entrant cannot secure.194 The $100 million plant is not an example of
such a barrier because anyone with the money can build one. The incumbent does not have any
special advantage. It may be that the addition of a third plant would drive price so low that it is
not worth building a plant. That, however, is an indicator that it is not socially worthwhile to use
society’s scarce resources to build another plant.195
Which definition is more appropriate? For most antitrust issues we are interested in knowing
whether incumbent firms can earn supra-competitive returns without attracting entry. From soci-
ety’s standpoint we would like to attract entry into various lines of business until the risk-adjusted
return from entry is competed down to the competitive level. The fact that significant investments
are required to enter an industry does not provide any information on this issue. Incumbents could
be earning a risk-adjusted competitive rate of return in an industry as a result of having made sig-
nificant risky investments over time. The fact that it is costly for another firm to enter the industry
is not evidence of a problem. If demand expanded unexpectedly incumbents would temporarily
earn supra-competitive returns as a result of raising prices. Either they could expand capacity by
making more investments or an entrant could do so. With today’s highly liquid capital markets
there is no basis for arguing that firms cannot secure the funding for investments that will yield a
more than competitive rate of return. Consequently, the first definition—an entry barrier means it
is “hard” to enter—does not advance the ball.
Suppose, however, that entrants lack some special advantages that the incumbents had. In
this case the incumbents would be earning a return from those special advantages. For example,
perhaps the incumbent has a patent that raises the cost of entry—maybe the entrant has to invest
in a costly design-around; maybe the entrant cannot produce as good a product; or maybe the
entrant will have higher costs of production. The patent is not necessarily a barrier to entry under
the second definition even though it may enable the incumbent to earn what seems, ex post, to be a
supra-competitive rate of return. It is possible that the entrant could obtain similar advantages for
itself by making a similar risky investment. But the patent may be a barrier—suppose the patent
is not really the result of risky investment but simply the result of a sloppy (but costly to overturn)
decision by the U.S. Patent Office.
To take another example, suppose the incumbent received tax breaks that enabled it to build its
plant for $50 million. If the entrant does not receive those same tax breaks it could never enter the
industry on the same cost terms as the incumbent. In effect, the incumbent would end up receiv-
194
See George Stigler, The Organization of Industry (1968). See also Carlton & Perloff, supra note 6, at
76-77. Unlike the Bain definition one does not need to inquire into whether the “barrier” gives rise to a supra-
competitive rate of return. One only has to ask whether the alleged barrier is something that entrants could obtain
at the same cost as the incumbent.
195
For a survey of court decisions pertaining to barriers to entry, see Werden, supra note 194.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 119
ing a return on its tax breaks. Whether one considers that return to be supra-competitive would
depend on the details of how the tax breaks came to be. A related example involves government
policies or regulations that make it harder for entrants to come in. For many years the Federal
Communications Commission (FCC) made it extremely difficult for firms to compete in long-
distance telecommunications. Only legal challenges forced the FCC ultimately to permit the entry
of MCI.196 The FCC prohibitions were a barrier to entry under the second definition.
U.S. v. Microsoft provides an interesting example since the case involved a two-sided market
(operating systems) and barriers to entry figured prominently.197 The government argued that the
stock of applications that had been written for Microsoft Windows was a barrier to entry that
prevented other firms from competing in the market for operating systems for Intel-compatible
computers.198 Applications developers write to Windows because a large number of end-users, and
end-users use Windows because it has a large number of applications. The district court agreed
with the government.199
This analysis of barriers to entry is not controversial under the first definition. Developing an
operating system that is attractive to end-users and software developers is expensive.200 However,
as mentioned above, the fact that this development is expensive does not really provide any useful
information. First, expensive is a relative concept. The costs of entering the two-sided market for
operating systems would need to be weighed against the profits that could be earned in this mar-
ket. Microsoft’s annual revenues from its desktop operating system products exceeded $8 billion in
2001.201 Second, the fact that it is expensive to enter does not provide any information on whether
entrants are deterred from making investments that could reduce any supra-competitive returns
that Microsoft realizes.
196
Gerald Brock, Telecommunication Policy for the Information Age: From Monopoly to Competition
(1994).
197
See United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001).
198
United States v. Microsoft Corp., Civil Action Nos. 98-1232 and 98-1233 (TPJ), Findings of Fact, Nov. 5, 1999,
at ¶ 36.
199
United States v. Microsoft Corp., Civil Action Nos. 98-1232 and 98-1233 (TPJ), Direct Testimony of Franklin M.
Fisher, Oct. 14, 1998, at ¶ 70.
200
For instance, it took about three years for Linus Torvalds and the open source programmers contributing to Linux
to develop version 1.0 of the Linux Kernel. See What is Linux, Linux Online (visited Mar. 22, 2002) <http://www.
linux.org/info/index.html>. It took more than four years for Microsoft to release version 1.0 of Microsoft Windows.
See Windows Timeline (visited Aug. 15, 2002) <http://www.technicalminded.com/windows_timeline.htm>. It took
more than two years of mutual effort by Microsoft and IBM to develop OS/2. See OS/2 1.0 (visited Aug. 15,
2002) <http://pages.prodigy.net/michaln/history/os210/index.html>. It took approximately four years to develop
LisaDesk, the first Apple operating system based on Alto research of Xerox Corporation. See History of Apple
Macintosh Operating System, (visited Aug. 15, 2002) <http://perso.club-internet.fr/jctrotot/Perso/History.html>.
201
Microsoft Corp. 10-K (FY2001) (downloaded Aug. 30, 2002) <http://www.sec.gov/Archives/edgar/
data/789019/000103221001501099/0001032210-01-501099.txt>.
PLATFORM ECONOMICS:
120 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
The government’s testimony and the district court’s opinion suggest that they were analyzing
barriers under the more stringent but more relevant second definition.202 They argued that
Microsoft had special advantages over rivals because it was the first entrant into the market for
operating systems for Intel-compatible computers.203 Applications developers wrote for Microsoft’s
operating system because it was the only alternative. Microsoft did not have to expend resources to
persuade application developers to write for its operating system versus other alternatives.
Let us consider this analysis through the lens of two-sided markets. Participation in a two-sided
market naturally requires the development of customers on both sides. The fact that Microsoft
has attracted a large number of applications is no more or less remarkable than the fact that it has
attracted a larger number of users. It is also no more remarkable than the fact that Sony has at-
tracted a large number of game developers, and that ATM networks have a large number of ATM
machines that connect with their networks throughout the country.
Indeed, the fact that a firm in a two-sided market has developed customers on one side is, from
society’s standpoint, a good thing. Not only do these customers benefit directly from the services
they obtain—they are inputs into the provision of services to customers on the other side. The fact
that Microsoft has a large stock of applications reflects the fact that applications developers are
receiving useful services from the features included in Microsoft’s operating system. It also reflects
considerable value to end-users who, as the government and district court correctly observed, value
these applications. There is no basis for suggesting that it is a bad thing for Microsoft to invest
resources in getting more applications written for Windows, anymore than it is a bad thing for
Nintendo to try to get more games written for its console, for Discover to try to get more mer-
chants to take its card, or for Goldman Sachs to try to find more companies that want to go on
the block.
By themselves positive feedback effects do not prevent entry. Indeed, many two-sided markets
have multiple providers. In the case of payment cards, there was successive entry by Diners Club
(1950), American Express (1958), Visa (1966), MasterCard (1966), and Discover (1986).204 In
the case of video games in the US, there was successive entry by Magnavox (1972), Atari (1975),
Coleco (1976), Fairchild (1976), Mattel (1979), Nintendo (1985), Sega (1989), Sony (1995), and
Microsoft (2001).205
It could be that Microsoft has special advantages that constitute a barrier to entry under the
second definition. But neither the government nor the district court put forward an analysis that
could distinguish these special advantages, if any, from the investments that Microsoft made in
getting and keeping both sides on board. The relevant question to ask concerning barriers to
202
See Findings of Fact, supra note 198, at 13-16.
203
Id, at 16. (“Microsoft never confronted a highly penetrated market dominated by a single competitor”).
204
See Evans & Schmalensee, supra note 10, at 10, 62-66.
205
See KENT, supra note 105, at xi-xvi.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 121
e ntry into a two-sided market is whether the incumbent has special advantages that could not be
replicated by entrants. That question needs to focus on the costs of getting into both sides of the
market. Saying that Microsoft has a lot of applications written for its operating system is no more
probative than saying that Dating Club B has a lot of women lined up at the door or that American
Express has many companies signed up for its corporate charge cards. Getting customers on both
sides is the name of the game in two-sided markets.
One way to think of this is to consider two operating system companies. Entrant 1 comes in
before Entrant 2. To get both sides on board, the company with the head start, Entrant 1, has to
spend $1 billion to get developers to write applications. If Entrant 2 had to spend $1.1 billion to
get both sides on board we would probably conclude that the entry barrier is fairly modest relative
to the risk-adjusted profits that could be earned in this business. If Entrant 2 had to spend $2 bil-
lion we might reach the opposite conclusion. In both cases we would want to consider these entry
barriers relative to prospective profits.
IBM’s experience in entering the operating system business with OS/2 is instructive. Shortly
after the introduction of Windows 2.0, OS/2 was completed. However, due to its high price
and incompatibility with other existing applications it was deemed a failure.206 If IBM had
made the same investment as Microsoft in getting both sides of the market on board it is not
clear that, with an equal or superior operating system, it would not have succeeded. In any case,
my point is not that there were or were not barriers to entry to the operating system market. It
is that by ignoring the economics of two-sided markets the government failed to conduct the
relevant inquiry.
The government also relied on the applications barrier to entry in its liability theory. It argued
that Netscape and Java would help erode the applications barrier to entry by allowing software
developers to write to a middleware layer that would sit between Windows and the applications.207
The words here matter. The phrase applications barrier to entry is a pejorative term that suggests
that the stock of applications served primarily as an entry barrier. The Government’s chief eco-
nomic witness in U.S. v. Microsoft has written:
That positive feedback effect—the applications barrier to entry—has made it difficult or im-
possible for rival operating systems to compete effectively with Microsoft by gaining more
than a niche in the market. New entry is not likely to erode Microsoft’s market share and
market power as long as the applications barrier to entry remains strong[.]208
206
Martin Campbell-Kelly, Not Only Microsoft: The Maturing of the Personal Computer Software Industry, 1982-1995,
75 BUS. HIST. REV. 103 (2001), available in NEXIS, at 12. See also United States v. Microsoft Corp., 253 F.3d
34, 55 (D.C. Cir. 2001).
207
See Direct Testimony of Franklin M. Fisher, supra note 199, at ¶¶ 36, 95.
208
David S. Evans et al., Did Microsoft Harm Consumers? Two Opposing Views 11 (2000).
PLATFORM ECONOMICS:
122 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
[T]he applications barrier would prevent an aspiring entrant into the relevant market from
drawing a significant number of customers away from a dominant incumbent even if the
incumbent priced its products substantially above competitive levels for a significant period
of time.209
The theory of two-sided markets sees the stock of applications in neutral terms—it is simply
the amount of demand satisfied on the developer side of the market and the value that can be
brought to user side of the market. All two-sided firms want to increase demand on both sides
and increase their advantages on both sides over rivals. Visa has more merchant acceptance than
American Express, Sony has more videogames written for its console than Microsoft, and Time
magazine has more readers than The New Republic.
ing clubs, indirectly in the case of other markets such as operating systems). Some numerical
examples can demonstrate the possible magnitudes of the effects. Using the Rochet-Tirole formu-
lation above, consider a two-sided market with 100 “buyers” and 100 “sellers,” which would lead
to 10,000 (100 3 100) transactions. Let us also assume that buyers (and sellers) have preferences
such that the 100th (or marginal) buyer has a 1 cent benefit per transaction, the 99th has a 2 cent
benefit, and so on, up to the 1st buyer, which has a $1.00 benefit per transaction. It is straightfor-
ward to show that the per-transaction benefit received by the average buyer is 50.5 cents.
The 100th buyer gets relatively little benefit from the 100 transactions it engages in, only
100 3 1 cent 5 $1.00. On the other hand, its decision to join has benefited each of the 100 sell-
ers. The 100th seller only benefits as much as the 100th buyer for that transaction, but everyone
else benefits by much more. The 100 transactions generated by the last buyer produces benefits on
the seller side averaging 50.5 cents per transaction, for a total benefit of $5.05. That is, the indirect
benefits to sellers are more than 5 times the direct ones to the 100th buyer. On average, across all
buyers and sellers, the indirect benefit is essentially equal to the direct benefit.210
The merger of two firms in a two-sided market is an obvious place in which competition regu-
lators should consider the efficiencies from the merger as well as its prospect for increasing prices.
ATM network mergers serve as a good example.211 In rule of reason cases the courts need to examine
the effect of the challenged practice on consumer demand on each side of the market and the indirect
effects from each side to the other side. Consider the Visa Check/MasterMoney litigation.212 Plaintiffs
claim that Visa and MasterCard have policies that result in a tie between credit cards and debit cards
and that without this tie the interchange fee to merchants would be lower. Let us suppose this is
correct. That cannot be the end of the analysis because one must analyze the effect of the practice on
the inextricably linked sides of the market. A forced reduction in the interchange fee to merchants
would reduce the stream of revenues to banks that issue debit cards; under competition, these banks
would increase the fees they charge for debit cards; that it turn would reduce the number of debit
cards held and used; and that in turn would reduce the value that merchants get from debit cards.213
210
The Rochet-Tirole model assumes buyers are not more likely to join if there are more sellers (or vice versa). If we
assumed the base of sellers were important to attracting buyers (and vice versa), the indirect benefits would be even
greater because a buyer joining the system would induce additional sellers to join (and so on), which would generate
additional indirect benefits on both the seller and buyer sides.
211
Robin A. Prager, ATM Network Mergers and the Creation of Market Power, Antitrust Bull. 355 (Summer 1999).
Some examples of ATM mergers include Peak-Minibank (1996), Exchange-Instant Teller (1996), and SCS-Pacific
Interchange (1992). See supra, at 361.
212
See In re Visa Check/MasterMoney Antitrust Litig.; Wal-Mart, et al. v. Visa U.S.A. Inc. and MasterCard
International, Inc., 280 F.3d 124 (2d Cir. 2001)
213
Based on the arguments presented at class certification the plaintiffs appear to be arguing that it is possible to
change one element of the pricing structure without having significant effects on the other elements. They argued
that “under the particular circumstances of the market at issue in this case, credit card interchange fees would not
have increased in the ‘but-for,’ untied world.” See In re Visa Check/MasterMoney Antitrust Litig.; Wal-Mart, et al.
v. Visa U.S.A. Inc. and MasterCard International, Inc., 280 F.3d 124, 154 (2d Cir. 2001).
PLATFORM ECONOMICS:
124 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
214
Frew and Jud, supra note 33, at 178.
215
Id.
216
Many countries have associations that serve those countries but belong to one of the global associations for the pur-
pose of having their cards accepted outside the country. For example, banks belonging to Groupement des Cartes
Bancaires in France issue the “CB” card. There is a domestic version only on Cartes Bancaires’ domestic network,
but most cards have Visa or MasterCard logos and can be used at merchants outside of France that accept Visa
or MasterCard. See Cartes Bancaires website (visited Sep. 2, 2002) <http://www.cartes-bancaires.com/GB/Pages/
FrameVie.htm>.
217
See Broadcast Music, Inc. v. CBS, 441 U.S. 1 (1979).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 125
Another justification for evaluating the IRF under the rule of reason is because it is a po-
tentially efficiency creating agreement among members of a joint enterprise. There are two
possible sources of revenue in the VISA system: the cardholders and the merchants. As a
practical matter, the card-issuing and merchant-signing members have a mutually depend-
ent relationship. If the revenue produced by the cardholders is insufficient to cover the
card-issuers’ costs, the service will be cut back or eliminated. The result would be a decline
in card use and a concomitant reduction in merchant-signing banks’ revenues. In short, the
cardholder cannot use his card unless the merchant accepts it and the merchant cannot ac-
cept the card unless the cardholder uses one. Hence, the IRF accompanies “the coordination
of other productive or distributive efforts of the parties” that is “capable of increasing the
integration’s efficiency and no broader than required for that purpose.”222
The Eleventh Circuit went on to find that “[a]n abundance of evidence was submitted from
which the district court plausibly and logically could conclude that the IRF on balance is procom-
petitive because it was necessary to achieve stability and thus ensure the one element vital to the
survival of the VISA system––universality of acceptance.”223 The Supreme Court declined to re-
view the Eleventh Circuit’s decision. Although the economic analysis of two-sided markets cannot
218
National Bankcard Corp. (NaBanco) v. VISA U.S.A., Inc., 779 F.2d 592 (11th Cir. 1986).
219
It relied on Broadcast Music, Inc. v. CBS, 441 U.S. 1 (1979).
220
One of Visa’s consultants went on to write one of the earliest papers on the economics of two-sided markets. See
William F. Baxter, Bank Interchange of Transactional Paper: Legal and Economic Perspectives, 26 J.L. Econ. 541
(1983).
221
National Bancard Corp. (NaBanco) v. VISA U.S.A., Inc., 596 F. Supp. 1231, 1260 (S.D. Fla. 1984).
222
National Bankcard Corp. (NaBanco) v. VISA U.S.A., Inc., 779 F.2d 592, 602 (11th Cir. 1986).
223
Id at 605.
PLATFORM ECONOMICS:
126 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
by itself say whether Visa’s interchange fee resulted in the optimal pricing structure from society’s
standpoint, the Eleventh Circuit correctly recognized the economic role of interchange fees in
devising a price structure that took both sides of the market into account.
The Royal Bank of Australia (RBA) reached a different conclusion in a recent investigation.224
It was concerned that Visa’s interchange fees were set too high, so as to encourage overuse of credit
cards in Australia. It relied on the findings in economic models of interchange fees that firms may
have private incentives to set the fees higher than socially optimal.225 The reasoning is that an indi-
vidual merchant may have an incentive to accept Visa cards if it expects to make sufficient incre-
mental sales from doing so, even if Visa cards are more expensive for the merchant than its other
alternatives. The RBA argued that these incremental sales were not a social benefit, as they came at
the expense of other merchants. For all merchants collectively, acceptance of Visa cards may have
generated little or no incremental sales. Thus, the argument was that Visa could exploit each indi-
vidual merchant’s willingness to pay, which derived from private not social benefits. In response to
these concerns, the RBA proposed a regulatory scheme for interchange fees that was based on an
allocation of cost to the issuer and acquirer sides; the scheme does not consider demand.
The RBA had not established that interchange fees were too high. It relied on the existence of
theoretical models that showed that they could be too high. But the same models also showed that
privately set interchange fees could be at the socially optimal level or lower. It makes no sense to
seek to lower interchange fees when we do not know if they are too high. All we know, and this
should come as no surprise in any two-sided market, is that one side (the merchants in this case)
would prefer to pay less. The RBA’s proposed regulatory scheme has no basis in the economics
of two-sided markets. The socially optimal price structure depends on a complex contribution of
costs and demand.226 There is no basis for focusing only on costs, nor is there a basis for assigning
costs to one side or the other. There is no economic basis for believing that the RBA method for
determining the interchange fee would increase or decrease the overall welfare of the consumers in
the two sides of the market.
E. Predation
The recognition that business strategies and their effects on consumers must be evaluated with re-
spect to both sides of the market has important implications for the analysis of predation. It may be
privately and socially optimal for prices on one side of the market to be below any possible measure
224
Reform of Credit Card Schemes in Australia IV, Finals Reforms and Regulation Impact Statement, Reserve Bank of
Australia, Aug. 2002.
225
Reform of Credit Card Schemes in Australia I: A Consultation Document, Reserve Bank of Australia, Dec. 2001, at
section 2.4.
226
See generally Ahlborn et al., supra note 21; Schmalensee (2002), supra note 4; Rochet & Tirole, supra note 4.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 127
of cost on this side. That is true not only during the initial stage in which most economists and
courts have recognized the virtues of “penetration pricing.”227 It is also true during the long-run
equilibrium of the industry. Complaints by entrants that they cannot get into one side because
the incumbent is engaging in predatory pricing—or has made predatory investments—are likely
to have little merit in two-sided markets. It also may be privately and socially optimal for firms to
make significant investments in one side even though these investments do not appear to generate
profits on that side. Again, this can occur even when the firm is mature.
To clarify the issues, let us consider extending the Brooke Group test of predatory pricing to
two-sided markets.228 The test has two prongs. First, a plaintiff alleging predation must show that
the defendant’s prices were “below an appropriate measure of… costs.”229 Thus, pricing must be be-
low cost to support a claim of predation, even though, in theory, there can be predatory prices that
are above cost. Second, the plaintiff must show that the defendant had “a reasonable prospect, or,
under §2 of the Sherman Act, a dangerous probability, of recouping its investment in below-cost
prices.”230 Finding that a competitor was harmed—even driven from the market—is not enough
under this test. Moreover, for a dangerous probability of recoupment to exist, low prices must
eliminate substantial competition in a way that persists even after a post-predation price increase.
For two-sided markets the pricing prong of this test should be based on the defendant’s pricing
level—the overall price to the two markets. This is simple and straightforward when both sides
of the market have the same unit of observation (person, transaction, product, etc.). One would
calculate whether the total price charged for a given transaction, let us say, exceeds marginal cost
(or, following the Areeda-Turner analysis, average variable cost).231
This analysis would apply to payment cards, dating services, B2B exchanges, real estate and
other matching-type markets. When the two sides of the market have different units of observation
one could normalize them to the same unit by basing the calculation on total revenue and total
variable cost. If total variable cost is less than total revenue then the defendant would fail the first
prong of the Areeda-Turner version of the Brooke Group test.
The recoupment prong is straightforward. For two-sided markets the court would need
to consider whether there is a dangerous probability that the defendant will raise prices high
enough and for long enough on both sides of the market to recoup its losses during the alleged
predatory phase. There is nothing novel about this for two-sided markets other than accounting
for the two sides.
227
See, e.g., Brooke Group Ltd. v. Brown & Williamson Tobacco Corp. 509 U.S. 209 (1993); Phillip Areeda & Donald
F. Turner, Predatory Pricing and Related Practices under Section 2 of the Sherman Act, 88 HAR. L. REV. 697 (1975).
228
See Brooke Group Ltd. v. Brown & Williamson Tobacco Corp. 509 U.S. 209 (1993).
229
Id. at 222.
230
Id. at 224.
231
See Areeda & Turner, supra note 227, at 702, 716-18.
PLATFORM ECONOMICS:
128 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
232
See Findings of Fact, supra note 198, at ¶ 250.
233
“The rare case of price predation aside, the antitrust laws do not condemn even a monopolist for offering its product
at an attractive price, and we therefore have no warrant to condemn Microsoft for offering either IE or the IEAK
free of charge or even at a negative price. Likewise, as we said above, a monopolist does not violate the Sherman Act
simply by developing an attractive product.” United States v. Microsoft Corp., 253 F.3d 34, 68 (D.C. Cir. 2001).
234
United States v. Microsoft Corp., Civil Action No. 98-1232, Complaint, May 18, 1998, at ¶ 16.
235
See the Global Platform website (visited Aug. 15, 2002) <http://www.globalplatform.org/news2.asp>.
236
See Global Platform Specification (visited Aug. 15, 2002) http://international.visa.com/fb/paytech/product-
splatforms/globalplatform.jsp>; and the Global Platform Brochure (downloaded Aug. 15, 2002) <http://www.
globalplatform.org/pdf/GlobalPlatform_Brochure.pdf>.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 129
Third, two-sided firms sometimes take actions to coordinate the behavior of their customers.
That occurs because standardization by one set of customers benefits the other set of customers.
For instance, recently B2Bs have been moving towards the standardization of information that
might significantly enhance and automate various procedures such as RFPs (request for proposals),
RFQs (request for quotes), fax requests, phone inquiries and purchase orders.237 NMMs, third-
party intermediaries whose primary purpose is to match corporate buyers and sellers, play a pivotal
role in this process.238 In U.S. v. Microsoft, Microsoft objected to letting computer manufacturers,
or for that matter end-users, selectively delete portions of the operating system code. One of its
justifications was that applications developers would then lack assurance on what features are avail-
able for them to rely on in the operating system on an end-user’s machine.239
Fourth, associations of competitors, joint ventures, and standard-setting bodies may arise to
provide coordination among competitors or among customers or among other important input
suppliers to either side of the market. Payment-card associations operate the network and set rules
that result in the determination of a pricing structure. Real estate agencies have associations that
operate the MLS.240 Microsoft collaborates with Intel and many hardware manufacturers to help
ensure that the Wintel platform attracts users and developers.241
These practices tend to be flypaper for antitrust complaints. The multi-product nature of
firms in two-sided markets can lead to a variety of issues surrounding “vertical relationships.” For
example, a firm might be accused of trying to dominate one side of a two-sided market to reduce
competition on the other side. Or to take another example, suppose a firm on one side of a two-
sided market wants to deliver customers who possess products A and B to the other side. One
way it might do that is to integrate A and B or require that customers purchase B as a condition
of purchasing A. The firm might be accused of tying in this instance—this is the plaintiff’s claim
in the Visa Check/MasterMoney.242 Although these sorts of vertical relationships may well result in
anti-competitive effects, it is necessary to evaluate their effects in the context of competition in
two-sided markets.
237
Roopen Roy & Abhijit Roy, Net Market Makers: The Winners in the B2B Play! Indianinitiative.com, Dec. 21, 2001,
(visited Aug. 21, 2002) <http://www.indiainitiative.com/indiainitiative_new/e-business/b2b_abhi_roopen.htm>.
238
Id.
239
Ronald A. Cass, Copyright, Licensing, and the ‘First Screen,’ 5 Mich. Telecommunications Tech. L. Rev. 35, 59-
64 (1999).
240
Real estate boards are non-profit organizations representing local real estate agents/brokers, which operate Multiple
Listing Services in local communities. For a definition, see Home and Real Estate Glossary (visited Aug. 28, 2002)
<at http://www.homes-and-real-estate.com/glossary/r.htm>.
241
Carmen Nobel, Microsoft, Intel Seek to Become Wireless Force, eWeek, Feb. 19, 2002 (visited Aug. 28, 2002) <http://
www.eweek.com/article2/0,3959,108489,00.asp>.
242
In re Visa Check/MasterMoney Antitrust Litig.; Wal-Mart, et al. v. Visa U.S.A. Inc. and MasterCard International,
Inc., 280 F.3d 124 (2d Cir. 2001).
PLATFORM ECONOMICS:
130 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Coordinated behavior also raises antitrust concerns. Whenever businesses get together there is
a worry that they are colluding to raise price or otherwise harm consumers. That may be just as true
in two-sided markets as it was for Adam Smith’s tradesmen.243 Nevertheless, there are legitimate
pro-competitive reasons for coordination among competitors or among customers and for having
institutions to help do this in two-sided markets.
V. CONCLUSIONS
Two-sided markets are an increasingly important part of the global economy. Firms that pro-
vide platforms for multiple customer groups are a critical part of many interrelated segments of
the computer industry. These “keystone” firms include Microsoft in operating systems for Intel-
compatible computers and Intel for computer chips. In most industrialized countries a large frac-
tion of payments takes place through firms and associations that provide platforms for merchants
and customers to exchange money. The increased importance of the Internet for household-to-
household, business-to-household, and business-to-business transactions and the emergence of
e-pay systems on the Internet will increase the fraction of payments going through commercial
payment platforms. And speaking of the Internet, although dot-coms are currently in a slump,
Internet-based businesses are sure to flourish over time and many of these are likely to be based on
a two-sided model.
But two-sided markets are not just in these high-profile sectors. They are dotted across the
economy. We began with perhaps a trivial example of dating clubs—discotheques, church clubs
for singles, and local village matchmakers could have served just as well. Others range from real
estate to video games to media firms. Some of the most recognizable brands in the world operate
in two-sided markets: Bloomberg, Century 21, Microsoft, Sony and Visa.
Two-sided markets and firms in those markets behave in ways that seem surprising from the
vantage point of traditional industries, but in ways that seem like plain common sense once one
understands the business problems they must solve. “Getting both sides on board.” “The chicken-
and-egg problem.” These are the mantras ones hears from the entrepreneurs in these industries, the
trade press that covers them, business gurus, and journalists. They contain important economic
implications.
Two-sided firms have to come up with the right price structure and the right investment
strategy for balancing the demands of the customer groups they must get and keep on their plat-
forms. That is a different problem than is faced by one-sided firms. It is also harder. One way to
see that is to recognize that different firms have chosen different price structures and have realized
different fortunes from their choices. American Express bet on a price structure skewed against
243
Adam Smith, an Inquiry into the Nature And Causes of the Wealth of Nations ¶ I.10.82 (Edwin Cannan
ed., 1904).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 131
merchants—it worked for some years but then got it into trouble.244 Meanwhile Visa surpassed
American Express, a firm that was once dominant and seemed unbeatable.245 Microsoft bet on a
price structure that catered to software developers. Bloomberg bet on a simple formula for its data
terminals—a flat fee for subscribers, and few charges for content providers.
There is no reason to believe that anticompetitive problems are more prevalent or less prevalent
in two-sided industries than other ones. Likewise, there is no basis for asking regulators to steer
clear of these industries or to spend extra effort on them. The message of this chapter, however, is
that antitrust analysis of these industries should heed the economic principles that govern pricing
and investment decisions in these industries. Prices do not and prices cannot follow marginal costs
in each side of the market. Price levels, price structures, and investment strategies must optimize
output by harvesting the indirect network effects available on both sides. By doing so businesses in
two-sided industries get both sides on board and they solve the chicken-and-egg problem.
244
Between mid-1980s and 1996 the American Express charge and credit card share dropped from more than 24
percent to 16 percent. By mid-1990s American Express realized the necessity to adopt a new business model. See
Evans & Schmalensee, supra note 10, at 185-193.
245
By 1996 Visa charge and credit card share was more than 45 percent compared to 16 percent held by American
Express. Id. at 174, 187.
PLATFORM ECONOMICS:
132 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
APPENDIX A
TABLE 1 Sources of Platform Revenue in Selected Two-Sided Platforms
Two-Sided Side That Gets
Industry Platform Side One Side Two Charged Little Sources of Revenue
Real Residential Buyer Seller Side One Real estate brokers derive income
Estate Property principally from sales commis-
Brokerage sions.1
Real Apartment Renter Owner/ Typically Side Apartment consultants and loca-
Estate Brokerage Landlord One ter services generally receive all of
their revenue from the apartment
lessors once they have successfully
found tenants for the landlord.2
Media Newspapers Reader Advertiser Side One Approximately 80 percent of
and newspaper revenue comes from
Magazines advertisers.3
Media Network Viewer Advertiser Side One For example, FOX earns half of its
Television revenues from advertisers.4
Media Portals and Web Advertiser Side One The average portal gets slightly
Web Pages “Surfer” over half of its revenues from ad-
vertisements. All other web pages
generally receive about a tenth
of their revenue from advertise-
ments.5
Software Operating Application Application Side Two For example, Microsoft earns at
System User Developer least 67 percent of its revenues
from licensing packaged software
to end-users.6
Software Video Game Game Neither – Both Both game sales to end users and
Game Player Developer sides are signi licensing to third party developers
Console ficant sources are significant sources of revenue
of platform for console manufacturers.7
revenue.
Payment Credit Card Cardholder Merchant Side One For example, in 2001, American
Card Express earned 82 percent of its
System revenues from merchants.8
Notes: 1Real Estate Brokers and Sales Agents, in Occupational Outlook Handbook, U.S. Department of Labor, Bureau of Labor Statistics
(visited Sept. 3, 2002), <http://www.bls.gov/oco/ocos120.htm>; 2Courtney Ronan, Apartment Locaters: How Do They Make Their
Money? Realty Times website, Jun. 30, 1998 (visited Aug. 22, 2002) <http://realtytimes.com/rtnews/rtcpages/19980630_aptlocator.htm>;
3
Lisa George and Joel Waldfogel, Whom Benefits Whom in Daily Newspaper Markets? NBER WORKING PAPER #7944 11 (October 2000);
4
Brian O’Keefe, Extra! Extra! Hot Media Stocks! Fortune, June 24, 2002 (visited Aug. 22, 2002) <http://www.pbs.org/wsw/news/fortunearti-
cle_20020620_01.html>; 5Clicking into Orbit, Technology Trend Quarterly Report, March 2000 (visited Aug. 15, 2002) <www.deloitte.nl/
downloads/website/clicking_into_orbit.pdf>; 61994 Worldwide Software Review and Forecast, IDC Report #9358, November 1994, 1995
Worldwide Software Review and Forecast, IDC Report # 10460, November 1995, 1996 Worldwide Software Review and Forecast, IDC Report
#12408, November 1996, 1997 Worldwide Software Review and Forecast, IDC Report #14327, October 1997, 1999 Worldwide Software
Review and Forecast, IDC Report #20161, October 1999, Worldwide Software Market Forecast Summary, 2001-2005, IDC Report #25569,
September 2001 ; 7David Becker, Revenue from Game Consoles will Plunge, Report Predicts, CNET.Com, Jul. 31, 2000 (visited Aug. 5, 2002)
<http://techrepublic-cnet.com.com/2100-1040-243841.html?legacy=cnet>; 8American Express Company Annual Report 2001 (visited Aug.
15, 2002), <http://www.onlineproxy.com/amex/2002/ar/pdf/axp_ar_2001.pdf>, at 35.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Economics of Two-Sided Markets 133
5 Two-Sided Markets
David S. Evans
ABSTRACT
This chapter addresses the analysis of market definition when the parties involved in an antitrust
or merger analysis include one or more two-sided platforms. We discuss how standard market
definition measures such as SSNIP tests, diversion ratios, and conditional logit demand analyses
have to be modified to account for the unique characteristics of two-sided platforms. We also
review how market definition of two- sided platforms was treated in recent US and EC case law.
I. INTRODUCTION
Nightclubs and online dating companies provide services to men and women who want to meet.
Credit cards enable consumers who have them and merchants that accept them to transact with
one another. Video game platforms, such as Sony PlayStation or Nintendo, provide software tools
that enable publishers to develop games and a device on which consumers can play those games.
In each case, the business provides a platform that enables two distinct but related groups of cus-
tomers to obtain value. These “two-sided platforms” or “two-sided markets” are the major form of
business organization in many industries.1
1
David S. Evans & Richard Schmalensee, Industrial Organization of Markets with Two-Sided Platforms, 3 Competition
Pol’y Int’l 150, 152 (2007) [hereinafter Evans & Schmalensee, Industrial Organization] (“Many diverse industries are
populated by businesses that operate ‘two-sided platforms.’ These businesses serve distinct groups of customers who
need each other in some way, and the core business of the two-sided platform is to provide a common (real or virtual)
meeting place and to facilitate interactions between members of the two distinct customer groups. Two-sided platforms
are common in old-economy industries such as those based on advertising-supported media and new-economy indus-
tries such as those based on software platforms and web portals. They play an important role throughout the economy
by minimizing transactions costs between entities that can benefit from getting together.”). Many platforms have more
than two groups of distinct users. See David S. Evans & Michael Noel, The Analysis of Mergers That Involve Multisided
Platform Businesses, 4 J. Competition L. & Econ. 663, 664 (2008) [hereinafter Evans & Noel] (“Many old industries
are based on [multisided platforms or] MSPs, ranging from village matchmakers that date from ancient times to adver-
tising supported newspapers introduced in the seventeenth century to payment cards introduced in the mid-twentieth
century. However, an increasing number of significant modern businesses are MSPs as a result of technological changes
that have drastically lowered the costs and increased the benefits of connecting diverse customer groups on a single plat-
form. These include most Internet-based businesses such as eBay, Facebook, and Google. These businesses are creating
new products and services such as social networking platforms and are disrupting existing industries such as advertising-
supported media.”). This chapter focuses on two-sided cases, but most concepts extend readily to multisided cases.
135
PLATFORM ECONOMICS:
136 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
In antitrust cases involving two-sided platforms, market definition and market power analyses
must take into account several economic issues that do not arise in other contexts. The two sides
of a platform business are closely linked, with interdependent prices and outputs and intertwined
strategies To understand the relevant competitive relationships, one must consider both sides of
the platform business.
For example, if an online dating service increased its prices to women, it would attract fewer
women to its site. As a result, it would lose male members and, with fewer men, even more women
would desert the service. Credit card issuers offer reward programs to persuade more people to
use credit cards. As credit card use increases, card issuers can convince more merchants to pay for
taking the card.
A platform can earn profits on either side. In practice, two-sided platforms often obtain most
of their incremental profits on one side and may provide services to the other side at prices below
incremental costs. For example, video game platforms earn most of their profits “on game develop-
ers through per-unit royalties on games and fixed fees for development kits and treat the gamers
side as a loss leader.”2
Most standard approaches to market definition, such as the SSNIP (small but significant and
nontransitory increase in price) test, diversion ratios, conditional logit demand analyses,3 and other
economic models and formulae, do not apply to two-sided markets without modification, occa-
sionally radical in nature.4 Many of these “standard tools of antitrust and merger analysis . . . were
developed based on the economics of single-sided businesses” and “do not necessarily apply in
ways that are material to the analysis of competition that involves multisided businesses.”5 In most
cases, that relationship does not hold without significant modification for two-sided platforms.
For example, one common approach—using the price-cost margin on one side to assess critical
2
Jean-Charles Rochet & Jean Tirole, Platform Competition in Two-Sided Markets, 1 J. Eur. Econ. Ass’n 990, 990-91
(2003) [hereinafter Rochet & Tirole, Platform Competition].
3
See, e.g., United States v. SunGard Data Sys., 172 F. Supp. 2d 172, 186-93 (D.D.C. 2001) (rejecting the govern-
ment’s market definition based on critical loss and SSNIP analyses); FTC v. Tenet Health Care, 186 F.3d 1045,
1050-51, 1053 (8th Cir. 1999) (relying on critical loss analysis).
4
David S. Evans, Competition and Regulatory Policy for Multi-Sided Platforms with Applications to the Web Economy, 2
CONCURRENCES 57 ¶ 28 (2008) (providing an overview of two-sided platforms); Evans & Noel, supra note 1,
at 675 (explaining that “the one-sided Critical Loss formulas for conducting the SSNIP test are wrong when applied
to two-sided markets”); Evans & Schmalensee, Industrial Organization, supra note 1, at 151 (“[P]rofit-maximizing
prices may entail below-cost pricing to one set of customers over the long run and, as a matter of fact, many two-
sided platforms charge one side prices that are below marginal cost and are in some cases negative. These and other
aspects of two-sided platforms affect almost all aspects of antitrust analysis—from market definition, to the analysis
of cartels, single-firm conduct, and efficiencies.”).
5
Evans & Noel, supra note 1, at 664 (“Each side of the MSP’s business influences and constrains its strategies on the
other side. Antitrust analysis that focuses on one side of the business in isolation from the other side is incorrect as a
matter of economics, and can lead to the wrong answer when indirect network effects are significant and are relevant
for assessing the practice at issue.”).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 137
loss—tends to understate the effects of a merger on prices compared with the two-sided market
formula.6 Another approach—estimating demand elasticities directly based on a standard one-
sided model—tends to overstate the effects of a merger on prices.7
This chapter describes how two-sided platforms operate, explains the basic economics of prof-
it-maximization for two-sided businesses, and notes some of the unique aspects of competitive
strategies when one of the rivals is two-sided. It summarizes key market definition issues, shows the
extent to which the use of standard techniques for two-sided platforms can result in bias, and ad-
dresses some basic conceptual issues for markets with two-sided platforms, such as how to measure
output and which firms to include in the analysis. Finally, it discusses cases and agency decisions
concerning two-sided markets.
The fundamental role of a two-sided platform in the economy is to enable parties to real-
ize gains from trade or other interactions by reducing the transactions costs of finding each
other and interacting. Two-sided platforms do this by matchmaking, building audiences,
and minimizing costs. Different platforms engage in these activities to different degrees.
Software platforms are mainly about minimizing duplication costs, advertising-supported
media in mainly about building audiences, and exchanges are mainly about matchmaking.
But they all seem to engage in each to some degree. All platforms help reduce costs by pro-
viding a virtual or physical meeting place for customers.10
6
Id, at 682.
7
Id.at 670.
8
See Evans & Schmalensee, Industrial Organization, supra note 1, at 151. See generally David S. Evans, The Antitrust
Economics of Multi-Sided Platform Markets, 20 YALE J. ON REG. 325 (2003) [hereinafter Evans, Antitrust
Economics]; David S. Evans, Some Empirical Aspects of Multi-Sided Platform Industries, 2 Rev. Network Econ.
191 (2003); Rochet & Tirole, Platform Competition, supra note 2; Jean-Charles Rochet & Jean Tirole, Two-Sided
Markets: A Progress Report, 37 Rand J. Econ. 645 (2006) [hereinafter Rochet & Tirole, Two-Sided Markets].
9
Evans & Schmalensee, Industrial Organization, supra note 1, at 151; David S. Evans & Richard Schmalensee,
Catalyst Code: The Strategies Behind the World’s Most Dynamic Companies 53 (Harvard Business School
Press, 2007) [hereinafter CATALYST CODE]; see Rochet & Tirole, Platform Competition, supra note 2, at 995.
10
Evans & Schmalensee, Industrial Organization, supra note 1, at 158 (explaining that platforms “minimize transac-
tion costs” through “matchmaking, audiencemaking, and . . . elimination of duplication”).
PLATFORM ECONOMICS:
138 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
For example, newspapers, magazines, and other types of advertiser-supported media use con-
tent to attract readers and then sell access to those readers to advertisers. Advertisers need readers;
media platforms reduce advertisers’ cost of reaching target audiences, facilitating profitable trans-
actions.11 Online auction websites such as eBay provide platforms enabling sellers of goods or ser-
vices find and consummate transactions with buyers. Importantly, each customer on one side tends
to realize more value when there are more customers on the other side.12 Shoppers value shopping
malls that have more stores that are of potential interest to them while retail stores value shopping
malls that have more potential customers.
Two-sided platforms usually perform three core functions to some degree.13 They serve as
matchmakers to facilitate exchange by making it easier for members of each group to find each
other (e.g., eBay, MySpace.com, NASDAQ). They build audiences because this makes it more like-
ly that members of a group will find a suitable match (e.g., Google, TiVo, Condé Nast, Fox News).
Finally, they provide shared resources and reduce the cost of providing services to both groups of
customers (e.g., Windows, Sony PlayStation, Xbox, SAP enterprise software, Linux, Palm OS).
Most two-sided platforms engage in all three core functions. For example, eBay makes it easier
for a person who has a particular item and a person who wants a particular item to find each other.
It does this in part by attracting a large number of buyers and sellers to its platform thereby mak-
ing thicker markets for each item. Finally, it provides a common platform for matchmaking that
reduces the costs of exchange for all participants.
A key feature of two-sided platforms is the presence of “indirect network effects.”14 Indirect
networks effects exist when the value that a customer on one side realizes from the platform in-
creases with the number of customers on the other side. A search platform is more valuable to ad-
vertisers if it is more likely that it will reach a larger number of potential buyers. It is more valuable
to users looking to buy something if there are more advertisers attracted to the platform because
that makes it more likely that the user will see a relevant advertisement.15 As a practical matter, it
is often the strength of these indirect network effects that determines whether being “two-sided”
matters enough to have a substantive effect on the results of economic analysis.16
11
See generally Simon P. Andersen & Stephen Coate, Market Provision of Broadcasting: A Welfare Analysis, 72 Rev.
Econ. Stud. 947 (2005).
12
Many two-sided platforms have positive indirect networks for both sides. In the case of advertising-supported me-
dia, advertisers value a platform more if it has more viewers but viewers may not value a platform more if it has more
advertising; the platform solves this asymmetry in valuation by paying viewers to join the platform in the form of
free content. Evans & Schmalensee, Industrial Organization, supra note 1, at 155-56.
13
Catalyst Code, supra note 9, at 7
14
Evans, supra note 4, ¶ 8 & n.10 (citing Rochet & Tirole, Two-Sided Markets, supra note 8).
15
See David S. Evans, The Economics of the Online Advertising Industry, 7 Rev. Network Econ. 359, 372-75 (2008).
16
In addition to traditional indirect network effects, externalities in use are also important: in the case of payment
cards a product cannot exist unless there is a cardholder and an accepting merchant. See Evans, supra note 4, ¶ 8 &
n.12 (citing Rochet & Tirole, Platform Competition, supra note 2).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 139
Two-sided p latforms’ key feature is the symbiotic relationship between the two sides; the
platforms must cater to two different customer groups simultaneously. To establish a two-sided
platform, the founders must solve the “chicken-and-egg problem”: customers on side A will not
participate without customers on side B while customers on side B will not participate without
customers on side A. The founders must be able to make credible commitments to one side that
if they show up at the platform, the other side will be there as well. In some cases these commit-
ments must be made virtually at the same time. For example, consumers will not carry a payment
card unless they know they can use it at merchants’ stores, and merchants and the processors who
service them will not incur the cost of accepting a card unless they know that enough customers
want to use it. In other cases these commitments occur sequentially. A shopping mall must first
convince retailers that enough shoppers will come once the retailers locate there. Then the mall
owner tries to attract shoppers to its completed mall.17
A related feature of two-sided platforms is the need to “balance” the demands of the two sides.
In setting prices, a two-sided platform needs to consider that charging a higher price to side A will
result in fewer A’s using the platform which in turn will result in fewer B’s getting value from the
platform. It also needs to consider that there must be enough A’s to be of interest to the B’s and
enough B’s to be of interest to the A’s in most cases. Setting store rental prices too high at malls
could result in shoppers not having access to the stores they most care to go to.
A two-sided platform’s profit-maximizing calculus is thus more complex than that of a tradi-
tional business. The two-sided platform needs to consider the demands of both sides, the inter-
relationships between these demands, the costs directly attributable to each side, and the costs of
running the platform. Many platforms consider both access fees to join the platform and usage fees
to interact on the platform in establishing prices that attract the optimal volume of transactions on
each side of the platform.
One side of a two-sided platform usually gets a better deal or receives service for free. For ex-
ample, searchers do not pay search engines, advertisers do. Buyers do not pay a fee on many online
transaction platforms, sellers do. Merchants, rather than cardholders, pay for most of the cost of
payment card schemes. Retail stores, not shoppers, pay to get access to a shopping mall. End users
pay for most software platforms, but software developers often get access to code for free. Women
often get discounts from dating services, while men seldom do.
There are many practical reasons for price differences between the two sides of a platform. A
two-sided platform has to attract enough members on each side to have sufficiently thick markets.
Most business-to-business exchanges failed in the early 2000s because, even though many buyers
17
See David S. Evans, How Catalysts Ignite: The Economics of Platform-Based Start-Ups, in Platforms, Markets And
Innovation (A. Gawker ed. forthcoming 2009) (discussing solving the start-up problem for two-sided platforms);
see also CATALYST CODE, supra note 9, at 89-91 (discussing two-sided platforms’ pricing and commitment);
Andrei Hagiu, Pricing and Commitment by Two-Sided Platforms, 37 (3) Rand J. Econ. 720 (2006) (same).
PLATFORM ECONOMICS:
140 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
had signed on, there were not enough sellers to make the exchange attractive.18 A two-sided plat-
form may therefore find that it has to keep prices very low to one side to attract enough partici-
pants to make the platform valuable to the other side. The side that is needed more or that is harder
to get receives a price break; conversely the side that gets the most value out of access to members
of the other side bears more of the cost. Economists have shown that the profit-maximizing prices
for two-sided platforms can result in one side getting a price that is less than the incremental cost
of providing a customer on that side and sometimes less than zero.19
Economic theory predicts that there will typically be a close relationship between prices and in-
cremental costs for traditional single-sided businesses. Competition drives prices down to marginal
cost20 and prices and marginal costs tend to move in the same direction. Deviations between price
and marginal cost are often taken as an indicator that firms have market power.21 In two-sided
businesses, however, there is not a close relationship between prices and marginal costs for the
goods or services provided on a particular side. Because the conditions for profit maximization de-
pend on demand elasticities and marginal costs for both sides and the cross-side network effects,22
in theory, a profit-maximizing two-sided platform will respond to an increase in cost of side 1 with
an increase in price on side 2 and no increase in price on side 1.
Some two-sided platforms set prices below marginal cost on one side to attract customers to
whom they charge prices above marginal cost on the other side. Competition between platforms
drives overall profits down. However, it is common to have a competitive equilibrium in which all
of the firms have adopted the same skewed pricing structure that recovers most of the costs from
one side while favoring the other side. The major payment card schemes in the United States, for
18
See Bob Tedeschi, Investing; How Killer B-to-B’s Went Into A Tailspin, N.Y. Times, May 7, 2000; Hazel Ward,
B2B Exchanges Fail to Deliver, ComputerWeekly.com, Nov. 29, 2001, http://www.computerweekly.com/
Articles/2001/11/29/183994/b2b-exchanges-fail-to-deliver.htm; Kate Hanaghan, B2B Exchanges—What Happens
Now?, Silicon.com, Jan. 23, 2002, http://software.silicon.com/webservices/0,39024657,11030680,00.htm.
19
See, e.g., Rochet & Tirole, Two-Sided Markets, supra note 8, at 659 (“[I]t is quite common for a platform to charge
below-cost (perhaps zero) prices to one side and high prices to the other.”); Mark Armstrong, Competition in Two-
Sided Markets, 37 Rand J. Econ. 668, 682 (2006) (noting that in some cases prices to one side of the two-sided
platform are “below cost”).
20
See Evans, supra note 4, ¶ 14 (citing Paul A. Samuelson & William D. Nordhaus, Economics (18th ed. McGraw-
Hill/Irwin 2005)).
21
See U.S. Dep’t of Justice & Fed. Trade Comm’n, Horizontal Merger Guidelines, § 0.1 (1992) [hereinafter
MERGER GUIDELINES], reprinted in 4 Trade Reg. Rep. (CCH) ¶ 13,104 (“Market power to a seller is the
ability profitably to maintain prices above competitive levels for a significant period of time.”); see also Dennis
W. Carlton, Market Definition: Use and Abuse, 3 (1) Competition Pol’y Int’l 2, 5 (2007) (“Roughly speaking,
‘market power’ means that the industry’s behavior deviates from perfect competition. One standard definition
of market power is the ability to set price profitably above the competitive level, which is usually taken to mean
marginal cost.”).
22
See Rochet & Tirole, Platform Competition, supra note 2, at 996-97.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 141
example, all have pricing structures in which merchants pay for the much of the cost of the system
while cardholders are subsidized through rewards and the float on their balances.23
A common condition for economic efficiency is that “the competitive price equals marginal
cost of production.”24 Nonetheless, this condition does not necessarily hold in the case of two-sid-
ed platforms in part because there is joint production of services to the two sides.25 For example, a
shopping mall provides services jointly to retailers and shoppers. In some cases, the product cannot
exist without both sides’ participation: A credit card provides transaction services only if there is a
person who uses the card for payment and a merchant who takes the card for payment.26
These considerations have implications for the analysis of market power. The analysis of either
side of a two-sided platform in isolation yields a distorted picture of the business. The low-margin
side will tend to appear highly competitive in the sense that prices are not significantly higher than
marginal cost. The high-margin side will tend to appear much less competitive because prices are
much higher than marginal cost. Competition between platforms may result in squeezing the mar-
gins down on the high-margin side but will not fundamentally alter the skewed pricing structure
which results from all firms’ seeking to perform the same balancing act. To analyze market power
and to understand competitive and anticompetitive strategies involving two-sided platforms one
has to consider the interplay between the two sides.
23
Evans, supra note 4, ¶ 15 & n.18 (citing David S. Evans & Richard Schmalensee, The Economics of Interchange Fees
and Their Regulation: An Overview, in Interchange Fees in Credit and Debit Card Industries: What Role
For Public Authorities? (Santa Fe: Federal Reserve Bank of Kansas City, May 2005)).
24.
See Dennis W. Carlton & Jeffrey M. Perloff, Modern Industrial Organization 70 (4th ed. 2005).
25
This point was first observed for the case of payment cards in a seminal paper that is a precursor to the two-sided
economics literature.” Evans, supra note 4, ¶ 16 n.19 (citing William F. Baxter, Bank Interchange of Transactional
Paper: Legal and Economic Perspectives, 26 J. L. & ECON. 541, 544 (1983)).
26
See Baxter, supra note 25, at 544 (“The mechanics of transactional services require that for every transaction in which
a purchaser becomes a maker of a check, there must be one- and precisely one-transaction in which a merchant be-
comes a payee; similarly, each use of a credit card by a card holder must be matched by precisely one act of acceptance
of the card (or, more accurately, the paper that the card generates) by a merchant.”). See generally Nat’l Bancard Corp.
v. Visa U.S.A. Inc., 779 F.2d 592, 594 (11th Cir. 1986) (discussing “the workings of the bank credit card industry”).
27
See Merger Guidelines, supra note 21, § 1.
PLATFORM ECONOMICS:
142 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
its harmful effects or reinforcing them.28 Market definition provides a screen that limits detailed
antitrust scrutiny to those situations in which one cannot count on competition to counter the
effect of a practice on price, output, innovation, or other measures of competitive performance.29
The business practice at issue must be considered in the context of both sides of the platform. A
major competitive constraint on practices that affect one side of a two-sided platform comes from
the impact of these practices on the platform’s other side. A business practice that increases price
and thus reduces the number of customers on one side of a two-sided platform will result in a de-
crease in demand on the other side for the reasons discussed above. Therefore, in assessing whether
a business practice could lead to a price increase on one side of the platform, one would have to
take into account both the reduction in demand on that side of the platform and the subsequent
negative feedback effects on the other side.30
The newspaper industry illustrates this point. Newspapers provide content that attracts readers
and sell access to these readers to advertisers. Most newspapers earn the preponderance of their
profits from advertisers.31 Most major daily newspapers are sold at a price that roughly covers the
incremental cost of printing and distributing those papers but does not contribute to the cost of
providing the content.32 When newspapers establish their profit-maximizing prices and produc-
tion levels, they necessarily consider the interrelated impacts on the reader and advertiser sides.33 If
a newspaper were to increase the cover charge to its readers, the number of readers would decrease,
resulting in the loss of advertising revenue.34 Any analysis of business practices would hazard miss-
ing important effects if it did not consider the interrelated sides of the business.35
Courts and agencies have taken differing approaches to defining relevant markets in news-
paper industry cases: some have considered the feedback effects between readers and advertis-
ers, while others have focused on just one side of the platform. For example, in United States v.
28
Id, § 2.
29
Cf. Carlton, supra note 21, at 22-23 (listing common mistakes in market definition).
30
Traditional market definition analysis focuses on the role of substitutes in constraining price—it does not focus, as it
must in two-sided platform markets, on the role of complementary products (the two interlinked platform products).
31
Evans & Schmalensee, Industrial Organization, supra note 1, at 156.
32
Id.; see Roger D. Blair & Richard E. Romano, Pricing Decisions of the Newspaper Monopolist, 59 S. Econ. J. 721,
721 & n.2 (1993) (confirming that newspaper cover price is below marginal cost and the sales revenue does not
cover the cost of ink and paper used in producing the newspaper) (citing studies showing that, in the 1980s, “a daily
newspaper with a circulation of 250,000 typically generated circulation revenue equal to only 75-80 percent of the
cost of ink and newsprint”).
33
See Blair & Romano, supra note 32, at 722 (“Given the feedback effect of circulation on advertising demand, the
publisher must be particularly concerned about the circulation price, which is not always under the publisher’s
control absent vertical integration. Given the feedback effect of circulation on advertising demand, the publisher
must be particularly concerned about the circulation price, which is not always under the publisher’s control absent
vertical integration.”).
34
See id.
35
See Chapter 5 for a discussion of the newspaper industry.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 143
Times-Picayune Publishing,36 the Supreme Court recognized that “every newspaper is a dual trader
in separate though interdependent markets; it sells the paper’s news and advertising content to its
readers; in effect that readership is in turn sold to the buyers of advertising space.”37 Taking into
account both sides of the platform business—its readers and its advertisers—the Court concluded
that Time-Picayune Publishing did not occupy “ a ‘dominant’ position in the newspaper advertis-
ing market in New Orleans.”38 By contrast, the U.K. competition authorities have acknowledged
“the two-sided nature” of the publishing industry in some cases, while considering “only one side
of the market” in other cases.39
36
345 U.S. 594 (1953).
37
Id. at 610.
38
Id. at 610-11, 627-28 (overruling the district court’s finding of dominance).
39
See John Wotton, Are Media Markets Analyzed as Two-Sided Markets?, 3 Competition Pol’y Int’l 237, 240-44
(Spring 2007) (discussing cases).
40
See Evans, Antitrust Economics, supra note 8, at 360-361. The two-sided considerations for implementing the SSNIP
test are still relevant in this case because a symmetric price increase will not only lower the direct demand for each
side but also lower the indirect demand through the feedback effects. Evans & Noel, supra note 1, at 666.
41
Evans & Noel, supra note 1, at 667.
PLATFORM ECONOMICS:
144 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
42
See Julian Wright, One-Sided Logic in Two-Sided Markets, 3 (1) Rev. Network Econ. 44, 48 (2004) (explaining
why a two-sided platform’s below-cost pricing does not indicate predation).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 145
43
See Evans, supra note 4, ¶ 25 & n.23 (citing Thomas R. Eisenmann, Geoffrey Parker & Marshall W. Van Alstyne,
Platform Envelopment 1 (Working Paper 2007), available at http://papers.ssrn.com/sol3/papers.cfm?abstract_
id=996852).
44
See Evans, supra note 4, ¶ 26.
45
See Fred Vogelstein, Gates vs. Google: Search and Destroy, Fortune, Apr. 18, 2005 (Google “does not sell its search
products [but] makes its money from the ads that accompany its search results”); see also Charles H. Ferguson,
What’s Next for Google, 108 (1) Tech Rev. 38 (2005) (discussing competition between Google and Microsoft).
46
Microsoft’s investment in advertising has included the launch of Microsoft adCenter, the company’s online advertis-
ing platform, and the purchase of aQuantive, a family of digital marketing service and technology companies, for
$6 billion. Press Release, Microsoft, Seventh Annual MSN Strategic Account Summit Celebrates Advertisers’ Role
in Microsoft Media Network Vision (May 3, 2006), http://www.microsoft.com/presspass/press/2006/may06/05-
03SAS7PR.mspx; Press Release, Microsoft, Microsoft Acquires aQuantive (May 18, 2007), http://advertising.mi-
crosoft.com/uk/microsoft-aquantive-announcement.
47
The history of software platforms shows that firms can start out vertically integrated by providing both the soft-
ware platform and its key complement such as “hardware, applications, and peripherals.” See David S. Evans,
Andrei Hagiu & Richard Schmalensee, Invisible Engines: How Software Platforms Drive Innovation
and Transform Industries 301 (2006) [hereinafter Invisible Engines] (explaining that the PalmPilot began as
an integrated bundle).
48
See Invisible Engines, supra note 47, at 301 (“[I]n video games both Sony and Microsoft acquired several high-
profile game developers before releasing their consoles so that both end users and other game developers could reli-
ably expect their consoles to have a number of high-quality games.”).
PLATFORM ECONOMICS:
146 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Finally, a two-sided platform may compete with a series of businesses that, when combined,
provide similar services to the same two groups of customers. For example, American Express and
Visa compete for cardholders and merchants. American Express is an integrated two-sided plat-
form that performs all of the necessary steps for meeting cardholder and merchant demand. By
contrast, Visa is a collection of separate entities that—when combined—perform services similar
to those provided by American Express. Working from the merchant level to the cardholder level,
there are acquirers that contract to take card receivables, merchant processors that handle the ac-
tual data processing and switching services, Visa that stands at the center of this platform as the
switch for transactions, card processors that handle data processing for banks, and banks that issue
cards to consumers.
Each of these market constellations results in different forms of competitive constraints and
possible counter strategies from rivals. The market boundaries that separate relevant from irrelevant
constraints are particularly complex. That fact cautions against placing great weight on mechanical
approaches to market definition that cannot account for competitive nuances and complexities.
A natural way to map out the contours of the market is to identify the groups of customers
served by the subject of the inquiry and its likely rivals, and then identify the various businesses
that serve these customers. The digital media business provides an example. Suppose one needs to
define the relevant product market for a media player such as Adobe’s Flash. As of 2007, Adobe
has three groups of customers— end users who receive its player for free; enterprises that pay for
its server; and content providers that use its tools to create content.49 These three groups of cus-
tomers are served to various degrees by (1) Apple’s iPod/iTunes, a vertically integrated platform
with its own media player; (2) Microsoft’s Windows Media Player technologies included both
in its client and server software; (3) RealPlayer, a stand-alone product for clients and servers and
the base for a media store that competes with iTunes; (4) software vendor Autodesk that provides
tools to content providers to make their content available in various formats including Windows
Media, QuickTime, or MPEG-2; and (5) various other businesses including Sorenson, Canopus,
and others.50 Not all these entities will necessarily be part of a properly defined relevant market.
However, to define relevant markets in cases involving two-sided platforms, one must consider
multiple types of potential competitors that may constrain the pricing and other strategies of the
subject of the inquiry.
The price-cost margin, also known as the Lerner Index of market power, is often used in merg-
er analysis.52 The Lerner Index equals “the difference between price and marginal cost” expressed as
a percentage of price.53 When a firm has maximized its profits, the Lerner Index equals the inverse
of the elasticity of demand.54 The elasticity of demand “is defined as the percentage change in
quantity that results from a 1 percent change in price.”55 For example, a 50-percent profit margin
corresponds to an elasticity of demand of two.56
If the elasticity of demand is very high (a large negative number), then the [demand] curve is
said to be very elastic. With a very elastic demand, a small price change induces a very large
change in the quantity demanded. If the elasticity is low (a number between –1 and 0), the
demand curve is inelastic, and a price change of 1 percent has relatively little effect on the
quantity demanded.57
The Lerner Index—calculated in this way for a single product—must be considerably modified
for use with two-sided platforms.58 This section explains why the Lerner Index and other tradition-
al merger analysis techniques must be modified and describes the two-sided version of these tests.
52
See Carlton & Perloff, supra note 24, at 93.
53
Id.
54
See id. at 92-93 (“[T]he price-cost margin depends only on the elasticity of demand the monopoly faces.”). To
maximize profits, a firm expands output to the point at which “the extra revenue from selling one more unit just
equals the extra cost of producing that last unit of output. That is, profit is maximized where marginal revenue equals
marginal cost.” Id. at 91 (emphasis in original). When incremental revenue is greater than incremental cost, “total
revenue increases as output expands,” but when incremental revenue is less than incremental cost, “total revenue
falls as output expands.” Id. at 90. When profits are maximized, the firm’s percentage profit margin is inversely pro-
portional to the elasticity of demand for the firm’s product. See id. at 92-93 (“The higher the elasticity of demand,
the closer is the monopoly price to the competitive price. Therefore, the key element in an investigation of market
power is the price elasticity of demand. Where the elasticity of demand is relatively inelastic, a monopoly markup
may be substantial[.]”). The Lerner Index is based on various assumptions that may not hold true in practice and,
as shown below, do not in fact hold true in the case of multisided platforms.
55
Id. at 92. When marginal and average variable costs are equal, the Lerner Index equals the contribution margin—
gross margin (revenue minus variable cost) expressed as a percentage of revenue—which is often available from
reported accounting data. See Evans & Noel, supra note 1, at 686.
56
If the profit margin is 50% or ½, its inverse is one divided by ½, which equals 2.
57
See Carlton & Perloff, supra note 24, at 92 (emphases in original).
58
See Evans & Noel, supra note 1, at 676 (“The one-sided Lerner Index is incorrect for MSPs [multisided platforms]
and relying on it significantly overestimates the true short-run own-price elasticity of demand and over-states Actual
Loss.”). There is no necessary theoretical relationship between marginal revenue and marginal cost on each side in
isolation for the same reason there is no necessary relationship between price and marginal cost. See Evans & Noel,
supra note 1, at 666. For many two-sided platforms, marginal revenue exceeds marginal cost on one side, but is less
than marginal cost on the other side. It is possible to extend the single-sided Lerner Index to an analogous concept
for two-sided platforms. See Evans & Noel, supra note 1, at 663 passim (exploring approaches).
PLATFORM ECONOMICS:
148 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
[t]he price increase question . . . for a hypothetical monopolist controlling the expanded
product group. In performing successive iterations of the price increase test, the hypotheti-
cal monopolist will be assumed to pursue maximum profits in deciding whether to raise the
prices of any or all of the additional products under its control. This process will continue
until a group of products is identified such that a hypothetical monopolist over that group
of products would profitably impose at least a “small but significant and nontransitory” in-
crease, including the price of a product of one of the merging firms. The [agencies] generally
will consider the relevant product market to be the smallest group of products that satisfies
this test.62
In most cases, the agencies “will use a price increase of five percent lasting for the foreseeable
future,” but may “use a price increase that is larger or smaller than five percent” depending on the
industry involved.63 In one-sided context, the agencies apply the SSNIP test through calculations
based on the Lerner Index or through direct statistical estimation of demand relationships.64
A two-sided platform “would like to find the prices for each side that maximize its profits.”65 As
it adds customers on one side through, for example, an acquisition of a competing firm, it would
ordinarily adjust prices to both sides. The standard one-sided SSNIP test would ignore the prices
for the two-sided platform’s other product, as well as the feedback effects between the two sides.66
59
This section compares two-sided and one-sided SSNIP tests. The one-sided SSNIP test’s recognized deficiencies are
also true in the two-sided context. See Chapter 1 for a full discussion of one-sided SSNIP test.
60
See Merger Guidelines, supra note 21, § 1.11.
61
Id.
62
Id.
63
Id.
64
See Chapter 1 of this volume for a discussion.
65
Evans & Noel, supra note 1, at 665.
66
See id. at 667.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 149
The application of a one-sided SSNIP test in cases involving two-sided platforms can result in a
“relevant market” that includes just one side of a two-sided platform, even though that other side
imposes significant constraints that would temper a price increase.67 It would place key aspects
of profit maximization and strategic interactions outside the “relevant market.”68 “[T]he market
would be drawn too narrowly and estimates of market concentration too high, because the stand-
ard approach fails to consider the tempering effects on price coming from the other side.”69
To illustrate this effect, consider a merger between two heterosexual dating clubs. The standard
one-sided SSNIP test would focus on one side of the platform, ignoring the other side and result-
ing in an overly narrow market definition.70 Using the standard one-sided SSNIP test, the change
in profits of the hypothetical monopolist in response to an increase of the admission fee for men
has two components: the monopolist gains from higher admission prices, but loses from the lower
volume of sales of dating services to men. However, the feedback effects between the two sides
amplify the monopolist’s losses from the decrease in the demand for dating services from men.71
As fewer men access the club, fewer women would be interested in joining the club.72 The decrease
in the number of women would in turn reduce the number of men, which would lead to a further
decrease in the number of women, and so on.73 A price increase deemed profitable under the one-
sided SSNIP test may turn out to be unprofitable under the two-sided SSNIP test.74
A two-sided extension of the SSNIP test could address these issues at least in theory. The ana-
lyst would start with the two-sided platform at issue and add its closest substitutes.75 The analyst
would then ask whether the combined platforms could increase the weighted average price by
a given percentage and thereby reduce the weighted average output of the platforms.76 At each
step, the hypothetical monopolist would be expected to “reoptimize prices across sides and across
platforms,” considering the feedback effects.77 Additional platforms can be added. This approach
becomes more difficult to implement when a two-sided platform faces diverse competitors with
different business models.78
67
Id.
68
See id.
69
Id.
70
See id.
71
See Evans & Schmalensee, Industrial Organization, supra note 1, at 159.
72
See id.
73
See id.
74
See Evans & Noel, supra note 1, at 669-70 (“Because the analyst’s estimate does not account for feedback effects, the
full impact of the price increase on demand is underestimated. As a result, antitrust markets necessarily will be de-
fined too narrowly, and merger analysis will overstate the increase in market power of merging parties and overstate
the predicted unilateral price effects of the transaction.”).
75
Evans & Noel, supra note 1, at 669-78.
76
Evans & Noel, supra note 1, at 669-78.
77
See Evans & Noel, supra note 1, at 674.
78
Evans & Noel, supra note 1, at 669-78.
PLATFORM ECONOMICS:
150 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
There are “many reasons to be weary of mechanical market definition exercises such as the
SSNIP test,” which “do not go away”—and in some cases are exacerbated—when two-sided plat-
forms are involved.79 For example, should one include both sides of the [two-sided platform]
business . . . in the market definition of just one side? The answer can be difficult and depends on
the situation at hand.”80 In discussing the SSNIP test, the Merger Guidelines refer to “adding or
removing ‘substitute products.’”81
The difficulty is that the sides of a given [two-sided platform] represent highly complemen-
tary products, and pricing decisions on both sides critically affect the [platform’s] profitabil-
ity. Yet the SSNIP test is silent on handling complementary products. We believe if the two
sides are very highly complementary and closely linked—for example, if the MSPs facilitates
transactions between the groups that occur in fixed proportions—and MSPs in an industry
all tend to serve the same two sides, then it can be reasonable to include both sides in the
market definition and the “transaction” as the product. However, in other industries MSPs
may all cater to the same side A customers but cater to very different kinds of side B custom-
ers. If the antitrust concern centered around the side A business, then including both sides
of all MSPs that share the A side in the market definition would open a Pandora’s box of
unrelated “B” types that make no sense under a single coherent market definition. Then
the market may need to be defined on the basis of side A only, but with the critical under-
standing that the B sides are an important constraint on behavior and that the formulas
presented here must be used to account for its influence.82
Another issue with the SSNIP test is that “we would expect a hypothetical monopolist of
several platforms to reoptimize prices across sides and across platforms just as we would expect
a hypothetical monopolist to reoptimize across products in a one-sided world.”83 The Merger
Guidelines suggest that the price increase deemed significant for market definition purposes will
generally be uniform across all products within the definition (for example, 5 percent), and “the
terms of sale of all other products [outside the proposed antitrust market] are held constant”
(that is, 0 percent price increase). In the case of [two-sided platforms], one could instead imag-
ine price increases that differ across sides or platforms, or consider a kind of 5 percent quantity-
weighted “average” price increase across sides, allowing the hypothetical platform monopolist to
reoptimize relative prices.84
79
See Evans & Noel, supra note 1, at 674.
80
Id.
81
Id.
82
Id.
83
Id.
84
Id.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 151
Critical loss is based on the incremental profit margin, while actual loss is based on the respon-
siveness of demand to changes in price of the product itself as well as that of substitutes.87 This
method “has won significant appeal both because of its simplicity and because of its easy measure-
ment of inputs” based on available data.88
A common back-of-the-envelope calculation for merger inquiries uses the Lerner Index to es-
timate the elasticity of demand for actual loss.89 However, the traditional Lerner Index for one side
of a two-sided platform does not consider the role of the other side of the platform in constraining
price.90 Suppose the analyst observes a profit margin of 25 percent. For a one-sided business, this
profit margin would imply that the elasticity of demand facing the firm was four.91 One could infer
85
See Evans & Noel, supra note 1, at 668; see also Barry Harris & Cento G. Veljanovski, Critical Loss Analysis: Its
Growing Use in Competition Law, 5 E.C.L.R. 213 passim (2003) (discussing the use of critical loss analysis in anti-
trust law); James Langenfeld & Wenqing Li, Critical Loss Analysis in Evaluating Mergers, 46 (2) Antitrust Bull.
300 (Summer 2001) (“The antitrust agencies and economists have increasingly focused on concepts of “critical
elasticity of demand” or “critical loss analysis” to define antitrust markets and evaluate the competitive effects of pro-
posed mergers.”). See generally Barry Harris & Joseph J. Simons, Focusing Market Definition: How Much Substitution
is Necessary?, 12 Res. L. & Econ. 207 (1989) (discussing the use of the critical loss concept “in practice”); Michael
L. Katz & Carl Shapiro, Critical Loss: Let’s Tell the Whole Story, 17 Antitrust 49 (Spring 2003) (discussing market
definition using critical loss analysis); Daniel P. O’Brien & Abraham L. Wickelgren, A Critical Analysis of Critical
Loss Analysis, 71 Antitrust L. J. 161 (2003) (explaining economics of critical loss).
86
Evans & Noel, supra note 1, at 668.
87
See id.
88
Id. at 669.
89
See Chapter 1 of this book.
90
Evans & Noel, supra note 1, at 666; see id. at 676 (“The one-sided Lerner Index is incorrect for [two-sided plat-
forms] and relying on it significantly overestimates the true short-run own price-elasticity of demand and overstates
Actual Loss.”).
91
If the elasticity of demand is the inverse of the profit margin, and the profit margin is 25%, then the elasticity of
demand is the inverse of 25%, 1/.25, or 4.
PLATFORM ECONOMICS:
152 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
from this relatively high elasticity of demand that significant substitutes constrain the ability of the
firm to increase prices.
For a two-sided business, a 25-percent profit margin would be consistent with less elastic
demand, because the two-sided business would also have to consider the effect of raising prices
on revenue from the other side and the resulting feedback effects. Thus a demand elasticity of less
than four plus positive feedback effects would result in the 25-percent profit margin. The use of
the Lerner Index will tend to overstate the elasticity of demand and understate the extent to which
a firm—or the putative hypothetical monopolist—could raise prices.92 All else equal, this mistake
results in underestimating the market power on one side of the two-sided platform and defining
markets too broadly on that side.93
It is possible to extend the simple version of actual loss analysis to two-sided platforms. The
analyst would start with the two-sided platform at issue and consider substitutes.94 The analyst
would have to estimate the feedbacks between the two sides in addition to the profit margins.95
Table 1 provides examples of the impact of feedback effects on actual loss estimation in cases in-
volving many competing two-sided platforms.96 It shows that the decision to expand or shrink the
market definition depends on the assumed feedback effects between side A and side B (i.e., side
A’s effect on side B and side B’s effect on side A). A decrease in the number of customers on side A
as a result of the price increase on side A causes a decrease in the number of customers on side B,
which in turn reduces the demand on side A.97
92
Evans & Noel, supra note 1, at 676.
93
Evans & Noel, supra note 1, at 670, 689. One should not, however, lose sight of the most fundamental mistake:
leaving out the second side of a two-sided market altogether.
94
Evans & Noel, supra note 1, at 669-78.
95
Evans & Noel, supra note 1, at 669-78. In performing a preliminary calculation based on readily observed data, the
analyst might have little basis to estimate the magnitude of the feedback effects.
96
To simplify the computation of the Critical Loss of output on side A, we assume that the two sides of the hypotheti-
cal monopoly platform are tied in fixed proportions. An example of such a platform is a transaction market like
credit card services where a transaction takes place between a customer on side A (card holder) and a customer on
side B (merchant). With the assumed level of profit margins of 25% on both sides, the critical loss in output on side
A is 17%.
97
See Evans & Schmalensee, Industrial Organization, supra note 1, at 159.
98
See Chapter 1 of this volume.
99
See Chapter 1 of this volume.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 153
for the product in question would be too small, because the models would not account for the
feedback effects from the other side of the platforms.100 These estimates will tend to yield an overly
narrow one-sided relevant market.101
It is possible, however, to conduct a statistical analysis of demand that recognizes the two-sided
nature of the market. Several empirical studies have done so.102 For example, Rysman studies the
feedback loop in the market for yellow pages.103 He estimates “two demand curves simultane-
ously”: . . . a consumer demand for directory usage as a function of [the demand for] advertising”
and “an inverse advertiser demand curve for advertising as a function of consumer usage and the
100
Evans & Noel, supra note 1, at 669-670.
101
Id.
102
See, e.g., Elena Argentesi & Lapo Filistrucchi, Estimating Market Power in a Two-Sided Market: The Case of Newspapers,
22 J. Appl. Econ. 1247 (2007) (analyzing network effects between readers and advertisers in the Italian magazine
industry with an estimation of market power); Pierre Dubois, Adriana Hernandez-Perez & Marc Ivaldi, The Market
of Academic Journals: Empirical Evidence from Data on French Libraries, 5 J. Eur. Econ. Ass’n 300 (April-May 2007)
(linking readers and authors in the academic publishing industry); Ulrich Kaiser & Julian Wright, Price Structure in
Two-Sided Markets: Evidence from the Magazine Industry, 24 Int’l J. Indus. Org. 1 (2006) (linking readers and adver-
tisers in the German magazine industry); Mark Rysman, Competition Between Networks: A Study of the Market for Yellow
Pages, 71 Rev. Econ. Stud. 483 (2004) [hereinafter Rysman, Competition] (establishing a link between readers and
advertisers in the yellow pages market); Mark Rysman, An Empirical Analysis of Payment Card Usage 55 J. Indus. Econ.
1 (2007) (linking cardholders and merchants in the payment card industry); see also Elena Argentesi & Marc Ivaldi,
Market Definition in Printed Media Industry: Theory and Practice (CEPR Discussion Paper No. 5096, 2005) [hereinafter
Argentesi & Ivaldi] (summarizing antitrust cases involving market definition in (two-sided) media industries).
103
Rysman, Competition, supra note 102, at 483.
PLATFORM ECONOMICS:
154 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
quantity of advertising.”104 He finds that consumer use of a directory “increases in the directory’s
level of advertising.”105 At the same time, retailer demand for advertising in a directory increases
“in the amount that consumers use the directory.”106 These results imply the existence of indirect
network effects between the two sides.107
There are two possible sources of revenue in the VISA system: the cardholders and the mer-
chants. As a practical matter, the card-issuing and merchant-signing members have a mutu-
ally dependent relationship. If the revenue produced by the cardholders is insufficient to
cover the card issuers’ costs, the service will be cut back or eliminated. The result would be
a decline in card use and a concomitant reduction in merchant-signing banks’ revenues.
In short, the cardholder cannot use his card unless the merchant accepts it and the mer-
chant cannot accept the card unless the cardholder uses one. Hence, the [interchange fee]
accompanies “the coordination of other productive or distributive efforts of the parties”
that is “capable of increasing the integration’s efficiency and no broader than required for
that purpose.”111
The court of appeals upheld the district court’s conclusion that the relevant market was “all
payment devices,” noting that the market was the same, whether viewed from the perspective
of the plaintiff (processing agent), the defendant (credit card system operator), or consumers
(cardholders).112
104
Id.
105
Id.
106
Id.
107
Id.
108
345 U.S. 594, 610-11 (1953).
109
779 F.2d 592 (11th Cir. 1986).
110
See id. at 600-02.
111
Id. at 602 (quoting Robert H. Bork, The Rule of Reason and the Per Se Concept, 75 YALE L.J. 373, 474 (1966)).
112
Id. at 604 & n.20.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 155
In United States v. First Data Corp.,113 the U.S. Department of Justice (DOJ) sued to block a
merger of First Data Corp. (First Data) and Concord EFS, Inc. (Concord).114 First Data controlled
NYCE Corporation, “the nation’s third-largest PIN debit network.”115 Concord operated “STAR,
the nation’s largest PIN debit network” that handled “approximately half of all PIN debit transac-
tions in the United States.”116
Debit card systems serve consumers who use cards for payment and merchants that accept card
payments.117
To execute a PIN debit transaction, a customer swipes a debit card at a [point of sale (POS)]
terminal and enters a PIN on a numeric keypad. After the PIN is entered, the POS termi-
nal transmits the transaction and bank card information to a “merchant processor,” which
acts as a conduit between the merchant and the various PIN debit networks. The merchant
processor sends the information to the appropriate PIN debit network, which switches the
transaction to the issuing bank’s “card processor.” The card processor accesses the bank’s
account database to verify the PIN and ensure that the customer has sufficient funds to pay
for the purchase. The card processor sends an electronic message to the PIN debit network
accepting or rejecting the transaction. The PIN debit network switches this reply back to
the merchant through the merchant processor to complete the transaction. The entire au-
thorization process takes place electronically in just seconds. At the same time, the merchant
acquirer “purchases” the transaction from the merchant, guaranteeing payment and facili-
tating settlement of the transaction.118
Many debit cards “can also execute ‘signature’ debit transactions in addition to PIN debit
transactions. Signature debit transactions are authenticated like credit card transactions,” except
the customer signs for identification at the merchant POS instead of entering a PIN.119
The DOJ defined the relevant market as “the provision of PIN debit network services.”120 The
DOJ explained that—compared to signature debit networks—PIN debit networks “offer a num-
ber of substantial advantages to consumers and merchants.”121
113
2004 U.S. Dist. LEXIS 13590 (D.D.C. 2004).
114
United States v. First Data Corp., No. 1:03CV02169 (D.D.C. 2003) (verified complaint).
115
Id. ¶ 3.
116
Id.
117
Id. ¶ 16.
118
Id. ¶ 17.
119
Id. ¶ 22.
120
Id. ¶ 25.
121
Id. ¶ 23.
PLATFORM ECONOMICS:
156 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
PIN debit networks are generally considerably less expensive to merchants than signature
debit networks, due to significantly lower interchange rates. PIN debit networks also pro-
vide a more secure method of payment than signature debit because it is much easier to
forge a person’s signature than to obtain an individual’s PIN; consequently, fraud rates for
PIN debit are substantially lower than for signature debit. Because of the increased security
of PIN debit, there is no need for the complicated and expensive charge-back procedures
that allow consumers to challenge signature debit transactions, thereby saving merchants
additional time and money. PIN debit transactions also settle instantaneously, guaranteeing
the merchant ready access to its receipts, whereas signature debit transactions usually take
a day or two to settle. Finally, PIN debit networks allow for faster execution than signature
debit networks. With a PIN debit transaction, customers can enter their PIN as soon as the
first product is scanned. By contrast, customers cannot sign for signature debit transactions
until after the entire order is totaled, prolonging the checkout process.122
In addition, unlike signature debit networks, PIN debit networks allow customers “to receive
cash back at the register when making a purchase.”123 Finally, customers using a PIN debit network
benefit from “the additional security provided by PIN verification as opposed to signature.”124
The DOJ recognized PIN debit networks as two-sided platforms:
The PIN debit network services market is characterized by significant network effects.
Financial institutions are more likely to join networks that are accepted by many merchants.
Conversely, merchants are more likely to accept networks that have many large financial in-
stitutions as members because the value of a particular PIN debit network depends in great
measure on the breadth of its acceptance and use.125
The DOJ asserted that the hypothetical monopolist test was appropriate in a two-sided market.
It explained: “There is no legal or economic support for the notion that the hypothetical monopo-
list test should be discarded simply because the PIN debit market is two-sided in nature . . . .”126
It concluded that “the two-sided nature of the PIN debit market does not limit the ability of a
network to profitably impose a moderate price increase on merchants.”127
122
Id.
123
Id. ¶ 24.
124
Id.
125
Id. ¶ 21.
126
United States v. First Data Corp., No. 1:03CV02169 (D.D.C. 2003) (plaintiff’s pretrial brief ).
127
Id. § V.C.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 157
While the [DOJ] applied a separate SSNIP test for both merchants and issuers, it expressly
accounted for the two-sided nature of payment network markets in its analysis. When the
[DOJ] analyzed the impact of a SSNIP on merchants, it considered the effect of such a price
increase on both merchant and card issuer behavior, including whether the SSNIP would pro-
duce a change in issuer behavior that would render the merchant price increase unprofitable.
Conversely, when the [DOJ] looked at the effect of a SSNIP on issuers, it examined whether
such a price increase might change merchant buying patterns to a sufficient degree to make
the SSNIP unprofitable. In theory, this approach could have produced different results for each
side of the market. The [DOJ’s] complaint focused on the presence of a PIN debit network
services market for merchants because it concluded that merchants were most likely to incur
harm from the merger. The [DOJ] did not state whether it believed that PIN debit network
services was also a relevant product market for issuers, but did state in its pleadings that “most
banks do not perceive PIN and signature debit networks as particularly close substitutes.”128
The government concluded that “[a] 5-10 percent increase in the fees the merchants pay for
PIN debit would not change any of the above . . . . Consequently, the overwhelming majority of
merchants would not reject or discourage customers from executing PIN debit transactions in
response to a moderate increase in the price of the product.”129 The parties settled before trial by
agreeing to divest the NYCE networks.130
In GTE Media Services v. Ameritech Corp.,131 in assessing the competitive effects of the defend-
ants’ conduct in the Internet Yellow Pages segment, the district court recognized the two-sided
nature of yellow pages businesses.132
The commercial nature of the defendants’ websites is revealed by the advertising revenues
the defendants generate when users in . . . interact with their Internet Yellow Pages websites.
Similar to publishers of the traditional, non-Internet Yellow Pages, the defendants directly
earn revenue by selling advertising space to advertisers. Therefore, advertising revenue in
128
Renata B. Hesse & Joshua H. Soven, Defining Relevant Product Markets in Electronic Payment Network Antitrust
Cases, 73 Antitrust L. J. 709, § IV.B.1 (2006); see also Eric Emch & T. Scott Thompson, Market Definition and
Market Power in Payment Card Networks, 5 Rev. Network Econ. 45 (2006) (discussing market definition in pay-
ment card networks markets).
129
United States v. First Data Corp., No. 1:03CV02169 (D.D.C. 2003) (plaintiff’s pretrial brief ). The DOJ
clarified that “consideration of the bank side of the market does not in any way suggest that the hypothetical
monopolist test leads to the wrong result.” Id. § V.C. This is perhaps best seen as an example of where the
five-to-ten percent SSNIP test defines a narrow market that seems at odds with the economics of the business
under consideration.
130
United States v. First Data Corp., 2004 U.S. Dist. LEXIS 13590, at *9-10 (D.D.C. 2004)
131
21 F. Supp. 2d 27 (D.D.C. 1998).
132
Id. at 38-39.
PLATFORM ECONOMICS:
158 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
the Internet Yellow Pages market substantially depends on the number of users accessing a
particular website because Internet advertisers are willing to pay higher advertisement rates
on websites with a higher volume of user traffic.133
The court concluded that the plaintiffs had adequately alleged antitrust violations, including “a
conspiracy to monopolize the national Internet Yellow Pages market.”134
By channeling users away from competing services, such as GTE’s SuperPages, the defend-
ants not only restricted competitors from generating advertising revenue but also secured
profitability for their website services. In this context, it is apparent that the purpose behind
these concerted acts was to “drive . . . rival[s] from the market by exclusionary or predatory
means.”135
In defining markets in the newspaper industry, some competition authorities have focused on the
characteristics of the publications’ readers. In Newspaper Publishing,136 the European Commission
(EC) clarified that different categories of newspapers should not be considered substitutes from
the point of view of buyers of advertising space because they “provide different ‘channels’ through
which to reach different socio-economic groupings of readers.”137 The Italian antitrust authority
took a similar position in Class Editori/Sole 24 Ore,138 holding that advertising in “daily papers
specialized in business and financial information” was a separate product market from advertising
in “newspapers of general information.”139 The Italian antitrust authority concluded that “comple-
mentarities” outweighed “substitutabilities” because of “the different characteristics of readership
between the two types of publications.”140
In at least one case, in analyzing the two-sided market for printed media, the antitrust au-
thority may have failed to “consider[] both sides of the market” when conducting econometric
analysis.141 In a case involving “two publishers or magazines and newspapers in France,” the French
competition authority sought to define the boundaries of the market for weekly magazines and
general information and conducted an econometric analysis to ascertain whether this definition
should include a larger number of titles than the definition used in previous decisions. The objec-
133
Id.
134
Id. at 42-45.
135
Id. at 45 (quoting Ass’n for Intercollegiate Athletics for Women v. NCAA, 735 F.2d 577, 585 (D.C. Cir. 1984)).
136
Case IV/M.423, Newspaper Publ’g (1994).
137
Id. ¶ 16 (noting the “very limited substitutability between the tabloid and quality segments”).
138
Provv. n. 3336, Class Editori / Il Sole 24 Ore (1995).
139
Argentesi & Ivaldi, supra note 102, at 6.
140
Id.
141
Id, at 8-9.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Two-Sided Markets 159
tive was to estimate the cross-elasticities between different titles in order to decide which ones were
to be included in market definition.142
Although the authority’s decision did not report its estimation results, its “very low estimates
of elasticities” could have resulted from “an inadequate econometric specification.”143 The French
authority’s econometric model, which did not take into account “the link between readers’ demand
and advertising demand,” has been criticized as “potentially misspecified” and having potential to
“lead to biased estimations of price coefficient and related elasticities.”144
Google’s acquisition of DoubleClick, cleared after in-depth investigations by the U.S. Federal
Trade Commission, the EC, and other competition authorities,145 involved interesting issues con-
cerning two-sided markets and is an example of a case in which focus on market definition can ob-
scure the dynamics of the business.146 The merger involved online advertising for web publishers.
Google is an integrated advertising platform that connects advertisers to publishers: it “provides
online advertising services for web publishers and advertisers in addition to operating a search-
engine that sells advertising space.”147 Web publishers sell space—and the viewers who see that
space—to advertisers.148 Publishers agree to program space to accept Google ads.149 Advertisers bid
on keywords that lead Google to insert ads on web pages that have those keywords.150 Google, act-
ing as an intermediary, splits the ad revenue with the publisher.151
“DoubleClick is principally a provider of server-based software tools and services for managing
online advertising for web publishers and advertisers.”152 It (1) licenses software and provides servic-
es to publishers to manage advertising on their web pages, and (2) licenses software to advertisers to
manage their advertising campaigns.153 DoubleClick uses certain efficiencies from having publishers
and advertisers use the same software as a selling point.154 Publishers use DoubleClick’s software to
manage advertising they sell through direct sales forces and through advertising networks.155 They
generally do not use the software with Google’s ad platform, which is a complete solution.156 Large
142
Argentesi & Ivaldi, supra note 102, at 8.
143
Id.
144
Id. at 8-9.
145
Evans & Noel, supra note 1, at 682.
146
Id.
147
Evans & Noel, supra note 1, at 682.
148
Id, at 683.
149
Id.
150
Id.
151
Id.
152
Evans & Noel, supra note 1, at 682.
153
See DoubleClick, DART for Publishers, http://www.doubleclick.com/products/dfp/index.aspx; DoubleClick,
DART for Advertisers, http://www.doubleclick.com/products/dfa/index.aspx.
154
Evans & Noel, supra note 1, at 684.
155
Id.
156
Id.
PLATFORM ECONOMICS:
160 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
publishers will typically use DoubleClick or a competing software package for much of their ad
space and will also use Google’s ad platform.157 Small publishers will often just use an integrated ad
platform, such as Google’s.158
Google’s acquisition of DoubleClick involved a complicated multisided market.159 As e xplained
above, both Google and DoubleClick are two-sided platforms. Google operates an integrated plat-
form; its advertising platform is the only thing that stands between the publisher and the advertiser.
By contrast, DoubleClick provides an input into a decentralized platform. A publisher uses a soft-
ware tool and one of several methods of intermediation, including its own sales force or advertising
networks. An advertiser uses a software tool and one of several methods of intermediation, including
buying agents and advertising networks.160 DoubleClick supplies the software inputs to both sides.
Any economic effects of the Google/DoubleClick transaction result from the acquisition by an
integrated platform of inputs into a decentralized platform.161 Key issues are the extent to which
the two platforms are substitutes taking both sides into account.162 If the two types of platforms
are in the same market, the inquiry should focus on the effect of the acquisition of an input into
one platform type on prices and other competitive strategies.163
VI. CONCLUSION
Market definition helps focus antitrust analysis on a relevant set of products and businesses. It
limits the amount of data gathering and provides context for consideration of possible competi-
tive effects. The fact that one or more subjects of the inquiry are two-sided platforms does not
fundamentally alter market definition analysis. However, the interdependence between the two
sides of a platform and the products and businesses relevant to both sides must be considered. The
standard one-sided approaches to market definition will tend to exclude the other side of the plat-
form, p lacing important two-sided considerations outside of the purview of the analysis and define
markets too broadly or too narrowly depending on the technique used. While these approaches
can be adapted for use in cases involving multisided businesses, market definition analysis in such
cases is likely to be more difficult in practice and subject to greater error than in cases involving
only single-sided businesses.
157
For example, Business Week found that “[n]early all of the major Web publishers already use DoubleClick’s services.”
See Robert Hof, Google Launches DoubleClick Ad Exchange, BUSINESS WEEK (Sept. 18, 2009).
158
For example, USA Today found that “[s]mall Web site operators have flocked to [Google’s] AdSense as a way to attract
advertising.” See Jefferson Graham, Google’s AdSense a Bonanza for Some Web Sites, USA TODAY, Mar. 10, 2005.
159
Evans & Noel, supra note 1, at 682.
160
Id, at 684.
161
Evans & Noel, supra note 1, at 685.
162
Id.
163
See Evans & Noel, Industrial Organization, supra note 1, at 682-89 (analyzing unilateral effects of the Google/
DoubleClick transaction using two-sided versions of traditional econometric analyses).
CHAPTER
Defining Markets that Involve
6 Two-Sided Platforms
David S. Evans and Michael D. Noel
ABSTRACT
A multi-sided platform (MSP) serves as an intermediary for two or more groups of customers
who are linked by indirect network effects. Recent research has found that MSPs are significant
in many industries and that some standard economic results—such as the Lerner Index—do not
apply to them, in material ways, without some significant modification to take linkages between
the multiple sides into account. This chapter extends several key tools used for the analysis of
mergers to situations in which one or more of the suppliers are MSPs. It shows that the applica-
tion of traditional tools to mergers involving MSPs results in biases the direction of which de-
pends on the particular tool being used and other conditions. It also extends these tools to the
analysis of the merger of MSPs. The techniques are illustrated with an application to an acquisi-
tion by Google in the online advertising industry.
I. INTRODUCTION
This chapter presents an empirical framework for examining market definition and unilateral ef-
fects in mergers in which one or more of the businesses that may be considered for the hypotheti-
cal market are multi-sided platforms (MSPs). MSPs provide goods or services to several distinct
groups of customers who need each other in some way and who rely on the platform to intermedi-
ate transactions between them (Evans (2003a, b), Rochet and Tirole (2003, 2006)).1 They typically
reduce transaction costs and thereby permit value-creating exchanges to take place that otherwise
would not occur (Evans and Schmalensee (2007a, b)). In particular they facilitate the realization of
indirect network externalities, and externalities in use, between the members of distinct customer
groups (Rochet and Tirole (2003)).
Many old industries are based on MSPs, ranging from village matchmakers that date from
ancient times to advertising-supported newspapers introduced in the 17th century to payment
cards introduced in the mid 20th century. However, an increasing number of significant modern
businesses are MSPs as a result of technological changes that have drastically lowered the costs and
1
Many multi-sided platforms have two primary sides such as advertising and readers, and much of the economic
literature has focused on the case of two-sided platforms (2SPs). We will discuss it as well later in this paper.
161
PLATFORM ECONOMICS:
162 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
increased the benefits of connecting diverse customer groups on a single platform. These include
most internet-based businesses such as eBay, Facebook, and Google. Such businesses are creating
new products and services such as social networking and are disrupting existing industries such as
advertising-supported media (Evans and Schmalensee (2007b)). MSPs are likely to become an in-
creasingly important part of merger and antitrust analysis in the years to come for several reasons.
• With the relative increase in the birth of MSPs more businesses will be based on this rather
than the traditional supply-chain model in which raw materials are turned into products
that are sold to wholesale and then to retail distributors.
• MSPs are prevalent in new industries that are going through consolidations based on merg-
ers and acquisitions which must be reviewed by antitrust authorities. Recent examples in-
clude the proposed merger of Euronext and the London Stock Exchange, and Google’s
acquisition of DoubleClick discussed in more detail later.
• Indirect network effects tend to lead to high levels of concentration (see Evans and Schmalensee
(2007a)) in industries which in turn leads to increased merger and antitrust scrutiny.
• Modern MSPs are sometimes locked in battle with other MSPs for control of an industry
and are disruptive forces in traditional industries. In both cases this leads to competitors be-
ing harmed and complaints being lodged.
• MSPs have business models that are not yet well understood and engage in highly complex
business strategies; unusual practices are suspect practices in our experience (Evans and
Schmalensee (2007b)).2
2
Indeed, some of the most visible antitrust cases around the turn of the 21st century have involved MSPs. These include
the Microsoft cases in the United States (see Microsoft v United States, Supreme Court Docet No. 01-236), the European
Community (see Microsoft v Commission, T-201/04), and Korea (KFTC v Microsoft, decision available at http://ftc.
go.kr/data/hwp/micorsoft_case.pdf), all of which involve various aspects of the Windows software platform; the credit-
card interchange fee cases in Australia(see Visa International Service Association v Reserve Bank Of Australia, N 973 of
2002), the European Union (see European Commission, “Interim Report I Payment Cards”, Apr. 12, 2006), the United
Kingdom (see Office of Fair Trading v MasterCard UK Members Forum Limited, CA98/05/05), the United States (see
United States of America v Visa U.S.A. Inc., Visa International Corp., And MasterCard International Incorporated, 98
Civ. 7076 (BSJ) ), and many other jurisdictions; mobile termination charges cases involving pricing to one side of a two-
sided market that have appeared in the European Community (see Vodafone / Oskar Mobile, Case No. COMP/M.3776;
Cellnet and Vodafone, Monopolies and Mergers Commission, Report to the Director General of Telecommunications,
December 1998; “Decrease of the Price of Fixed-to-mobile Calls”, Autorité de Regulation des Telecommunications,
Press Release, 6 November 2002; Competition Commission (2003), “Vodafone, O2, Orange, and T-Mobile: Reports
on References under Section 13 of the Telecommunications Act of 1984 on the charges made by Vodafone, O2, Orange
and T-Mobile for terminating calls from fixed and mobile networks”); and the U.S. Department of Justice case involving
realtor.com (see United States of America v. National Association of Realtors, Civil Action No. 05 C 5140); and actual
and proposed mergers involving stock exchanges (e.g. NYSE and Euronext; Euronext and London Stock Exchange;
London Stock Exchange and Germany’s Deutsche Börse; London Stock Exchange and NASDAQ).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 163
The standard tools of antitrust and merger analysis, which were developed based on the eco-
nomics of single-sided businesses, do not necessarily apply in ways that are material to the analysis
of competition that involves multi-sided businesses. Each side of the MSP’s business influences and
constrains its strategies on the other side. Antitrust analysis that focuses on one side of the business
in isolation from the other side is incorrect as a matter of economics, and can lead to the wrong
answer when indirect network effects are significant and are relevant for assessing the practice at is-
sue (Evans (2003a), Wright (2004)). This chapter shows how the standard tools used for analyzing
market definition and unilateral effects for mergers need to be modified when the parties are MSPs.
The analysis of market definition and power has obvious extensions to other areas of antitrust.
We present an empirical framework that can be used to handle situations in which one or more
MSPs may be the subject of the merger analysis. Section 2 provides an informal discussion of the
analysis of market definition, market power, and unilateral effects for situations that involve MSPs
and how that analysis differs from that for situations that only involve single-sided firms. Section 3
then presents our formal analysis for the special case of two-sided markets. Section 4 considers a se-
ries of examples and simulations to highlight the benefits of pursuing the correct analysis. Section
5 applies the analysis to analyzing unilateral effects for a particular example: Google’s purchase of
DoubleClick. Section 6 presents brief conclusions.
3
To keep matters simple we consider the case where each side is charged a membership fee as in Armstrong and Wright
(2007). MSPs generally involve platforms on which interactions take place customers face an access fee and a usage
fee although they may choose to make some of those fees zero.
PLATFORM ECONOMICS:
164 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
The platform would like to find the prices for each side that maximize its profits by taking these
considerations into account. As Rochet and Tirole (2003) observe, one can think of these as de-
termining the absolute and relative levels of prices. Three key results hold for two-sided platforms
based only on the assumptions that there are two distinct customer groups, there are positive ex-
ternalities between members of those groups, and a two-sided platform provides a good or service
that facilitates exchange of value between the two customer groups in the face of these externalities:
• First, each optimal price depends on the price elasticities of demand for both sides; the na-
ture and intensity of the indirect network effects between each side; and the marginal costs
that result from changing output of each side. For the special case considered by Rochet and
Tirole (2003) the profit-maximizing prices are given by:
A B B
A
p
5 A c, p 5 A c (1)
1 B 2 1 1 B 2 1
• Second, an increase in marginal cost on one side does not necessarily result in an increase in
price on that side relative to the price of the other side. From (1) we have
p A A
5 (2)
p B B
which means the price ratio between two sides depends only on the ratio of elasticities (and
not inverse elasticities), but not on marginal cost.
• Third, the profit-maximizing price for one side may be below the marginal cost of supply for
that side or even negative. A common situation analyzed by Armstrong (2006) is when the
platform in effect buys A’s who are valued by B’s.
More generally, the relationship between price and cost is complex, and the simple formulas
that have been derived for single-sided markets do not apply.
Several results that are relevant for the analysis of market definition and unilateral effects follow
immediately from these results. In describing these results we distinguish between mergers between
“symmetric” MSPs which are defined as MSPs that serve coincident sides and “asymmetric” MSPs
which are defined as MSPs that do not have at least one side in common.
The widely used Lerner Index
p2c 1
5 (3)
p
where is the usual own-price elasticity of demand, does not accurately summarize the profit-
maximizing equilibrium for MSPs when applied to a particular side. This condition does not
consider the linkage between the two sides and as a result does not reflect the profit-maximizing
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 165
equilibrium condition for a two-sided platform. For the special case considered by Rochet and
Tirole (2003) the two-sided version of the Lerner Index is
p A 1 p B 2 c 1
5 (4)
A
p 1 p B
1 B
A
Merger analyses such as the critical loss and diversion ratio that are based on the one-sided
Lerner Index are therefore not correct when applied to multi-sided businesses.4
Merger simulation models are also misspecified when they fail to consider the multi-sided
nature of the business.5 Consider the standard conditional logit equation that underlies the basic
models for differentiated-product industries:
When applied to one side of a two-sided industry this equation fails to consider the feedback
effects between the two sides—in other words that the demand by side A depends on the number
of customers on side B—and that one must consider the demand by sides A and B simultaneously
to properly account for all the feedback effects.
The SSNIP test for defining a relevant market does not apply without significant modifica-
tions when any of products involved in the analysis are produced by an MSP. Consider the case
of a merger between two symmetric MSPs that serve the same customer groups A and B. To
define the market an analyst proceeds by starting with the merger of the products that serve de-
mand for, say, side A because that is the focus of the competition concern.6 The set of products
is expanded until a hypothetical monopolist over that set of products could raise price by, say,
5 percent or more on each of those products. That set of products then defines the market for
analysis. However, by ignoring side B the analyst fails to consider that the hypothetical price
increase reduces the number of side A customers available to side B, which thereby reduces the
prices that side B customers will pay, and furthermore reduces the number of side B customers
available to side A, which in turn reduces the prices that side A customers will pay. The link
between sides A and B reduces the profitability of any price increase. The market defined by a
one-sided application of the SSNIP test is necessarily drawn too narrowly from an otherwise
4
We discuss this in detail below.
5
See Werden, Gregory J & Froeb, Luke M (1994); Baker, J., and T. Bresnahan (1985); Feldman, R. (1994); and Nevo,
A. (2000).
6
For a discussion of the SSNIP test see US Department of Justice Merger Guidelines available at http://www.usdoj.
gov/atr/public/guidelines/horiz_book/hmg1.html. In particular see Section 1 on Market Definition, Measurement
and Concentration.
PLATFORM ECONOMICS:
166 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
properly conducted single sided approach because it fails to consider the tempering effects on
price coming from the other side.7
The mistake though is more profound. The purpose of market definition is, in part, to help fo-
cus the economic analysis on a relevant but finite set of products and competitive relationships for
analysis. For industries in which the multi-sided effects are sufficiently strong, market definition
that excludes one of the sides of an MSP results in the failure to consider multi-sided strategies and
market linkages. Failure to consider those multi-sided relationships can result in Type I and Type
II errors: failing to recognize practices that may be harmful because of the two-sided relationships
and condemning practices that are innocuous in a two-sided context.
The remainder of this chapter provides the toolkit for conducting merger and unilateral effects
analyses for MSPs.
7
The direction of this bias can be definitively signed when the platforms are symmetric, or alternatively when the price
changes deemed significant (e.g. 5%) are symmetric across products, which is standard in market definition exercises.
The bias can go in either direction when both 1) asymmetric platforms are involved and 2) asymmetric price changes
are considered, as may occur in many merger analyses. Other opposing biases arise from particular calibration tech-
niques commonly used in one-sided analyses. These will be discussed in detail later.
8
See Harris and Simons (1989), O’Brien and Wickelgren (2003) and Katz & Shapiro (2003). See Harris and
Veljanovski (2003) for discussion of the use of critical loss analysis in antitrust proceedings.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 167
in the price of all its products unprofitable – to “Actual Loss” (AL) – the predicted percentage loss
in quantity that the monopolist would suffer if it did increase prices on all its products by X % .
With symmetric one-sided firms, the well known formulas are given by CL 5 X /( X 1 M ) and
AL 5 X (OWN 2 CROSS ), where M is the percentage markup, and OWN and CROSS are the own and
cross price elasticities respectively. The relevant market is found when Actual Loss equals Critical
Loss for a hypothetical monopolist of the given set of products in the proposed antitrust market.
If Actual Loss exceeds Critical Loss, the relevant market is expanded to include more substitutes.
Otherwise, it is contracted.
Critical Loss Analysis has also been used to estimate unilateral price effects from proposed
mergers. In this context, one inquires whether the merger of the parties would lead the merged
firm to raise prices by X % or more. If the test shows that Actual Loss exceeds Critical Loss, the
merged entity would not find an X % price increase in all its products profitable. Otherwise, it
would find it profitable. This provides only a first look at whether there are significant unilateral
effects and more sophisticated techniques would ordinarily be used to confirm this initial finding.
The one-sided calculations are trivial, and the technique has won widespread appeal both be-
cause of its simplicity and easy measurement of inputs.9
9
One could also calculate and compare profit totals directly across hypothetical and real worlds. The sum of firm
profits before is 5 2Mpq, and the joint profit after is i 5 2pq(M 1 X)(1 1 X(OWN 2 CROSS )). We, however,
maintain the Critical Loss formulation throughout the paper because of its appeal, its transparency (over black box
profit comparisons), and its tangible metric – namely, how many sales a firm can afford to lose when it raises price.
PLATFORM ECONOMICS:
168 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
it takes for customers who experience a price increase directly to respond, but before any feedback
effects commence. (The feedback effects, we know, will cause additional indirect responses over time
as relative and absolute platform sizes change.)10 Since the analyst’s estimate does not account for
feedback effects, the full impact of the price increase on demand is underestimated. As a result, anti-
trust markets will be defined too narrowly, and merger analysis will overstate the increase in market
power of merging parties and overstate the predicted unilateral price effects of the transaction.
Lerner Bias. An opposite bias can occur when the analyst uses observed markups to calibrate
the own elasticity of demand based on the one-sided Lerner Index. This bias overstates the true
short-run own-price elasticity of demand. In fact, it yields an elasticity estimate that is even larger
than the true long-run own-price elasticity of demand. We define long-run to be as long as neces-
sary for all meaningful feedback effects to be worked out, from the consequent changes in platform
size.11 In this case, the resulting market definition would be too broad and predicted unilateral ef-
fects of a transaction would be too small.
The situation is more complex with asymmetric platforms. When the test considers equal price
changes (e.g. 5%) across all of the hypothetical monopolist’s products, the biases are as above. However,
if the asymmetry is significant and if the analyst allows for differential price changes across products,
the direction of the overall bias depends on the analytical method and also the relative and absolute
sizes of the price increase(s) considered. Differential price changes are more likely to be a consideration
in the analysis of unilateral effects, where antitrust concern may center around particular products,
rather than in market definition analyses, where the price increases deemed to be significant are gener-
ally taken as uniform (e.g. 5% or 10%) under the SSNIP test. Imagine a merger between a low-margin
and high-margin (in absolute terms) platform and imagine there is concern that the prices will increase
at the low margin platform only. Given properly estimated short-run elasticities, the one-sided calcula-
tion underestimates the loss on the low margin platform, but misses entirely the positive gain to the
high-margin platform. If enough customers would switch to the high-margin platform, the one-sided
calculation understates the unilateral price effects of the transaction. This possibility turns out to be
significant in our discussion of the Google-DoubleClick merger in Section V.
Our exposition returns to the symmetric case. Call the platforms firm 1 and firm 2, and
the two groups of customers A and B. Note that platform symmetry implies P A 5 P1A 5 P2A
and P B 5 P1B 5 P2B ; we do not assume symmetry across the two sides, P A P B, which are
typically quite different in practice. Also, define total quantity across platforms on side s as
Q S 5 Q 1S 1 Q 2S.
10
In practice, most likely the analyst will estimate some unknown mix of a short-run and long-run elasticity (i.e. once
all feedback effects have been worked out), since indirect network effects from platform size changes are not properly
accounted for. The direction of bias is the same.
11
Our definitions of “short” and “long” are defined to be the periods in which the direct price responses and m eaningful
indirect price effects take place, respectively. They do not relate to calendar time and are different from the concepts
of short- and long-run (1/2 one or two years) generally used by antitrust practitioners.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 169
Imagine that a hypothetical monopolist of both two-sided platforms wished to increase prices
to the consumers on side A by ∆P A and to consumers on side B by ∆P B. Similar to the one-sided
case, the gain on inframarginal sales due to the price increase on side A would be:
However, unlike the one-sided case, the marginal loss due to the price increase on side A now
consists of two components:
the loss on side A and the loss on side B due to the price increase on A.12 Here, marginal costs
are denoted by C A and C B , and ∆Q s represents the loss in total quantity on side s over the time
period that is relevant for the merger review (typically one or two years). Similar equations are
derived for the price increase on side B. Summing up and equating the gains to the losses, we get
the Two-Sided Critical Loss Formula in response to changes in the prices on side A and B of X A %
and X B % respectively:
s 5 A,B
s s
R (X 1
( )
M s )
DQ s
Q s
1 s
R X s
50
where R s 5 Q s P s, the revenue earned from side s. Critical Loss is the set of percentage quan-
tity reductions on side A, DQ A / Q A , and side B, DQ B / Q B , that would leave the hypothetical
monopolist profits unchanged.13
The Actual Loss to the hypothetical MSP monopolist depends on several factors. As with
one-sided firms, Actual Loss depends on the (short-run) own-price and cross-price elasticities of
demand. Higher own price elasticities tend to increase Actual Loss since a price increase at a
given platform results in relatively more customers switching away from that platform. On the
other hand, higher (short-run) cross price elasticities between the monopolist’s platforms tend to
decrease Actual Loss because relatively more of the customers that switched away from one plat-
form can be recaptured with the monopolist’s other platform.
12
The merger guidelines state that when determining the relevant antitrust market “the terms of sale of all other
products [outside the proposed antitrust market] are held constant”. In general, we would expect a hypothetical
monopolist to adjust prices on side B as well as on side A. This and subsequent formulas account for both cases.
13
A special case is when the two sides are tied together in a fixed proportion, e.g. a transaction market like credit card
services, where a transaction takes place between a customer on side A (a card holder) and a customer on side B
(a merchant). Suppose the proposed antitrust market is the transaction, which by definition includes both sides.
Given Q 5 Q A 5 Q B and X A 5 X B 5 5% , equation (8) collapses to: |DQ / Q | 5 X /( X 1 M AB ) where
M AB 5 (P A 1 P B 2 C A 2 C B ) /(P A 1 P B )) . This is the familiar one-sided Critical Loss formula where the “prod-
uct” is the transaction with a composite price of p A 1 p B.
PLATFORM ECONOMICS:
170 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Unlike one-sided firms, Actual Loss also depends on the strength of the indirect network
externalities that customer groups provide to one another. Recall that an MSP is only successful
because it is able to bring two distinct customer groups together in significant numbers. When an
MSP increases its price on side s for platform i, there is the usual contraction in demand on side
s, as with one-sided firms. But now because there are fewer side s customers, the platform is less
valuable to side r customers. This causes a contraction on side r as well. The feedback effects take
over, causing another contraction of s side, then r side again, and so on. The stronger the external-
ity across groups, the greater the demand contractions will be after a price increase and the greater
is the Actual Loss, all else equal.
The exact Actual Loss formula depends on the specific demand form chosen. We consider, as
an example, an isoelastic demand function adapted to include the special features of MSPs.14 Let
qis 5 as qir 2 ds qjr 1 uis where uis 5 ms 2 bs ris 1 gis psj
i j, s r, where qis 5 ln(Qis) is the log quantity demanded by side s customers at platform
i, pis 5 ln(Pis ) is log price to side s on platform i, and parameters as , bs, and ds are non-negative.
Since we assume competing platforms, gs is non-negative as well.15 Given the log-log form, the
b’s and g’s are the usual short-run own and cross price elasticities for side s customers respectively.
Recall that we define short-run to include the time needed for the direct response of customers
impacted by the price increase but too short to include any of the indirect network feedback effects
that follow. The indirect network externalities (the marginal value one side puts on the presence of
the other at a platform) are captured by as, the within-platform cross-side externality, and ds , the
cross-platform cross-side externality.16 The as , for example, is the percentage change in quantity
demanded by side s at platform i in response to a one percent change in the quantity of platform i’s
side r customers. We assume as . ds, bs . gs, and as 1 ds , 1 to ensure stability of the system.17
Solving for the reduced form equations of the qis ’s (as a function only of prices), and taking
the total derivative of each qis , we derive the formula for Actual Loss to side A customers after price
increases to side A and side B of X A % and X B % respectively:
A
(gA 2 bA) X A 1 (aA 2 dA)(gB 2 bB )X B
L 5 (9)
1 2 (aA 2 dA)(aB 2 dB)
14
The results of the article carry through to other commonly used demand formulations.
15
This demand system can be derived from a standard exogenous differentiated goods approach with an additional
term included that is increasing in the consumption of consumers on the other side.
16
The same formulas apply whether firms multi-home or single-home. While it does not change the structure of the
demand system, it would be expected to impact estimates of the parameters.
17
The first condition ensures demand on the other side of both platforms does not increase after a price increase and
demand contraction at both platforms on the first side. The second ensures demand does not increase on both sides
after a price increase on one side by both platforms. The third prevents exploding demand following a price increase
on one side by one platform.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 171
LA is the percentage Actual Loss on side A at each platform and at both platforms collectively
LA 5 DqiA 5 Dln Q A 5 DQ A / Q A (since platforms are symmetric), once all feedback effects have
been worked out. The formula for the Actual Loss on side B is similar.
We see the two-sided Actual Loss formula for side A is a generalization of the one-sided for-
mula. When all cross-side externalities are zero, so that each group’s demands do not depend on the
numbers of customers on the other side at either platform, the formulas for LA and LB coincide with
their one-sided counterparts. When two-sided externalities are present, however, it is necessarily the
case with symmetric platforms that the correct two-sided formulas give larger Actual Loss estimates
than the one-sided versions, given otherwise unbiased estimates of the short-run price elasticities.
There are two reasons for the “Estimation Bias” here. First, an increase in price on side A alone
causes not only a reduction in demand on side A but also a reduction in demand on side B, since
the B group values a platform less when there are fewer A’s. So, unlike the one-sided case, there are
losses on both sides (LA , 0 and LB , 0).
Second, and importantly, there is a multiplier effect in two-sided industries that magnifies
the immediate loss on side A and on side B over time. As discussed above, when side A contracts
after its price increases, the platform is less valuable to its B customers and the B side contracts.
But fewer B’s decreases the value of the platform to the A customers and now the A side contracts
further. Both sides contract in turn, and as a result the Actual Losses are larger than they would be
in a one-sided world where absolute and relative platform size does not matter.
The multiplier effect can be seen in the denominator of the two-sided Actual Loss formulas.
The denominator is necessarily less than one, so the two-sided Actual Loss formulas necessarily
yield larger estimates of loss than the one-sided formulas, for a given set of parameters. The biased
one-sided versions overestimate the hypothetical monopolist’s ability to raise prices.
To calculate whether a given price increase or increases are likely to be profitable, one sim-
ply compares Critical Loss and Actual Loss, by substituting the values of Ls into DQ s / Q s in the
Critical Loss equation (8). If the left hand side of (8) is negative, the price increase(s) will not be
profitable. If positive, the price increase(s) will be profitable.
In merger analyses, the test would be performed using just the set of products or services that
the newly merged firm would control. The merged entity is de facto the hypothetical monopolist
and assumptions about possible cost efficiencies post-merger are easily worked into the analysis.
In market definition exercises this test is repeated many times, each time expanding or contracting
the proposed antitrust market, until a market is found where Critical Loss would equal Actual Loss
for a hypothetical monopolist of all the products in that market. Given a market definition, the
market shares of the “in” firms are calculated and then used as proxies for market power.
There are many reasons to be wary of mechanical market definition exercises such as this
SSNIP test, or in relying on artificially discrete market boundaries.18 However, for better or worse
18
See Carlton (2007) for further discussion.
PLATFORM ECONOMICS:
172 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
the SSNIP test is commonly used, and our purpose here is just to show that the one-sided formulas
are simply wrong when MSPs are involved. The underestimation of Actual Loss means relevant an-
titrust markets defined on the basis one-sided formulas will be narrower than they would be under
the correct two-sided calculations, and estimates of market power too high.
It is generally preferable to estimate all the parameters of the demand system whenever reliable
data exists.19 In practice, data availability is such that often parameters must be calibrated with lim-
ited information. Theoretical restrictions, like the Lerner Index that relate markups and elasticities,
can be used to calibrate parameters that otherwise may be difficult or impossible to estimate. Other
theoretical restrictions, like the Slutsky Symmetry rule that relates cross-price elasticities, can be
used to ensure logical consistency across certain parameters.
The relationship between markups and elasticities is especially important here because markups
appear in the Critical Loss formula and the elasticities appear in the Actual Loss formulas. This re-
lationship should be consistent with economic theory.20 In calibration exercises, it is now common
practice to use the (one-sided) Lerner Index to estimate the own price elasticity from the observed
markup when the latter cannot be estimated from data, perhaps because data are lacking or the
analyst has insufficient time. This turns out to be a problem in two-sided settings. The one-sided
Lerner Index is incorrect for MSPs and relying on it overestimates the true short-run own price
elasticity of demand and overstates Actual Loss.
It is well known in the one-sided case that the percentage markup equals the inverse of the
own price elasticity at the profit maximizing output. In the two-sided case, we can derive the first
order conditions for profit maximization. Each platform i chooses prices on each side to maximize:
→
max
A B
i 5
pi ,pi
q
s 5 A,B
i
s
( p )* ( pis 2 cis ) (10)
19
There are several empirical papers that estimate two-sided demand systems in particular industries in order to estab-
lish the existence of indirect network effects. Although generally not in the context of merger or market definition
analysis, these highlight the importance and challenges of proper demand identification in a two-sided context. For
example, see Rysman (2004) establishing a link between readers and advertisers in the yellow pages market, Rysman
(2007) for cardholders and merchants in the payment card industry, Argentesi and Filistrucchi (2005) for readers
and advertisers in the Italian magazine industry with an estimation of market power, Wright (2004) and Kaiser and
Wright (2006) for linking readers and advertisers in the German magazine industry and Dubois, Hernandez-Perez,
and Ivaldi (2007) linking readers and authors in the academic publishing industry. Argentesi and Ivaldi (2005)
recap antitrust cases involving market definition in (two-sided) media industries. Using limited data for the French
magazine market, they include an estimation of subscriber demand elasticities using the price of advertising as an
instrument, suggesting a two-sided linkage. Elasticity estimates differ with and without the instrument.
20
Some authors disregard the Lerner index as too simple and unrealistic for application in specific real-world antitrust
cases (see Scheffman and Simons (2003) and Harris (2003)). We agree with Katz and Shapiro (2003) that there
should be a presumption of the theoretical markup-elasticity relationship pending evidence to the contrary. It is
possible, however, to estimate both markups and elasticities independently of one another directly from the data and
sidestep these issues.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 173
Davidand
David S. Evans S. Evans andD.Michael
Michael Noel D. Noel
The two first order conditions simplify to:
Risis ε+iisssR+
Ris + Ris M
s
i MRiirs εMii irsε+iisrssRi=M0i rε ii =r 0sr
s ss r r sr
(11) (11)
Ri 1 Ri Mi ii 1 Ri M i ii 5 0 (11)
for swhere:
for s = A,B, = A,B, where:
for s 5 A,B, where:
(α δ + α δ )γ + (1 − α α − δ δ ) β
εiiiiss 5
ss
= ε iiss(α=Aδ B + α BAδ AB)γ s +B (1A− αs Aα B − δ AAδ BB) β s A B s
(1 − (α A (−1δ−A()(ααA B− −δ δA )(α − δ B ))(
B ))(B1 − (α
1 − (α A B+ +δ δA )(
A + δ A )(α
α + δ B ))
B )) B
ss ss
We callWe ε iisscall
We andiiεεiiand
call sr and iisrεthe
sr long-run own-price elasticity and the long-run cross-side price elastic-
ii the long-run own-price
elasticityelasticity and the long-run cross-side
ii the long-run own-price and the long-run cross-side
ity, in absolute value, once all feedback effects have been worked out. The i subscripts on the ’s
price pricethese
elasticity,
indicate elasticity,
inare absolute in absolute
elasticitiesvalue, with value,
oncerespectall once all feedback
feedback
to changes effects
in thehaveeffects ofhave
pricebeen theworkedbeen
side worked
out. by firm
s good out. i only.
The
Thefirst order conditions
i subscripts the ε’s
onindicate doindicate
not reproduceare theelasticities
one-sided Lerner Index formula. They do not
The i subscripts on the ε’s these arethese elasticities with respect with to respect
changes to changes
in the in the
even uniquely identify the values of the own price elasticities from the markups, but rather only
price of price
the side
constrain of the
the sideby
srelationship
good s good
firm iby
between firmthe
only. i only.
own and cross price elasticities and the indirect network
externalities.
Theseefirst The first order conditions do not reproduce the one-sided Lerner Index
To theorder
“Lerner conditions
Bias” most doeasily,
not reproduce
imagine for thea one-sided
moment symmetry Lerner Index across sides aA 5 aB and
dA 5formula.
formula. dB. The
They They
domarkup
not even doequation
not evenfor
uniquely uniquely
side s then
identify identify theofvalues
simplifies
the values to M
the iof
own
s
5the (own
price
ss
ii 1 price
ii ) elasticities
sr 21
elasticities . Now note that
ss
the long-run own price elasticity, ii , is necessarily larger than b (the usual short-run own price
from thefrom
markups,
elasticity),
the markups,
due tobut therather
butonly
indirect
rather only constrain
constrain
network
the relationship between
the relationship
effects. The long-runbetween cross-sidethe
the
andown and
ownelasticity,
price sr
ii , which is
cross
zero cross
price priceone-sided
in aelasticities
truly elasticities
and the and theisindirect
indirect
industry, network network
greater than zeroexternalities.
in absolute value here.21 Consequently, for
externalities.
a given set of elasticity estimates, equilibrium markups are lower that the one-sided Lerner Index
wouldToimply. see theTo The seeconverse
“Lerner the “Lerner
Bias” Bias”
most
is that amost
foreasily,given easily,
imagine imagine
for a moment
set of markups for a moment
observed in thesymmetry
symmetry across
data, the across
true short-
run
sides αAown-price
sides
= αB αand elasticities
A = δαAB=and δB. δThe of demand
A= δ
are
B. The markup
markup lower equation
equation than the one-sided
for side for
s then sLerner
sidesimplifies Index
then simplifiesto would to suggest. The
reason isS that the indirect network externalities penalize price increases more, so the short-run own
M = ( ε −ss1 + ε sr ) −1 . Now note that the long-run own price elasticity, ε ss
, is
S
= (ε iielasticity
M i price ss sr
. Now
i+ ε ii ) ii must ii benote that the
especially lowlong-run
in order own markups atε iia given
price elasticity,
to support
ss
, is level.ii
ss
The bias is strong 2 since ii . 0 , the one-sided estimate is even larger than the true long-run
necessarily
necessarily larger larger
than thanusual
(the β (the usual short-run
short-run own price elasticity), dueindirect
to the indirect
own-price elasticity, βand this results in a biasown price
in the elasticity),
opposite due toMarket
direction. the definitions will be
set too
network large, or
network
effects. The thelong-run
effects. expected
The long-run price increases
cross-side pricefrom
cross-side a merger
price
elasticity, ε iisr will
elasticity,
, which ε iiassumed
be sr
, which
is zero in too small.
is azero
trulyin a truly
The comparative statics of the two-sided markup equations give21 insight into the problem.
one-sided one-sided
As with industry, industry,
the one-sided is greater is than
greater
calculation, zerothan zero in
ainhigher
absolute absolute
value here.
short-run ownvalue21 here. Consequently, for a
Consequently,
elasticity b (which for aincreased both
given setgiven set of elasticity
of elasticity estimates, estimates,
equilibrium equilibrium
markupsmarkups are lower arethat
lower the that the one-sided
one-sided
Lerner
21 Lerner
Index
Recall isIndex
thiswould would
imply.
the long imply.
The
run change The converse
converse
in quantity is that
on atisplatform
side rfor athat fori due
given a given
set set ofin markups
ofa markups
to change price ofobserved
observed
the side s at platform
i, once all feedback effects have been worked out. There is no direct effect of side s prices on side r demand, but there
in the
in theisdata, thedata,
true the true short-run
short-run own-price own-price elasticities
elasticities of demand of demand
are lower arethan
lower
thethan the
an indirect effect of side s prices on side r demand for two-sided platforms, via changes in side s demand.
one-sided
one-sided Lernerwould
Lerner Index Indexsuggest.
would suggest. Theisreason
The reason is that
that the the indirect
indirect networknetwork
PLATFORM ECONOMICS:
174 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
iiss and iisr ) lowers markups. But unlike the one-sided case, a higher short-run cross price elastic-
ity of demand g also results in lower markups (again increasing iiss and iisr ). The reason is that a
higher cross price elasticity causes a higher proportion of the side A customers who switch after a
side A price increase to buy from platform j rather than buy nothing instead. Since platforms are
two-sided, the relative value of firm i’s platform to the side B customers falls when more A’s move
to j, causing a demand contraction on the B side as well. The feedback effects begin, and platform
i contracts further.
Greater indirect network externalities also work to decrease markups. First, by raising price a
platform triggers the market shrinking feedback effect causing its demand to repeatedly contract
on each side. It is easiest to see this by noting Mis 5 (1 2 a)/b when d 5 0 in the symmetric case.
A greater cross-side own-platform externality a penalizes price increases more and acts to lower
equilibrium markups.
Second, and surprisingly, a higher cross-side cross-platform externality d also acts to lower
markups. To see this, note that Mis 5 (1 2 (a 1 d))/b as g → b in the limit in the symmetric
case. The parameters a and d now appear as a sum, rather than a difference (as they did in the
Actual Loss formulas). Only platform i is changing price, not the hypothetical monopolist of
both products. A higher d means that in response to an increase in the price to side A at platform
i, more side B customers switch to platform j rather than to nothing at all. This makes j more
attractive to side A, causing the market shrinking feedback effects to begin, and hurting platform
i. Thus, higher values of both a and d penalize price increases more and lead to lower markups
in equilibrium.
Overall, for a given set of short-run own-price and cross-price elasticities, two-sided pressures
lower equilibrium markups below that which the one-sided Lerner Index would predict. One-
sided estimates of short-run own-price elasticity derived from markups overstate the true elasticity
and understate the profitability of potential price effects. We now turn to a demonstration of the
potential “Estimation” and “Lerner” Biases from using one-sided formulas in situations involving
symmetric MSPs.
22
We could also assume price increases on both sides A and B, and the same qualitative results still hold.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 175
It is useful to rearrange the Critical Loss equation (8), substituting in LA and LB for DQ A / Q A
and DQ B / Q B , to get:
XA1MA X B 1 MB
A
LA 1 (1 2 ) B
LB 5 21 (12)
X X
which holds when Critical Loss equals Actual Loss exactly. It is clear that the
equation is really a weighted sum of the ratios of Actual Loss to Critical Loss on
each side, where the weight is given by
R A X A
5
R AX A 1 R BX B
We refer to the left hand side of this equation, in absolute value, as the Actual-Critical Loss
Ratio (ACR). If the ACR is greater than 1, Actual Loss exceeds Critical Loss, and the price increase
would not be profitable. The relevant market would be expanded. If ACR is less than 1, the price
increase is more than profitable and the relevant market contracted. The ACR represents a measure
of the closeness of the Actual and Critical Loss.23 Note that if the industry were truly one-sided,
so that LB 5 0 after a price increase on side A, this equation reduces to the ratio of Actual Loss to
Critical Loss that would be applicable with one-sided firms.
To begin, assume that the short-run own-price elasticities are well estimated and the analyst
does not depend on the (one-sided) Lerner Index. We see the one-sided calculations are biased
towards a lower ACR, and the bias grows as the indirect network externalities become more im-
portant. Figure 1 plots the ACR against the own-price elasticity b for several different values of
the cross-side own-platform externality a. Recall the parameter a represents the short run percent-
age change in a platform’s side A business due to a 1% change in demand on its side B business,
and vice versa. The figure includes both the ACR calculated using the one-sided formula (which
assumes a and d are both zero), and also using the correct two-sided formula for MSPs. Other
parameters are held constant, and the price increase to side A is taken to be 5%.24
First, and as we would expect, the ACR increases as the short-run own-price elasticity increases
in both calculations for a given a. There is an indirect effect of higher b through markups that
increases Critical Loss, but the direct effect of b that increases Actual Loss dominates, and the ACR
rises with b.
The calculated effect of the own-platform cross-side externality a on the ACR, however, de-
pends on the calculation being used. In the two-sided calculation, a higher a is associated with
stronger feedback effects, which result in greater Actual Loss and greater ACR following a price
23
The derivation assumes XA 0 and XB 0. When XB 5 0, the second term reduces to (R B M B / R A X A )LB .
24
The assumption of 5% is often used but arbitrary.
PLATFORM ECONOMICS:
176 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
0.8
1.5
0.6
Actual/Critical Ratio
0.4
alpha=0.0
1
0.2
0.4
0.6
.5
0.8
0
increase, all else equal. In contrast, the ACR falls under the one-sided calculation. This is because
the Actual Loss calculated with the one-sided formulas ((bA 2 gA ) X A ) is independent of a, but
20
the Critical Loss is greater since observed equilibrium markups are lower with higher a. The gap
between the ACRs under the two calculations grow with higher a, and with higher b.
Figure 2 plots the “critical beta” under each calculation against the cross-side own-platform
externality a. The critical beta is the value of the own price elasticity, conditional on the other
15
tion of the proposed antitrust market would be called for under the one-sided calculation when an
expansion of the proposed market should be called instead. In merger analyses these would be the
10
combinations of a and b for which the analyst would conclude Regionfrom the one-sided calculation that
of Error
the hypothetical monopolist in a market definition analysis or the merged entity in the unilateral
effects analysis would significantly increase prices when in fact it could not.
The figure shows that as the indirect network effects become stronger, the region of error grows
5
wider. The corresponding figure for the “critical gamma”, with bCritical
Two-Sided held fixed
Beta (not shown), yields the
same conclusion.
1
0 .1 .2 .3 .4 .5 .6 .7 .8 .9
Alpha
One-Sided Calculation Two_Sided Calculation
20
15
Region of Error
5
0 .1 .2 .3 .4 .5 .6 .7 .8 .9
Alpha
(Gamma = 1, Delta = 0)
In these examples, it is assumed that the short-run elasticities are properly estimated; only the
formulas used differ. However, if the analyst relies on the (one-sided) Lerner Index to estimate the
own price elasticity of demand, and then uses the usual one-sided formulas, the bias goes in the
other direction. The analyst will overestimate the short-run and long-run own price elasticities24of
the MSPs and will overestimate Actual Loss. The one-sided Critical Loss calculation is unchanged
(because it is based on observed markups) and so the ACR will be erroneously high. In other
words, profitable price increases will not be expected for even relatively low values of the true
short-run own-price elasticity b. The critical betas and the region of error are plotted in Figure 3,
and this time the critical betas are lower under the one-sided calculation than under the two-sided
one. In the region of error, the analyst who relies on the one-sided calculations would recommend
an expansion of the antitrust market when a contraction should be called for, or would conclude
that merging platforms could not profitably raise prices after the merger when it fact they could.
The asymmetric platform case, given by formulas set out in the appendix, is more involved.
When platforms are asymmetric but the price increases considered on any side are equal across
platforms, the biases move in the same directions as in the symmetric case. However, when plat-
forms are asymmetric and price increases considered are also allowed to differ across products,
sided calculation than under the two-sided one. In the region of error, the analyst who
relies on the one-sided calculations would recommend an expansion of the antitrust
PLATFORM ECONOMICS:
178 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
market when a contraction should be called for, or would conclude that merging
platforms could not profitably raise prices after the merger when it fact they could.
FIGURE 3 Region of Error, Mismeasured b in One-Sided Calculations
10
Critical Beta
0 .1 .2 .3 .4 .5 .6 .7 .8 .9
Alpha
(Gamma = 1, Delta = 0)
25
Across all four potential price changes of a given equal amount, the “average” bias from those price changes collec-
tively goes in the same direction as that in the symmetric case, which in turn depends on the analytical method used.
However, in the asymmetric case there is variation in the size of these biases and with sufficiently asymmetric firms,
some may have the reverse sign. Because differential price changes act as weights in Actual Loss (equation A.4), the
overall bias can go in the opposite direction of the symmetric case if the reverse-signed biases are sufficiently highly
weighted.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 179
One-sided calculations do not account for this possibility, and as a result, understate the profit-
ability of this price increase. If we consider a price change at only the high-margin platform, we
can get an opposite bias. The profitability of the price increase would be overstated by a one-sided
calculation as long as the gain from expansion of the low-margin platform cannot compensate for
the contraction of demand at the high-margin platform. Thus, when asymmetric platforms and
asymmetric price changes are considered, the direction of the biases depends on the specific ques-
tion being asked as well as the method used.
26
As of the writing of this paper the merger is under review by the Federal Trade Commission and the European
Commission.
27
The authors have been consultants to several parties that have sought close scrutiny of the transaction on the part of
the relevant competition authorities and have argued that the transaction could significantly lessen competition in
various markets.
28
Advertisers value the demographic and other characteristics of viewers in addition to their sheer number. Thus views
by high-income residents of Boston are more valuable to the local BMW dealer than views by low-income resident
or high-income residents of Topeka.
29
The more developed part of the online advertising business consists of search. Search-engines supply advertising
inventory on their search-results pages and thereby compete with regular publishers for advertiser dollars. Search
engines are also relevant to Web 2.0 because they are the way people often find websites either through organic
search results or through advertising.
PLATFORM ECONOMICS:
180 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Publishers create advertising inventory by designing their web pages to accept graphical, text,
or video ads in various portions of the page. The amount of inventory they can supply then
depends on the viewers they attract to these pages. This supply is highly heterogeneous for two
primary reasons. First, like newspapers, some space is seen as more desirable than others – the top
right is more desirable than the bottom left because people are more likely to pay attention to the
former than the latter. Second, some viewers are worth more to advertisers than other viewers, and
the technology of online advertising enables publishers and advertisers to establish prices for view-
ers with particular characteristics.
Like all sellers and buyers, publishers and advertisers require ways to identify optimal trading
opportunities and to establish transaction prices. There are two major ways that this “interme-
diation” occurs. First, it can occur directly through bilateral exchanges between publishers and
advertisers. eBay for example, may sell Nokia the right to present an advertisement in a particular
spot to viewers with specific characteristics by having its sales agents deal directly with Nokia’s
buying agents. Second, it can occur indirectly through multilateral exchanges between publishers
and advertisers using advertising networks. Hearst Publishing may sell ValueClick – an advertising
network – advertising inventory from its various online newspaper and magazine properties, which
ValueClick will then sell to advertisers who want to reach the kind of viewers that Heart Magazines
has. The advertiser in this case typically buys access to a type of viewer –“fashion conscious young
women in upscale locations” – but has not specifically bought space on Cosmopolitan’s website.
The extent to which advertisers and publishers use direct and indirect methods of distribution for
advertising varies. Smaller ones typically rely only on indirect methods because it is not economi-
cal to carry the cost of salespeople and purchasing agents. Google’s AdSense network is especially
popular with small advertisers and publishers in part for this reason.
This section concentrates on large web publishers, which account for the preponderance of
advertising revenue and large advertisers, which account for the preponderance of online spend-
ing.30 A significant part of the advertising inventory bought and sold by these large advertisers and
publishers involves bilateral exchange. Large publishers either have direct sales forces or hire third-
party sales reps to sell their ad inventory. Likewise, large advertisers have purchasing agents or, more
often, use media buyers at their advertising agency to purchase ad inventory. Advertising inventory
sold this way is said to be “reserved”. Large publishers often sell their “premium ad space” this way.
Large publishers often also rely on other intermediaries to sell ad inventory indirectly that they
have not “reserved” for advertisers directly. These intermediaries are called “advertising networks”.
Publishers may use ad networks because they are more efficient than a direct sales force for some,
or all, of their ad inventory; or because they have excess inventory that they have not sold directly,
30
A large publisher refers to one which is sufficiently large to use stand-alone tools to serve their directly sold ads and
any remnant ads sold through non-integrated and/or integrated ad networks. This includes at least the top 500
publishers, and accounts for the majority of advertising revenues. Smaller publishers do not find it economically
efficient to hire direct sales forces. They usually rely on an ad network to sell their ad inventory.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 181
perhaps because of spikes in viewers. Advertisers use ad networks because it is another way to reach
viewers. Hard estimates are difficult to come by, but advertisers and publishers we have talked to
seem to agree that somewhere around 30 percent of advertising revenue for large publishers is sold
indirectly, as well as more than 50 percent of advertising impressions (the number of viewers of an
ad); this figure appears to vary significantly across publishers depending on the way they have chosen
to manage their advertising business. On average, advertising inventory sold indirectly costs less than
advertising inventory sold directly because the space is less desirable than the directly sold space.
There are several aspects of advertisers’ management of their online campaigns, as well as pub-
lishers’ management of their sales of advertising inventory. As discussed previously, advertisers and
publishers need intermediation services. For bilateral exchanges they may use some combination
of in-house and third-party providers, while for multilateral exchanges they mainly use advertising
networks. Advertisers and publishers likewise need management, reporting, and technology solu-
tions such as those offered by DoubleClick and aQuantive. These tend to be server-based software
that can help manage advertising inventory and campaigns that may involve millions of ad impres-
sions (that is, views of an ad by an individual) a day. These server-based software tools are highly
sophisticated mission-critical applications.
Large publishers usually use a publisher tool such as DoubleClick’s Dart for Publishers
(DFP).31,32 This tool is typically hosted on a web server maintained by the provider. The publisher
hardcodes links to the publisher tool to fill the ad space for which it wants to use the management,
reporting, and serving capabilities of the publisher tool. It will also typically integrate the publisher
tool into many other aspects of the website technology and business practices. As an example of
how such publisher tools are used, consider a visit to www.cnn.com. When the entertainment
page of www.cnn.com is clicked, a decision is made as to which ad to present to the user and, once
chosen, the ad is displayed so the user can see it, and the publisher can earn some money. The
publisher – in this case www.cnn.com – uses its publisher tools to check whether the particular ad
space that the user is about to see has been “reserved”, and if not, whether there is an ad network
that can fill that inventory space. The publisher tool then retrieves the chosen ad and presents it to
the user. This entire process happens in the blink of an eye. Publishers will typically only use one
publisher tool (i.e., they “single-home” on tools).
Large advertisers and advertising agencies often have an advertiser tool such as DoubleClick’s
Dart for Advertisers (DFA).33 Large advertisers typically manage advertising on hundreds of web-
sites and across numerous products using many methods of online advertising. This tool helps
them manage these various campaigns. Advertisers usually use one advertiser tool although adver-
tising agencies may use several. However, single-homing does appear to be the norm.
31
See http://www.doubleclick.com/products/dfp/index.aspx
32
A handful of mega-large publishers such as MSN have their own proprietary tools but most others use a third-party
tool.
33
See http://www.doubleclick.com/products/dfa/index.aspx
PLATFORM ECONOMICS:
182 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
A few providers offer more or less complete solutions for advertisers and publishers. Google’s
AdSense/AdWords platform is one of these. Publishers can hardcode ad space to Google’s AdSense34
which takes care of everything – selling the inventory, managing the ad space, serving ads to the
viewer, and sending the publisher a portion of the proceeds after taking a commission.35 Advertisers,
likewise, can buy space from the Google Content Network36 through AdWords37 (which bundles
Google’s search-based and contextual-based advertising products). Yahoo! and Microsoft offer
similar all-inclusive solutions. These solutions have all resulted from leveraging the technologies
developed for search-based advertising – especially the keyword bidding auctions – to the buying
and selling of publisher ad inventory. Some other ad networks are also integrated to lesser degrees;
they may offer publishers serving technologies so that publishers can hardcode ad network into
particular space. Many large publishers use an integrated platform for contextual ads and an unin-
tegrated platform (based on a particular publisher tool) for non-contextual ads; the unintegrated
platform is used to access multiple standalone ad networks, as well as ads sold directly.
Online advertising is highly differentiated. It comes in different forms such as graphic ads
that mix pictures and text, all text ads, and video ads; appears in different places on websites; and
is targeted to viewers in various ways. It is also bought and sold in different ways which leads to
differentiation in terms of the distribution channels used by advertisers and publishers. The un-
derlying economics and driver of substitution are fairly straightforward. Publishers are interested
in maximizing the rate of return on investment in their ad inventory. Advertisers are interested in
maximizing the rate of return on investment from their advertising spending- which ultimately
means getting consumers to buy things. Advertiser and publisher decisions on ROI are interlinked
because online advertising is fundamentally a two-sided business, as discussed below.
Surveys of large websites show that DFP is used by somewhat more than half of these websites.
Larger sites are more likely than smaller sites to have DFP. DFP has about 63 percent share based
on page views.38 DFA holds roughly a 26 percent share of non-search advertiser tools.39 Google is
34
See https://www.google.com/adsense/login/en_US/
35
In the online advertising Traffic Acquisitions Costs (TAC) refers to what an advertising platform pays for traf-
fic. Google pays TAC to publishers in return for contributing their advertising inventory to the Google Content
Network. The publisher receives TAC. In this case 1-TAC is the percent commission paid by the publisher to
Google for selling its ad inventory. If Google pays the publisher 80% of the revenue that Google receives from the
advertiser the publisher has paid a commission of 20% to Google.
36
See https://adwords.google.com/select/afc.html
37
See http://adwords.google.com/select/Login
38
The survey conducted by LECG shows that AdSense is used on 50% of all websites sampled. DoubleClick is used
on 63% of all websites sampled. A second survey conducted by Keystone Strategy for American viewers shows that
AdSense is used on 45% of surveyed sites while DoubleClick is used on 61%.
39
This is based on aQuantive’s estimates of shares for itself and DoubleClick in advertiser tools (among providers that
sell those as standalone non-integrated products), adjusted for those advertiser tools that are sold on an integrated
basis (with the ads), by providers such as AdSense.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 183
the leading integrated ad platform. Estimates suggest that Google accounts for 51 percent of the
ad revenue through the indirect channel and 27 percent overall. The unintegrated ad networks ac-
count for an estimated 45 percent of ad revenue through the indirect channel and 25 percent of
web publisher ad spending overall.40
Following the acquisition, Google would have control of a key input into the unintegrated
channel for selling ads directly and indirectly. It would have the ability to coordinate the price of
DFP, and through DFP, the overall price of distribution through the direct sales and through the
ad networks.41 The question we consider is whether the combined entity would have the incentive
to alter prices significantly. Two-sided effects are important for analyzing this question because
changes in the prices of the inputs alter the advertiser’s demand for space as well as the publisher’s
demand for selling through a particular method.
Table 1 reports the data and sources we have relied on. The key data is that on margin, which
shows that both Google and DoubleClick earn roughly 75 percent gross margins. Ad inventory
sells on average, with great dispersion, for about $2 per thousand impressions (that is, views by
users). Google sells on a cost per click basis but obtains the equivalent of approximately 40¢ per
thousand impressions based on the evidence we have seen.42 Publishers typically pay 40¢ per thou-
sand impressions on average to the standalone ad networks and 5¢ per thousand impressions to
the publisher tool provider.
We want to ask whether the merged Google-DoubleClick entity would have the ability to
increase the price of DFP by X%, holding all else constant. This was a central point of concern
with the merger. Google’s AdSense is an integrated product, and so charges higher markups than
DoubleClick’s DFP. This suggests it might be advantageous for Google to raise the price of DFP
and shift publishers to its AdSense network.43 In the analysis, we take X to be either 5% or 10%.
Since the platforms are inherently asymmetric, the analysis uses the formulas set out in the ap-
pendix. Ex ante, the direction of bias from using the one-sided calculation in the two-sided setting
is unknown.
40
Source: Keystone Strategy.
41
We focus on price effects as is traditional in this sort of analysis. Google could engage in a variety of other strategies
post-acquisition that could involve the exercise of acquired market power but that do not necessarily involve increas-
ing prices.
42
The revenue paid by advertisers by publishers is estimated at around $2 CPM of which publishers keep approxi-
mately 80%. See http://www.mydigitallife.info/2006/10/21/google-adsense-giving-publishers-average-of-78-reve-
nue-share/ and http://www.webmasterworld.com/forum89/6913.htm
43
In the two-sided calculation, we attribute 2/5th of the Google unit price and revenues to the publishing side and
3/5th to the advertising side, to match the DoubleClick ratio. These are the expected side-specific prices that Google
would charge had advertisers paid publishers directly for the advertising space and each paid Google a commission
for the value of the tools and intermediation services provided. This avoids the need for negative prices and negative
externality parameters in the calibration, which adds transparency.
PLATFORM ECONOMICS:
184 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Given the limited data available, calibration is the most feasible approach.44 For the one-sided
calculation, there are four parameters to estimate (all on the publisher side), using a combination
of data and restrictions based in economic theory. First, we use the theoretical Lerner Index to fix
the own-price elasticities given the observed markups (two restrictions). As discussed, the use of
the one-sided Lerner Index in a two-sided setting yields a biased estimate of the true price elastici-
ties. We also assume Slutsky symmetry, which equates quantity-weighted cross-price elasticities
across platforms (one restriction). We then just need to calibrate one of the cross-price elasticities
and we could calculate Critical Loss and Actual Loss. However, instead we employ the common
and useful technique in calibration exercises to set Critical Loss equal Actual Loss (the fourth re-
striction) and then back out the values of the cross-price elasticities (connected together by Slutsky
symmetry) that would exactly equate CL and AL.
44
Given enough time a competition authority could possible obtain sufficient data from members of the industry and
through surveys to obtain estimates from demand-system estimation.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 185
In the two-sided model there are sixteen parameters total in the demand system that we need
to estimate using a combination of data and economic theory, i.e. sixteen restrictions. First, we use
the two-sided version of the Lerner Index given by equations (11). Conditional on the externality
parameters, this yields a linear relation between each own-price elasticity and its corresponding
cross-price elasticity. We keep Slutsky symmetry, a similar symmetry for the externality parameters,
and seek out the “most symmetric” solution. We assume two-sided effects are conservatively small,
with own-platform cross-side externalities of 10%. The restrictions are explained in more detail in
the footnote.45
In total, these add to fifteen restrictions. Again, we equate Critical Loss and Actual Loss as the
sixteenth restriction, and back out the own- and cross-price elasticities (connected together by the
other restrictions) that would exactly equate CL and AL.
Following common practice, we report the easily interpretable “critical diversion ratios” and
“critical switching levels” rather than the b’s and g’s themselves.46 The critical diversion ratio is
defined as the fraction of the quantities lost at DoubleClick’s DFP that needs to be immediately
recaptured by Google’s AdSense platform, for a price increase in DFP to be profitable once the in-
direct network feedback effects work themselves out. It is effectively the quantity-weighted ratio of
the cross-price elasticity (of AdSense’s quantity with respect to the price of DFP) to the own-price
elasticity (of DFP). The Critical Switching Level is defined as the fraction of the total quantities
of DFP that needs to be immediately recaptured by AdSense after a price increase for the increase
to be profitable once all feedback effects are worked out. It is calculated as the AdSense-quantity-
weighted cross-price elasticity to total DFP quantity.
45
The two-sided Lerner Index accounts for four restrictions, each relating an own-price elasticity with a corresponding
cross-price elasticity. We assume Slutsky symmetry that ensures consistency in the cross-price elasticities on each side
(four restrictions) and a similar restriction to ensure consistency across the indirect network externalities on a given
side (four restrictions). Our results do not change qualitatively if we loosen the assumption of Slutsky symmetry,
however. We assume Google’s own-platform cross-side externality is 0.1 on each side, and its cross-side cross-plat-
form externality 0.05 on each side (four more restrictions). Under Slutsky symmetry in the two-sided parameters,
the corresponding values for DoubleClick are 0.1 and 0.005. We also assume the “most symmetric solution” across
sides (one restriction). This is because, conditional on the externalities, many combinations of b and g are techni-
cally possible. A relatively low b on one side could exist and satisfy the markup condition if the corresponding g is
extremely high. Note that the effect of a change in b is about 1/a times the effect of a change in g in the markup
equations. But this causes an opposite stress on the other side, in order to equate AL and CL (see below). This tends
to force the other b to be relatively low, and the other g to be extremely small – generally negative with a similarly
large magnitude as the g on the first side. The more reasonable solution is that which yields the most symmetric b’s
and g’s, conditional on all the other constraints. In a fully symmetric model this ensures a symmetric solution; in an
asymmetric model the b’s and g’s still differ across sides. It is implemented by minimizing the sum of squares of the
g’s across sides (alternatively, the sum of squares of the b’s across sides.) The total number of calibrated values plus
restrictions adds to 15. We could then fix one of the b’s or one of the g’s to complete the model. Instead, we force
CL5AL as the sixteenth restriction, and back out the critical b’s and g’s (tied together by the above constraints) that
would result in CL equaling AL.
46
See Shapiro (1996).
PLATFORM ECONOMICS:
186 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Table 2 reports the critical diversion ratios and critical switching levels under our best data as-
sumptions, for both the one-sided and two-sided calculations, and for critical price increases (X)
of 5% and 10%.
Given the high markups in the industry, and the large differential in absolute margins between
AdSense and DFP or DFA, the results show generally that it is likely the Google-DoubleClick
entity would be able to increase the price of DFP by five or ten percent, even when the amount
switching between platforms is very low. That is, Google-DoubleClick would need only a few of
the customers it loses after increasing price at DFP to switch to AdSense, instead of switching to a
competing platform or not buying at all (given by the critical diversion ratio). It needs even fewer
of its total publishing customers to switch to AdSense (given by the critical switching level).47
While all the numbers are low due to the nature of the proposed merger, the critical diversion
ratios and switching levels implied by the one-sided results turn out to significantly overestimate
the two-sided calculations. In other words, the one-sided calculation understates the potential for
unilateral price effects from the merger. Whereas the one-sided numbers allow for a small possibil-
ity that the price increase would not be profitable, the two-sided calculations suggest that Google
is likely to raise prices on DFP even if virtually no customers switched to AdSense.
This creates a “region of error” between the critical diversion ratios. If the actual diversion ratio
is below the one-sided figure and above the two-sided figure, it would result in different conclu-
sions depending on the calculation. The one-sided calculation would suggest profitable price in-
creases, the two-sided calculation would not. The same is true for actual switching levels.
The finding of profitable price increases with an effectively zero cross-price elasticity in the
two-sided calculation might be surprising. Merging firms in a one-sided world only raise price
above its formerly profit-maximizing level if doing so directly causes customers to switch to its
other products, i.e. a large enough cross-price elasticity. The reason why zero switching levels are
plausible in the two-sided case is because of the indirect network externalities. By raising the price
47
We present this analysis to demonstrate the effect of two-sided considerations on standard diversion ratio analysis.
Whether and under what circumstances it is appropriate to conclude from this sort of analysis that a merger should
be blocked is much more complicated issue which we do not address in this article.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 187
of DFP, DoubleClick loses publishers and, via the feedback effects, also loses advertisers, then
more publishers, more advertisers and so on. As the DoubleClick platform shrinks and its value
deteriorates, advertisers and publishers switch to other platforms, including Google’s AdSense.48
The increase in profits to Google from this switching, which is driven by changes in platform size
rather than changes in relative prices directly, makes the price increase on DFP profitable. Hence,
48
Customers who are sensitive to platform size, even if perfectly price inelastic, make this switch.
PLATFORM ECONOMICS:
188 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Google benefits from contracting and devaluing the low (absolute) margin DoubleClick platform
to increase the relative value of its high (absolute) margin AdSense platform. 49
Table 3 performs robustness checks and reports critical diversion ratios over a wide range of
data assumptions.50 The same patterns hold for critical switching levels. The two-sided calculation
is often zero, and always zero for 5% price increases, given the high markups involved.
In summary, we have demonstrated that one-sided and two-sided calculations yield different
critical diversion ratios (and critical switching levels) and open up a region of error between them.
The size of this region of error depends on the data and externality assumptions, but is non-trivial
as a general rule. If the actual diversion ratio falls in between the one- and two-sided calculations,
the analyst would give the wrong recommendation about the unilateral price effects of the pro-
posed Google-DoubleClick merger.
49
The fact that Google can profit from DFP price increases that are purely contractionary suggests that other non-price
methods of shrinking the relative size of the DoubleClick platform can also be profitable. Quality degradation is an
example. More likely, however, Google could tie intermediation services to its tools software on the DoubleClick
platform, as it currently does on AdSense. This would foreclose third party networks and direct sales channels for
intermediation, and effectively convert DoubleClick from a low-margin platform to a high-margin platform like
AdSense.
50
Except as noted, all other data assumptions remain the same.
51
See Klein et al. (2006), Evans and Schmalensee (2005a, b).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 189
We are dubious that the light generated by market definition analysis in markets involving
MSPs is worth the candle. The proper analysis is difficult and highly dependent on proper mod-
eling and data. Mistakes that result in narrow markets that assume away clear multi-sided effects
are quite costly. Although the analysis of unilateral effects is also challenging, our advice to practi-
tioners is to avoid formalistic market definition exercises and to use one’s candles to illuminate the
effects of the merger on prices and other dimensions of competition.
s 5 A,B i 5 1,2
s s
R i (X i 1
(
M is )
DQ is
Qi s )
1 Ris X is 5 0 (A.1)
The N platform formula differs only in the number of summed terms. This formulation allows
different price changes at each platform on each side for use in a variety of questions.
For Actual Loss, consider again the isoelastic demand system given by
for i,j 5 1, 2, i j and s 5 A, B, where lower case q’s and p’s represent log values. Assume all
parameters are positive (though this need not be the case) and ais . djs, bis . gjs, ais 2 djs , 1 and
ais 1 djs , 1 to prevent the two-sided effects from exploding.
Since this a demand system, ownership patterns do not matter for estimation. Also, if there are
two platforms that both sell to side A, it is not necessary that they both sell to side B. The two “side
B’s” can refer to completely different goods. It may also be that one platform sells to sides A and B,
whereas another firm sells only to side A, and a third firm only sells to side B.
The reduced form value of qsj as a function only of parameters and prices, accounting for all
the feedbacks in the system, is:
qis 5 [Γis (a jsu jr 2 d jsu ir 1 u js ) 1 is (a isu ir 2 d isu jr 1 u is )] / is (A.3)
where
uis 5 mis 2 bis pis 1 gis p js
Γis 5 2aisdir 2 dis arj
is 5 (1 2 ajsair 2 djs dir )
is 5 is (1 2 aisair 2 dis djr ) 1 Γis(ajsair 1 djs dir )
PLATFORM ECONOMICS:
190 Part Two : Two-Sided Antitrust Economics Essays on Multi-Sided Businesses
Totally differentiating, and replacing all dpis with X is and dqis with Lis , we derive the Actual
Loss formula:
The analyst then compares Actual Loss and Critical Loss as described in the text. The analysis
can further be extended to N platforms. The Actual Loss formula is derived from the following
system of 2N equations:
N
qi 5 s
d q
k51
s
ik k
r
1 uis (A.5)
Quantity at platform i on side s depends on the quantity at each platform k (including i )
on the other side r (the dik’s), and the price at each platform (including i ) on side s (the gik’ss).52
Substituting each qir equation into each qis equation, the system can be rewritten in matrix form:
Ds s 5 s for each side s, where Ds is a N3N matrix with ij th element l ijs equal to
N
s
l 5 1(i 5 j) 2
ij d d (A.6)
k51
s
ik k
r
where 1(i 5 j) is an indicator function equal to one when i 5 j and zero otherwise. Also, s is
a N-vector with ith element equal to qis, and Fs is an N-vector with ith element equal to fis , where:
N
s
f 5 ui 1
i
s
d d (A.7)
k51
s
ik k
r
Actual Loss is calculated for each qi by total differentiation, i.e. Ls 5 (∆s )21 d s, for s 5 A,B,
s
52
We have replaced the notation ai from the two-platform version with dii’s and the coefficient on own price from bi
with gii’s.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Defining Markets that Involve Two-Sided Platforms 191
To close the model, we note that the first order conditions defining markups in the two-
platform symmetric case are the same as those in the 2-platform and N-platform asymmetric cases,
since each set of equations is platform specific.
As before, the calculated Actual Loss amounts X sj are substituted into the Critical Loss equa-
tion and the ACR is checked. (Alternatively, one can fix all parameters except one, and use the
ACR 5 1 identity to back out the “critical” value of the remaining parameter.) The comparative
statics from the symmetric and two-platform cases carry through to the N firm asymmetric case.
Part Three
WEB BASED PLATFORMS
CHAPTER
Economics of the Online
7 Advertising Industry
ABSTRACT
Online advertising has grown rapidly and accounts for about 7 % of US advertising spending.
It is projected to increase sharply as more consumers spend time online on their personal com-
puters and as additional devices such as mobile phones and televisions are connected to the
web. This chapter describes how the online advertising industry works. The industry is populated
by a number of multi-sided platforms that singly or in combination facilitate connecting ad-
vertisers to viewers. Search-based advertising platforms, the most well-developed of these, has
several interesting economic features that result from the combination of keyword bidding by
advertisers and single-homing.
I. INTRODUCTION
Online advertising began in 1994 when HotWire sold the first banner ads to several advertisers.1
Revenue in the United States grew to an estimated $7.1 billion in 2001 or about 3.1 percent of
overall advertising spending. The dot-com bust destroyed or weakened many of the early online
advertising industry players and reduced the demand for online advertising and related services.
The industry regained momentum by 2004 as the business model for “Web 2.0” came to-
gether.2 A number of businesses such as Advertising.com, Google and ValueClick emerged that
facilitated the buying and selling of advertising space on web pages. Many web sites settled on
the traditional “free-tv” model: generate traffic by giving away the content and sell that traffic to
advertisers. Of the 20 top web sites, based on unique visitors, in the United States, 14 generate
the preponderance of their revenues from the sale of advertising inventory—the eyeballs that view
1
Barbara K. Kaye and Norman J. Medoff, Just A Click Away: Advertising on the Internet (Massachussetts: Allyn and
Bacon, 2001).
2
“Hundreds of Internet companies have emerged since the dot-com crash, looking to capitalize on a resurgent online
advertising market. Companies in this new wave -- known as Web 2.0 -- have focused on online collaboration and shar-
ing among users. They hope to attract millions of users and become the next YouTube, which was acquired by Google
Inc. earlier this year for $1.65 billion.” See Is ‘Web 2.0’ Another Bubble? The Wall Street Journal, December 27, 2006.
193
PLATFORM ECONOMICS:
194 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
space allocated for promotions—to advertisers.3 In 2007, advertisers in the United States spent
more than $21 billion advertising on websites, about 7 percent of all U.S. advertising spending.4
The portion of advertising that is done online is expected to increase significantly over time as
web-based content and services expand and people are able to access the web through more devices
such as mobile telephones and televisions. The valuations that the capital markets are placing on
businesses related to online advertising are consistent with this prediction. Google has had a nearly
five-fold increase in its market value from August 2004 when it was valued at $29 billion to $136
billion in April 2008. During 2007 several companies in the online advertising market were pur-
chased at multiples of 10-15 times annual revenues.5
The online advertising industry burst into the public eye in 2007. Google’s sky-rocketing stock
price and its forays into industries such as word processing software, online payments, and mobile
telephones drew significant attention. More than 500 articles on Google appeared in the New York
Times, Wall St. Journal and the Financial Times during the year. The U.S. Federal Trade Commission
and the European Commission launched in-depth antitrust investigations into Google’s acquisi-
tion of DoubleClick, which provides software technology and services to online advertisers and
publishers.6 Privacy concerns also came to the fore in 2007 as consumers, government agencies
and the media started focusing on the massive amount of personal data that online advertising
companies were storing and using.7
This chapter describes how the online advertising industry works, focusing on several economic
aspects of this business. Although the online advertising industry has revolutionized many aspects of
an age-old business, it is important to understand, as we present in Section 2, that the new industry
has much in common with the old. The unique features of online advertising include the use of
Internet-based technologies and data collection mechanisms to target and track specific individuals
3
Of these 14, five also use a subscription model to supplement revenues. Out of the remaining six, four use the mer-
chant model, one uses the auction model (eBay.com), and one is a not-for-profit (wikipedia.org).
4
eMarketer, US Online Advertising: Resilient in a Rough Economy, March 2008. Estimates of advertising spending,
especially online advertising spending, should be viewed as indicative.
5
Google announced in May 2007 that it would purchase DoubleClick for $ 3.1 billion which is more than 10 times
DoubleClick’s revenues according to one account. Louis Story and Miguel Helft, Google Buys an Online Ad Firm for
$3.1 Billion, New York Times, April 14, 2007. Microsoft purchased aQuantive at a multiple of about 13. Peter Galli,
Microsoft’s aQuantive Buy Shows Big Ad Plans,, eWeek.com, May 18, 2007. Yahoo paid some $680 million for 80
percent share of Right Media which generated about $35 million in revenues in 2006. Michael Liedtke, Yahoo snaps
up Right Media for $680M , USA Today, April 30, 2007.
6
European Commission Press Release, Mergers: Commission opens in-depth investigation into Google's proposed take
over of DoubleClick, November 13, 2007; Google SEC Filing, Form 8-K, sec.gov, May 29, 2007. Also, the United
States Senate held hearings on this acquisition. See An Examination of the Google-DoubleClick Merger and the Online
Advertising Industry: What Are the Risks for Competition and Privacy?,” Senate Judiciary Committee, September 27,
2007. The FTC cleared the transaction on December 20, 2007. See Statement of Federal Trade Commission concerning
Google/DoubleClick, FTC File No. 071-0170, http://www.ftc.gov/os/caselist/0710170/071220statement.pdf.
7
Vidya Ram, EU Turns Spotlight On Google, Forbes, May 28, 2007; Steve Lohr, Google Deal Said to Bring U.S.
Scrutiny, New York Times, May 29, 2007; Darren Waters, Google privacy policy ‘is vague’, BBC News, May 31, 2007.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 195
and to automate the buying and selling of advertising inventory. Like modern finance, online ad-
vertising relies heavily on advanced economic and statistical methods. These topics are discussed
in Section 3, which focuses on search-based advertising- the most well developed part of online
advertising business to date- and Section 4, which examines online display advertising, a rapidly
evolving part of the business that involves advertising on any web sites that draw traffic. The online
advertising industry is highly complex, undergoing a series of rapid changes, and could well result in
a high degree of concentration, if not monopoly, in the intermediation of advertising inventory and
the control of personal data. Section 5, presents concluding remarks, and explains why the online
advertising industry will remain at the center of public policy debate for many years to come.
zine price sharply to increase circulation, and instead earned revenue from selling advertisements.
Revenue and profits increased from this pricing innovation. Most magazine publishers quickly
followed, and today that is how most earn their profits. The radio industry initially struggled with
a subscription-based model, but several stations discovered the power of advertising and the rest
quickly followed. Television followed the same path.11
Advertising agencies emerged in the mid 19th century as brokers between newspapers and ad-
vertisers. The first agency in the United States started in 1841. Its agents bought large amounts of
newspaper advertising space at a discounted price and then resold it to advertisers. At first the ad-
vertisers designed the ads and the agency just placed them, but later on advertising agencies started
designing the ads and providing other services. The business model that eventually developed in-
volved giving creative and marketing services away in return for commission on the media buying.
11
Evans and Schmalensee, supra note 10; Microsoft Encarta supra note 9
12
For a general introduction to the economics of advertising see Kyle Bagwell, The Economic Analysis of Advertising,
Mark Armstrong and Rob Porter (eds.), Handbook of Industrial Organization, Vol. 3, North-Holland: Amsterdam,
2007, pp. 1701-1844.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 197
Nielsen Media Research’s data on demographics and what is being viewed. Broadcasters and ad
agencies negotiate prices based on Nielsen’s numbers, and the outcomes of these negotiations in
turn determines which programs remain on the air.13 Furthermore, contracts between advertis-
ers and TV networks usually include a rating guarantee. Should ratings of the program in which
the ad is shown fall short of the agreed level, TV networks would provide extra ad time to the
advertisers.14
Traditional media that use content to attract viewers have adopted two different models. In
the subscription/advertising model the publisher charges viewers a fee to obtain access to the
content, and advertisers a fee to obtain access to the viewers. Many newspapers and magazines
follow this model. They then balance the demand from advertisers and subscribers to maximize
revenues. Some magazines, e.g., the Economist, have adopted reader-friendly strategies with high
reader fees but sparse advertising. Others have adopted advertiser-friendly strategies, e.g., Vogue,
with lower reader fees and more advertising, some of which makes reading the magazine difficult.
In the free-media model the publishers do not charge viewers for access to the media at all, and
in fact try to distribute the media as widely as possible. They earn all of their revenues and profits
from the sale of advertisements. Free radio and television have embraced this model in the United
States. However, there are many free newspapers and magazines that have adopted the free-TV
model. Pay-television and satellite radio have adopted the mixed subscription/advertising model.
These different business models are now better understood as a result of the work on multi-sided
platforms.15
13
Franklin M. Fisher, John J. McGowan, and David S. Evans, The Audience Revenue Relationship for Local Broadcast
Stations, Bell Journal of Economics (Autumn 1980); Nielsen Starts Watching the Ad Watchers, Journalism.org,
November 2, 2006; Collecting and Processing the Data, Nielsen Media Research, http://www.nielsenmedia.com.
14
For example, “[I]f NBC Universal [did] not deliver the viewers it [had] promised advertisers, it would have to offer
them compensatory commercial time…” (Stuart Elliott, Olympic-Size TV Audience for the Athens Games?, New York
Times, August 13, 2004. Also “ABC, CBS, NBC, Fox and the CW network sold $9.3 billion in prime-time ads for
this season. In the process, they sold about 80 percent of their time, holding back some to give advertisers should
ratings fall short of guarantees.” (Meg James and Alana Semuels, Lower ratings could pinch TV ads, Los Angeles
Times, December 12, 2007.
15
Rochet and Tirole, supra note 10; Simon P. Anderson & Jean J.Gabszewicz, The Media and Advertising: A Tale of
Two-Sided Markets, Handbook of Cultural Economics, Elsevier, 2006; Simon Anderson and Stephen Coate, Market
Provision of Broadcasting: A Welfare Analysis, Review of Economic Studies, Vol 72, No. 4, October 2005, pp. 947-972.
PLATFORM ECONOMICS:
198 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
TABLE 1 US Ad Spending, 2002 - 2006 (Billions). Display includes display ads,
rich media, and sponsorship. Rich media includes interstitials.
Search Display Classifieds E-mail Other Total
2002 $0.90 $3.42 $0.90 $0.24 $0.54 $6.0
2003 $2.56 $2.99 $1.24 $0.22 $0.29 $7.3
2004 $3.74 $3.65 $1.73 $0.19 $0.29 $9.6
2005 $5.13 $4.25 $2.13 $0.25 $0.75 $12.5
2006 $6.76 $5.41 $3.04 $0.34 $1.35 $16.9
% Change: 651% 58% 238% 41% 150% 182%
2002-2006
Source: IAB Internet Advertising Report, 2002-2006 Full-Year results. Available at http://www.iab.net/insights_research/1357.
based advertisements. Table 1 reports U.S. advertising spending for 2006 in these categories and
growth between 2002 and 2006.
Search-based advertising accounted for the largest portion with 40 percent followed by dis-
play-related advertising with 32 percent in 2006 (of which 22 percent was display advertising, with
rich media and sponsorship accounting for the remaining 10 percent.) All segments have grown
significantly in the last few years with search having the highest growth rate.
In many ways, on-line advertising is similar to traditional advertising. Publishers use content
to attract viewers and then sell advertisers access to those viewers. Advertisers can display text
(like classifieds), graphics (like magazines) and video (like television) ads in the space supplied by
the publishers. On one level, one can think of the web as just adding more advertising inventory,
much like displaying ads on televisions in the back of taxis. Indeed, in some ways the introduction
of online advertising was a less radical innovation than the introduction of other media. After all,
television enabled advertisers to reach mass audiences with video ads while the web is relying on
quite traditional methods of presentation.
Three radical innovations, however, distinguish online from off-line advertising. The first has
transformed the service obtained by the advertiser: the Internet provides a highly efficient mecha-
nism for delivering ads to individual users and collecting information for targeting ads to those
users. The second has transformed the process of buying and selling advertising space: the Internet
has enabled the development of more efficient intermediation markets for advertising—the key-
word bidding system used for search and contextual advertising is the most mature example. The
third is leading to economies of specialization: traditional publishers provide content for attracting
viewers and sell advertising space to advertisers; online publishers are increasingly turning the sell-
ing of advertising space over to specialized advertising platforms such as Google or advertising.com.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 199
FIGURE 1 A search on Google for “luxury hotels London” shows the resulting web-
page divided into three areas: 1) organic search results, 2) paid search
results, and 3) more paid search results.
PLATFORM ECONOMICS:
200 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
results for some search engines. A search query may generate a series of pages of search results, and
each page may have ads on the right-hand side if there have been buyers for the space.
A “search-based advertising platform” (search-ad platform for brevity) attracts viewers to its
pages largely by displaying the organic search results from its search engine. It allocates a portion
of the page for the purpose of selling advertising space, divides this place into slots (there are typi-
cally 8-10 per page) which it sells to advertisers. There are two key technological innovations that
underlie this process, both of which depend on advanced economic and mathematical methods,
and which ultimately help determine the nature of the market structure.
16
When users bid on keywords, those keywords are set to a certain match option. These options include a broad match
(which displays the ad when customers search for words in the keyword list in any order and possibly with other
terms), a phrase match (which displays the ad when a customer’s search query includes all keywords in the exact
order given, even if the query has other terms that precede or follow the phrase), and an exact match (which makes
an ad eligible when a search includes the specific keyword or phrase, in order, and without any other terms in the
query). Negative keywords can also be added so that an ad will not be displayed if a search query contains a negative
keyword.
17
See Benjamin Edelman, Michael Ostrovsky, and Michael Schwarz, Internet Advertising and the Generalized Second
Price Auction: Selling Billions of Dollars Worth of Keywords, American Economic Review, v. 97(1), March 2007,
pp.242-259.; Microsoft, What’s the Cost?, http://advertising.microsoft.com/search-advertising/cost. Also see Hal
R. Varian, Position Auctions, Working paper, University of California, Berkeley, 2006.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 201
ad to a searcher. Estimating these types of quality measures is especially difficult for advertisers and
keyword combinations for which the search-ad platform has no experience.18
The keyword bidding process and the quality score algorithm together determine both the
CPC advertisers pay and the slots they receive. Search-ad platforms sometimes provide bidders
guidance on what they would have to pay to get particular slots. For example, Google estimates
that bidders would have to pay an estimated $3.04 to get the third slot for “luxury hotels London”
and $0.05 to get the third slot for “competition economists.”19
These technologies affect the market structure in two ways as we will see below. First, the key-
word bidding system can give rise to demand-side scale economies. There is essentially a liquidity
effect arising from larger platforms having thicker markets for keywords. Second, platforms that
have superior technologies can earn more from additional searchers and therefore bid more for
traffic thereby accelerating positive feedback effects. Before discussing these features in more detail
we survey the current state of competition in the search-based advertising business.
18
The exact methodology of each search ad platform and the precise dates of implementation are not public but
there have been significant differences across platforms. Google appears to have had a quality score since at least
August 2005 (http://adwords.blogspot.com/2005/12/new-addition-to-quality-score.html). Yahoo does not appear
to have relied on these quality measures until its Panama platform was implemented in February 2007. Microsoft
used Yahoo’s platform until it launched its own search ad platform in the United States in May 2006. Microsoft’s
platform appears to have always had some form of quality scoring. See http://blogs.msdn.com/adcenter/ar-
chive/2006/08/23/715430.aspx discussing an “enhancement” to “existing guidelines around relevance and quality”
in April 2007. There may of course still be significant differences in the effectiveness of the quality scores across
platforms.
19
See Varian (2006) for an analysis of the equilibrium bidding for slots. Also see Michael Schwarz and Konstantin
Sonin, Efficient Actions in a Dynamic Auction Environment (2005).
20
We discuss evidence on this below.
PLATFORM ECONOMICS:
202 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
The industry is highly concentrated with an HHI of over 3,000 based on search traffic and
higher based on advertising revenue.21 The size distribution of firms is also highly skewed. Measured
by 2006 advertising revenue, the largest platform (Google) is nearly three times as large as the
second-largest platform (Yahoo).22 The rankings by search traffic are slightly less skewed.
Most consumers have, or can easily obtain a search engine. The typical computer comes with
a search toolbar preinstalled in the browser. A 2007 study examined PCs sold by OEMs who ac-
counted for at least 1 percent of U.S. home and small office sales in 2006. All of the PCs had a
search toolbar preinstalled, with Google having the greatest number of installations.23 It is also easy
21
The HHI is a standard measure of concentration used in merger analysis. It ranges from a high of 10,000 for a pure
monopoly to 0 for a perfectly competitive industry. Industries with HHIs in excess of 1,800 are considered to be
sufficiently concentrated that US antitrust authorities look closely at mergers in these industries. See Federal Trade
Commission, 1992 Horizontal Merger Guidelines[with April 8, 1997, Revisions to Section 4 on Efficiencies], http://
www.ftc.gov/bc/docs/horizmer.shtm.
22
Neither Google nor Yahoo break out their search advertising revenues from their contextual advertising revenues, so
this ratio is for both search and contextual advertising combined. Google’s worldwide 2006 search and contextual
advertising revenue is estimated at $10.5 billion and Yahoo’s worldwide 2006 search and contextual advertising
revenue is estimated at $3.7 billion. Estimates come from the Lehman Brothers Internet Data Book, April 2007,
pp. 81, 85. The Google to Yahoo ratio based on these revenue figures is about 2.9, while the ratio for search traffic
of Google to Yahoo’s in November 2007 was 1.7.
23
Google was found on three of the seven PCs, Yahoo! on two, AOL on one, and Windows Live on one.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 203
to add additional search toolbars, such as the Windows Live Toolbar and the Yahoo! Toolbar. Thus,
consumers can use multiple search engines if they want and it is relatively easy to do so.24 Looking
among users of Google, Yahoo and Microsoft’s search engines over the course of one month, more
than a third used more than one search engine. But the extent of multihoming differed signifi-
cantly across platforms—only about 40 percent of users of Google used one of the two other search
engines over the course of a month, while the comparable figures were over 70 percent for Yahoo
users and 80 percent for Microsoft users.25 That could come from their using multiple engines for
the same search, or different engines for different searches.26 Nevertheless, most people I have asked
(acquaintances plus several polls of audiences) use a single search engine primarily and tend to use
other ones for idiosyncratic reasons. Moreover, only rarely in my experience do people use multiple
search engines for a single query. Thus in the remainder of the paper I will assume that users “single-
home” on a search-ad platform using the terminology from the two-sided platform literature.
On the advertiser side, multihoming on search ad platforms varies significantly across advertis-
ers. Although representative data are not publicly available, discussions I have had with industry
participants indicate that all of the largest advertisers, such as Amazon and WalMart, will adver-
tise on all of the three major search ad platforms. The value of a click from one search engine is
independent of the value of a click from another search engine since these are different lead op-
portunities if people single-home for a given search. All else equal, there is therefore no reason
to pay for clicks only from a single search engine.27 The major incentive not to use an additional
search-ad platform is the cost of setting up that platform and monitoring ad campaigns on it.
Those costs, which are largely fixed, will be amortized over fewer clicks on a smaller search ad plat-
form. Discussions I have had with industry participants indicate that medium and small search
advertisers will use Google uniformly but are less likely to use Yahoo and, especially, Microsoft. In
24
Google has programmed its search toolbar to make it difficult for users to easily switch. When a user attempts to
change the default search, Google blocks the switch by default and briefly displays a notification in the bottom right
corner of the screen. If the user clicks on the notification before it disappears, a popup window provides the user
with the option to disable Google’s “default search protection”. This does not appear to be the case for Yahoo or
Microsoft.
25
comScore MediaMetrix, Digital Calculator Report, January 2008. Note that the extent of multihoming for users
overall is lower than for any individual platform because a given user that multihomes between say Google and
Yahoo will be counted as multihoming twice in the individual platform multihoming figures and only once in the
overall figure.
26
There are also “meta” search engines such as Excite, Dogpile, and Metacrawler that deliver results from several en-
gines. These are not widely used as of 2007.
27
This statement is not meant to suggest that there are no substitution possibilities from the standpoint of the adver-
tiser. Search generates leads which can translate into sales with some probability. However, advertisers have a num-
ber of different methods available to them to generate leads of which search-based advertising is only one way. We
would expect that advertisers will invest in different methods of lead generation up to the point where the returns
are equalized at the margin. They may therefore increase their investment in platforms that have lower CPCs than
platforms that have higher CPCs for given keywords.
PLATFORM ECONOMICS:
204 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
addition, advertisers both large and small that do use multiple platforms will tend to place fewer
bids on the smaller platforms. In aggregate, this leads to significantly more bids on Google than
on Yahoo or Microsoft.28 We discuss the fixed costs of maintaining search advertising campaigns
in more detail below because they can have important implications for positive feedback effects.
28
Google AdWords Vs. Microsoft AdCenter, Brick Marketing, October 27, 2007, http://payperclickjournal.com/google-
adwords-vs-microsoft-adcenter/10/27/2007/; Henry Blodget, The Real Reason Yahoo’s Revenue Per Search Stinks,
Silicon Alley Insider, Oct 2, 2007, http://www.alleyinsider.com/2007/10/the-real-reason.html.
29
See Yahoo and Procter & Gamble Develop Interactive Traffic Building Promotion, Google and Yahoo News and Press
Releases, May 8, 1996.
30
For a discussion of the evolution of search-based advertising see John Battelle, The Search: How Google and Its Rivals
Rewrote the Rules of Business and Transformed Our Culture (Penguin Group 2005), in particular Chapter 3; also
see Michael Cho et al, Search-based Online Advertising (Working Paper, 2005) people.ischool.berkeley.edu/~hal/
Courses/StratTech07/Tech/Tech05/C-report.doc.
31
Saul Hansell, Google Keeps Tweaking Its Search Engine, New York Time, June 3, 2007. Google has described
PageRank, its system for ranking web pages, as the “heart” of its software. (Google, Our Search: Google Technology,
http://www.google.com/technology/, downloaded December 2007)
32
An interesting area of exploration concerns differences in market structure internationally. Google is the dominant
search provider globally and in many countries. It has reportedly more than a 90 percent share of search traffic in
Argentina, Brazil, and Spain. However, it is not the largest player in China where Baidu has 61 percent of the search
traffic or Japan where Yahoo! has 50 percent.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 205
1st Yahoo Yahoo Yahoo Yahoo Google Google Google Google Google
(S: Inktomi A: (S: Google (S: Google: (S: Google, (S: Own, (S: Own, (S: Own, (S: Own, (S: Own,
Own) A: Own) A: Own) A: Overture) A: Own) A: Own) A: Own) A: Own) A: Own)
2nd AltaVista AltaVista Microsoft Google Yahoo Yahoo Yahoo Yahoo Yahoo
(S: Own (S: Own, (S: Inktomi, (S: Own, (S: Google, (S: Own, (S: Own, (S: Own, (S: Own,
A: DoubleClick) A: DoubleClick) A: Overture) A: Own) A: Own) A: Own) A: Own) A: Own) A: Own)
3rd Excite Lycos Google Microsoft Microsoft Microsoft Microsoft Microsoft Microsoft
(S: Own, (S: Fast Search (S: Own, (S: Inktomi, (S: Yahoo’s (S: Yahoo’s (S: Own, (S: Own, (S: Own,
A: Excite’s & Transfer A: Own) A: Overture) Inktomi, Inktomi, A: Yahoo’s A: Own) A: Own)
MatchLogic) A: Own) A: Yahoo’s A: Yahoo’s Overture)
Overture) Overture)
Notes: Yahoo acquired Inktomi in 2003; In 1999 and 2000, AltaVisa used DoubleClick for banner ad (not ads
matched to user search queries).
Source: Data for years 1999-2003 compiled from the following trade press articles: Sherman Fridman, Yahoo’s Popularity Continues Strong-
Report, Newsbytes New Network, December 22, 1999; Altavista Climbs in Search Referrals, PR Newswire, July 12, 2000; Yahoo Still on
Top in Portal Battle, but MSN Creeping Up, Agence France-Press, May 30, 2001; Google Challenging Yahoo as top Global Internet Search
Engine: Study, Agence France-Presse, April 30, 2002; comScore Media Metrix Launches Breakthrough System to track Actual Consumer
Search Queries, comScore press release, April 28, 2003; Search Sites Play a Game of Constant Catch-Up, New York Times, January 31, 2005.
Data for years 2004-2005 from David Hallerman, Search Marketing: Players and Problems, eMarketer, April 2006. Data for years 2006-2007
compiled from comScore reports.
It is useful to focus on the industry since the dot.com bust which marked the collapse of many
of the early startups and the emergence of the advertising-supported Web 2.0 industry. Since that
time the search-based advertising industry has become increasingly concentrated as Google has
outdistanced its rivals. The gap between Google and rest of the search firms has increased substan-
tially—the ratio of the largest firm (Google) to the next largest (Yahoo) has increased from 1.3 to
3.1 over the same period. Table 4 reports the growth rates in search traffic and advertising revenue
from 2001 to 2007. In the US, Google has exceeded the growth rates of the second two largest
platforms in search in 7 of the last 10 quarters and in revenue in 9 of the last 10 quarters (through
the second quarter of 2007).
Google’s high share combined with trends indicating that its lead is expanding raise several
interesting economic questions: is search-based advertising subject to winner-take-all competition;
how easily can more efficient providers displace the incumbent; and to what extent can differentia-
tion on either platform side permit viable competition among several platforms? We turn to these
questions next.
access to those eyeballs to advertisers. However, several features of the technologies underlying
search-ad platforms give rise to unique features.
33
There would not seem to be any economic or technological obstacle to consumers using multiple search engines.
They are most likely to use the search engine that they expect will provide the best search results which may be some
combination of organic results and ads. However, it would seem possible that a market structure could arise with
differentiated platforms in which consumers find it useful to rely on several platforms or in which they have greater
incentives to use a meta-search platform that combines results from several platforms. Therefore the existence of
single-homing may be endogenously determined by the market structure.
34
In reality there are a variety of ways to generate leads, beyond search-ad platforms, and there may be some substitu-
tion between these different methods. Moreover, there may be diminishing returns to the value of leads for a variety
of reasons including diseconomies of scale in production and distribution.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 207
In fact, there are significant differences in CPCs for keywords across the search-ad platforms.
None of the platforms reports these measures, but industry estimates generally place Google sig-
nificantly above Yahoo. One estimate places Google’s worldwide CPC at around $2.00, nearly
three times its estimate of Yahoo’s CPC of around $0.75.35
35
See Not Out of the Woods But A Step In The Right Direction, Bear Stearns, October 17, 2007. Other estimates of rev-
enue per search (RPS), which is equal to the CPC times the CTR for each search (not the CTR for each advertiser),
place Google substantially above Yahoo. Mark Mahaney, an analyst in Citigroup, estimated that in 2006 Google
made 4.5 cents to 5 cents on every search, while Yahoo generated only 2.5 cents to 3 cents a search (See Miguel
Helft, A Long-Delayed Ad System Has Yahoo Crossing Its Fingers, New York Times, February 5, 2007); Caris & Co.
analyst Tim Boyd estimates that Yahoo made on average between 10¢ and 11¢ per search in 2006, and Google made
between 19¢ and 21¢ per search (See Catherine Holahan, Why Yahoo’s Paname Won’t Be Enough, BusinessWeek, Dec
26, 2006); Justin Post, an analyst with Merrill Lynch, estimated that in 2006, each US search generates 4¢ for Yahoo
and 11¢ for Google (See Robert D. Hershey Jr., Sunny and Gloomy Signs at a Web Crossroads, New York Times, Nov
19, 2006.); A recent article states that most analysts estimate that Yahoo!’s RPS is about 30 percent less than Google’s
(See Henry Blodget, The Real Reason Yahoo’s Revenue Per Search Stinks, Silicon Alley Insider, Oct 2, 2007.). The dif-
ferences between these estimates is likely due to whether one looks at searches only within the U.S. or worldwide,
and whether one looks at only searches for which there are ads or all searches. But within each estimate, Google is
placed significantly above Yahoo.
36
Although search-ad platforms have to make investments in developing search engines and other technologies it does
not appear that there are significant scale economies that would, by themselves, limit the market to one or just a few
players. Search-ad platforms are readily scalable by adding servers and communication. Eisenmann estimates that
a search-based advertising platform could break even with about 7.5 percent of the global market. See Eisenmann,
supra note 16.
PLATFORM ECONOMICS:
208 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
“liquidity”—the volume of buyers and sellers that could reach mutually profitable trades—is well
documented for exchanges.37 Without enough liquidity markets are too thin and unsustainable.
In the case of search-ad platforms the advertisers are the buyers of access to Internet users
while the searchers are selling that access through the ad search platform. More advertisers and
more searches increase the likelihood of profitable matches given that advertisers and searchers are
heterogeneous. It is useful to consider why in more detail.
Searchers obtain more relevant ads when there are more advertisers.38 Suppose an individual is
in Germany and needs a SIM card. She types in “Germany SIM cards” into Google. On September
26, 2007 she would have seen ten ads, of which eight are directly relevant to her query. If she typed
the same query into the smaller MSN search engine on that same day she would have seen eight
ads of which two are directly relevant to her query. That phenomenon is general: search-ad plat-
forms with more advertisers will generally deliver more relevant ads to the searcher; that statement
is particularly true for less common keyword combinations, for which there is a thinner advertising
market. Since many searchers are looking to buy things, the larger platform is more valuable to
them and they are therefore more likely to use the larger platform all else equal.39 (In addition each
searcher benefits from other searcher’s queries because the search engine captures information that
makes search results more relevant to subsequent searches.)
Advertisers also value more searchers. Consider an advertiser that earns $50 per unit on the sale
of widgets. It has access to platform 1 and platform 2 where the first platform has 10 times as many
searchers as the second. On average every click generates a sale 20 percent of the time. Suppose
that it pays $0.50 per click to obtain the 3rd slot on each platform. Platform 1 sends 200 clicks per
week generating 40 sales and platform 2 sends 20 clicks per week generating 4 sales. Then the ad-
vertiser earns $2000 in revenue per week from its campaign on platform 1 for which it pays $100
and earns a profit of $1900; it earns $200 in revenue per week from its campaign on platform 2,
37
See, for example, Kenneth D. Garbade and William L. Silber, Structural Organization of Secondary Markets: Clearing
Frequency, Dealer Activity and Liquidity Risk, The Journal of Finance, Vol. 34, No. 3 (June 1979), pp.577-593;
See also Nicholas Economides, Network Economics with Application to Finance, Financial Markets, Institutions &
Instruments, Vol. 2, No. 5 (1993a), pp. 89-97.
38
Many searchers are looking to buy something and therefore may value relevant advertising. It has been report-
ed that about 40 percent of search queries involve potential commercial transactions. See Thomas Eisenmann,
The Economics of Internet Advertising: Implications for the Google-DoubleClick Merger, Presentation for AEI-
Brookings Joint Center, July 2007; see also Dai, et al, Detecting Online Commercial Intention, World Wide Web
Conference, Edinburgh, Scotland, 2006.
39
Search engines typically capture data on searches and accumulate this over time. (Maria Godoy, Google Records
Subpoena Raises Privacy Fears, NPR.org, January 20, 2006. They use this information to improve the ability of the
search engine to deliver relevant results. Some studies have found that there are not significant differences in the qual-
ity of search results across the major platforms despite the extreme differences in the number of searchers. Thus it
would appear that direct network effects from search are limited. However, search and click-through histories enable
search-ad platforms to estimate CTRs better and therefore gives rise to another potentially significant scale effect.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 209
for which it pays $10 and earns a profit of $190. The advertiser therefore values access to the large
platform more than it values access to the small platform, even though it does not value a click or
a searcher on one more than the other.40
The existence of fixed costs, together with the difference in platform value documented above,
may have an effect on the economics of the search-ad platforms, given the CPC pricing structure,
especially for smaller platforms. Advertisers incur two costs of running campaigns that are inde-
pendent of the number of clicks. First, they incur costs of setting up the platform, installing soft-
ware, and learning how to use it.41 Consequently, the advertiser must exceed a minimal volume of
advertising (or more specifically, a minimal level of incremental profits) from this campaign before
contracting with another search ad platform. Platform set-up costs discourage smaller advertisers
from joining smaller platforms. As discussed above, many smaller advertisers advertise only on
Google or only on Google and Yahoo.
Second, advertisers incur costs of running a campaign on keywords. They have to make deci-
sions on the bids and monitor the performance of the campaign.42 These tasks generally cannot be
automated fully and therefore require humans. Thus, the advertiser must exceed a minimal volume
of clicks on a campaign before mounting it on an ad platform that has been set up. To take the
example above, if it cost $200 per week to monitor a campaign for widgets the advertiser would
run the campaign on platform 1 but not platform 2. Therefore, campaign monitoring costs also
discourage advertisers from mounting campaigns on platforms that have fewer search queries. As
discussed above, even larger advertisers will tend to maintain smaller campaigns on the smaller
search ad platforms.
These considerations lead to a positive feedback loop between the search and advertiser sides.
To see this, consider starting in a situation in which two platforms have equal numbers of searchers
and advertisers. Now suppose platform 1 has an exogenous increase of 10 percent in search traffic
with platform 2 holding steady. That will result in some advertisers joining platform 1 that had
previously found it unprofitable to join either platform, and in some advertisers mounting cam-
paigns on platform 1 that they had previously decided not to mount on either platform.
Platform 1 now has more relevant ads for searchers. We would expect that some searchers
would switch from platform 1 to platform 2. That in turn would increase the volume of advertising
on platform 1. One could go through the same argument with an exogenous increase in advertis-
40
This statement is true except for the situation in which it pays is maximum value per click of $10. Then it earns zero
profit from either platform and is indifferent between them. While the second-price auction in theory is designed
to get bidders to pay their maximum values in reality it is will not do so perfectly and we disregard this extreme
situation.
41
There is also an activation fee for the major search-ad platforms. Google and Microsoft each charge $5; Yahoo has
no fee for the “Self Serve” version, but charges $199 for an assisted setup.
42
Maintaining an advertising campaign requires choosing the right keywords and fine tuning them, modifying bids,
selecting the best landing pages, revising ad text, and monitoring account statistics such as clicks, impressions, CTR,
average CPC, average position, and conversion rate.
PLATFORM ECONOMICS:
210 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
ing on the other platform. In both cases the effect of an advantage on one side becomes magnified
as a result of the positive feedback effects.
Platform 1 obtains a further advantage as it obtains more advertisers. As a result of the keyword
bidding system, an increase in advertisers may increase the bids on keywords. Consider first a situa-
tion in which we would expect the two platforms to secure identical CPCs. Platforms 1 and 2 have
the same 20 bidders interested in 10 slots; the slots are as valuable on platform 1 as on platform 2;
and the keyword auction is equally efficient in the sense of getting the bidders to reveal their highest
values. In this case we would expect the auction to result in the same bids for the same slots. Now
suppose that the number of advertisers on platform 1 increases exogenously by two advertisers, while
the number of advertisers on platform 2 is unchanged. If the new advertisers are situated similarly
to the existing advertisers (e.g., if they are all drawn from the same distribution), then it is likely that
one of the new advertiser’s optimal bid will place it in one of the top 10 slots. Suppose, for example,
the new advertiser falls into the sixth slot. This has two main positive revenue effects for the platform.
First, the new advertisers in slots 6 through 10 all have higher bids than before (because the new 10th
place slot is taken by the advertiser previously in slot 9, and so on). With the higher bids come higher
payments to the platform. The second effect comes from the new advertiser in slot 6 having a higher
bid than the prior advertiser in slot 6. This higher bid increases the bid of every advertiser in slots 1
through 5, because their optimal bids depend (positively) on the level of the bidder below them.43
All else equal, positive feedback effects would lead one ad search platform to achieve a mo-
nopoly position.44 The largest platform would always realize the highest CPC and provide the
largest overall value to searchers and advertisers in the aggregate. The platform that provided the
highest quality search engine for users and ad platform for advertisers would necessarily win the
market. We would expect, though, that these positive feedback effects would diminish with the
size of the platform. That is because the fixed costs become less important as the number of search
queries expand, the value of additional bidders declines as the number of bidders increases, and
more keywords have thick markets as the number of searchers expand.
43
See Varian, supra note 25.
44
This is the classic tipping story for markets with indirect network effects. See Brian Arthur, Increasing Returns and the
New World of Business, Harvard Business Review 74 (July–August 1996): 100–109; Michael Katz and Carl Shapiro,
Systems Competition and Network Effects. Journal of Economic Perspectives 8 (Spring 1994): 93–115. Tipping does
not occur in many real-world markets as discussed in David S. Evans and Richard Schmalensee, The Industrial
Organization of Markets Based on Two-Sided Platforms, Competition Policy International, Spring 2007.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 211
auction, or if platform 2 was better at predicting CTR, and therefore in maximizing revenue from
page placement, or if platform 2 was able to deliver more relevant ads (and therefore better leads)
that advertisers would pay more for. Together, these advantages would enable platform 2 to obtain
a higher CPC for a given keyword and to obtain a higher CTR by offering more relevant ads to
searchers. (We refer to this as an “RPS|S” advantage to denote that it is RPS controlling for size.)
Let us consider the implications of the RSP|Scale differences for the dynamic competition
among platforms. One way for a smaller ad platform to catch up with a larger ad platform is to buy
traffic. In fact, a substantial portion of Google’s search ad traffic comes from third-party sites. AOL
and Ask.com, the next largest search sites after Google, Yahoo, and MSN, use Google to supply
paid search advertising results. Combined, they accounted for about 10 percent of search traffic in
the United States among the top five search engines, compared to Google’s 61 percent.45 Google,
Yahoo, and Microsoft have bid for a number of these exclusive search contracts with third-party
publishers. Google has won most of the head-on competition for which public data is available.
For example, of the top 10 web sites that offer sponsored search results, and are not operated by
firms that also own search engines, 8 use Google to provide paid search results.46
Consider the situation in which the only difference between two platforms is that platform
2 has more search traffic than platform 1. Moreover, there are decreasing returns to scale so the
RPS increases but eventually flattens out with respect to scale (RPS’(S)>, RPS”(S)<0, and both
eventually equal zero). Platform 2 has a higher RPS than Platform 1 but Platform 1 has a higher
increase in RPS at its smaller size for any increase in search traffic. Then for sufficiently large
increases in traffic Platform 1 will receive a larger increase in revenue than Platform 2 and it will
have an incentive to outbid Platform 2. (For smaller increases in traffic, however, that increase
taken alone, can be more valuable to the larger platform.) 47 More generally, Platform 1 could
45
Discussions with industry analysts suggest that 25 percent or more of Google’s search ad revenue comes from part-
ner sites.
46
Web sites owned by Google, Yahoo, Microsoft and Baidu were not considered. The website aol.com was considered,
even though Google owns a 5 percent stake in AOL. Similarly, mapquest.com, which is owned by AOL, was consid-
ered. The website ebay.com uses Google for paid search ad listings outside of the United States but uses Yahoo in the
United States. Go.com also uses Yahoo. Lycos.com uses Ask.com, which in turn uses Google for paid search ad listings.
Amazon uses its A9 search site, which uses Microsoft for search. Download.com uses Search.com. Search.com is a
“meta” search site and includes unpaid search results from Google, Ask.com, Microsoft and others. The paid search ad
results on Search.com are frequently referenced to Google and we have counted it among the publishers using Google.
47
For example, suppose the smaller platform had 10 searches with an RPS of 50 cents and the larger platform had 30
searches with an RPS of 100 cents and that an increase of 10 searches would increase the smaller platform’s RPS to
70 cents and the larger platform’s RPS to 105 cents. Even under these assumptions where the smaller platform gets a
much larger increase in RPS, the increase in profits from getting the additional 5 searches is still larger for the large
platform than the small one ($6.75 versus $5.50). In the long run, the smaller platform would also consider the ag-
gregate benefits of accumulating additional traffic from multiple sources. The value of an increment in traffic comes
not only from increasing its RPS on its existing traffic and the value of the new traffic, but also from the additional
value that can result from future deals that increase its RPS on all of its traffic.
PLATFORM ECONOMICS:
212 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
Value of ∆RPS
(resulting from P1
∆T) for existing
traffic
Value
of ∆T
at new
RPS
∆T ∆T
potentially equalize its RPS with platform 2 if it can buy enough search traffic to eliminate the
difference between the two. That would seem possible if the search queries are expanding over
time and the RPS scale effect eventually vanishes. (Note, also, that the smaller platform could
buy traffic by paying users of Platform 2 to switch as opposed to entering into deals with pub-
lishers to carry its search tool bar.)
Now suppose that platform 2 receives a higher RPS at every level of search query traffic as
shown in Figure 2. In this case, platform 2 has a bidding advantage over platform 1 at all sizes.
Platform 1 can not achieve the same RPS by buying traffic. Moreover, if the difference between
the curves is sufficiently large, platform 2 will continue to have an advantage even if the RPS curve
flattens. Even if the smaller platform gets a greater increase in RPS from buying the same amount
of traffic, it faces two disadvantages relative to the larger and more effective platform: (1) the value
of the incremental traffic at the new RPS is still substantially lower than the other platform and (2)
the value of the increase in RPS is applied to a smaller volume of existing traffic than for the larger
platform. It would seem difficult for a smaller platform to catch up with a much larger platform
that has also has a substantial RPS advantage at every given scale.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 213
48
For example Future, plc is a UK company that focuses on specialized magazines. These include Cross Stitcher,
Guitarist, PC Gamer, Fast Car, Revolver, Mountain Biking UK, and Disney Girl.
49
Mark Rysman, Competition Between Networks: A Study of the Market for Yellow Pages, Review of Economic Studies (2004).
PLATFORM ECONOMICS:
214 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
countries for which data were analyzed.50,51 The fact that it currently receives a CPC that is more
than 2 times larger than its nearest rival also suggests that it has a significant RPS advantage over
its rivals that enables it to bid more than its rivals for incremental traffic.52 In fact, Google has won
most of the significant competitions to provide search toolbars or paid search listings on websites
in the last couple of years, including a deal with AOL, one of the largest search sites of any signifi-
cance besides Google, Yahoo and Microsoft, which operate their own search ad platforms.53
These facts would suggest that the positive feedback and RSP|S effects for Google are signifi-
cant. To coexist, competing platforms would have to narrow the scale differences, narrow any ef-
ficiency differences that lead to a higher RPS for a given scale, and differentiate themselves to com-
pensate for remaining scale and efficiency disadvantages. Competing platforms could also grow
and potentially overtake Google by creating superior search engines that provide more relevant
results for users and superior advertising engines that, among other things, deliver more profitable
sales to advertisers. Such improvements in quality on the two sides of the market may explain how
Google was able to overtake Yahoo which was the initial leader in search-based advertising.
50
Source; comScore, “qSearch 2.0 Key Measures Report,” November 2007.
51
The fact that Yahoo lost its lead in search is of course not consistent with a pure positive feedback explanation for
Google’s success.
52
See supra 37.
53
Google also has deals with Fox, AOL, IAC and Dell. On August 7, 2006, Google and Fox Interactive Media entered
into a multi-year contract making Google “the exclusive search and keyword targeted advertising sales provider for
Fox Interactive Media.” As part of the agreement Google will pay Fox at least $900 million so long as Fox is able to
meet given traffic requirements and other commitments. See http://www.google.com/press/pressrel/ir_20060807.
html. In December 2005, Google concluded a $1 billion deal with AOL which among other things would allow
“AOL… to sell additional ads for its search engine also [which is] powered by Google.” See http://www.news.com/
AOL-to-stick-with-Google/2100-1030_3-5998600.html. Google and IAC (the parent company of Ask.com) ex-
tended their sponsored search and advertising agreement in a move which is worth an estimated $3.5 Billion to IAC
over the next 5 years. See http://www.nytimes.com/2007/11/06/technology/06diller.html. Dell also recently (May
2006) concluded a deal with Google which “means millions of Dell computers will leave the factories with Google
software already installed on them.” See http://news.bbc.co.uk/2/hi/technology/5019416.stm.
54
Search-engines are a special type of publishers. They use search result pages to attract viewers and then sell access to
these viewers to advertisers.
55
Source: comScore My Metrix report, November 2007.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 215
media companies such as The New York Times and CNN that have established online extensions,
new media companies such as You Tube, blogs such as technorati.com,56 social network sites such
as Facebook; and web portals that operate their own ad platforms including MSN and Yahoo. The
ad inventory on these websites is often called “display” and often looks much like one sees in print
media. Unlike the suppliers of ad inventory on search results pages, the suppliers of display adver-
tising inventory usually charge based on the cost-per-thousand viewers (CPM); that pricing model
may change as discussed below.
The display advertising business has a complex business ecosystem. Unlike search which is
based on an integrated two-sided platform that sits between the viewer and the advertiser, display
advertising consists of several different platform models. These platform models differ according to
the extent to which intermediation between the publisher and the advertiser is integrated into the
publisher that supplies the inventory and the extent to which software tools that link advertisers
and publishers are integrated into either the publisher that supplies the inventory or the interme-
diary between the publisher and the advertiser. The provision of advertising to viewers is therefore
often provided by two or more interconnecting two-sided platforms. Part A provides an overview
of the display advertising business. Part B describes the current market structure, including the role
of different two-sided platforms. Part C discusses ways in which this industry, which is currently
going through significant changes, may evolve.
56
David S. Evans, Economics of the Blogosphere, The Catalyst Code, August 6, 2007, http://www.thecatalystcode.
com/theconversation/blog/2007/08/06/economics-of-the-blogosphere/.
57
Web pages can also include languages such as JavaScript that generate HTML code when a page is called.
PLATFORM ECONOMICS:
216 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
Page Top
Top Top
Left
Content Right
Middle
Bottom Bottom
Left Content Right
Bottom
The key technological difference between traditional display advertising and online display
advertising involves the ability to target ads to particular consumers in real time. In the traditional
advertising industry targeting is primarily accomplished through the development of media that
use content to attract particular socioeconomic and demographic groups (from broad groups such
as people between 29 and 55 to narrow groups such as fly fishermen). Traditional advertisements
are seen by all views of the media containing the ad. Online advertising makes it possible to serve
an advertisement specifically for the viewer of the web page based on knowledge of the viewer’s
socioeconomic characteristics or even what sorts of products the viewer is considering purchasing.
Online advertisers and publishers currently engage in targeting to a relatively modest degree.
An advertiser will buy the front page of Yahoo, which is not that different from buying advertising
on a television show—the advertiser is interested in reaching those audiences but is not targeting
beyond that. Many of the investments currently made in online advertising (including the pur-
chases of ad networks and ad serving and behavioral targeting providers), however, reflect a belief
that more sophisticated targeting will become increasingly important.
Unlike traditional advertising, online ads can use the IP address of the computer that has
signed on to the website to focus on narrow geographic areas including ones with particular
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 217
s ocioeconomic characteristics. Publishers also have the ability to determine what ads viewers have
seen recently on their sites and limit duplicate messages. They can also determine what content the
user has recently been interested in—for example, an auto advertiser may be interested in reaching
viewers who have recently looked at the auto reviews, but with online advertising these viewers can
be shown ad not only on the auto section but on other content pages they may be viewing. Viewers
for whom the publisher has registration or other identifying information offer further targeting
opportunities. Contextual advertising platforms, as discussed below, provide some additional tar-
geting by considering whether certain keywords such as “luxury hotels London” occur on the web
page. It is expected, however, that online advertising will increasingly make greater use of viewer
data—whether actual or inferred—to better target more ads to viewers and thereby make online
advertising make effective for advertisers. We discuss such “behavioral targeting” below in examin-
ing future developments in the online advertising business.
Like all sellers and buyers, publishers and advertisers require ways to identify optimal trading
opportunities and to establish transaction prices. There are two major ways that this “interme-
diation” occurs. First, it can occur directly through bilateral exchanges between publishers and
advertisers. eBay, for example, may sell Nokia the right to present an advertisement in a particular
spot to viewers with specific characteristics by having its sales agents deal directly with Nokia’s
buying agents. Second, it can occur indirectly through multilateral exchanges between publishers
and advertisers using advertising networks. Hearst Publishing may sell ValueClick—an advertis-
ing network—advertising inventory from its various online newspaper and magazine properties.
ValueClick may then sell advertisers access to particular kinds of viewers by pooling relevant ad-
vertising inventory from Hearst and other publishers. In this case, the advertiser buys access to a
type of viewer—for example “fashion conscious young women in upscale locations”—but is not
necessarily guaranteed that it will reach these women on Hearst’s Cosmopolitan website.
The extent to which advertisers and publishers use direct and indirect methods of distribution
for advertising varies. Smaller publishers and advertisers typically rely only on indirect methods be-
cause it is not economical to carry the cost of salespeople and purchasing agents. Larger publishers
and advertisers—which account for the preponderance of online display advertising sales—often
use both direct and indirect methods. Larger publishers use direct methods to sell their premium
space—such as the space on the first web page and high quality space on some of the interior web
pages—and this space is then said to be “reserved” for an advertiser. They also use indirect methods
to lower quality space such as the bottom space on interior pages, or advertising inventory that has
not been reserved either because there was an unexpected spike in viewership which increased the
supply of inventory, or because they did not sell all of their advertising inventory at the prices sought
in direct sales. Advertising inventory that has not been reserved is sometimes called “remnant”. We
discuss the relative pricing and use of directly and indirectly sold online advertisements below.
Advertisers and publishers rely on various services which they either provide themselves or pur-
chase from others. Large advertisers and advertising agencies often license an advertising package
PLATFORM ECONOMICS:
218 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
as aQuantive’s AdManager.58 They typically manage advertising on hundreds of websites and across
numerous products using many methods of online advertising. The advertising software helps
them manage these various campaigns and instructs the server to deliver advertising copy down-
stream into available inventory.
Large publishers usually license a publisher tool such as DoubleClick’s DART for Publishers
(DFP).59,60 Publisher tools are highly sophisticated software programs. They are usually integrated into
many other aspects of the website technology and business practices. The publisher hard codes links to
the publisher tool to fill the ad space for which it wants to use the management, reporting, and serv-
ing capabilities of the publisher tool. When a viewer clicks on a web page the publisher tool makes a
decision, based on preprogrammed rules, as to which ad to present to the viewer. The publisher tool
checks whether the particular ad space that the viewer is about to see has been “reserved”, and if not,
whether there is an ad network that can fill that inventory space. The publisher tool then retrieves the
chosen ad, or directs the advertiser or ad network’s server to retrieve the ad, which is then shown to the
user. This entire process happens in the blink of an eye. A few very large publishers rely on publisher
tools they have developed themselves rather than licensing a publisher tool from a third party.
Publishers also use a variety of advertising networks for indirectly sold advertisements as we
discuss below.
• Publishers supply content on their websites that attracts viewers and sell the resulting adver-
tising inventory to advertisers directly or indirectly.
• Advertising networks serve as intermediaries between publishers and advertisers by aggregat-
ing the demand and supply of advertising inventory. (Other intermediaries such as third-
party sales representative also operate two-sided intermediation platforms.)
• Advertising software tool companies usually supply both software tools for publishers and
software tools for advertisers; they at least try to generate positive feedbacks between the
publisher and advertiser side.
Table 5 lists the top two firms in each of these categories and estimates of their revenues.
58
See http://www.atlassolutions.com/services.aspx and http://www.doubleclick.com/products/dfa/index.aspx.
59
See http://www.doubleclick.com/products/dfp/index.aspx.
60
A handful of mega-large web publishers such as MSN have their own proprietary tools but most others use a third-
party tool.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 219
Notes: i) Rankings and revenues of publishers are for the United States only. All other rankings and revenues are based on world-
wide data.
ii) The Advertising Age figures for publisher revenue are industry estimates.
iii) Google does not report contextual revenue publicly. The given figure is a rough estimate based on discussions with
industry experts.
iv) The revenue number for ValueClick is for the company’s Media segment, which includes Display Advertising, Lead
Generation Marketing, Email Marketing and E-commerce, so the display ad revenue may be somewhat overstated.
v) The revenue figures for DoubleClick and 24/7 RealMedia are for the respective companies’ entire revenue, including but
not necessarily restricted to revenue from the provision of software tools.
vi) Atlas’ revenue figure is from aQuantive’s ‘Digital Marketing Technologies (DMT)’ segment. In its annual report, aQuan-
tive states that “[o]ur DMT segment consists of Atlas, a provider of digital
marketing technologies and expertise.”
PLATFORM ECONOMICS:
220 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
• Large integrated online publishers have their own publisher software tools for operating
their websites. In addition to their direct sales forces, they also have advertising networks
that sell their non-reserved inventory to advertisers.
• Large nonintegrated publishers have their own direct sales forces. However, they usually
license a third-party publisher tool and work with one or more third-party advertising net-
works to sell their non-reserved inventory to advertisers.
• Small nonintegrated publishers typically do not have their own direct sales forces. They
usually rely on an advertising network that has its own publisher tools. Once they paste a
direction to that advertising network in their advertising space they need do nothing more.
Detailed data on the sales of these three types of publishers is not available. However, discus-
sions with industry experts suggest that the substantial majority of display ad revenue are for ads
placed on the first two types of publishers. The largest publishers— , AOL, MSN and Yahoo—are
integrated, fall under the first category, and collectively account for ad revenue that is likely on the
same order of magnitude as the nonintegrated publishers.
There are two major kinds of advertising networks.
• Contextual ad networks use the same software that they use for managing their search-based
advertising campaigns to manage their contextual-based advertising campaigns. Advertisers
use this software to bid on keywords on the publishers that rely on the contextual network.
The contextual ad network then insert ads provided by the advertiser on those websites based
on the advertiser’s bid and the content of the sites on its network.61 Advertisers also have some
ability to select inventory based on the website or based on various characteristics of users.
• Most online networks do not use contextual targeting. These standard online advertising
networks specialize in aggregating advertising inventory from a variety of publishers that
corresponds to some group for which there is demand from advertisers. They then sell ad-
vertisers the ability to present their ads to some number of viewers in that group without
necessarily guaranteeing that the ads will be seen on any particular site. As mentioned below
some of these ad networks are using behavioral targeting techniques to deliver more relevant
ads for viewers and advertisers.
Table 5 lists the leading advertising networks worldwide. Google operates by far the largest
contextual advertising network and has the largest advertising network overall, with ad revenues
61
The contextual platforms also used other characteristics for targeting in addition to content.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 221
that are likely three times those of its next closest competitor in 2006. Other leading firms include
Advertising.com with revenue of $455 million is 2006, and ValueClick with revenue of $383 mil-
lion in 2006. A number of even smaller competitors, such as Tribal Fusion, BlueLithium, Quigo
and Tacoda are not publicly traded and financial data on their ad revenues are not available.
Finally, the leading software tools providers make tools for both publishers and advertisers even
though these tools are in some sense very different software. They have at least attempted to per-
suade publishers and advertisers that they would benefit from their tools because the other side of
the market is using those tools. On the DART for Publishers web page, for example, DoubleClick
states, “[o]ur position at the center of the digital media landscape means we can deliver unparal-
leled buyer/seller integration, including zero discrepancies on ads trafficked from DoubleClick’s
DART for Advertisers solution to its DART for Publishers solution. This puts money directly into
your pocket and simplifies billing and reconciliation.”62 A key development in 2007 was that the
leading software tool providers were purchased by other platforms. Google bought DoubleClick,
Microsoft bought aQuantive, and WPP—a large advertising agency—bought 24/7 RealMedia.
DoubleClick is generally regarded as the largest tools provider,63 especially for major publishers.
Table 5 lists the leading tools providers worldwide. In 2006, DoubleClick had revenue of about
$300 million. Tools providers not listed in the table include MediaPlex and ValueClick.
There is a close relationship between the form of intermediation used by publishers and their
use of publisher tools and advertising networks. Publishers that want to sell directly must have a
third-party publisher tool or provide their own software. Publishers that want to sell some space
indirectly must use an advertising network. Comprehensive data are not available on the portion
of directly and indirectly sold advertising inventory. Advertisers and publishers I have talked to,
however, seem to agree that somewhere around 70 percent of advertising revenue for large pub-
lishers is the result of direct sales and the remaining 30 percent of advertising revenue the result
of indirect sales through contextual or traditional advertising networks. Because indirectly sold ad
inventory is much cheaper than directly sold ad inventory, the advertising networks account for a
higher percentage—perhaps around 50 percent—based on the number of “impressions” (the view-
ers that see an ad). However, some individuals in the industry put the direct sales figure as high as
95 percent of revenue for large publishers.
62
See http://www.doubleclick.com/products/dfp/index.aspx.
63
See http://www.redherring.com/Home/pages/print/posts/?bid=28262ee6-7dba-4f55-bb53-cd1a4e1d52c0&mode=Full.
PLATFORM ECONOMICS:
222 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
of the viewers; and the type of intermediation used. As is common in business-to-business transac-
tions large publishers often engage in individual negotiations with advertisers and discount prices
off of a rate card. The contextual ad networks typically charge advertisers on a CPC basis which is
determined from the keyword bidding process. Not surprisingly, large publishers receive payments
that are in the same range as for advertising networks for the same amount of space (three or four
contextual ads typically appear in the same space), on the order of $2 CPM.
The publishers have to pay for intermediation services when they are not using a direct sales
force to sell the ad. Google, the leading contextual network, is believed to keep about 15-20 per-
cent of advertiser revenues.64 The standard advertising networks typically keep 20 percent or more,
although the negotiations are highly individualized.65 The tool providers also impose a transaction
cost on the buying and selling of advertising inventory. In all cases the advertising networks keep
a portion of the revenue they receive from advertisers as a commission. For directly sold ads the
publisher tools provider will usually charge about 5 cents per thousand impressions served and the
advertiser tool provider about 7-8 cents per thousand impressions.
64
Across all of Google’s partner traffic, including both contextual and search, in the 12 months from Q4 2006 to Q3
2007, Google retained a 16 percent share. This share likely varied by type of traffic and publisher. See Google 2006
10-K Form, Google 10-Q form for the quarterly period ended September 30, 2007. To the extent the percentage
retained differs significantly between search and contextual, then this figure may over or understate.
65
In discussions with industry experts, I have heard the share retained by the ad network may be as high as 50 percent
in some cases. Typically, the ad network offers the publisher a fixed CPM guarantee and retains whatever increment
it can sell the inventory for.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Economics of the Online Advertising Industry 223
narrowly defined inventory.66 However, they also have the opportunity to reduce what appear to
be rather high transaction costs for buying and selling advertising inventory. Based on the figures
reported above the transactions costs incurred by large publishers and advertisers is on the order of
20 percent or more of the sales price after the ad networks and publisher tool providers are taken
into account. That is much higher than intermediation costs in other industries—0.1-1.7 percent
for financial exchanges67 and as little as 1-2 percent for some real estate transactions.68
The second development increases the value of electronic exchanges and is therefore could
help these exchanges achieve a critical mass of liquidity. Data-intensive behavioral targeting can
significantly increase the likelihood that an ad will reach a consumer to whom that ad is relevant
and who is likely to translate into a sale for the advertisers. As this technology improves the value
of selling ads through automated processes rather than through direct selling would increase. This
sophisticated targeting faces at least two challenges. The first is that it is technologically difficult
to process the relevant data and serve the advertisement so that it is part of the web page when the
viewer first looks. The second is that behavioral targeting relies on extensive personal information
the use of which has drawn increased scrutiny by regulators and the media.69 Online advertising
companies currently collect and store significant data through the insertion of cookies on personal
computers and other techniques. It is possible that these activities will be curtailed.
The market structure that would result from these developments would have one or more
advertising platforms sitting between publishers and advertisers. Trading platforms typically have
significant scale economies and positive feedback effects. That results in a small number of viable
players. In the case of financial exchanges the globalization of markets combined with the develop-
ment of electronic trading has led to consolidations and increased concentration. Several factors
suggest that there would be stronger forces leading to the emergence of a single advertising platform
than has been the case for financial exchanges. First, behavioral targeting requires the use of histori-
66
These are B2B exchanges. We know from the dramatic collapse of B2Bs in the early 2000s that obtaining liquid-
ity, especially from the supply side (publishers in the case of online advertising) is the most serious challenge. See
Silicon.com, B2B exchanges - What happens now?, January 23, 2002; Hazel Ward, B2B exchanges fail to deliver,
Computer Weekly, November 29, 2001; Bob Tedeschi, Investing; How Killer B-to-B’s Went Into A Tailspin, New York
Times, May 7, 2000.
67
Data are for 2004. See Competition in the EU securities trading and post-trading-Issue paper, European Commission,
2006.
68
Data are for 2002-04. See Estate agency market in England and Wales, OFT, 2004.
69
The FTC has heard privacy cases against several web properties, including the online advertiser ValueClick, which
was ordered in March 2008 to pay a fine of $2.9 million “to settle Federal Trade Commission charges that its ad-
vertising claims and e-mails were deceptive and violated federal law.” See http://www.ftc.gov/opa/2008/03/vc.shtm.
The FTC has also published a report on privacy issues: Protecting Consumers in the Next Tech-ade: A Report by
the Staff of the Federal Trade Commission, March 2008, available at http://www.ftc.gov/os/2008/03/P064101tech.
pdf. Further, there have been moves towards legislation in the states of New York and Connecticut that would
restrict the use of personal information for online advertising. See http://www.nytimes.com/2008/03/20/business/
media/20adco.html.
PLATFORM ECONOMICS:
224 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
cal information on consumers often pulled from their past behavior as observed by the advertising
platform. Larger platforms could secure advantages that would be hard for smaller platforms to
replicate. Second, behavioral targeting may have some of the long-tail effects that have given large
search platforms advantages. As the supply and demand is sliced more finely, larger platforms have
thicker markets in every niche. Third, effective behavioral targeting is likely to require significant
technical advances as well significant computer capacity. The IP rights in particular could establish
barriers to smaller players.
Winner-take-all competition stories, however, tend to work better in theory than in practice
in many industries. While it seems implausible that there will be many commercially significant
advertising platforms, two factors identified in the multi-sided platform literature could lead to
several viable platforms. The first is whether it is possible for advertising platforms to differenti-
ate themselves sufficiently from their rivals from the standpoint of advertisers or publishers while
maintaining sufficient liquidity. The second concerns the extent to which advertisers or publishers
can multi-home. There would not appear to be any obstacles to advertisers using multiple adver-
tising platforms. Publishers currently rely on their publisher tools to multi-home with multiple
alternative channels. If publisher tools become integrated into the advertising platform the abil-
ity to multi-home is likely to decline, thereby reducing competition (except for exclusives) at the
advertiser level.
The development of advertising platforms for trading ads and eyeballs would alter market struc-
ture in the publishing industry as well. Most online and offline media companies are vertically
integrated into selling advertising. They have their own sales forces which, for the larger publishers,
sell the bulk of advertising. As advertising platforms evolve, publishers may specialize in attracting
content and leave the advertising business to the platforms. Many small publishers now do just that.
b usinesses are providing more content and services that people want to spend time viewing. The
number of Internet users worldwide increased from 741 million in December 200670 to 816 mil-
lion in December 2007.71 Younger people in particular are spending more time on social network-
ing sites such as Facebook and online gaming sites such as Pogo.com.
In addition, more of the devices that people use for consuming content will have an Internet
connection in the future. More than 1 billion PCs will be in use worldwide by the end of 2008.72
There are 3 billion mobile phone subscribers worldwide and nearly 2 billion smart phones in use.73
An increasing portion of those mobile phones will have internet connections and browsers as does
the iPhone. It is expected that many televisions will have Internet connections as well.74 Other
devices may have internet connections as well. Amazon’s e-book reader has an always-on wireless
feature that would enable it to deliver targeted ads as well.
Finally, many of the advertising-supported web-based platforms have encouraged developers
to write applications that work with their platforms. These applications may also carry advertising.
For example, Google has encouraged developers to write applications using Google Maps but has
reserved the right to insert advertisements into those applications.75
These changes are important for understanding the current structure and evolution of the on-
line advertising business. They are also critical for numerous public policy issues that have emerged.
Will a single ad platform emerge or will several remain viable? What are the consequences of alter-
native market structures for ad platforms for a web economy that is increasingly based on selling
eyeballs to advertisers? Data is central to these ad platforms. Historically, communication providers
such as the telephone companies have been highly respectful of privacy rights. The ad platforms
have business models that are based on collecting and hoarding highly personal data from indi-
viduals. The implications of that have not been fully thought through by either consumers or
policymakers.
70
comScore MyMetrix Custom Media Trend Report, June 2007.
71
comScore MyMetrix Key Measures Report, December 2007.
72
See http://www.forrester.com/Research/Document/Excerpt/0,7211,42496,00.html
73
Associated Press, Yahoo revs up mobile phone - Plan hopes to attract new users, more ads, The Star-Ledger, January 7,
2008; Alan Duerden, CIOs On The Back Foot Over Handheld Device Security, The Banker, October 1, 2007.
74
See http://tvdecoder.blogs.nytimes.com/2008/04/16/tv-but-with-connections/.
75
See http://www.google.com/accounts/TOS
CHAPTER
The Online Advertising Industry:
8 Economics, Evolution, and Privacy
David S. Evans
ABSTRACT
Online advertising accounts for almost 9 percent of all advertising in the United States. This share
is expected to increase as more media is consumed over the internet and as more advertisers
shift spending to online technologies. The expansion of internet-based advertising is transform-
ing the advertising business by providing more efficient methods of matching advertisers and
consumers and is transforming the media business by providing a source of revenue for online
media firms that compete with traditional media firms. The precipitous decline of the newspa-
per industry is one manifestation of the symbiotic relationship between online content and on-
line advertising. Online-advertising is provided by a series of interlocking multi-sided platforms
(also known as two-sided markets) that facilitate the matching of advertisers and consumers.
These intermediaries increasingly make use of detailed individual data, predictive methods,
and matching algorithms to create more efficient matches between consumers and advertisers.
Some of their methods raise public policy issues that require balancing providing consumers
more valuable advertising against the possible loss of valuable privacy.
1
The top websites were based on ComScore MediaMetrix rankings. A researcher reviewed each website to determine
whether it presented advertising on the home page or immediate branches of the home page. They determined
whether advertising was the most significant source of revenue based on considering other apparent sources of rev-
enue and the business model followed by the site. The appendix provides the details. The top 100 websites accounted
for 41 percent of all page views on the web according to ComScore.
226
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Online Advertising Industry: Economics, Evolution, and Privacy 227
of Commerce, 2008), which is becoming an ever larger portion of overall commerce. Web-based
sellers use online advertising to drive consumers directly to their sites where they can browse for
goods and services and purchase them with a few clicks. Online advertising accounted for 8.8
percent of all advertising in the United States in 2008 (Hallerman, 2008). That share is expected
to grow as people spend more time with new online media such as social networking and sites that
offer user-generated content; as more traditional media such as television is transmitted over the
internet; and as more advertising is delivered to browser-enabled mobile phones.2
Internet-based advertising is the source of a “gale of creative destruction” (in the words of
Schumpeter, 1942) that is sweeping across the advertising and media landscape, especially in the
United States. Newspapers, particularly, are losing readers and advertisers to web media supported
by online advertising. That has lead to a downward spiral as indirect network effects work in
reverse. The market caps of the major publicly traded newspaper businesses in the United States
declined by 42 percent between January 2004 and August 2008, compared to a 15.6 percent in-
crease for the Dow Jones Industrial average over that same time period (Blodget, 2008). With the
additional pressure of the financial crisis, newspapers in several major cities including Denver and
Seattle have closed down and others including the New York Times are in distress. More generally,
online advertising is disrupting all aspects of the global advertising industry, which had estimated
revenues of $625 billion in 2007 (Minton, 2007), from how creative work is done, to how adver-
tising campaigns are run, and to how advertising is bought and sold.
Online advertising methods are, arguably, leading to significant reductions in transactions
costs between merchants and consumers. The methods enable merchants to deliver information
that is targeted to those consumers who value the information the most and are most likely to act
on it. An oft-quoted line in the advertising business states ruefully: “Half the money I spend on
advertising is wasted. The only trouble is I don’t know which half.” The new techniques replace a
sledgehammer with a scalpel. In doing so, they collect and analyze detailed information about how
people use their computers—raising difficult issues concerning the expectation of privacy and the
regulation of the online advertising industry.
No research has yet examined the value of the productivity improvements created by on-
line advertising technology but they appear significant. Consider a business that sells saltwater
fishing rods to people who enjoy fly fishing. The traditional approach to matching this buyer
and seller involved the creation of a magazine, such as FlyFisherman, with content that attracts
the relevant people. In contrast, the online approach relies on a variety of techniques to match
an advertising message to a consumer. A search engine indexes web results that are relevant
2
Moreover, a consortium of cable television providers are developing methods for delivering targeted advertising to
cable subscribers over their networks; although the advertisements are delivered over the closed cable networks rather
than over the internet the targeting methods and business models are similar to those used for online advertising.
Tim Arango, Cable Firms Join Forces to Attract Focused Ads, NY Times, Mar. 10, 2008.
PLATFORM ECONOMICS:
228 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
to a consumer who types in the phrase “saltwater fishing rod,” and with this information, the
search engine can sell ads to sellers of saltwater fishing rods. Contextual advertising on web
pages could do the same thing. A consumer who visits a blog for fly fishermen could be pre-
sented with an advertisement. Developing behavioral targeting techniques, discussed below,
can also identify individuals who are interested in fly fishing and determine whether they are
looking around the web for information that would suggest they might be in the market for
a saltwater rod.
This chapter presents the evolution of the online advertising business. It examines the sup-
ply of online advertising “inventory,” which equals the space times the views of that space; the
demand for that inventory; and intermediaries that operate between the sell and buy sides. It
also explores some of the key developments such as behavioral targeting for matching advertis-
ing messages to consumers and considers some economic aspects of the privacy issues that these
technologies raise. Aside from providing a survey of an important new segment of advertising
this essay suggests, at various points, that many of the interesting questions, and economic puz-
zles, about the advertising industry—offline and online—remain to be addressed. For example,
online advertising is a “two-sided market” (Rochet and Tirole, 2003; Anderson et al., 2005), as
is advertising generally. Intermediaries operate platforms that facilitate advertisers and consum-
ers connecting with each other. Innovative intermediaries operate exchanges and face the critical
liquidity issues discussed in the market microstructure literature (O’Hara, 1998; also see Evans
and Schmalensee, 2009).
engine between helping people find things quickly and keeping eyeballs trained on the site to make
money. The search engines later moved to the cost-per-click model. GoTo.com—which is now
owned by Yahoo!—introduced many of the key technological and business model innovations in
the next three years (“GoTo to Overture,” 2005; “History of Pay Per Click, 2007). These included
adopting the cost-per-click approach to pricing and the use of auctions to allocate the advertising
spots on the page showing results of the search.
During this same period, fairly traditional methods of advertising were mimicked on the
web. These included web versions of business directories similar to the yellow pages such as
yellowpages.com; web versions of newspaper classified ads such as Craigslist; and web versions
of direct mail and telephone marketing such as CheetahMail. These web-based vehicles were
charged for in ways that were similar to their traditional counterparts with the exception of
Craigslist, which provided a significant amount of free advertising to consumers. The remainder
of this chapter does not discuss web-based directories or email advertising because they do not
raise particularly novel issues.
Online advertising revenue has increased steadily over time in absolute terms and as a fraction
of all advertising revenue. Consistent figures are available since 2000. They show that U.S. online
advertising has increased from $8.1 billion in 2000 to $21.2 billion in 2007 and from 3.2 percent
of all advertising to 8.8 percent over that time period (based on Interactive Advertising Bureau
Press Releases 2000-2007). The relative mix of online advertising has also changed. Table 1 shows
the evolution of various online advertising formats from 2000 until the first half of 2008. In 2008,
search and display-related ads were the leading advertising formats with 44 per cent and 21 per
cent share of the total revenue, respectively. However, ads that are linked to the results of a keyword
search have grown explosively from 1 percent of online advertising revenue in 2000 to 44 percent
in 2008, while display ads dropped from 48 percent of total revenue in 2000 to 21 percent of total
revenue by the first half of 2008.
In 2009, fifteen years after its birth, the online advertising industry remains in considerable
flux. The delivery of online advertising exhibits rapid technological change. At the same time, new
economic structures are emerging and business relationships among the key players are chang-
ing. For example, in 2008, Google completed its acquisition of DoubleClick, which was a major
technology provider, and Microsoft entered into negotiations, ultimately aborted, to buy Yahoo!
Nevertheless, certain features of the “online advertising ecosystem” have become clear and provide
a useful framework for this discussion.
On one side of the business are advertisers that want to reach consumers and consider
online advertising as a possible way to do that. On the other side are consumers who may or
may not be receptive to receiving advertising messages. In between are various intermediaries.
The fully integrated intermediaries touch consumers and advertisers directly. The search-based
advertising platforms (Evans, 2008) are examples: they bring consumers to their search results
pages and sell access to these consumers directly to advertisers; their platforms integrate all
PLATFORM ECONOMICS:
230 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
Ad Formats Definitions:
Banner is a space (usually rectangular) on a Web page that shows the advertiser's message.
Sponsorships represent custom content and/or experiences created for an advertiser which may or may not include ad elements.
(e.g., reskinning a section of a web site with the advertiser's branding).
Rich Media refers to advertisements that incorporate animation, sound, and/or interactivity in any format.
Slotting fees the fee charged for premium ad placement and/or exclusivity.
Digital Video format includes commercials that appear in live, archived, and downloadable streaming content.
Search refers to advertising by paying Internet companies to list and/or link their company site domain name to a specific search word or
phrase. It includes paid listings (text links appear at the top or side of search results for specific keywords), contextual search (text links
appear in an article based on the context of the content, instead of a user-submitted keyword), and paid inclusion (guarantees that a
marketer's URL is indexed by a search engine).
Classifieds: posting a product or service in an online listing for a fee.
Lead Generation are referrals to qualified purchase inquiries.
Email ads include banner ads, links or advertiser sponsorships that appear in commercial email communication.
Interstitials are ads displayed during a transition from one Web page to the next.
Source: Interactive Advertising Bureau Annual Reports and Press Releases, 2000-2008
the necessary technology for doing this. Many intermediaries are partly integrated. Publishers
such as reuters.com bring consumers to their sites and have direct sales forces that sell adver-
tising inventory directly to advertisers. But these publishers also typically rely on technology
providers (“ad servers”) that handle passing ads from the advertisers to the publishers’ adver-
tising spaces as well as advertising networks which aggregate online advertising inventory and
sell it to advertisers. Finally, some publishers are highly specialized and contract most tasks
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Online Advertising Industry: Economics, Evolution, and Privacy 231
out. That is true of blogs—even large ones—which rely on ad networks such as Google’s
Content Network to sell ads for them and to provide the relevant technology. Figure 1 shows
the relationship between the various agents.
Almost all of the participants in intermediation between advertisers and consumers operate
multi-sided platforms, sometimes working with agents for the advertiser or consumer. For exam-
ple, media-buying firms work for advertisers and advertising agencies on the “buy side” and with
publishers on the “sell side.” This results in an industry of interlocking multi-sided platforms.
Some of these platforms have more “sides” than just buy and sell. Facebook, for example, operates
a software platform (Evans, 2009) that encourages developers to write applications that also enlist
advertisers and consumers.
As with advertising generally, a key feature of online advertising is that consumers are paid with
content and services to receive advertising messages and advertisers pay to send these messages.
A fundamental question, not addressed here, is why this particular pricing and reward structure
has held over long periods of time and across many different types of advertising. Among other
things, the answer would help to illuminate the extent to which advertising is a method for reduc-
ing transactions costs between buyers and sellers or a source of imperfection that distorts decision
making (Bagwell, 2002).
Eyeball
Liberto.it, Spiegel.de,
Attracting eyeballs with content Publishers
FT.com, engadget.com
PLATFORM ECONOMICS:
232 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
example, search engines know the keywords a user requested, and publishers know the content
of the page the user is looking at. Both may know recent search or browsing behavior. Ads can
be customized based on this information.
These features can make online advertising a more efficient matchmaking vehicle for advertis-
ers and viewers than offline advertising. Advertisers can target their messages to those consumers
for whom the messages are most relevant and who are most likely to buy as a result of receiving
this message. Viewers are more likely to receive messages that are relevant and valuable to them.
An implication of this observation is that there are economic incentives for advertising and viewing
to move online.
TABLE 2 The top 20 web properties with ads and the content they use
to attract viewers
2008 Internet
Advertising
Rank Property Content Revenues ($M)
1 Fox Interactive Media entertainment video, news, social networking, image 900
hosting, games network
2 Yahoo! Sites search results and various applications (news, email, 3430
weather forecast)
3 Google Sites including search results, email, maps, user-generated videos, 7430
YouTube blogs
4 Microsoft Sites* search results, email, entertainment videos, music, news 1970
5 AOL LLC news, entertainment, email, search results, greetings 1360
6 FACEBOOK.COM social networking site 130
7 eBay online auction and shopping site for mostly used goods 2370
9 Comcast Corporation TV listings, Free TV episodes, cable television services
10 Viacom Digital entertainment news, videos, music clips, TV listings,
reviews
11 Time Warner - (excl. movies, TV schedule, videogames, cable television
AOL) services,
14 Amazon Sites online shopping site, daily blog, customer reviews
15 EA Online strategy video games
16 Ask Network search results, public generated questions and 480
answers, maps, news
17 Photobucket.com LLC image and video hosting, sharing applications
19 BEBO.COM social media network, music, videos, applications
21 Cox Enterprises Inc. news, entertainment news, motor vehicles marketplace
22 Disney Online entertainment videos for kids, games, music library, 270
fairy tale creating applications
24 United Online, Inc floral product, internet services, social networking
25 Glam Media news and content on fashion, health and life style
26 ESPN sports news and videos, TV listings, sports results and
rankings
Notes:
1) Source: comScore Media Metrix data for February 2008.
2) Rankings are based on page views. Only web properties with ads are included. Some properties included well-known sites which are
therefore not broken out separately; MySpace is included in the Fox properties and accounts for 98 percent of page view
3) Content information is determined through direct website reviews.
4) Although not all of Microsoft sites supply advertisements, e.g., MICROSOFT.COM, MSN.COM gains most of its revenue from displaying
ads and accounts for 95% of all page views of Microsoft sites.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Online Advertising Industry: Economics, Evolution, and Privacy 235
9%
8%
Percent of Total Page Views
7%
6%
5%
4%
3%
2%
1%
0%
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96
Rank
Most of the other large suppliers are partly integrated. They usually rely on providers of pub-
lisher-serving technology for the software and communications system that receives copy from
advertisers and then make decisions on where and when to insert that copy into inventory. They
typically have their own sales forces that distribute their inventory to advertisers. But they also rely
on advertising networks to sell space that they cannot or do not want to sell themselves (see the
discussion below on price discrimination). Smaller websites ordinarily do not have enough volume
to support a sales force. They often rely on an advertising network which may also provide the
necessary technology. For example, a publisher can paste html code supplied by the advertising
network into the part of the webpage the publisher wants to sell; that code will retrieve and insert
an advertisement.
There are at least three main sources of supply of online advertising inventory, each of which
results in different methods for selling advertising.
First, search engines generate search-results pages. Search engines need to enable users to
find what they are looking for quickly and to move from the search results page to the desired
webpage. In the late 1990s, search engines struggled to find the right balance between providing
valuable search results to attract eyeballs and selling advertisers access to these eyeballs. Most
ended up dividing the search results page into the “organic search results” that are based on the
relevance of the web page to the keywords entered by the user; and the “paid search results”
PLATFORM ECONOMICS:
236 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
which are clearly demarcated text advertisements that also look like search results. Although or-
ganic search results are valuable to advertisers,3 search engine providers do not charge advertisers
for these listings. Instead, advertisers often hire “search engine optimization” companies, such as
Performics, to increase their rankings in the search results. Each page typically has around ten
slots available for a paid search advertisement. Advertisers bid on a cost-per-click for these slots
and the search provider allocates the slots, roughly speaking, on the basis of expected revenue it
will receive—that is, cost-per-click times expected clicks) (Evans, 2008; Varian, 2007). To maxi-
mize revenues from the scarce space, the search engines can use the cost-per-click bids and the
expected number of clicks that the ads will receive to allocate the spots. Projecting the number
of clicks and the effects of different allocation mechanisms is a difficult problem; the various
search engine providers have solved that problem to varying degrees and their relative success
in doing so explains in part differences in the revenue they receive for each search conducted
(Evans, 2008).
Online media sites provide content that is broadly similar to what consumers get from
traditional media. In fact, many traditional media companies have established web sites which
use some material that is also provided through offline channels. Prominent examples in-
clude cnn.com, nytimes.com, mtv.com, and cosmopolitan.com. Other sites such as Yahoo!
redeploy content from a variety of sources such as newswire services. Finally, some sites, such
as youtube.com and drudgereport.com, provide content only online and often do so inno-
vatively. With the exception of the video sites, these sites allocate portions of their pages to
advertisements much like newspaper pages do. They generally sell the ads on a cost-per-mille
basis—that is, based on how many eyeballs see their pages—through their own sales team and
through advertising networks.
Figure 3 provides a sample layout: as with newspapers ads, different spots are perceived as
more valuable because they are likely to receive more attention from consumers. Conversations
with knowledgeable industry participants indicate that ads in the top half of the page garner a
cost-per-mille of around $12.50 while ads in the bottom half of the page realize about half of that.
Banners and skyscraper ads typically go for more than $12.50 cost-per-mille. The more desirable
space tends to be sold directly, while the less desirable space tends to be sold through advertising
networks. These ad networks pay between 15 and 45 percent of the cost-per- mille for a display ad
on the top half of the page depending on the quality of the space.
3
Search engines can be viewed as (at least) three-sided platforms based on third-party content obtained from the web,
users looking for this content, and advertisers who seek access to these users. In principle, search engines could charge
third-party content providers for the indexing and listing services made available but thus far have generally chosen not
to. In March 2004 Yahoo launched a paid inclusion program that guaranteed listings on the Yahoo! search engine for
commercial websites in return for payment. This scheme was not popular with website marketers or the public and was
discontinued. Microsoft and AskJeeves charged for inclusion for a time but also discontinued this practice.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Online Advertising Industry: Economics, Evolution, and Privacy 237
Page Top
Middle
Bottom Bottom
Left
Content Right
Bottom
People go to social networking sites to obtain updates on what their friends are doing, to up-
date their own pages, and through these and other activities to communicate with their friends.
These sites have allocated some space for advertising and have sold this inventory to advertisers.
This space is sometime known as the “dead zone” because of the lack of attention that social net-
working users pay to it. The average cost-per-mille payments to social networking sites are often
less than $0.50. The social networking sites, and on-line advertising businesses, are working on
advertising methods that work more effectively with social networking communities. Google, for
PLATFORM ECONOMICS:
238 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
example, has developed a technology that makes it possible for advertisers to identify and target
“influencers” in social networking communities to distribute messages (Helft, 2008). Whether
social networking sites can achieve cost-per-mille payments for their viewers that are comparable
to other types of on-line advertising remains to be seen. The promise of advertising on social net-
working websites is that word-of-mouth referrals are the primary influence on purchase decisions
(BIGresearch, Simultaneous Media Usage Study (SIMM12), June 2008), and perhaps online com-
munities can be used to create a source for such referrals.
The supply of advertising inventory is highly heterogeneous. It differs in size, the likelihood
that consumers will pay attention to it, and the characteristics of the viewers. Not surprisingly, the
price that advertisers pay per thousand views varies from a few cents to several hundred dollars in
the case of high-income professionals. (This heterogeneity and price distribution is also true for tra-
ditional advertising.) Controlling for quality, however, suppliers of advertising engage in extensive
“value-based pricing”—which economists call price discrimination.
The search-based advertising platforms, in principle, use second-price auctions to allocate slots
and in this way seek to extract higher payments from those willing to pay more (Varian, 2007;
see also Edelman et. al., 2007). Indeed, we might expect that the rents earned from the advertiser
side would be bid away, at least partly, on the viewer side through competition among platforms
(Rochet and Tirole, 2003; Armstrong, 2006), including through explicit subsidies to viewers to
join the platform.4 That has happened to a degree. Microsoft adopted some mechanisms for pro-
viding incentives to viewers in 2008; for instance, the Windows Live search site has a button
labeled “Search a lot, Earn a lot” which describes incentives for searching that include free Xbox,
Microsoft software, frequent flier miles, cash back and more (see <http://www.live.com/?form=M
XCA00&kwid=019409a179d80c4ba7fe55c739b4f11c>, visited on November 28, 2008). Yahoo!
has also entered into incentive arrangements for people to download toolbars that rely on its search
engine. For example, Yahoo shares revenue with companies that get people to search with Yahoo
and those companies in turn provide benefits—such as donations to charity—to individuals who
download a toolbar that contains the Yahoo search feature (see <http://affiliates.freecause.com/
index.cgi?action=about>, visited on November 28, 2008). Google has not followed suit. As of
2008, there does not appear to be robust price competition to persuade searchers to use alternative
4
In general, multi-sided platforms can have profit-maximizing equilibrium price structures in which one or more
sides have zero or negative prices (Rochet and Tirole, 2006). In the case of search-based advertising platforms the
pricing on the advertiser side is fixed insofar as the platform adopts the standard auctions approach. These platforms
compete, however, in a number of dimensions including the amount and quality of the search results provided to the
users and the ease of use of the platform for users and advertisers. Such competition is similar to free television which
competes through the attractiveness of the content rather than cash transfers to viewers. The search-based advertis-
ing platforms can dissipate the rents from the second-price auctions of advertising space spending on research and
development and other expenditures that improve quality for users and advertisers. They can also dissipate it through
distribution deals such as those discussed in the text.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Online Advertising Industry: Economics, Evolution, and Privacy 239
engines available to them. A different kind of competition for attention does take place. Google,
Microsoft and Yahoo! compete to become the default search engine on the browser, on various tool
bars that are distributed, and in publisher web sites. Although consumers can readily change these
default settings the search engine providers presumably pay for these because enough consumers
stick with the default. These deals provide incentives to entities that can provide search engine
defaults for their users to compete harder to these users by providing greater quality.
Online publishers engage in price discrimination as well. This practice is similar to other busi-
ness-to-business transactions which are often characterized by individual negotiations, and the
online practices are similar to the offline practices. In both cases, heterogeneous supply is being
matched up with heterogeneous demand through bilateral exchanges and with limited public
information. Media companies can take space they do not choose to sell directly and make it avail-
able to intermediaries that sell it indirectly. Larger online publishers typically have contracts with
several advertising networks and have technology that selects the highest price for inventory in
more or less real time. Online advertising networks are discussed in more detail below.
extent they can reinforce each other (“The Medium Is The Massages,” 2002). Berndt et al. (2002)
find that 57 percent of the 28 pairs of the cross-elasticities they estimated indicated the methods
were, on net, substitutes and the remainder were complements--although typically weak ones.
The emergence of online advertising has provided another type of media for advertisers. The
online methods of advertising, however, usually have close counterparts offline. A GM Corvette ad
on cnn.com is not visually that different from a GM Corvette ad in USA Today and GM will pay
for both ads based on the number of viewers. A Yahoo! ad for pizza restaurants on the Upper West
Side of Manhattan is not that different from an ad for a pizza restaurant in the Manhattan “Yellow
Pages” telephone directory. Hardcopy mail and e-mail solicitations for vacuum cleaners are similar,
and both entail payment based on results. Of course, potentially important differences may arise
in the value received by the advertiser from the comparable online and offline message. Online,
the GM ad viewer can click through to a website; someone from out-of-town can find the pizza
restaurant on Yahoo; and the e-mail solicitation for vacuum cleaners can take a buyer to a website
to purchase the vacuum and have it shipped.
Different methods of online advertising are potentially substitutes or complements both with
each other and with other forms of advertising. For example, if an advertiser wants to reach a large
number of individuals to introduce a new product, it can buy a banner ad on the Yahoo! home
page which has 48 million visitors a day (based on comScore Media Metrix report for February
2008) or it can buy a couple of 30-second television spots on Fox TV’s American Idol which had
more than 20 million viewers for its early 2008 episodes. A typical advertiser will place display ads
on multiple web sites as well as several television shows to reach a large audience. However, as these
methods are all designed to reach large numbers of people, they are probably substitutes at the
margin. The advertiser might complement this campaign by buying keywords on one or more of
the major search engines so that consumers that see the product can search for it online, learn more
about it, and possibly purchase it. Of course, investigating the fine structure of demand requires
more than such ruminations. The extent to which different methods of offline and online advertis-
ing substitute or complement for each other depends on the nature of the particular campaign, the
objectives the advertiser seeks to pursue, the aggregation across advertisers of many possibly vary-
ing demand relationships, and other factors. The net degree of substitution or complementarities
between different advertising methods is an open empirical question.
Advertisers typically hire firms to design and execute advertising campaigns. Although these
firms often have relationships through a conglomerate, such as WPP, they have become more spe-
cialized over time (Berndt et al., 2008). Increasingly one firm does the creative work and plans the
campaign while another firm engages in buying and placing media. While some firms have arisen
that specialize in online advertising, most creative work and media buying is performed by adver-
tising firms that manage both offline and online work for advertisers. Within the online part of the
business, advertisers or their agents also purchase various technologies for distributing online ad-
vertisements to suppliers of advertising inventory and measuring the success of online campaigns.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Online Advertising Industry: Economics, Evolution, and Privacy 241
It appears likely that online advertising provides two potentially significant economic efficien-
cies. First, a promising conjecture is that online advertising allows the economy to reduce the
amount of resources devoted to creating content for aggregating and sorting potential buyers.
Society may not need to invest as much in magazines, newspapers, and other media whose main
purpose is aggregating the right eyeballs for advertisers. Although consumers value that content
they have not had to pay for its cost of production. Second, online advertising likely increases the
accuracy of the match between the buyer and the seller. The seller has greater ability to target con-
sumers that are likely to buy and the consumer is more likely to receive useful messages and less
likely to receive time-consuming but irrelevant messages.
In both cases one can argue that there are some losses that also need to be considered. The news
media gathers and reports news through professional journalists in some cases scattered through
the world including war zones. One can argue that the news media provides an important public
service in a democratic society and that its value exceeds what individuals or advertisers may pay
for it. Moreover, as with any technological change some people gain and others lose. Those who
strongly prefer the touch and feel of newspapers will lose if not enough people are willing to sup-
port the costs of these newspapers. In addition, online advertising may provide more efficient
matching and delivery of ads but that begs the longstanding question of whether advertising is
providing people with valuable information that helps them make better buying decisions, or
whether it is getting people to buy based on deceptive information or by persuading them to do
things that they will soon regret.
insurance category. It could infer from a user’s IP address and browsing behavior that the user is
probably a woman (from browsing behavior) who lives in a well-off suburb (from the IP address),
in a region with low accident and theft rates (from the IP address), and is over the age of 25 (from
browsing behavior). The intermediary might identify potential sales targets by just using simple
screening methods such as these or may use more sophisticated statistical methods to predict the
sort of the prospects the insurance company is looking for. Either way the advertising intermediary
can use these methods to determine whether a particular individual who is browsing a website at
a particular moment has exhibited the web browsing behaviors and personal characteristics that
make that person a good target for an ad. It can make that decision almost instantaneously and
then insert an ad into advertising space on the page that the target is viewing. The advertiser would
typically pay a premium over standard online advertising rates for views by these targeted individu-
als because the likelihood that they will “convert” the view by this “qualified prospect” into a sale
and thus their expected profits is higher.
Although behavioral targeting is an area of intense innovation, as of 2009 only a small portion
of the advertising revenue earned by publishers results from selling these sorts of behaviorally tar-
geted advertisements.5 Two factors limit the current deployment of this seemingly efficient method
of advertising.
First, since behavioral targeting narrows the group of people that see an advertisement the
likelihood that these individuals will ultimately purchase the product has to be high enough
to offset the reduction in the number of people that view that advertisement. On average only
about 1 out of 400 viewers click on a given ad (Marketing Sherpa, 2008) and only a fraction of
those viewers purchase the product. Unless behavioral targeting is sufficiently precise advertis-
ers may prefer to reach a larger group of individuals all else equal. Reaching 10,000 people of
whom 1/1000 (or 10) will ultimately purchase a product is better than reaching 2,000 people
of whom 1/500 (4) will ultimately purchase the product. There is a tradeoff between precision
and reach. In this example, all else equal, a behavioral screen that targeted a fifth of the potential
population would have to identify people that were more than five times as likely to buy to be
better than targeting everyone. Precision is limited by the amount and quality of data that are
available. (In addition, behavioral targeting is not relevant for brand advertising that is generally
aimed at a broad audience to influence their views on a company or a product rather than to
make a direct sale.)
5
Behaviorally targeted advertising is mainly used by advertising networks. Advertisements placed by advertising
networks account for a small portion of the advertisements on publisher web sites and only a portion of the
advertisements placed by advertising networks rely on behavioral targeting. Based on discussions with industry
participants the fraction of revenue for advertisements placed by advertising networks is small and behavioral
targeting smaller still.
PLATFORM ECONOMICS:
244 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
Second, the advertising platform that implements a behavioral targeting campaign must
have access to a large enough universe of viewers to find enough candidates to make the cam-
paign worthwhile to the advertiser. There are fixed costs of designing and executing advertising
campaigns; the advertiser needs to make enough sales to recover these costs and make a return.
Suppose that it costs $10,000 to design a campaign and the expected incremental profit from
each sale that results from this campaign is $10. Then the campaign would have to generate an
expected 1000 conversions to recover the fixed costs. If only 1 out of 1000 consumers that are
exposed to an ad is converted to a sale, the campaign would need to reach at least 1,000,000
individuals for it to break even. Consider a behavioral targeting campaign that converts 1 out of
500 consumers by targeting the 20 percent of the potential universe of people that are the most
likely buyers. To yield 1000 conversions that campaign would need to reach 2,500,000 people.
The total sample needed depends generally on the conversion rate and the targeting rates which
depend on each other. Many advertising networks are not large enough to engage in highly re-
fined behavioral targeting while some of the largest advertising platforms have not yet deployed
highly refined targeting.
Several developments may increase the use of behavioral targeting methods. With improve-
ments in predictive techniques and the availability of data on viewers behavioral targeting would
become more precise and advertisers would increasingly prefer targeting to reach. If advertis-
ing networks that use behavioral targeting methods increase their scale, or if larger advertis-
ing platforms increasingly deploy behavioral targeting and data collection methods, they will
obtain enough viewers to make behavioral campaigns economically efficient. Privacy concerns
discussed below could limit the development of behavioral targeting. Consumers may resist
having advertising platforms collect detailed information about their browsing behavior and
government regulations may limit the ability of advertising intermediaries to collect these data.
Several firms have developed or are proposing to develop electronic exchanges (for one exam-
ple, see the discussion the DoubleClick Advertising Exchange at <http://www.doubleclick.com/
products/advertisingexchange/index.aspx>, visited on November 29, 2008). These exchanges con-
nect web-based publishers (or their agents) that would make their advertising inventory available
and advertisers (or their agents) that would want to purchase this inventory. Advertising networks
that have an excess supply or demand of advertising inventory could also participate. The advertis-
ing inventory would be auctioned off in real time and the exchange then handles the delivery of
the advertising from the advertiser to the publisher. No data on the size of these exchanges is avail-
able, but based on discussions with individuals in the online advertising business, as of the end of
2008 it does not appear that the existing exchanges have achieved a critical mass of liquidity that
would enable them to sustain themselves. Indeed, few such business-to-business exchanges have
achieved critical mass in any area of business—most have died (Evans, 2009). It remains to be seen
whether the advertising exchanges will ultimately face a similar fate. (Harris, 2002 discusses the
role of liquidity for an exchange and Evans and Schmalensee, 2009 discuss the role of critical mass
for multi-sided platforms.)
of 21 percent between February 2004 and February 2008 (according to ComScore Media Metrix).
There are low financial barriers to the formation of these sites as a result of web technologies. This
increased supply of advertising inventory puts downward pressure on advertising rates, promises
to reduce the returns that traditional media can get from advertising, and therefore potentially
reduces the quality-adjusted supply of content by traditional media. (The increase in the quality
of advertising could increase or decrease the demand by advertisers which could therefore have a
partly offsetting effect on rates.)
Third, online advertising increases the supply of online content which provides a substitute
for traditional content. The potential returns from online advertising encourage entities such as
Yahoo! and MSN, as well as traditional media, to present various kinds of content online that
consumers used to consume mainly offline. In addition, of course, viewers are moving from offline
to online media because, as with the move from radio to television, they simply like the content
better along certain dimension, which can include the ease, flexibility, and interactive dimensions
of access. Thus far the evidence shows that as a result of online content consumers have substituted
away from radio and newspaper content but not television content. The US Census reports that
persons above 12 years of age spend 32% more time watching that cable and satellite television
than they did in 2001. The usage of radio and traditional newspapers has declined by 3% and
15%, respectively.6
These trends are likely to accelerate as more content is available either online or through other
technologies that enable the sellers of advertising inventory to employ online methods. Some tra-
ditional media may make a fairly smooth transition to the online world. The television industry
could change its delivery method so that programming is delivered primarily through internet
protocols—what is known as IPTV—and implement online technologies to serve ads. Verizon’s
FiOS Television, which is based on internet technologies, had 2.5 million Internet subscribers and
1.9 million television customers in the United States as of December 31, 2008.7 The production
and programming sides of the business could operate much like they do today although many
of the business methods, and assorted institutions, for selling and delivering advertising would
change. The same is true for radio which is already being delivered over the internet. One cannot
consume internet radio on the morning drive to work at the moment but that may change once
there is more extensive wireless internet coverage and cars become equipped with internet-enabled
electronics.
The forces of creative destruction have hit the newspaper industry the hardest and the earliest.
The industry is in free fall as a result of the self-reinforcing migration of readers and advertisers
6
US Census Bureau, Table 1089: Media Usage and Consumer Spending: 2001 to 2011, http://www.census.gov/
compendia/statab/tables/09s1089.pdf.
7
“Verizon FiOS-Fact Sheet,” Verizon News Center (available at http://newscenter.verizon.com/kit/fios-symmetrical-
internet-service/all-about-fios.html, last visited on March 24, 2009).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Online Advertising Industry: Economics, Evolution, and Privacy 247
online. Figure 4 shows the circulation for all U.S. daily newspapers, their advertising revenues, and
the market capitalizations of the Standard & Poor’s 500 Publishing Index—all indexed to 100 in
2002. There has been a precipitous drop in circulation and advertising revenue and the S&P 500
indicates that the market expects dramatic reductions in the future profit stream of the newspaper
industry. As mentioned earlier several major daily newspapers in the United States have closed
their doors. Although many newspapers have developed web versions that attract significant num-
ber of viewers, their increased online advertising revenues have only partly compensated for their
loss of traditional advertising revenues. As radio and television move to internet-enabled platforms
they could find the same decline in advertising revenues with the same impact on the economics
of providing traditional content.
The industrial structure of the online advertising industry could evolve in a variety of ways. One
possibility is that the industry will have a highly concentrated set of intermediaries at its center,
110
90
70
50
30
10
2002 2003 2004 2005 2006 2007 2008
Year
Ad revenue and publishing index are adjusted for inflation, all quantities indexed to 2002 base year. In 2002 ad revenue of all US newspapers
equaled $44 billion, and the total US daily newspaper circulation was 50.7 million.
S&P 500 Publisher Index is composed of the prices of the common stocks of the following companies: Gannett Co., Inc., The McGraw-Hill
Companies, Inc., Meredith Corporation, The New York Times Company, and Washington Post Company. It is weighted by market cap.
In 2002, the total market cap of these companies was $44.1 billion.
Source: Advertising Age, Newspaper Association of America, Washington Post Company Annual Reports.
PLATFORM ECONOMICS:
248 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
with many content providers around this core. This result would be akin to the U.S. stock exchange
system which has 5 major exchanges that account for 92 percent of transaction volume.8 Another
possibility is that the industry will have many intermediaries at its center. Some of the intermediaries
will focus on mass advertising while others will focus on niches. The ultimate structure depends on
the relative importance of several factors: the strength of indirect network effects and scale econo-
mies on one side, and the possible benefits of specialization of knowledge in certain areas.
8
Data as reported by Arcavision. Note that the 5 exchanges are associated with either Nasdaq or NYSE Euronext. The
shares of each of the five exchanges are: NASDAQ-29.9%, NASDAQ TRF-19.8%,NYSE Arca 18.1%, NYSE 17%
and NYSE TRF 7%.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Online Advertising Industry: Economics, Evolution, and Privacy 249
The collection of these data has proved controversial, however, and raises some delicate public
policy issues. Consumer privacy advocates and regulators have criticized the search-engine provid-
ers for capturing and storing data (Dye 2009). In response, these providers have agreed to reduce
the length of time for which they store data with individual identifiers. Google, for example, has
reduced this period from two years to nine months.9 The collection of data, as well as behavioral
targeting, has attracted lawsuits and legislative inquiries. NebuAd and Phorm are good examples.
The two behavioral targeting firms enlisted Internet Service Providers into their advertising net-
work and used browsing data from these Internet Service Providers to target advertisements. They
quickly became the subject of legislative inquiries in the United States and the United Kingdom
and many of the Internet Service Providers withdrew as a result of the controversy.10
The use of personal data for targeted advertising raises several public policy issues. If people
had ownership over information about themselves, and there was a competitive market for it, they
could decide whether to sell it to an online advertising business. In making this decision, people
could take into account the value of receiving possibly more relevant ads, as well as any other com-
pensation they might get. They would also take into account any costs from the possible leakage
of their private information.
To a degree, consumers have some control over their private information. A user can choose
not to use websites that insert “cookies” that collect data on her machine (or choose to delete such
“cookies” on a regular basis). Web browsers have increasingly provided mechanisms for consum-
ers to control the retention of information on their browsing history and manage their cookies.
Reviews of the recently released browsers in the market, IE 8.0 and Google Chrome 2.0, emphasize
new features like private browsing and search suggestions. 11 To the extent the consumers exercise
these choices they put competitive pressure on online advertisers to account for the value that
consumers place on data about themselves. Consumers can also avoid websites and web services
scan and store personal content. Some users may avoid Google Gmail because it scans and records
the content of emails while others may value the targeted ads that result from Google’s ability to
examine the content of the emails.
Nevertheless, three potential problems arise that could warrant the consideration of govern-
ment intervention into the treatment of privacy by the online advertising industry.
First, there is the usual imperfect information problem that is often used to justify consumer
protection efforts (Stiglitz and Walsh, 2002). Consumers may not know that information is be-
ing collected and stored. It appears likely that few consumers knew that Google stored each user’s
9
“Google to dump user data earlier,” BBC News (September 9, 2008), available at <http://news.bbc.co.uk/2/hi/
technology/7605801.stm>
10
See Paul (2008).
11
<http://www.microsoft.com/windows/internet-explorer/features/overview.aspx?tabid=1&catid=1>,
<http://www.google.com/chrome/intl/en/features.html#>
PLATFORM ECONOMICS:
250 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
search history, including details sufficient to identify the IP address of the user’s computer, for two
years before this was the subject of news stories (for example, “Google’s growth raises privacy con-
cerns,” 2008). Even with the publication it remains unclear how much consumers know about the
extent to which various online advertising related businesses collect information on them.
Second, consumers may agree (either tacitly or explicitly) to provide private information with-
out anticipating that this information would be sold to other vendors who might combine it with
other information about them. For example, a consumer may feel differently about providing
search data associated with her IP address to a search engine provider if she knew the search engine
provider would sell the data to another vendor that had figured out a way to associate her IP with
her personal information including name, address, and telephone number.
Third, competition among advertising platforms may not necessarily result in the optimal pro-
vision of privacy. When there is perfect information competition can usually be relied on to yield
the optimal provision of quality to the consumer. Online advertising intermediaries are multi-
sided platforms that compete simultaneously for advertisers and viewers. Whether this competi-
tion results in the optimal provision of privacy, and the extent to which it would do so in a highly
concentrated market, would need to be investigated carefully (see Rochet and Tirole, 2006) for an
analysis of similar issues in credit cards).
In principle, of course, consumers could learn about the collection and use of their informa-
tion. But this incurs costs. Likewise, the online advertising businesses could increase efforts to en-
gage in transparent contracts with consumers. But this incurs costs, too. The critical public policy
question is how property rights—including those enforced through regulation—over private data
should be assigned.
Public policymakers in the United States and the European Community are grappling with
these issues (Federal Trade Commission, 2006, 2007; European Commission, 2008). Too strin-
gent regulations could harm consumers. After all, the online advertising industry benefits consum-
ers in several ways. It increases the likelihood that they will receive relevant ads and decreases the
likelihood that they will waste time on irrelevant ads. Moreover, it promises to reduce the costs of
advertising to businesses, and some or all of these costs would be passed on to consumers in the
form of lower prices. On the other side, overly lenient regulations could also harm consumers.
Consumers could incur the costs of having private information disclosed and potentially misused,
and incur the costs of reducing their use of the web because of concerns over privacy. Regardless of
whether it their private information is disclosed consumers may not like receiving ads that reflect
too much knowledge about them even if it is only a software program on a remote server that has
that knowledge.
Resolving the privacy dilemma is important for innovation in behavioral targeting. Innovators
will benefit from knowing what information they can collect and how they can use it without risking
lawsuits, being pilloried in the press, and being hauled in front of Congress. Consumers will benefit
from balancing the benefits of receiving relevant ads against the cost of losing valuable privacy.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Online Advertising Industry: Economics, Evolution, and Privacy 251
XI. APPENDIX
Statistics for Online Advertising of the Top 100 Ranked Websites
Of the top 100 ranked website:
percent of total page viewed 47.84%
40
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Online Advertising Industry: Economics, Evolution, and Privacy 253
41
CHAPTER
Web Economy, Two-Sided Markets
9 and Competition Policy
David S. Evans
ABSTRACT
The web economy has grown rapidly in the last decade. Online businesses have several key fea-
tures that are important for understanding the pro-competitive and anti-competitive strategies
they may engage in. The two-sided markets literature helps elucidate many of these strategies. It
also provides guidance for the antitrust analysis of market definition and exclusionary practices
for web-based businesses.
Ten years ago a tweet was something a bird did. We generally did not poke our friends. When we
sent an email about buying a car we would have jumped out of our chairs if an advertisement for
BMW all of a sudden popped up on our screens. And our mobile phones did not have application
stores. Change has occurred rapidly following the development of the commercial internet but it
has a long way to go. It takes years for entrepreneurs to come up with ideas, for businesses to start,
and for industries to evolve and sort themselves out. It will take a decade or two, perhaps, for things
to settle down. Competition policy will find itself increasingly grappling with mergers, exclusion-
ary practices, and collusion as the web economy becomes more prominent, as it matures, and as it
goes through a period of significant turmoil.
A decade ago we would have finished the phrase “two-sided” with “coin” and not markets and
we would have thought that a person who engaged in multihoming had a place in the country.
Like the web economy the study of two-sided platforms has grown rapidly since its birth at the
turn of the century. The literature has flourished with many theoretical papers and much empirical
research. Most major competition authorities around the world are using the two-sided market
approach in a broad range of cases from credit cards to shopping malls.
Competition policy matters involving web-based businesses will provide fertile ground for us-
ing two-sided analysis. In this paper I want to use the lens of two-sided markets to describe some
key features of the analysis of market definition, market power, and exclusionary practices particu-
larly as they relate to entry.
254
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Web Economy, Two-Sided Markets and Competition Policy 255
text-based ads on search-results pages and a bit more with contextual ads on its publisher network.
It is not making much money from many of the other businesses that it has gotten into and it has
largely abandoned its efforts to make money from selling traditional ads for radio and television. It
is an important player, nevertheless, for three key reasons. First, it’s the dominant player in helping
people find things on the web and that is terribly important. Second, it gets billions of dollars of
revenue a year from search and contextual ads and this money helps it fund grander ambitions.
Third, it does not like other firms controlling parts of the web because that gets in the way, or can
get in the way, of it selling more advertising.
The iPhone is the next stop. What makes the iPhone so important is that it is a platform for
developing applications. There are more than a 100,000 applications that people can download
for their phones. More of these are being written every day and most are web-based applications.
These applications are helping to transform other industries—below we will see the example of
Square which may disrupt payments.
Facebook is a six-year-old company that is run by a twenty-five-year-old. It has taken over the
social networking space but it is actually a late entrant in that segment. There was a flurry of social
networking sites that started in the late 1990s such as Six Degrees of Separation. Then Friendster was
the king of the segment in the early 2000s. MySpace toppled Friendster and then Facebook has leap-
frogged MySpace. Like YouTube, Twitter and other sites that attract eyeballs, Facebook really has not
figured out how to make huge dollars from the huge traffic that it generates around the world. The
idea is advertising. The problem is that when you go visit your friends you do not necessarily want
to have a stranger trying to sell you male enhancement drugs. If you are an advertiser like Procter
and Gamble you may not want your ads appearing next to pictures of drunk and half-naked people.
How do these web celebrities get along in the neighborhood? Google doesn’t like Facebook be-
cause it cannot get its spiders (which pull content for its search engine) onto this large and growing
part of the web. Facebook likes the iPhone because it is a great device for people to go visit their
friends. Google and Apple used to get along so well that Google had a couple of representatives
on the Apple board. But then Google could not stand having Apple control a large part of the
mobile space for inserting ads. As a result, Google unfriended Apple and they are now engaged in
the mobile platform wars.
enough viewers to be counted around the time that YouTube began back in 2005. Almost all of them
are gone now. More generally, most platforms fail to take off. The growth of these platforms is driven
by network effects and they face a critical mass problem. The problem is analogous to an exchange
which can survive only if it gets enough liquidity. Much of the work that web entrepreneurs like the
founders do when they start is trying to figure out ways to get enough customers on board the plat-
form to take off. YouTube as an example of tackling the critical mass problem. It had to figure out
how to get enough people to upload videos and how to get enough people to view those videos and
how to get both of those groups onboard the platform in enough numbers to ignite the platform and
to get rapid growth. Businesses that get enough liquidity (such as Hulu and YouTube) ignite while
those that don’t (such Revver, another video sharing site) impode. It is as if there is an invisible wall.
Once the entrepreneur pushes the platform through that wall the platform can take off.
We have known for a long time in a two-sided literature that exclusivities are a way to solve the
chicken and egg problem. The reverse is true as well. One way for an incumbent firm to prevent
a new firm from taking off is to make it hard for it to get enough critical mass by entering into
exclusives with enough of at least one major customer group.
B. Free
Many of the web-based platforms discussed above are free to at least one group of customers. That
is a well studied and documented phenomena in the two-sided literature. The platform generates
value by getting one or more customer groups together. It can be profitable to charge one group of
customers nothing just to get them on board so that the platform can charge another group of cus-
tomers for access to them. We often see this offline. There are many free newspapers, for example.
We often see « free » online because the marginal cost of serving another user is zero. Thus, you do
not have to pay for your Facebook page. That social network makes money by selling your eyeball
to advertisers and selling complementary goods like virtual gifts.
There is a tendency in competition policy cases to ignore the customers that are getting things
for free. That happens for two reasons. One is that analysts tend to equate the business with the
money side. We have seen this with Windows. An important group of customers for Microsoft are
software developers that use the Windows APIs. Microsoft has chosen to provide most of its ser-
vices to these customers for free. But those customers get a lot of value from Microsoft and should
be considered in any welfare calculation. Another reason that analysts ignore the free side is that
the traditional methods of market definition focus attention on a single group of customers even
though the members of the two groups have welfare that is inextricably intertwined.
web business like Pandora or Skype or Twitter, you would mainly spend your time writing software
code. All of these businesses on the web are based on thousands of lines of code. This software can
be locked down so no one can get access to it. It can also be opened up so that others can use the
features of it ; in that case it becomes a software platform that can support the development of
complementary applications. Many of the web-based businesses have started software platforms
by opening up their code. You might wonder how Firefox managed to cut Internet Explorer down
to size. Much of it had to do with encouraging developers to write applications that increased the
value of the Firefox browser. Facebook has done the same thing and there is an active developer
community on Facebook with more than 500,000 applications written so far. Farmville is an
example of a Facebook application. Google Maps has become such a powerful product because
Google made the APIs available so developers could build applications that integrate mapping.
The software platform model is transforming the web. Almost every major web property has
followed that strategy. It is propelling rapid growth and innovation.
From a competition policy perspective these applications that have been built on top of these
platforms cut two ways. For one, they are the source of great value. The developers benefit from the
platform directly because it makes it possible for them to engage in innovation, to write applica-
tions for your iPhone or Facebook or for Google for example, at a low cost. The consumers of the
applications also benefit. But these applications also pose a possible barrier to entry. It is the old
chicken and egg story. It takes a lot of effort to compete with the incumbent platforms that already
have both sides onboard. As a result, if you want to compete with the iPhone, you have to cope
with the fact that the iPhone has a 100,000 and counting applications.
pretty much everywhere at zero marginal cost. The stock price of TomTom and Garmin which is
another navigational device maker plummeted last October after the Android phone came out
with this navigational device.
to draw hard market boundaries and when the perspective of this market provides a distorted view
of the competition that is actually taking place in the real world.
There is a growing consensus among economists that the current tools for market definition
face numerous problems especially when there is product differentiation which there almost always
is. The hypothetical monopolist test is quite difficult to implement reliably. The results are highly
dependent upon whatever assumption the economist is making about the shape of the demand
schedule and the sequence of the products considered among other things. All those problems be-
come much more difficult when the markets are two-sided. Market definition, for example, does
not deal very well with the complementary products that characterize two-sided markets. Many of
the problems that critical loss has in one-sided markets become an order of magnitude harder in
two-sided markets.
For the web economy competition policy will do better to rely on methods that are less for-
mulaic and based more on qualitative research into the nature of competition in the business
ecosystem.
As with any new kind of business the web businesses will provide creative ways of monopoliz-
ing unlawfully that we have not thought of. There will also be pro-competitive explanations that
we have not imagined for practices that look suspicious. I mentioned critical mass earlier. Firms
use many methods to achieve critical mass and obtain platform ignition. Firms can use these
same sorts of methods to prevent their rivals from getting critical mass. It may not take much
to prevent a rival from achieving ignition. That is something for competition authorities to be
concerned about.
Google will likely become the testing ground for the next decade of antitrust analysis of the
web economy. The economics in this company is really pretty simple. It makes money from ad-
vertising. That means it wants spaces to put ads and it wants eyeballs to look at those ads. The
more space and the more eyeballs they have the better. My suspicion is that Google enters other
parts of the web primarily to ensure that nothing comes between them, the advertising space, and
the eyeballs looking at that space. We are seeing this now in mobile. Advertising and eyeballs are
going to mobile devices. Google has started the Android operating system and launched it own
line of phones to help ensure access to mobile advertising inventory. As Google keeps entering
other parts of the web ecosystem, we are going to return to the an debate. Is Google a dominant
firm, leveraging its way into other businesses. Or is this a double marginalization story where a
dominant firm in one market can make consumers better off by either making a related market
competitive or extending its own dominance into that market. Thus, is it an anti-competitive
story or an efficiency one.
The web economy is intellectually interesting, the issues are complicated, and there is much
room for debating whether practices are good or bad. All of this will make the practice of competi-
tion policy most intriguing and perhaps profitable, for good or not, for many years to come.
CHAPTER
Antitrust Issues Raised
10 by Global Internet Economy
David S. Evans
ABSTRACT
Web-based businesses are increasingly the subject of antitrust concerns. Plaintiffs in the United States
have sued eBay for tying its online payments service to its transaction service. Multiple jurisdictions in
the European Community have claimed that Apple has violated the competition laws by limiting the
ability of its music player to play music from competing music stores and limiting the ability of com-
peting music players to play music purchased from its music stores. During 2007, although the U.S.
Federal Trade Commission decided not to block Google’s acquisition of DoubleClick after a lengthy
investigation, it expressed its intent to “closely watch these markets” involved in online advertising.
The web economy poses two major challenges to competition authorities. The law and economics
for analyzing the multi-sided platforms that dominate the internet sector is not well developed. At the
same time the web-economy is evolving very rapidly and in ways that are sure to result in antitrust
complaints and investigations. Competition authorities and courts will need to exercise great care in
balancing the protection of consumers from anticompetitive behavior against causing harm from
interfering in complex businesses that are both rapidly moving and not fully understood.
Web-based businesses are increasingly the subject of antitrust concerns. Plaintiffs in the United
States have sued eBay for tying its online payments service to its transaction service.1 Multiple ju-
risdictions in the European Community have claimed that Apple has violated the competition laws
by limiting the ability of its music player to play music from competing music stores and limiting
the ability of competing music players to play music purchased from its music stores.2 During
†
This Essay was previously published in the Northwestern University Law Review Colloquy on May 5, 2008, as David
S. Evans, Antitrust Issues Raised by the Emerging Global Internet Economy, 102 NW. U. L. REV. COLLOQUY 285
(2008), http://www.law.northwestern.edu/lawreview/ colloquy/2008/13/.
1
See, e.g., Complaint, Malone v. eBay, Inc., No. 07-01882-JF (N.D. Cal. Apr. 4, 2007).
2
Consumer Council of Norway, Complaint against iTunes Music Store (Jan. 25, 2006), available at http://forbrukerpor-
talen.no/filearchive/Complaint%20against%20iTunes%20Music%20Store.pdf; Helena Spongenberg, Apple’s iTunes
Risks EU Backlash After Norway Ruling, EU OBSERVER, Jan. 25, 2007, http://euobserver.com/871/23334; Apple:
European Consumer Organisations Join Forces in Legal Dispute over iTunes Music Store, VERBRAUCHERZENTRALE,
Jan. 22, 2007, http://www.vzbv.de/go/dokumente/571/8/36/index.html; Free-DRM Music by iTunes, but EC Starts
Official Investigation, EDRI.ORG, Apr. 12, 2007, http://www.edri.org/edrigram/number5.7/ituned-free-drm;
Increased Competition Pushes iTunes Towards Interoperability, FORBRUKEROMBUDET, Feb. 11, 2008, http://
www. forbrukerombudet.no/index.gan?id=11038653.
261
PLATFORM ECONOMICS:
262 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
2007, although the U.S. Federal Trade Commission decided not to block Google’s acquisition of
DoubleClick after a lengthy investigation, it expressed its intent to “closely watch these markets”
involved in online advertising.3
Of course, competition policymakers have not just discovered the Web. In 1998, shortly after
the start of the commercial Internet three years earlier, the U.S. Department of Justice and vari-
ous states filed an antitrust case against Microsoft for engaging in various practices related to Web
browsers.4 The European Commission started an investigation of Microsoft’s practices related to
media players that stream music over the Internet in 2001.5 However, the Microsoft cases mainly
involved the use of the company’s market power in personal computers to influence competition
in Web-based markets that threatened it. The matters involving Apple, Google, and eBay concern
market power in Web-based products and services themselves.
The Internet economy is likely to raise antitrust concerns—and possible demands for regulation—
for years to come. Gargantuan global firms have emerged, which will likely attract scrutiny by compe-
tition authorities and by policymakers concerned with competition issues. The companies mentioned
above, for example, have shares in putative antitrust markets that rival those held by Microsoft.6
Apple has more than a 70% share of paid music downloads in the European Union,7 Google has
more than an 80% share of search queries in Europe,8 and eBay has more than a 90% share of auction
site page views in France, Germany, Italy, Spain, and the UK.9 Competition authorities and private
parties can challenge the practices of these leading firms under the antitrust laws of most jurisdictions.
Such challenges are especially likely under European Community law and decisional practice, which
impose special obligations and significant scrutiny on firms that have market shares as low as 40%.10
3
Fed. Trade Comm’n, Statement of Federal Trade Commission Concerning Google/DoubleClick (Dec. 20, 2007),
available at http://www.ftc.gov/os/caselist/0710170/071220statement.pdf.
4
The D.C. Circuit Court of Appeals upheld in 2001 portions of a lower court ruling in 2000 that found Microsoft
guilty of violating section 2 of the Sherman Act. See United States v. Microsoft Corp., 253 F.3d 34, 45 (D.C. Cir.
2001) (citing United States v. Microsoft Corp., 87 F. Supp. 2d 30 (D.D.C. 2000)).
5
In 2007, the European Court of First Instance upheld the Commission’s 2004 decision that Microsoft had violated
Article 82 EC by tying its media player software to its dominant operating system. See Case T-201/04, Microsoft
Corp. v. Comm’n, 2007 WL 2693858 (Sept. 17, 2007).
6
These are not necessarily relevant antitrust markets, but they are ones that competition authorities could plausibly
adopt.
7
See Anna Jenkinson & Don Jeffrey, Apple, Record Companies Restrict Music Sales, EU Says (Update6), Bloomberg.Com,
Apr. 3, 2007, http://www.bloomberg.com/apps/news?pid=20601085&sid=aiMdS7KddG14&refer=europe.
8
comScore, MyMetrix qSearch 2.0 Key Measures Report (Dec. 2007). comScore is a “global Internet information
provider” that gathers data on Internet usage trends. comScore, Who We Are, http://www.comscore.com/about/
default.asp (last visited Aug. 12, 2008). comScore’s data analyses are based on its panel of over two million users. Id.
In recruiting its panelists, comScore attempts to ensure that “[a]ll demographic segments of the online population
are represented in the comScore Global Network, with large samples of participants in each segment.” comScore,
comScore Methodology, http://www.comscore.com/method/method.asp (last visited Aug. 12, 2008).
9
comScore, MyMetrix Key Measures Report (Dec. 2007).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Issues Raised by Global Internet Economy 263
Moreover, many Web-based firms have complex business models and arrangements. Separating the
merely complicated from the nefarious will take courts and competition authorities time to sort out.
This chapter describes the economics and technology behind the Web-based economy and how these
features will influence Internet competition policy in the years to come.
Part I provides a birds-eye view of the Web-based economy. Although this sector is evolving
quickly, its contours are beginning to take shape and we can be reasonably confident that several
globally dominant firms will play significant roles. Part II describes the economics of the Web-
based economy. The key businesses are what economists call “multi-sided platforms” that serve
several distinct but interdependent customer groups. Google, for example, serves people who are
searching the Web, advertisers who want to reach these users, and application developers who are
using Google’s software to develop complementary products. The leading multi- sided platforms
for the Web are often built on “software platform” technologies that make portions of their code
available to software developers who write applications that benefit users of the software platform.
Part III considers the competition that arises in the Web-based economy. The appearance of domi-
nant firms in key sectors will ensure ongoing scrutiny, and the nature of the economics and tech-
nology of these businesses will result in ongoing disputes over their practices.
The Web economy poses two major challenges to competition authorities. Currently, the law and
economics for analyzing the multi-sided plat forms that dominate the Internet sector are not well de-
veloped. At the same time, the Web economy is evolving very rapidly and in ways that are sure to result
in antitrust complaints and investigations. Competition authorities and courts will need to exercise
great care in balancing the protection of consumers from anticompetitive behavior against causing
harm from interfering in complex businesses that are both rapidly moving and not fully understood.
and uniform resource locators (URLs).11 For simplicity, this chapter refers to the Internet as the
physical communication system and to the Web as all products and services that rely on this com-
munication system.
The key innovations behind the Internet started occurring in the early 1960s, and the key in-
novations behind the Web came in 1989.12 The commercial Web began roughly in 1995 with the
introduction of browsers that made Web navigation easy for regular computer users; this therefore
encouraged the formation of businesses geared to a mass audience. There was much prognostica-
tion in the late 1990s on how the Web would evolve and enormous optimism that it would lead
to quick fortunes as network effects—more users make a site more valuable which leads to more
users—propelled early entrants to monopolies.13 Those hopes were seemingly dashed in 2001,
when the market capitalizations of most Web-based firms plummeted and vast numbers of these
firms vanished.14 It became apparent that many of these firms had not developed business models
that allowed them to make money from the visitors who came to their sites. In the aftermath of the
burst bubble, a robust Web-based economy has emerged that is creating completely new services,
from social networking to behaviorally targeted advertising, and also disrupting many traditional
businesses from media to telephony.15 As we will see below, advertising revenue for delivering traf-
fic has become the major driver for a significant portion of the Web economy.
A large portion of the population in industrialized countries touches the Web-based economy
daily. In 2007, 55% of the population in the European Union16 and 71% of the population in
the United States had access to the Internet.17 In December 2007, 78% of all Internet users in the
11
Other Internet services include online gaming, Voice-over-IP, email, instant messaging, file sharing, and other com-
munication services. Most of these services are bundled into Web sites.
12
See David C. Mowery & Timothy Simcoe, Is the Internet a U.S. Invention?—An Economic and Technological History
of Computer Networking, 31 RES. POL’Y 1369, 1371–76 (2002).
13
See, e.g., Carl Shapiro & Hal R. Varian, Information Rules: A Strategic Guide to the Network Economy
13 (1999) (“When the value of a product to one user depends on how many other users there are, economists say
that this product exhibits network externalities, or network effects . . . . Technologies subject to strong network ef-
fects tend to exhibit long lead times followed by explosive growth. The pattern results from positive feedback: as the
installed base of users grows, more and more users find adoption worthwhile.”).
14
See, e.g., Suzanne McGee, New Economy Sours on Venture Capitalists, Wall St. J., Jan. 2, 2001, at R6; Easy.com Easy.
gone, Economist, June 16, 2001, Special Section: A Survey of the New Rich, at 6; Living in Freefall, Economist,
Nov. 18, 2000, at 117–18.
15
See Is ‘Web 2.0’ Another Bubble?, Wall St. J. Online, Dec. 27, 2006, http://online.wsj.com/article/
SB116679843912957776.html.
16
Internet World Stats, European Union Internet Usage Stats and Population Statistics, http://www.internetworld-
stats.com/stats9.htm (last visited Apr. 19, 2008). Population penetration is defined as the number of Internet users
divided by the total population of the region. An Internet user is defined as “anyone currently in the capacity to use
the Internet,” namely, that he has access to an Internet connection point and can use the technology. Internet World
Stats does not adjust the figures to exclude children, illiterate people, or any other segment of the population.
17
Internet World Stats, North America Internet Usage Stats and Population Statistics, http://www.internetworldstats.
com/stats14.htm#north, (last visited Apr. 19, 2008).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Issues Raised by Global Internet Economy 265
United States and 69% worldwide used webmail;18 39% of all Internet users in the U.S. and 47%
worldwide used instant messaging.19 According to a 2007 survey of Americans who frequently use
the Internet, 84% used a search engine.20 The same survey found that the average respondent spent
thirty-three hours per week on Web products and services in 2007 compared with sixteen hours
viewing television.21
Web-based products and services are consumed primarily through the personal computer,
which is the primary Internet-connected device in most countries.22 However, it is widely expected
that most mobile phones around the world will soon be connected to the Internet.23 This increased
portability will increase the amount of time people can access the Web because people usually have
their mobile phones with them all the time. It will also increase the use of Web-based products
and services in lesser developed countries because mobile phones are less expensive than personal
computers and more widely held.24
One must be modest in speculating on the future evolution of the Web. The Internet and the
Web are very new technologies by historical standards. One could not have reliably forecasted the
development of electricity at a similar point during its development. The recent dot-com bust
teaches how wrong smart and financially motivated people can be about the business prospects of
a new technology. Nevertheless, five features of the Web-based economy appear to have emerged
18
See comScore Media Metrix, Web-Based Email Usage for December 2007, http://www.marketingcharts.com/wp/
wp-content/uploads/2008/02/comscore-yahoo-microsoft-google-email-web-based-us-worldwide-december-2007.
jpg (last visited Aug. 12, 2008).
19
See comScore Media Metrix, Internet Messenger Usage for December 2007, http://www.marketingcharts.com/
wp/wp-content/uploads/2008/02/comscore-yahoo-microsoft-google-instant-messenger-us-worldwide-decem-
ber-2007.jpg (last visited Aug. 12, 2008).
20
IDC, U.S. Online Consumer Behavior Survey Results 2007, Part I: Wireline Internet Usage 1 (2007) (on
file with author).
21
Id. at 52.
22
The major exceptions are Korea and Japan, where many consumers use their mobile phones to connect to the
Internet and where sophisticated Web-based mobile phone services have been introduced. See Michael Fitzpatrick,
Why Mobile Japan Leads the World: A Combination of an Urban Lifestyle and Infrastructure Advantages Mean that the
Fixed Internet Is Being Left Behind by the Mobile, Guardian (London), Sept. 27, 2007, at Tech. News & Features 3.
23
See, e.g., Press Release, comScore, Mobile Phone Web Users Nearly Equal PC Based Internet Users in Japan (Sept.
20, 2007), available at http://www.comscore.com/press/release.asp?press=1742; Press Release, Ipsos, Mobile Phones
Could Soon Rival the PC as World’s Dominant Internet Platform (Apr. 18, 2006), available at http://www.ipsos-na.
com/news/act_hit_cntr.cfm?id=3049&Region=us& PDF_name=mr060418-1b.pdf; see also Darren Waters, Why
the Future Is in Your Hands, BBC NEWS, Feb. 18, 2008, http://news.bbc.co.uk/2/hi/technology/7250465.stm.
24
See Press Release, Ipsos, supra note 23. Also, cell phone unit shipments were nearly 1.2 billion units in 2007 while
the number of PCs shipped globally was just over 270 million. See Press Release, Gartner, Gartner Says Worldwide
PC Market Grew 13 Percent in 2007 (Jan. 16, 2008), available at http://www.gartner.com/it/page.jsp?id=584210;
Ann Steffora Mutschler, SIA: 2007 Global Chip Sales Up Modestly by 3% to $256B, EDN, Feb. 1, 2008, http://www.
edn.com/article/CA6527975.html? industryid=47037.
PLATFORM ECONOMICS:
266 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
that are critical for understanding how this industry will evolve, the competitive strategies used in
this sector, and their implications ultimately for antitrust policy.
Many Web businesses follow the traditional advertising-supported media model. Content
is used to attract traffic. Access to that traffic is sold to advertisers. The content is usually made
available for free so that advertising is the primary source of revenue and profits. Many of the lead-
ing Web properties follow this approach.
Table 1 lists the top ten Web properties in the United States based on the number of visitors
to the properties’ Web pages. All but three of these sites are primarily supported by advertising:
Amazon and eBay are funded through transaction mechanisms while Wikipedia is funded by
voluntary donations.
Google and Ask primarily sell advertising on their search results pages. AOL, Fox, Microsoft,
Time Warner, and Yahoo primarily sell advertising on their various Web properties. The emergence
of the advertising-supported media model for Web sites is one of the most significant changes since
the dot-com bust and is the revenue source behind what is sometimes called Web 2.0.
Transaction platforms play a key role. The Web has resulted in the development of a number
of transaction platforms that reduce the costs of connecting buyers with sellers and consummating
25
Press Release, comScore, comScore Media Metrix Releases Top 50 Web Rankings for January, at tbl. 3 (Feb. 18,
2008), available at http://www.comscore.com/press/release.asp?press=2067.
26
Properties include all of the sites owned by an entity, including search engines, international sites, and sites acquired.
For example, the “Google Sites” include Google.com and other international homepages; search pages for specific
categories such as news and images; applications such as Gmail, Google Maps and Google Earth; and sites owned
by Google such as Picasa and YouTube.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Issues Raised by Global Internet Economy 267
trades between them. These platforms earn most of their revenues and profits from transaction fees.27
For example, although eBay began by helping consumers sell second-hand goods to other consum-
ers who wanted to buy those goods, it has evolved into a broad platform for connecting consumers
and businesses. Other e-commerce sites, such as Amazon, have started moving from directly selling
merchandise on their own behalf to providing a platform for connecting businesses and consumers.
Social networking is a critical innovation. Social networking has emerged as a new form
of communication and interaction among individuals. MySpace, for example, has attracted 69
million users worldwide, who post information about themselves on the site and use it to stay in
contact with friends and to make new acquaintances. A related phenomenon is that a great deal
of the content on the Web is generated by users. Although sites such as YouTube, MySpace, and
Facebook rely on advertising to make money as traditional media firms do, they expend few re-
sources on actually creating or purchasing any content themselves—they focus mainly on inducing
others to provide this content.
Demand- and supply-side economies of scale tend to lead to certain segments of the Web
being dominated, on a national and often a global basis, by a few large firms.28 On the demand
side, some Web-based platforms create more value for each customer as they obtain more custom-
ers. A transaction platform such as eBay, for example, is more valuable to buyers because it has
more sellers and more valuable to sellers because it has more buyers. Moreover, buyers and sellers
appreciate being able to reach others across borders.
On the supply side, some Web-based platforms—Google, for example—also incur significant
fixed costs in developing and maintaining their software platforms and in acquiring computer
server and storage capacity for their activities. The average cost of providing products and services
is lower for larger platforms, which can amortize these fixed costs over a larger customer base.
Demand- and supply-side economies of scale result in larger firms being able to offer greater
value to consumers at lower costs, as discussed in more detail below. These economies of scale do
not necessarily result in a monopoly, but they do tend to limit the number of viable firms in a
27
These transaction platforms usually have integrated payment platforms. eBay has PayPal, Alibaba has AliPay, and
Google’s commerce site has Google Checkout. In 2007, 25% of eBay’s net transaction revenue came from the use of
its payment platform PayPal, rather than its auction platforms and communications segment (for example, Skype).
28
Demand-side economies of scale result when a product or service becomes more valuable as more people use it. This
results from direct and indirect network effects. In the case of direct network effects, “the benefit to a network user
depends directly on how many other users are hooked up to the network.” In indirect network effects, “the benefit
to a user arises indirectly because the number of users of the network affects the price and availability of comple-
mentary products.” See Dennis W. Carlton & Jeffrey M. Perloff, Modern Industrial Organization 392–93
(4th ed. 2005). Supply-side economies of scale result when the long-run average total cost of providing a good or
service falls as the quantity of output increases. Id. at 36–40.
The mere fact that a firm has a large share of a segment does not necessarily imply that it has economies of scale or
network effects, and in fact, some of the shares discussed below are likely the result of the fact that these firms were
just better than their rivals. However, for the reasons discussed below, it is apparent that indirect network effects,
and to a lesser extent cost-based economies of scale, are important for these businesses.
PLATFORM ECONOMICS:
268 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
segment.29 In some segments, it is in fact unclear how many viable firms will remain and whether
they will evolve towards monopoly.
Web platforms support many Web firms. These giant global Web-based firms provide plat-
forms for other Web-based businesses. They make software services available so other businesses
can provide complementary services. For example, the music service iLike.com and the online
Scrabble platform Scrabulous are two of the most popular Facebook applications, raising the value
of the Facebook platform as a whole. Google makes its popular mapping software available to de-
velopers who are writing applications based on Google’s maps. Web platforms also provide services
that these businesses rely on. Many Web-based small businesses depend on auction sites such as
eBay or e-tailers such as Amazon to make sales. Most blogs rely on Google to sell ad space on their
sites to advertisers who want to reach blog viewers.30 These Web platforms are examples of the
multi-sided platforms that we describe in more detail below. The Web-based businesses that rely on
these platforms provide complementary products and services that make the Web platform more
valuable and help drive revenue to it.
29
As firms become larger, indirect network effects and economies of scale from further expansion may diminish, and
congestion and managerial economies may counter the other benefits from size. Furthermore, multiple firms may
coexist in a segment despite indirect network effects and economies of scale if they can differentiate their products
and services from each other.
30
See, e.g., Posting of David Evans to Catalyst Code, http://www.thecatalystcode.com/theconversation/blog/2007/
08/06/economics-of-the-blogosphere/ (Aug. 6, 2007).
31
Multi-sided platforms are also called “two-sided markets” by economists, even though they are not markets—at
least as markets are defined in antitrust. See, e.g., Jean-Charles Rochet & Jean Tirole, Two-Sided Markets: A Progress
Report, 37 RAND J. Econ. 645 (2006). Most Web-based businesses are at least two-sided because they are transac-
tion platforms (such as eBay, Amazon, and Alibaba), which connect buyers and sellers; social networking sites (such
as MySpace and Facebook), which connect friends; or advertising-supported sites (such as CNN.com and social
networking sites), which connect viewers and advertisers.
32
See David S. Evans, The Antitrust Economics of Two-Sided Markets, 20 Yale J. on Reg. 325, 328 (2003); David S.
Evans, Some Empirical Aspects of Multi-Sided Platform Industries, 2 Rev. Network Econ. 191 (2003); Jean-Charles
Rochet & Jean Tirole, Platform Competition in Two-Sided Markets, 1 J. Eur. Econ. Ass’n 990 (2003).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Issues Raised by Global Internet Economy 269
value-creating exchanges. They tend to arise where there is some value from getting multiple sides
together but transactions costs or other obstacles stand in the way. eBay, for example, drastically
lowered the cost of exchange between buyers and sellers of second-hand goods.
Multi-sided platforms usually perform each of three interrelated core functions to some
degree.33 First, they serve as matchmakers to facilitate exchange by making it easier for members of
each group to find each other. That can be for love (Matchmaker.com) or money (eBay). Second,
they build communities (or audiences) because this makes it more likely that members of a group
will find a suitable match. Facebook provides value in part because people are more likely to find
people they want to meet and because advertisers can reach a large audience. The value of the
platform grows as the audience grows. Third, they provide shared resources and reduce the cost of
providing services to multiple groups of customers. This is an especially important characteristic of
the software platforms discussed below.
One key feature of multi-sided platforms is the presence of indirect network effects.34 This
means that the value that a customer on one side realizes from the platform increases with the
number of customers on the other side. Consumers looking to buy something value a search
engine more if it provides advertisements that are more relevant to their search, while companies
value advertising on a search engine higher if they are more likely to reach potential consumers.
Another key feature is that multi-sided platforms must cater to multiple, distinct customer
groups simultaneously. To establish a two-sided platform, for example, the founders must solve a
chicken-and-egg problem: customers on Side A will not participate without customers on Side B,
but customers on Side B will not participate without customers on Side A. YouTube had to pursue
people who wanted to post videos, people who wanted to watch videos, and advertisers who want-
ed to reach these viewers. These features make the profit-maximizing calculus for a multi-sided
platform more intricate than for a traditional business. A firm operating one of these platforms
must consider the demands of all sides, the interrelationships between these demands, the costs
directly attributable to each side, and the costs of running the platform.
Further complicating this calculus is the fact that the profit-maximizing prices for multi-
sided platforms can result in users on one side getting a price that is less than the incremental
cost of serving that customer, and sometimes the price for users on one side may even be less
than zero.35 The side that is “needed more” or that is “harder to get” may receive a price break;
conversely, the side that gets the most value out of access to members of the other side likely
33
See David Evans & Richard Schmalensee, The Industrial Organization of Markets with Two Sided Platforms, 3
Competition Pol’y Int’l 158–59 (2007).
34
See, e.g., Michael L. Katz & Carl Shapiro, Systems Competition and Network Effects, 8 J. Econ. Persp. 93 (1994).
35
See Mark Armstrong, Competition in Two-Sided Markets, 37 Rand J. Econ. 668, 668–70 (2006); Geoffrey G.
Parker & Marshall W. Van Alstyne, Two-Sided Network Effects: A Theory of Information Product Design, 51 Mgmt.
SCI. 1494, 1498 (2005); Rochet & Tirole, supra note 31, at 659.
PLATFORM ECONOMICS:
270 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
bears more of the cost.36 As an empirical matter, many multi-sided platforms make their money
from one side and make access to the platform available to another side for a price that does not
cover the cost of provision.37 Facebook, for example, is free to users and makes money by selling
advertising.38
There are several major classes of industries in which most if not all of the businesses are
based on multi-sided platforms. These include advertising-supported media, such as newspapers,
magazines, radio, and television; payment services, such as credit and debit cards; exchanges, such
as auction houses, commodity exchanges, and financial exchanges; and dating and matchmak-
ing businesses, such as singles bars and matchmaking services. Another major class consists of
industries that have software platforms as their underlying technology.39 These include computer
operating systems, mobile telephones, personal digital assistants, and video game consoles.40 Many
Web-based businesses are also included.
36
See David S. Evans & Richard Schmalensee, Catalyst Code: The Strategies Behind The World’s Most
Dynamic Companies 71–92 (2007).
37
When there are more than two sides, at least one side must make money. Id.
38
See Facebook, About Facebook, http://www.facebook.com/about.php (last visited Apr. 23, 2008).
39
See generally David S. Evans, Andrei Hagiu & Richard Schmalensee, Invisible Engines: How Software
Platforms Drive Innovation and Transform Industries (2006).
40
See id. at 1–2.
41
The owners of Scrabble have objected to this take on their game. See Heather Timmons, Scrabble Tries to Fight a
Popular Imposter at Its Own Game, N.Y. TIMES, Apr. 7, 2008, at C7.
42
An operating system is a type of software platform that manipulates the computer hardware in addition to providing
code that supports other software and hardware applications.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Issues Raised by Global Internet Economy 271
may reside on other servers that are attached to the Internet. This has resulted in what is sometimes
called “cloud computing,” in which the software platform and possibly the application primarily
reside on several interchangeable computers that the individual user accesses through the Internet.
Google’s search-based advertising platform is an example. The search engine that individuals use
to conduct search queries, much of the software that advertisers rely on for advertising campaigns,
and much of the software that publishers rely on for inserting advertisements into their Web pages
reside on vast interconnected but indistinguishable “server farms” that Google operates around
the world.
Social Friends
Networks
Application
Own Search Users
Google
Advertisers Advertisers
Advertisers
Application
Developers
Own
Buyers Google
Advertisers Content
Network
Advertisers
Google
Transaction Advertisers
Platforms Searchers
Sellers Publisher
Own
Advertisers Websites
PLATFORM ECONOMICS:
272 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
Google’s advertising platform enables companies to insert ads based on keywords used in a
search query, in which case the ad appears on the search results page, or based on the keywords
found in a Web site that belongs to Google’s network of Web publishers.43 Google’s search engine
makes money by drawing traffic to its search results pages, where it sells and places advertis-
ing. That search engine also helps people find Web-based businesses—including publishers and
e-tailers—that are not paid advertisers. Those businesses benefit from Google’s search engine, but
Google does not charge them for being listed in the organic search results that appear on the left-
hand side of the search results page. Google also provides advertising services to Web publishers.
Those publishers make space available for Google to insert ads; Google sells that space to advertis-
ers and pays the publishers a portion of its ad revenues.
Many of the entities that Google interconnects with are also multi-sided platforms. Web
publishers operate two-sided platforms in which they use content to attract viewers and sell
access to those viewers to advertisers. Many small publishers, including blogs, rely entirely on
Google to sell their advertising space. Many large publishers use Google to sell some portion of
their advertising space; some of them also have Google search boxes and receive payments from
Google for advertising revenue that results from their visitors clicking on ads on Google’s search
results pages. Social networking sites are similar to Web publishers in using advertising to make
money. The site attracts traffic by providing social networking and makes money by selling that
traffic—and data related to individual users—to advertisers through platforms such as Google.
Google’s advertising platform also intersects with eBay’s transaction platform. eBay buys adver-
tising on Google’s search results pages to attract potential buyers to the various products and
services on eBay. In addition, eBay makes advertising space available to Google and receives
payments in return.
Google makes its APIs available to software developers that are writing programs to provide
other services. In return, Google reserves the right to insert advertising on those services. Since
January 2007, developers have written around 20,000 “gadgets” used across 100,000 Web sites.
These mini-applications use the Google Gadgets API and can run on various Google platforms
(e.g., Google Calendar, iGoogle, Google Desktop, Blogger, Google Maps, and Orkut). Often
they can be embedded in other Web pages and run on other third-party applications (e.g.,
MyAOL). Developers can also create map applications on their Web sites using the Google
Maps API. For example, using the Google Maps API, Orbitz added“Orbitz Updates” to its site,
a map which shows real-time user-reported weather, traffic, parking, and wait-line conditions at
U.S. airports.
43
See David S. Evans, The Economics of the Online Advertising Industry 37 (Jan. 2008) (unpublished manuscript,
on file with the Northwestern University Law Review).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Issues Raised by Global Internet Economy 273
44
I am using the term “category” to refer to commonly known types of products or services, such as social networking,
portals, instant messaging, and online auctions. These categories do not necessarily correspond to markets because
products and services in one category can substitute to some degree for products and services in another category.
Moreover, the extent of cross-category substitution can vary over time; for example, in my experience people are
increasingly using social networking sites such as Facebook as their entry point when they sign on to the Web rather
than a traditional portal such as Yahoo.
45
I have chosen these three because, as discussed below, they reflect the most important types of platforms for the fore-
seeable future. However, the same basic points apply to instant messaging, online payment systems, and webmail.
These categories do not necessarily correspond to relevant antitrust markets, and any analysis of market definition
should properly consider the two-sided issues mentioned above. As noted below, these shares are based on categories
as reported by comScore and may overstate or understate the true significance of the leading companies. First, the
shares of the main search engine providers (Google, Yahoo, Microsoft, and Baidu) are understated because com-
Score includes searches that are done within Web sites such as eBay in its calculation of search shares, even though
eBay is not generally used for Internet searches. Second, the calculation of search query shares further understates
the search revenue share for Google because Google earns a higher revenue per search than Microsoft or Yahoo.
See Miguel Helft, A Long-Delayed Ad System Has Yahoo Crossing Its Fingers, N.Y. Times, Feb. 5, 2007, at C1; Search
Marketing Communications, http://cohn.wordpress.com/category/revenue-per-visit/ (Feb 7,2008).
46
Facebook’s implied market value after investments by Microsoft and Chinese billionaire Li Kashing is $15 billion;
that suggests that at least some investors are betting that Facebook will become the leading social network. See
Thomas R. Eisenmann & Brian Feinstein, Facebook Platform 1 (Harvard Bus. Sch. Case Study No. N2-808-128,
2008); Suzy Jagger, Li Ka-shing Makes Big Impression on Facebook, Times Online (London), Mar. 29, 2008, http://
business.timesonline.co.uk/tol/business/industry_sectors/technology/article3642805.ece.
PLATFORM ECONOMICS:
274 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
France eBay 99% Delcampe 0% Google 82% Microsoft 3% Skyrock 78% Facebook 7%
Germany eBay 99% Yatego GmbH 0% Google 80% eBay 6% StudiVZ 43% Schueler 12%
Italy eBay 99% Bidplaza 0% Google 85% Telecom Italia 3% Netlog 27% Badoo 23%
Spain eBay 94% Mercado-Libre 3% Google 93% Microsoft 2% Fotolog 50% Metroflog 9%
South Korea NHN 65% Lycos 16% CyWorld 97% Paran.com Club 0%
UK eBay 98% Price-drop tv 0% Google 74% eBay 6% Bebo 40% Facebook 28%
USA eBay 96% Bidz 1% Google 53% Yahoo! 17% MySpace 62% Facebook 21%
Worldwide eBay 57% Taobao 14% Google 62% Yahoo! 13% MySpace 17% Facebook 15%
47
comScore, Mymetrix Key Measures Report (Dec. 2007); comScore, Mymetrix Qsearch 2.0 Key Measures
Report (Dec. 2007).
48
The search figures are shares as reported by comScore. They include searches on Web sites where searches are primar-
ily or exclusively used to search within the site rather than generally on the Internet. They also include searches on
sites where advertising is not featured. Shares of search engines used for general searches on the Internet and shares
of search engines relevant to search advertising will likely be higher than the shares reported here.
49
Social networking shares are also as reported by comScore. They include blogging sites such as Blogger. If these sites
are excluded, the shares of the market leaders would be higher.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Issues Raised by Global Internet Economy 275
31, 2008. Following this decline, Microsoft announced its desire to acquire Yahoo on February 1,
2008. Despite its past success and enormous user base, some analysts concluded that Yahoo could
not succeed on its own.50 Other Web giants have also encountered problems. The growth of eBay
has slowed, and the company was undergoing a shakeup in management in early 2008. It faces
increased competition from Amazon, Google, and other Web properties that provide transac-
tion platforms for businesses. The capital markets have also expressed profound uncertainty over
Google’s growth. Its share price fell by 34% between January 2, 2008 and April 17, 2008, the day
it announced its first quarter earnings.51 Google reported a 30% increase in quarterly profits that
day, and its stock increased by 23% by April 22, 2008.52
50
See, e.g., Michael Liedtke, Microsoft Woo Yahoo?, St. Paul Pioneer Press, Feb. 12, 2008, at C1; Claudine
Beaumont, Fading Yahoo! Needs Microsoft Rescue, Telegraph, Feb 2, 2008, http://www.telegraph.co.uk/money/
main.jhtml?xml=/money/2008/02/01/bcnmicro601.xml; Grouchy Geek, http://grouchygeek.blogs.fortune.cnn.
com/2008/02/01/yahoo-needs-microsofts-help/ (Feb. 1, 2008, 11:56 a.m.).
51
See Google Finance, Google Inc., http://finance.google.com/finance?chdnp=1&chdd=1&chds=1&chdv=1&ch
vs=maximized&chdeh=0&chdet=1208462400000&chddm=28934&q=NASDAQ: GOOG&ntsp=0. In fact,
Google’s stock price had been stumbling since November, 2007. See Posting of Jim Kerstetter to Cnet Tech News
Blog, http://www.news.com/8301-10784_3-9920376-7.html (Apr. 16, 2008, 11:06 PDT).
52
See Press Release, Google, Google Announces First Quarter 2008 Results (April 17, 2008), avail able at http://www.
google.com/intl/en/press/pressrel/revenues_q108.html; Google Finance, Google Inc., http://finance.google.com/fi
nance?chdnp=1&chdd=1&chds=1&chdv=1&chvs=maximized&chdeh=0&chdet=1208894400000&chddm=156
4&q=NASDAQ:GOOG&ntsp=0.
53
In keeping with the global focus of this chapter, “complainants” is used to refer to parties that complain to a com-
petition authority as well as plaintiffs in private actions, which is the dominant form of antitrust enforcement in the
United States.
54
See Case T-219/99, British Airways plc v. Comm’n, 2003 E.C.R. II-5917, ¶¶ 211, 223–25; Case COMP/38.233,
Wanadoo Interactive, Commission Decision of July 16, 2003, ¶ 227.
PLATFORM ECONOMICS:
276 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
with the Commission in its Microsoft judgment.55 Under U.S. law, firms that have market shares
in excess of sixty percent are often considered to have monopoly power.56 Although it has become
more difficult for plaintiffs to prevail on various monopolization theories in the United States, the
caselaw on tying products together and offering discounts for bundles of products continues to
provide significant opportunities for plaintiffs to pursue cases.
The existence of indirect network effects and economies of scale means these platforms are
competing in “winner take all” and “a few winners take all” markets. That leads to aggressive
struggles to win market share at the expense of rivals. Competition authorities worry and rivals
complain, however, when tough business tactics succeed in reducing the rivals’ sales, thereby fore-
closing them from the market. Complicating matters, competition authorities and courts have
difficulty distinguishing procompetitive from anticompetitive business practices for multi-sided
platforms. For example, as mentioned previously, multi-sided platforms often charge prices that
are below cost to customers on one or more sides of the platform. In some circumstances, these
low prices drive out competition as a result of what business strategists refer to as “envelopment.”57
Rivals who lack the money-making side of the platform that subsidizes the money-losing product
cannot survive.
Given the probable evolution of the Web-based economy, several competition and regulatory
issues are likely to arise in the coming years as a result of this antitrust scrutiny.
The emergence of impregnable monopolies. There are likely to be concerns over the seeming
monopolization of certain segments. It is possible that the Web economy will see a constant churn-
ing of its leading players. The fact that eBay and Yahoo have lost their once apparent impregnabil-
ity is consistent with the view that dominance is fleeting. However, the evolution of the Web econ-
omy thus far is also consistent with the evolution of other industries in which it takes time for the
winners to emerge.58 If so, it is possible that a handful of firms will have near-monopoly positions
in certain segments and that those positions will be protected in part by indirect network effects
and the economies of scale resulting from the ability to average fixed software and hardware costs
55
See Harry First, Strong Spine, Weak Underbelly: The CFI Microsoft Decision, Global Competition Pol’y, Sept.
28, 2007, http://www.globalcompetitionpolicy.org/index.php?id=555&action=907; Renata B. Hesse, Microsoft
and the Court of First Instance: What Does It All Mean?, Global Competition Pol’y, Oct. 3, 2007, http://www.
globalcompetitionpolicy.org/index.php?id=562&action=907.
56
See Gerald F. Masoudi, Deputy Assistant Attorney Gen., Antitrust Div., U.S. Dep’t of Justice, Some Comments
on the Abuse-of-Dominance Provisions of China’s Draft Antimonopoly Law, Address Before the University of
International Business and Economics Competition Law Center Conference on Abuse of Dominance: Theory and
Practice, (July 21, 2007), available at http://www.usdoj.gov/atr/public/speeches/225357.htm.
57
See Thomas R. Eisenmann, Geoffrey Parker & Marshal W. Van Alstyne, Platform Envelopment 3 (Harvard Bus. Sch.
Tech. & Operations Mgmt. Unit, Working Paper No. 07-104, 2007), available at http://ssrn.com/abstract=996852.
58
See David S. Evans, Daniel D. Garcia Swartz & Bryan G. Martin-Keating, The Failure of E-Commerce Business:
A Surprise or Not? 3 EUR. BUS. ORG. L. REV.1, 7 (2002); Michael Gort & Steven Klepper, Time Paths in the
Diffusion of Product Innovations, 92 ECON. J. 630, 631, 642 tbl.3 (1982) (noting the decline in number of firms).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Issues Raised by Global Internet Economy 277
across larger communities. American antitrust policy recognizes that such monopoly is the reward
for successful investment and innovation.59 European Community-based competition policy views
competition as the more desirable outcome, and when that is not possible, it imposes significant
obligations on the dominant firm.60
Leveraging into adjacent markets. The structure of the Web ecosystem makes it likely that
dominant firms will seek to move into related markets for complementary products or services.
Because these firms are based on software platforms, it is relatively easy to add new features and
services. For example, Google introduced its Google Checkout payment service in competition
with PayPal by extending its software platform, integrating code into Google Product Search, and
bundling Google Checkout for merchants into AdWords for advertisers. Moreover, the leading
Web platforms often provide complementary services. It is a well-established economic proposi-
tion that a monopoly could make a greater profit if it also owned complementary monopolies or
if it could replace these complementary monopolies with competitive markets.61 Therefore, assum-
ing that competition is not feasible, we would expect the dominant firms to attempt to establish
monopolies across other segments. This could happen through mergers or through one dominant
firm challenging another, as Google is doing with eBay.
Access to facilities. Issues involving access to other platforms and to the intellectual property
that enables one platform to work together (interoperate) with another are likely to be raised when,
as is the case in the European Community, the law is conducive to doing so.62 One set of issues con-
cerns access to a “closed platform.” The Apple iTunes/iPod platform is largely closed. Apple does
not encourage interoperability. Indeed, Apple seeks to prevent other music stores from making
music available for iPods and other companies’ devices from playing music from iTunes. Although
it could obtain indirect network effects from an open strategy, it has chosen a tightly integrated
59
See Verizon Commc’ns Inc. v. Law Offices of Curtis V. Trinko, LLP, 540 U.S. 398, 407 (2004).
60
Firms that are dominant are subject to the provisos of Article 82 of the EC Treaty, which has been interpreted to
prohibit various forms of tying, refusals to deal, bundled price rebates, pricing below cost, and other activities.
See Richard Whish, Competition LAW 202–08 (5th ed. 2004) (discussing examples of abuse of dominance).
In many of these cases, the practices are essentially prohibited per se if the firm is dominant—has a share of an
antitrust market that is higher than forty percent or so. See Case T-219/99, British Airways plc v. Comm’n, 2003
E.C.R. II-5917, ¶¶ 211, 223–25; Case COMP/38.233, Wanadoo Interactive, Commission Decision of July 16,
2003, ¶ 227.
61
See Michael A. Salinger, Introduction to Chapters VII and IX of Augustin Cournot, Researches into the Mathematical
Principles of the Theory of Wealth, 4 COMPETITION POL’Y INT’L 274, 280–82 (2008) (“Today, the funda-
mental distinction between horizontal and vertical effects is widely accepted by antitrust practitioners.”).
62
The OpenGroup, a consortium that aims to facilitate interoperability, explains that interoperability is the ability to
both exchange information and to use it. “Without a way to exchange information . . . high-tech systems literally
can’t communicate with each other. And, if they can’t communicate, they can’t work—interoperate—with each
other.” Open Group, Interoperability Matters, http://www.opengroup.org/bus_area/interoperability/info1/IBinfo1.
htm#what (last visited Apr. 10, 2008).
PLATFORM ECONOMICS:
278 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
software-hardware business model. This issue is central to the recent European Community cases
against Apple.63 Other platforms close themselves in particular dimensions. Facebook, for exam-
ple, does not allow search engines to crawl its Web site, and as a result, the content on this social
networking site is not available to searchers.
Another set of issues relates to portability of data. Web platforms derive benefits from the data
they collect in a variety of ways. eBay’s “Feedback Forum” provides quality information on sell-
ers that is valuable to buyers. Users input a great deal of valuable personal information into social
networking sites such as MySpace and Facebook. Google retains data on search queries that it can
use to refine searches and deliver ads. In these cases one could imagine competitors seeking access
to this information under an “essential facilities” theory under European Community law.64 One
could also imagine competition policy cases over restrictions that prevent users from exporting
their data to competing sites. This battle has already begun, as Facebook rivals are currently lobby-
ing publicly for the portability of social networking data.
Tying and bundling. As the current spate of cases suggests, it is probable that leading Web
platforms will face complaints over tying of various forms.65 It is also a common business strategy
for software platforms to expand by adding features. They face low marginal costs in doing so, they
can sometimes provide efficiencies by integrating features together or making it easier for consum-
ers to obtain them more conveniently, and they can aggregate demand over users who may value
one feature but not another.66 The D.C. Circuit Court of Appeals found that it was appropriate
63
See Eric Bangeman, EU Commissioner Criticizes iPod-iTunes Tie-in, Ars Technica, Mar. 11, 2007, http://arstech-
nica.com/news.ars/post/20070311-eu-commissioner-criticizes-ipod-itunes-tie-in. html; Keith Regan, iTunes Draws
Ire of EU Commissioner, Macnewsworld, Mar. 13, 2007, http://www.macnewsworld.com/story/56255.html.
64
“[T]he essential facilities doctrine imposes liability when one firm, which controls an essential facility, denies a
second firm reasonable access to a product or service that the second firm must obtain in order to compete with
the first.” Alaska Airlines, Inc. v. United Airlines, Inc., 948 F.2d 536, 542 (9th Cir. 1991). For a review of the im-
plementation of the essential facilities doctrine by the European Union, see James Turney, Defining the Limits of the
EU Essential Facilities Doctrine on Intellectual Property Rights: The Primacy of Securing Optimal Innovation, 3 Nw. J.
Tech. & Intell. Prop. 179 (2005).
65
For example, in September 2007, the European Court of First Instance upheld the European Commission’s ac-
cusations against Microsoft of illegal tying of the Windows operating system and Windows Media Player. See
Case T-201/04, Microsoft Corp. v. Comm’n, 2007 WL 2693858 (Sept. 17, 2007). Further, in January 2008, the
Commission opened an investigation into the alleged tying of Internet Explorer and other software products to
the Windows operating system. See Press Release, European Comm’n, Antitrust: Commission Initiates Formal
Investigations Against Microsoft in Two Cases of Suspected Abuse of Dominant Market Position (Jan. 14, 2008)
available at http://europa.eu/rapid/pressReleasesAction.do?reference=MEMO/08/19. The Commission has also
censured Apple for its tying of the iPod music player and the iTunes music store. See Bangeman, supra note 63;
Regan, supra note 63.
66
See Evans, Hagiu & Schmalensee, supra note 39, at 303–36. Demand aggregation is most easily seen with
newspapers: many people only read portions of the newspapers; however, by offering various features, readers find
enough content to persuade them to subscribe to the newspaper.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Antitrust Issues Raised by Global Internet Economy 279
to apply a rule-of-reason legal standard to tying for software platforms67 because of the p ossibility
68
that there were particularly compelling efficiency explanations for the practice. Whether the
U.S. courts treat the Web-based companies as software platforms, however, remains to be seen;
these companies have different business models and practices than Microsoft did with Windows,
which was the subject of the D.C. Circuit decision. Moreover, the European Community’s Court
of First Instance has reaffirmed the Community’s formalistic approach to tying in the Microsoft
judgment.69
Envelopment and predation. Multi-sided platforms are a bit like clumsy giants stepping on
other creatures as they move through the ecosystem. Although they may crush competitors inten-
tionally, this may also happen as a natural byproduct of legitimate pricing and design decisions.
Multi-sided platforms—and this is particularly true with the leading Web platforms—give many
features and services away, often for the purpose of attracting traffic. They can readily crush compa-
nies that charge for features and services that they offer for free. But it is not only the little guy that
faces concern from this strategy. Google Checkout can undercut eBay’s PayPal because Google,
unlike eBay, can obtain more advertising revenue from having an efficient payment method and
secure data that it can use to target ads better. Likewise, Google’s ability to subsidize software with
advertising poses a threat to Microsoft and other software companies that charge for software;
Microsoft is moving rapidly into online advertising just so that it can have a source of revenue
similar to a key rival. Nevertheless, one would expect that tying, bundling, and pricing strategies
that foreclose rivals will lead to competition policy investigations and prosecutions.
IV. CONCLUSION
At the inception of most new industries, hundreds of firms enter.70 They battle it out over time.
A few winners usually emerge—firms that have secured economies of scale or that have benefited
from superior management or both. This pattern has been repeated numerous times over the
course of the second industrial revolution that started after the Civil War. Consider the automobile
industry. It began with the invention of the automobile by Karl Benz in 1885. The first commercial
automobile company in the United States was Duryea, which entered the market in 1893. As of
1908, 253 automobile companies were competing in the U.S. This was whittled down to the big
67
“A rule-of-reason analysis requires first checking whether the facts of a given case suggest that anticompetitive tying
is a possibility and then weighing those anticompetitive effects with the benefits resulting from a tying policy.” See
David S. Evans, A. Jorge Padilla & Michele Polo, Tying in Platform Software: Reasons for a Rule-of-Reason Standard
in European Competition Law, 25 World Competition 509, 514 (2002).
68
See United States v. Microsoft Corp., 253 F.3d 34, 93 (D.C. Cir. 2001).
69
See Christian Ahlborn & David S. Evans, The Microsoft Judgment and Its Implications for the Competition Policy
Towards Dominant Firms in Europe, 75 Antitrust L.J. (forthcoming 2008).
70
See Gort & Klepper, supra note 58, at 631.
PLATFORM ECONOMICS:
280 Part Three : Web Based Platforms Essays on Multi-Sided Businesses
four—General Motors, Ford, Chrysler, and American Motors—by 1960. Many other industries
followed a similar course.
Antitrust scrutiny often follows consolidation. The leaders in the industry have large market
shares, which, under longstanding antitrust practice, makes them vulnerable to claims of unlaw-
fully maintaining or acquiring a monopoly or running afoul of other antitrust laws that have a
market power screen. AT&T was hit with its first major antitrust case in 1911, twenty-six years
after it opened the first telephone exchange. IBM faced its first antitrust case forty-seven years after
it received patents for the punch card machine.
We can expect the Web-based industries to follow the same trajectory, and thus far they have.
Massive entry has taken place. As with many new industries, we remember the YouTubes that suc-
ceeded but we forget that Google Video and hundreds of other start-ups tried and quickly failed.
There are some differences, however, which suggest that antitrust controversies will arise more
quickly than they did for other industries. The first is speed. Although the notion of “Internet
Time” may have been exaggerated, it is true that Web-based firms can achieve leading positions in
many countries around the world very quickly. The second is complexity. Almost all of the leading
Web-based firms have intricate multi-sided business models. The third is interconnectedness. The
Web economy is interconnected, which leads to dependencies and rivalries that can create conflict
and antitrust complaints.
As a result, the competition authorities and courts will have a challenging set of issues to deal
with concerning the Web-based economy in the years to come. The future will bring merger cases,
as firms seek to consolidate to achieve economies of scale and indirect network effects; refusal-to-
deal cases, as closed platforms deny others access to their communities; predation cases, as rivals
complain about free offerings that foreclose or destroy them; tying cases, as platforms use software
platform technologies to add features and functions, which in some cases will foreclose their rivals;
and exclusive dealing cases, as platforms lock up traffic to achieve indirect network effects. Courts
and competition authorities should exercise care in balancing the need to protect long-run social
welfare against the need to stop anticompetitive strategies in this highly dynamic and complex part
of the economy.
Part Four
PAYMENT CARDS
CHAPTER
ABSTRACT
The simple payment card has been around since at least the beginning of the twentieth cen-
tury. Hotels, oil companies, and department stores issued cards before World War I. In response
to customer requests, Sears began offering lines of credit in 1910 to customers of “unquestion-
able responsibility,” although the Sears card came more than a decade later. Some large retailers
gave cards to their wealthier customers that identified them as having a charge account with
the store. By the 1920s, several department stores allowed cardholders to pay off their bills in
monthly installments. Metal “charge-plates” with embossed consumer information were intro-
duced by department stores in 1928. During the 1920s as well, oil companies issued “courtesy
cards” for charging gas. By the end of World War II, charge cards were no longer a novelty, but
they were about as far from the cards of today as barter was from coin.
“Perhaps you would like to see what our credit cards are like,” [Doctor Leete asks of his guest, Julian
West]. “You observe...that this card is used for a certain number of dollars....The value of what I
procure on this card is checked off by the clerk, who pricks out of these tiers of squares the price
of what I order.” [Mr. West, from Boston, has awakened 113 years in the future, in the year 2000.]
—Edward Bellamy, Looking Backward, 2000–1887 (1888)
282
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses More than Money 283
Merchant coalitions were another source of competition. Hotel owners balked at the fee they
had to pay on charge cards. In 1956, the American Hotel Association established the Universal
Travelcard. It didn’t charge participating hotels and rental car agencies any fee but billed cardhold-
ers the same $30 fee as Diners Club. The National Restaurant Association, with sixty thousand
members, signed on. One might wonder how they could get by without the 70 percent of the pie
from merchant fees that Diners Club received. Part of the answer is that they didn’t do central
billing. Each hotel billed its customers directly, although the card association did ensure payment.
There was a railroad card and an airline card as well.
Some banks had also entered the payment card business by the 1950s. But their cards—some-
times called “shopper” cards—targeted a different cardholder and merchant base. These bankcards
were typically held by “housewives,” as the newspapers of the day put it, and could be used only
at retail stores in the locality the bank did business. Franklin National Bank started one of the first
shopper cards on Long Island in 1951, and one hundred or so banks—mainly small, suburban
banks in the Northeast—followed. While one hundred may seem like a large number, one should
keep in mind that in 1951, there were 13,455 commercial banks as well as many more credit
unions and savings and loans. Banks generally charged retailers the then-standard 5 to 7 percent
merchant fee. Cardholders reportedly didn’t pay any direct fees and, as with the other card pro-
grams, were supposed to pay their monthly bill in full. These cards were thus charge cards, though
the press of the day called them credit cards.
Of the early major entrants, only Diners Club survived the decade as a stand-alone company.
It bought Trip-Charge in 1956 and Esquire’s card program in 1958. The bankcards failed mainly
because they had trouble signing up merchants, and only twenty-seven of the shopper cards were
still in existence by 1957. The Universal Travelcard and the Gourmet Magazine Club card were
swallowed in 1958 by a new competitor that appeared during what became a critical year for the
future history of the card industry.
Diners Club also expanded overseas in the mid-1950s, using franchise agreements to extend
its reach to Europe. As in the United States, European hotel trade associations posed strong resist-
ance to the travel and entertainment (T&E) cards, going so far as to expel members who accepted
payment cards that required a merchant discount. Some hotels in England and Switzerland chose
to flout the prohibition; they accepted the T&E cards and formed their own association. They also
went a step further: the newly formed hotel association created the BHR credit card in the 1950s,
which evolved into the EuroCard in the mid-1960s.
The travelers cheque, invented by an American Express employee and offered beginning in
1891, was a significant advance over the money order. The cheques came in multiple denomi-
nations just like cash and had the dual-signature system (sign when you obtain, and sign when
you cash) that remains the major security device to this day. Initially, people could cash travelers
cheques only at American Express offices; later, they could cash them directly at merchants. The
major selling point for consumers was security: American Express guaranteed payment, but it
also assumed responsibility for lost or forged checks. The product was a highly profitable one for
American Express for over a century, although its popularity has declined steadily over the last
three decades. The profits all came from individuals. A consumer purchasing $500 in travelers
cheques, say, would pay American Express a fee in addition to the $500, and American Express
would continue to earn interest on the amount invested in the cheques until they were cashed.
Stolen or misplaced cheques that were not cashed or replaced would be pure profit.
A hundred years after its formation in Buffalo, New York, American Express was the world’s
largest travel agency and operated the world’s largest private mail service. Between its cheques and
travel offices, it was profiting enormously from the boom in international travel following the end
of the Second World War. The number of American Express travel agencies grew from fifty at the
end of the war to nearly four hundred ten years later. The company sold approximately $6.5 billion
worth of travelers cheques in 1951, and by the end of the decade claimed to control 70 percent of
the U.S. travelers cheque business.
The Diners Club charge card was a new competitor for American Express. By 1953, American
Express had begun planning its response. It considered buying Diners Club in 1956, but rejected
the idea. The company finally entered the charge card industry on October 1, 1958, with 17,500
merchant locations and 250,000 cardholders. It achieved this scale quickly by buying the Gourmet
Magazine Club card and the Universal Travelcard. Within seven months of launching its card op-
erations, American Express had over 600,000 cardholders. By late 1960, it had a charge volume of
over $584 million and 750,000 cardholders.
American Express adopted a slightly different pricing policy than Diners Club. It initially set
its annual fee $1 higher (in 1958 dollars) than Diners Club’s $5, thereby suggesting that it was the
more “exclusive” card (in 2008 dollars, the American Express annual fee was $42 while the Diners
Club fee was $35). But it set the initial merchant discount slightly lower than Diners Club’s 7
percent: 5 to 7 percent for restaurants, according to their sales volume; and 3 to 5 percent for the
recalcitrant hotel industry, depending on the hotel guest’s charge level.
American Express struggled at first. Even a charge card involves extending credit for a time,
and American Express, unlike the founders of Diners Club who have been in the consumer cred-
it business, had no experience doing this. By 1961, with losses mounting, it considered selling
the business to Diners Club, but decided that such a sale might not pass muster with the Justice
Department. Instead, American Express hired George Waters, later known as the “Father of the
Card,” to run its card operations. Waters started putting pressure on customers who had not
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses More than Money 287
sent their payments in on time. And he raised the annual fee to $46, and later to $56. Despite
the increased fees, the American Express payment card system continued to grow. By the end
of 1962, there were 900,000 American Express cardholders who could use their cards at 82,000
merchant locations. In 1962, almost four years after its launch, the card operation posted its
first (small) profit.
Today, American Express accounts for 13 percent of the dollars transacted on payment cards.
By that measure, it is almost a quarter of the size of the Visa system (see Table 1.1 in chapter 1)
and almost one-half the size of MasterCard. Diners Club, on the other hand, shrunk to almost
nothing within the United States. The charge card pioneer spread itself too thin in the 1960s and
early 1970s in an attempt to counter the inroads made by American Express. In particular, Diners
Club tried unsuccessfully to follow the American Express model by expanding into travel clubs and
travel agencies, but it lost money on both endeavors and lost sight of the newly emerging competi-
tion from the bankcards.
development of the “Everything” credit card program it had started in 1967. When Citigroup
quickly dropped the Everything card and had not introduced a replacement by 1978, however, the
Justice Department relented, and Citigroup bought Carte Blanche back again. By that time, Carte
Blanche’s share of credit card volume had declined to less than 1 percent.
American Express moved past Diners Club to become the industry’s volume leader in 1966.
Diners Club continued to decline throughout the 1960s, in part because it lacked the travel of-
fices that American Express used to distribute its card during the industry’s early days. American
Express also had a better T&E brand as a result of its travel offices and travelers cheques. Diners
Club attempted to meet American Express head-on through travel and reservation system acquisi-
tions, but it failed to make those profitable. Diners Club was sold to Citigroup in 1981. After the
sale, Diners Club shifted its focus to affluent business travelers, trying to follow American Express’s
successful up-market strategy. The final indignity for this pioneer came in 2008 when Discover, the
newest and smallest of the payment networks, bought Diners Club International for $165 million
(Citigroup retained the franchise to issue Diners Club cards in the US but then sold even that to
Bank of Montreal in 2009.)
And the statewide BankAmericard became the worldwide Visa card.
ers’ checking accounts. Many business travelers and wealthy households had one of these cards,
but most Americans didn’t. Data for 1966 are not available, but even by 1970 only 9.2 percent of
households had one of the T&E cards.
Interstate banking regulations and other hurdles made it difficult for Bank of America to
compete head-to-head with the three T&E cards. To take its card national, the California bank
decided to franchise. In 1966, it announced that it would license its BankAmericard program
to selected banks across the country. Each bank would operate the program independently us-
ing the BankAmericard name; merchants signed up by the franchisees would have to accept all
BankAmericards, allowing consumers to use their BankAmericards at any participating merchant.
Bank of America charged the franchisees a royalty of up to 0.5 percent of cardholder volume and
an entry fee of about $132,000.
Unlike T&E cards, bankcards did not charge cardholders membership fees, earning revenue
from finance charges and merchant discounts instead. For example, the Chase Manhattan Charge
Plan, introduced in 1958, charged cardholders 1 percent of the revolved (that is, unpaid) balance
every month, while charging merchants 2 to 6 percent of sales depending on volume.
The BankAmericard franchise was not limited to the United States. Major banks in coun-
tries such as Canada, Columbia, Italy, Japan, Mexico, Portugal, Spain, the United Kingdom, and
Venezuela signed up as international BankAmericard franchisees around the same time as the
domestic franchise system launch in 1966. In 1968, MasterCard also expanded internationally
by forming alliances with EuroCard, the European card association mentioned earlier, and Banco
Nacional in Mexico. The alliances allowed the MasterCard network and foreign networks to inter-
operate, but preserved each card as a distinct brand. In addition, MasterCard expanded in Asia by
gaining member banks in Japan.
In the United States, within two months of the Bank of America franchising announcement,
American Express, Carte Blanche, and Diners Club responded by offering their own franchise op-
portunities to banks. The American Express bankcard differed from the standard card: it offered a
minimum $9,000 line of credit. American Express would split revenues with the banks: banks got a
commission for signing up cardholders and revenues from credit provided by the card; and existing
American Express cardholders could be converted to the bank program. American Express didn’t
charge additional franchise or licensing fees. Carte Blanche priced its franchise at $45 for every $5.3
million in bank assets, with a $26,429 minimum fee. Diners Club offered to franchise its card for
that same fee to banks with less than $5.3 billion of assets and for $52,858 for banks with $5.3 bil-
lion or more. While the historical evidence is sketchy, it does not appear that the various efforts at
franchising these cards attracted any takers within the United States. (American Express also planned
to franchise its Uni-Card credit card across the country in 1968. It had a million cardholders and
eighteen thousand merchants in New England, New York, New Jersey, and Pennsylvania. Instead of
following through on those plans, though, it sold the Uni-Card program back to its original owner,
Chase Manhattan, in 1969, and Chase converted these cards to BankAmericards in 1972.)
PLATFORM ECONOMICS:
290 Part Four : Payment Cards Essays on Multi-Sided Businesses
For many banks, there were significant negatives to the franchise system. Major banks, includ-
ing Wells Fargo in California and Chase Manhattan in New York, were not eager to sign up to
issue someone else’s card. The successful franchise systems we’re familiar with—McDonald’s, the
Athlete’s Foot, or Mail Boxes Etc.—typically involve a prominent brand name with outlets oper-
ated by unknown local entrepreneurs. Although some franchisees can become quite successful with
multiple locations, they generally have little ability or desire to promote their brand name over the
franchisor’s. This was not the case with the major banks.
Developing a proprietary card system was another option for banks. As we mentioned,
Citigroup, in addition to owning Carte Blanche, had started its proprietary Everything card in
1967. Because Citigroup held a national banking charter and had customers across the country
who were potential payment cardholders, it initially hoped to develop the Everything card into a
national brand. Other banks found this option unattractive. While a national charter was, legally
speaking, not necessary to issue credit cards around the country, some banks were reluctant to
expand out-of-state.
Many banks found the answer in co-opetition. Banks competed for merchants and cardhold-
ers; banks cooperated at the card system level by setting operational standards. Despite the dismal
experience of the 1950s, many banks decided to start cards in the 1960s. They compared the
problems of going it alone to the benefits of cooperation. The calculus led them to form associa-
tions. Five banks in Illinois founded the Midwest Bank Card. By January 1967, nearly six hundred
banks in Illinois, Indiana, and Michigan had joined; some of these issued one of the five original
members’ cards. There were also two Michigan associations. Other banks across the country fol-
lowed. Three New York City banks started the Eastern States Bankcard Association in June 1967.
The state and local banking groups began to develop ties with other groups. The Interbank Card
Association started in 1966. Early on it included banks in Buffalo, Pittsburgh, Milwaukee, Seattle,
and Phoenix. At the same time, several banks in California started the Western States Bankcard
Association, issuing cards under the Master Charge service mark. By 1967, the California banks
issuing Master Charge had joined the Interbank Association. By February 1968, Interbank had
286 banks in at least seven states.
It became apparent during 1968 that two competing national networks of banks had emerged:
the BankAmericard franchise system, and the Interbank cooperative system. Banks—and groups
of banks—started aligning with one or the other. “Just about every bank in the card field,” said
Business Week, “is convinced that it must join one or the other network.” Bankers Trust in New
York City went with BankAmericard and franchised the card to other banks in the New York
area. Meanwhile, Citigroup converted its Everything card to Master Charge and joined Interbank.
Chemical and Manufacturers Hanover joined that association as well. For the most part, the larger
banks had chosen Interbank over BankAmericard. In contrast to the BankAmericard franchise
model, Interbank charged only a “modest” entrance fee and a small annual fee to cover the operat-
ing costs of the joint enterprise. And as noted, banks would be selling a brand they jointly owned,
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses More than Money 291
rather than that of another bank. This was an important point for banks that harbored hopes of
future national expansion when interstate banking restrictions were lifted—though in hindsight,
that was still more than three decades away.
BankAmericard was not doing too badly by many measures. Under its franchise model, it had
about 27 million cardholders and about 565,000 merchants by 1970, a sizable jump from 1.8 mil-
lion cardholders and 61,000 merchants in 1966. But it was in the process of being overtaken by
Interbank, which had attracted most major banks. And the franchisees were restless.
The franchisees quickly went from restless to rebellious. They had grown in importance to the
system and wanted a voice in its future. In 1970, faced with this revolt as well as with operational
problems, Bank of America agreed to convert the system into a membership-owned corporation:
National BankAmericard, Inc. (NBI). NBI wasn’t an ordinary stock corporation; instead, its mem-
bers had voting rights that couldn’t be bought or sold. Initially, NBI had 243 charter members,
including Bank of America, Bankers Trust, and First Chicago as well as numerous smaller com-
munity and regional banks.
The brief period 1966–1970 turned out to be critical for the future development of the pay-
ment card industry. Three alternative business models battled against each other. The go-it-alone
model had been the one used by American Express, Diners Club, and Carte Blanche for charge
cards. Bank of America adopted this model in California, Citigroup tried it with its Everything
credit card, and American Express tried it with its Uni-Card credit card. Bank of America and the
three T&E card companies tried the franchise model. Finally, the co-opetitive model was used
by Interbank and many other associations of banks across the country. The web of interstate and
branch banking restrictions played a crucial role in the battle among these models.
By the end of the 1960s, the franchise model was dead in the United States. And the go-it-
alone companies and the co-opetitives had gone in different directions. Sticking to what they knew
best, the go-it-alones decided not to issue credit cards. American Express had ventured into credit
with its Uni-Card, but decided to get out and did not try again for twenty years. The co-opetitives,
on the other hand, focused on issuing cards with a revolving line of credit. (Debit cards were soon
added, but took until the mid 1990s to have an impact.)
The basic idea behind co-opetition was clear as early as the Midwest Bank Card. The five
founding members competed with each other for cardholders and merchants in the Chicago area.
They cooperated in two related respects. They agreed to make their systems “interoperable.” A
First National Bank of Chicago cardholder could use her card at every merchant who had signed
up with any of the five banks. A Harris Trust Company merchant could accept as payment any
card from these banks. What the banks lost in helping their competitors, they more than gained
in making their own card more appealing to cardholders and merchants. Interoperability forced
cooperation in another way. When a First National Bank of Chicago cardholder bought something
at a merchant who was affiliated with Harris Trust Company, Harris Trust had to get reimbursed
by First National. These banks faced the same problem as in the BankAmericard franchise: system
PLATFORM ECONOMICS:
292 Part Four : Payment Cards Essays on Multi-Sided Businesses
cooperation was essential to process the slips of merchant receipts that were growing exponentially
with the number of participants in the system.
The two national associations encountered similar issues to Midwest Bank Card, only on a
grander scale. MasterCard—which started out as the Interbank Card Association in 1966, changed
its acceptance brand to Master Charge in 1969, and finally became known as MasterCard in
1979—and Visa—which began as BankAmericard in 1958, switched to National BankAmericard,
Inc. in 1970, and settled on Visa in 1976—took cooperation further than any of the regional as-
sociations. First, they established rules and processes for settling transactions. Part of this involved
how the merchant and the cardholder banks divvied up the transaction proceeds. Some of the
regional cooperatives had initially exchanged at par so that the cardholder’s bank reimbursed the
merchant’s bank for the entire transaction and the cardholder’s bank didn’t get any of the merchant
fees. (The banks may have simply been applying the check model of par exchange, which they
soon found unsatisfactory for cards.) MasterCard and Visa both settled on an interchange fee—a
percentage of each transaction that the merchant’s bank gave to the cardholder’s bank.
Another area of cooperation was on the card brand. The banks that belonged to Interbank
decided early on to use the Master Charge brand. That—and not the individual bank’s name—was
most prominent on the cards from the late 1960s and early 1970s. Visa replaced BankAmericard
as the brand name for that system in 1976. The co-opetitive felt that “standardizing the somewhat
confusing array of blue-white-and-gold cards issued under different names in twenty-two countries
around the world” would lead to greater acceptance of the card. Focusing on the system’s brand
involved a trade-off basic to the co-opetitive model: choosing between doing things at the system
versus the member level. Overall, the co-opetitives chose to do most things at the member level.
In a few pages we will tell the story of how the bank associations died. Yet they endured for
about four decades and were responsible for the remarkable growth of the payment card industry
and for shaping the industry we know today.
tion, volatile interest rates, and economic stagnation. The economy went through several severe
recessions in the 1970s and early 1980s (see Table 1). The economy also experienced accelerating
inflation, jumping from 4.2 percent in 1972 to a peak of 9.4 percent in 1981. This led to an in-
crease in (nominal) interest rates. (Economists distinguish between nominal interest rates, which are
quoted by lenders, and real interest rates, which are lower than nominal rates by the rate of inflation.
Real rates adjust for the impact of inflation on purchasing power over time; nominal rates tend to
vary with the rate of inflation, all else being equal.) The one-year Treasury bill rate climbed from
4.9 percent in 1971 to 14.8 percent in 1981. As interest rates climbed, state usury laws made credit
card lending a bad business proposition. Banks need a spread between the finance rate they charge
consumers and their own cost of funds to cover their operational costs, including fraud and defaults,
and make a profit. As inflation rose, that spread generally narrowed and even became negative in
states with low caps on loan rates. Banks lost money on the credit they had already made available
on cards and became unwilling to extend further credit in the face of these losses. Usury laws, and
a Supreme Court decision about them, helped shape the card industry in the 1970s and beyond.
raised their credit criteria for issuing cards and imposed higher membership fees. For example,
because Arkansas imposed a low cap on consumer loan rates, banks in that state had to set their
standards relatively high and offer credit to very few people in order to hold down costs. As a result,
charge-off rates in Arkansas were low, but so was the use of credit cards. (Charge-offs are credit card
balances that have been written off as losses for tax purposes.)
Usury laws had a significant effect on the development of a national card industry because they
limited the ability of banks to market their cards on a national or even regional basis. Interest rates
that were lawful in one state were unlawful in another. A bank therefore couldn’t market a card na-
tionally or regionally and capture scale economies from wide distribution. The few banks that issued
in multiple states had to incur the expense of administering multiple card programs with different
terms in each state. This wasn’t just an issue of processing extra paper; the banks had to adjust credit
standards and collection criteria according to the finance charges they could assess in each state.
A Supreme Court decision changed the rules of the game in 1978 and helped create na-
tional competition for payment cards. First of Omaha Service Corporation, a subsidiary of First
National Bank of Omaha, began to apply interest rates that were legal in Nebraska, but higher than
the Minnesota rate ceiling, to its Minnesota credit card customers. Marquette National Bank of
Minnesota challenged this practice. The Supreme Court sided with First of Omaha. In Marquette
National Bank v. First of Omaha Service Corp., the Court ruled that as a national bank, First
National Bank of Omaha “may charge interest on any loan at the rate allowed by the laws of the
State where the bank is located.” The Court also said that a bank’s “location” refers to the state in
which the bank is chartered, regardless of the states in which it solicits customers.
The Marquette decision led to three major developments. First, nationally chartered banks start-
ed issuing credit cards from states with less-restrictive usury laws. Citigroup, for example, moved
its credit card operations from New York, which at the time had an interest rate cap of 12 percent
on balances greater than about $1000, to South Dakota, which had raised its interest rate ceiling to
19.8 percent. Many other banks moved their operations to Delaware, which eliminated interest rate
caps in early 1981. By 1987, many large banks legally resided in Delaware, including Bankers Trust,
Chase Manhattan, Chemical, Manufacturers Hanover, Morgan, and Marine Midland.
Second, in an attempt to attract or retain such movable card operations, some states began to
modify their usury laws. In 1980, the same year that Citigroup announced it would be moving its
operations to South Dakota, the New York State Senate passed a bill that eliminated interest rate
ceilings on most types of loans, except annual rates on credit cards, which remained at 25 percent.
By 1988, the majority of states still had some form of an interest rate cap on credit card loans, but
many had raised their ceilings.
Third, less balkanization from state credit restraints set the stage for marketing payment cards
on a nationwide basis. Citigroup led the way in the late 1970s and early 1980s: it expanded nation-
wide through acquisitions and mass-mail credit card solicitations. Throughout the 1980s, many
other banks launched national credit card campaigns as well, including Bank of America, Chase,
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses More than Money 295
Continental Illinois, First Chicago, and Manufacturers Hanover. By permitting nationwide com-
petition, Marquette enabled issuers to realize scale economies in marketing and processing costs,
thereby making payment cards more readily available to consumers across the country. Thus, even
though Arkansas still caps interest rates at 5 percent above the Federal Reserve discount rate today,
consumers in Arkansas have a wide range of card choices from national issuers.
B. Duality
Federal antitrust laws also had a major effect on the evolution of the payment card industry in
the 1970s. Visa’s rules initially prohibited MasterCard-issuing banks from issuing its cards or han-
dling its merchant paper. In July 1971, one of Visa’s charter members, Worthen Bank and Trust
Company of Little Rock, Arkansas, filed an antitrust suit, claiming that Visa’s prohibition amount-
ed to an illegal group boycott. While this case was eventually settled out of court, Visa remained
exposed to similar lawsuits. In 1974, it asked the Antitrust Division of the U.S. Department of
Justice for a business clearance review—in effect, a letter of approval for a rule that would prohibit
dual membership by card-issuing and merchant-acquiring banks. After a year of consideration, the
division declined to grant clearance, citing insufficient information. Without the division’s sup-
port, Visa removed all restrictions on dual membership in mid-1976. The age of what has become
known as “duality” began, and members of each system rushed to join the other.
Competition among issuers increased sharply as new members scrambled to sign up consumers
for their second card. From mid-1976 to mid-1977, the number of cardholders increased by 11.7
percent for MasterCard and 13.1 percent for Visa—significantly higher growth than in preceding
years. Banks that were previously exclusive to MasterCard could now sign up their merchants for
Visa (and vice versa) as well as compete generally for merchants. Merchant acceptance grew sharply
from mid-1976 to mid-1977, by 24.3 percent for MasterCard and 17.3 percent for Visa, again
much higher than in preceding years.
One year after the restriction was removed, twenty of the nation’s twenty-two-largest banks
that issued cards had become dual. With increased system overlap, Visa and MasterCard twice dis-
cussed the possibility of merging system infrastructures in the 1980s, but decided against the move
both times. With duality, the difference between Visa and MasterCard began to blur, at least as
far as consumers were concerned. Some dual members offered common agreements, billing state-
ments, and credit lines. Some used common advertisements for their dual cards.
Over time, almost all member institutions joined both associations, and the increasing overlap
in membership led to some decline in competition between the two systems. Dual membership
lost some of its appeal in the late 1990s. The Justice Department, ironically, claimed in a case filed
in 1998 that MasterCard and Visa were violating the antitrust laws by having dual membership.
A federal district court judge disagreed. By then, though, meaningful duality had withered, as the
associations sough allegiance, and in the next decade the associations were disbanded.
PLATFORM ECONOMICS:
296 Part Four : Payment Cards Essays on Multi-Sided Businesses
a basic credit check, and some didn’t even do that much. Moreover, they could not provide secure
delivery of cards. Chicago postal clerks were caught hoarding unmailed cards to sell on the black
market; small-time criminals stole cards out of mailboxes; even the Mafia got involved, trafficking
in stolen cards and working with dishonest merchants to submit false sales slips. As a result of all
this, the levels of fraud and defaults in the early 1970s were quite high. (In April 1970, the Federal
Trade Commission banned the mass mailing of credit cards.) The economics of two-sided markets
tells us that we should not be too quick to claim that these efforts were bad business, however.
Many platforms invest in getting one or both sides on board; it is unclear whether banks could
have ignited their platforms in some other way without incurring comparable losses.
Added to the turmoil of fraud and consumer delinquency were the thorny issues of how to
coordinate the various banks in the system. Who was responsible for fraud and unpaid cardholder
bills, and how would the various parties settle with one another? Recall that in the early 1970s, the
newly formed bank associations did not have computer systems to smooth transactions; everything
was done with paper and postage.
Although bankcard issuers had started to leave the red ink behind by 1972, they still found it
difficult to earn the same returns from credit card lending as they did in their other lines of lending.
From 1974 (the earliest year with available data) through 1980, the rate of return on assets (net
before-tax earnings divided by assets) on credit card lending ranged from 1.61 percent (in 1980) to
3.09 percent (in 1977), for an average of 1.53 percent. In contrast, banks earned far higher average
rates of return on other forms of lending over that same period: 2.26 percent on installment loans,
2.48 percent on real estate loans, and 3.04 percent on commercial loans. The discrepancies are
especially surprising given that credit card loans are not secured with physical assets, whereas the
other loan types are. Credit card loans were thus riskier to make, but earned banks a lower return
throughout the 1970s.
to have saved members over $110 million in fraud prevention in its first year of operation. Visa
then set about building BASE-II, which computerized the entire transaction process and solved
the other major member headache: the physical interchange of paper among members. BASE-II
went online in 1974. MasterCard made similar investments to move its systems off paper and onto
computers, with its own BankNet and I-Net systems.
The payment card industry was poised for growth. Interest rate ceilings were no longer as seri-
ous a problem. Banks’ cost of funds fell dramatically after the 1980–1982 recession, and usury
laws were less prevalent, less severe, and after the Marquette decision, less important. Many banks
had mature card programs with customers whose credit behavior they had observed for some time.
The 1970 Fair Credit Reporting Act helped to ensure that creditors had access to accurate credit
reports, increasingly at the national level. And the card industry had learned from experience how
to reduce losses from cardholders who didn’t pay their bills.
Credit cards were well-understood products that could make money for banks in several ways.
Banks that issued cards charged membership fees. (Many banks had instituted these fees in response
to a 1980 Federal Reserve requirement that they hold reserves against outstanding credit card debt.)
They also earned income from merchants: they received a percentage of each transaction through
the interchange fee, which was around 1.6 percent in the early 1980s. And they received finance
charges from cardholders who revolved their balances. Banks that acquired transactions from mer-
chants earned revenue from fees that were usually based on percentages of transactions volume.
The payment card industry grew dramatically over the remainder of the 1980s, during one
of the longest peacetime economic expansions in U.S. history. Between 1982 and 1990, overall
consumer spending rose from $4.0 trillion (about $48,000 per household) to $5.6 trillion (almost
$60,000 per household). Retail spending increased from $2.1 trillion (about $25,000 per house-
hold) in 1982 to $2.7 trillion (around $29,000 per household) in 1990. Restaurant spending al-
most doubled from $140 billion ($1,700 per household) to $262 billion ($2,800 per household).
These increases naturally provided more opportunities to use payment cards. Dollar transactions
on credit and charge cards increased from $181 billion in 1982 to $467 billion in 1990. Part of
this increase reflected growth in spending on cards by people who had cards. The average monthly
charge on cards increased from $313 per household in 1977 to $461 in 1992. (Data are not avail-
able from the early 1980s.) Another part reflected growth in the number of households with cards.
The percentage of households with at least one card increased from 43 percent in 1983 to 62 per-
cent in 1992. (More people started carrying multiple cards so the average amount charged per card
did not increase as much as the average amount charged per household.)
People not only charged more. They borrowed more. Between 1982 and 1990, the total
amount of consumer credit outstanding rose from $747 billion to $1,186 billion, and the average
consumer credit outstanding per household increased from almost $9,000 to just under $13,000.
Thus, the total consumer credit outstanding grew at an average annual rate of 6 percent. Credit
card debt grew even faster: outstanding loans on credit cards grew at an average annual rate of 21
percent, from about $58 billion to $263 billion, increasing from about $680 to slightly more than
$2,800 per household in 1982 and 1990, respectively
Not surprisingly, the increased demand for payment and credit services, combined with a
more favorable financial environment for payment card issuers, generated an enormous increase
in the supply of payment card services. Existing issuers from Citigroup to American Express is-
PLATFORM ECONOMICS:
300 Part Four : Payment Cards Essays on Multi-Sided Businesses
sued more cards. More important, banks that wanted to get into the card business could issue
the established MasterCard and Visa brands, and consumers could use the cards they issued
instantly at the millions of merchants who already took these two brands. Between 1981 and
1991, about 4,200 financial institutions became issuing members of the Visa association. (The
pool of possible entrants expanded as a result of 1982 legislation that made more institutions
eligible for federal deposit insurance.) Several of these entrants are particularly noteworthy, and
we will discuss them shortly.
On the other side of card transactions, acquirers signed up more merchants during the 1980s—
both new merchants who were part of the wave of new business formation and existing merchants
who hadn’t taken cards before. From 1982 to 1990, the number of U.S. merchants accepting
Visa increased by 37 percent. By 1991, more than 2.5 million merchants accepted Visa cards.
MasterCard experienced a similar growth in merchant acceptance.
The increased supply of card services for customers and merchants came from three other
notable sources. Sears, Roebuck and Co. started the Discover Card in 1985 (and went national in
1986). Less than two years after the signature orange-and-black Discover Card was released, there
were twenty-two million cards in circulation—more than Citigroup had accomplished after two
decades—and $4.7 billion in receivables, ranking it third among all credit and charge cards in the
United States. Over 700,000 merchants accepted the card by 1987.
Less than two years after the Discover Card launch and about twenty years after selling its cred-
it card (the Uni-Card), American Express launched the Optima credit card. Yet what American
Express thought would be a successful competitor to the bankcard issuers quickly turned into a
disaster. Default rates skyrocketed and losses soared, costing the company hundreds of millions of
dollars. This sophisticated company’s problems make clear the complexity of the credit card busi-
ness. American Express assumed that their charge card holders would be good credit risks—after
all, they paid their bills on time. What American Express failed to recognize, though, was that
many of these charge card holders were using their cards for company business for which they were
reimbursed. They were less conscientious when they had to pay the bills themselves. There were
enough defaults to sink the Optima card in red ink. By October 1991, during a recession that hit
white-collar workers particularly hard, about 8 percent of Optima’s receivables were charged off.
The credit arm of American Express had to take a $382 million charge against third-quarter earn-
ings in 1991, including $158 million to restructure the credit card operations and $223 million to
add to credit loss reserves. In addition, about 1,700 employees were laid off. Altogether, American
Express reported a 93 percent drop in its third-quarter income for 1991. By early 1992, Optima’s
credit losses topped out at 12 percent of receivables. American Express eventually turned its credit
cards into a success in the 1990s, as we will see, but it wasn’t easy.
Meanwhile, the pioneers of the payment card industry, Diners Club and Carte Blanche, had
become marginal players in the United States by the end of the 1980s. Together, they had a share
of less than 2.5 percent of payment card transaction volume.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses More than Money 301
In the 1980s, several significant entrants into the card business came in through MasterCard
and Visa. Three giant nonfinancial firms—AT&T, General Electric, and General Motors—decided
to get into the card business. MasterCard was much more enthusiastic about these nonbanks
becoming part of its association than Visa was, so many of these firms issued MasterCards at first.
These firms either bought banks that were members of MasterCard or Visa, or they entered into
contracts with banks that were association members. Either way, it was the industrial giants whose
names were prominently featured on the cards.
Many of the nonbanks’ cards were tied to the business of the firm that sponsored the card. “We
saw our relationships with twenty-two million calling–card customers in jeopardy,” stated Paul
Kahn, head of AT&T Universal Card Services Corp. Thus, AT&T linked its card to its primary
business; the payment card could function as a calling card. A General Motors card allowed card-
holders to accumulate points that could earn a rebate on a General Motors car. General Electric’s
card, however, had no such links. Instead, it offered general reward coupons that were redeemable
at various retailers. The nonbank firms gained many cardholders and became some of the largest
programs in the early 1990s. In fact, two of the top ten bankcard issuers in the early 1990s (by
charge volume from 1991 to 1994) were nonbanks: AT&T Universal and Household Bank (which
issued a General Motors card).
Some, but not all, of the nonbank card issuers were able to sustain this early growth. Still,
the General Motors card is issued by Household (owned by HSBC) and enables cardholders to
accumulate points toward a General Motors car. In 2009, General Electric Capital Financial was
the twelfth-largest issuer, by charge volume, of general purpose cards; it issued cards for all four
networks. AT&T, in contrast, sold its card portfolio to Citigroup in 1998, though the AT&T
Universal card continues to function as an AT&T calling card.
Two other kinds of cards were introduced in the 1980s that were notable marketing successes.
The first was proposed by one of the bank associations. In 1978, Visa introduced affinity programs
that for the first time, allowed a nonmember’s name or logo to be displayed on the face of the card.
In 1980, however, Visa banned new affinity card programs, arguing that they tended to dilute the
Visa brand, but it allowed the existing programs to continue. Five years later, faced with apparent
payment card saturation, both Visa and MasterCard began allowing new affinity programs. After
only one year, 296 clubs, charities, professional associations, and other nonfinancial organizations
had developed Visa and MasterCard affinity programs. By 1989, there were over 2,000 affinity
card programs in the United States, ranging from the Sierra Club to the National Football League
to the UCLA Alumni Association. Affinity cards involving for-profit companies are also known as
“cobranded” cards.
The other new type of card was introduced by Citigroup. Citigroup and American Airlines
announced the AAdvantage bankcard in April 1987, marketing it to about six million members of
American’s AAdvantage frequent-flier program. At the time, AAdvantage was the largest-frequent
flier program, and Citigroup was the largest credit card issuer in the United States with fifteen mil-
PLATFORM ECONOMICS:
302 Part Four : Payment Cards Essays on Multi-Sided Businesses
lion cards issued. The card had a $58 annual fee and credit lines of up to around $58,000, which
led some industry expert to assume that Citigroup was “competing for the same kinds of customers
as American Express Co.’s Green and Gold Cards.” By early 1993, about 1.5 million AAdvantage
credit cards were in circulation.
These three payment card innovations in the 1980s—the nonbank issuers, affinity and co-
branded cards, and frequent-flier rewards cards—each contributed to a dramatic rise in consumer
card use over the decade. By 1991, there were almost twice as many cards per household compared
to five years earlier.
in issuing these kinds of cards. As we describe in more detail in chapter 8, Visa embarked on a
campaign to convince banks that they could make money from debit cards and to convince card-
holders to use these cards. Visa made a commitment to promote a new debit card brand—Visa
Check—through extensive national advertising. Around the same time, Visa staff tried to convince
banks that the interchange fee revenues they would receive from transactions on the cards they is-
sued would make these cards profitable. (Visa’s debit card interchange fees were set slightly lower
than its credit card interchange fees at the time.) MasterCard followed with a similar product and
strategy after initially pursuing an unsuccessful PIN debit strategy.
The bank that operated the EFT networks faced a different business problem: they had got-
ten many cards into the hands of households, but didn’t have merchants who were willing to take
those cards. During the 1980s, banks had installed ATMs and issued cards to their customers that
allowed them to take out cash and conduct other banking transactions. Associations of banks
formed that enabled customers of one bank to use their cards at the ATMs of other banks. The
EFT networks didn’t have a system for authorizing card transactions by signature. But they could
authorize and settle transactions if merchants had PIN pads that were connected to the ATM
switches. And most of the EFT systems required banks to allow ATM cards to be used for retail
transactions. To get merchants on board, the EFT networks had to convince merchants to install
PIN pads. They did this by setting an interchange fee that was much lower than that charged by
the card associations. This resulted in merchants with PIN pads paying a much lower merchant
discount for PIN debit transactions than for signature debit transactions.
Most banks chose to put the marks of either MasterCard or Visa (for signature debit) and
the EFT networks (for PIN debit) on the same card. This resulted in synergies between signature
and PIN debit. The existing base of ATM cards made it easier for banks to issue Visa Check
cards, while the Visa Check promotions helped persuade people to carry and use these cards—
in both modes—for paying for things. Despite these synergies the card associations and the
EFT networks waged war to capture consumer transactions. That and the competition between
MasterCard and Visa for banks to issue their debit cards is one of the subjects of a later chapter
on the system wars.
The efforts to increase debit card use, by both Visa and the EFT networks, were successful.
By 2002, debit cards accounted for 29 percent of both all payment cards and all payment card
volume. Of the $572 billion worth of transactions on debit cards in 2002, about two thirds were
signature debit and one third was debit. Debit cards became an important part of the package
that banks provided to consumers. In addition to providing interchange fee revenues, these cards
helped banks acquire checking account customers who then typically used the bank for various
services ranging from brokerage accounts to personal loans to mortgages.
While the rapid growth of debit cards was by far the most notable change in the payment
card industry in the 1990s, four other developments deserve mention. The first was a financial
innovation known as “securitization,” which enabled credit card lenders to sell credit card debt to
PLATFORM ECONOMICS:
304 Part Four : Payment Cards Essays on Multi-Sided Businesses
other institutions that could consolidate many different kinds of debt from many different l enders.
Credit card issuers could use securitization income to expand their businesses. Securitization also
allowed assets to be moved from the card issuers’ balance sheets, thus lowering the capital reserves
they were required to hold. In addition, credit card issuers could reduce their risk from cardhold-
ers defaulting on payments—a particular worry in the event of an economic downturn. Without
securitization, an individual issuer, especially those that focused on credit card lending, had dif-
ficulty diversifying away their risk exposure. Securitization allowed lenders to diversify their risk
and thus to extend credit deeper into the pool of relatively risky consumers (that is, consumers
with poor credit histories). Industry experts estimate that $61 billion in retail and bank credit card
loans, roughly 10 percent of outstanding balances on store and bank credit cards, were securitized
in 2001. (This development came later than, but is similar to, the securitization of mortgages.
Most mortgage lenders resell their loans to companies that consolidate loans from many lenders
and then diversify the risk through complex financial arrangements.) Securitization also made it
easier for companies to both enter credit card lending (because they could diversify their risk) and
get out of the business (because they had a market for the loans they had made).
The second key development occurred in the banking industry. We saw in the previous chap-
ter that legislative changes resulted in a massive consolidation of banks. This had effects on the
payment card industry. Consider the fifty-largest banks ranked by total assets in 1990. Through
mergers and acquisitions, these fifty banks were consolidated into eighteen banks by 2003. For
example, Chase Manhattan took over Chemical Bank in 1996. Bank One acquired First USA
and First Chicago in 1997 and 1998, respectively, and Bank of America—one of the largest
bank credit card issuers—merged with NationsBank in 1998. This automatically led to con-
solidation among payment card issuers. Moreover, there was substantial consolidation within
the card industry through portfolio sales. AT&T Universal sold off its portfolio to Citigroup
in 1998, for instance, and MBNA purchased the SunTrust Bank portfolio in 1999. Additional
shifts have occurred as some issuers have substantially increased their share of sales volume. For
example, MBNA more than doubled its share of Visa and MasterCard credit card volume from
4.5 percent in 1990 to 10.8 percent in 1999. The payment card industry became more concen-
trated—the top ten bankcard issuers accounted for 74 percent of Visa and MasterCard volume
in 1999 versus 44 percent in 1990.
A third development concerns the co-opetitives. During the 1990s, each co-opetitive began
aggressively encouraging its members to dedicate themselves to one association at the expense of
the other association. By the end of the 1990s, both MasterCard and Visa began to offer formal
partnership programs, providing benefits to banks that agreed to focus their card business on just
one system.
Finally, one can’t talk about the 1990s without mentioning about the emergence of the com-
mercial world-wide web in the middle of the decade. Cards were the perfect payment instrument
for online transactions and quickly became the main way to pay on the web.
Golden Anniversary
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses More than Money 305
30%
25%
20%
15%
10%
5%
0%
70
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
Note: General-purpose payment cards include credit, charge, PIN debit, and signature debit cards.
Sources: Bureau of Economic Analysis; and The Nilson Report (various 1970–2003 issues).
1
Market caps as of July 17, 2009.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses More than Money 307
Citigroup did just that. Several banks including Citi wanted Visa to give the bank brand
more prominence on the card. Up to that point in time MasterCard and Visa cards highlighted
the brands of the association. That made sense during the years in which all banks benefit from
the expansion of these card brands in the United States and, indeed, around the world. But as the
banks became bigger and more ambitious rivals, and the MasterCard and Visa brand names were
engraved in the minds of the public, some of the banks wanted the ability to give themselves a
larger piece of the real-estate on the plastic card. Despite Citi’s threat to bolt Visa refused to relent
to Citi’s demand over this or other issues. This national issuer, made good on its threat and started
that year to move its U.S. card volume over to MasterCard. That was a financial jolt to Visa and to
its members who had to bear a greater part of the cost of the system.
Ironically, it is easier to manage a club with many small members than one which a handful
of members who are trying to seek advantages for themselves. These “clubs” had that worked well
for many decades became a victim of their own success as some their members became very large.
The associations had been flypaper for litigation almost from their beginnings but the risks
had become both more palpable and larger in the early days of the new century. Even without
the management challenges MasterCard and Visa had little choice but to kill the associations and
reincarnate themselves as public equity corporations.
The antitrust laws of the United States and many other countries are rightfully suspicious when
firms that compete with each other get together and agree to do things. The most severe antirust
sanctions are imposed on “cartels” in which competitors get together to fix prices or divide mar-
kets. Yet the U.S. Supreme Court has recognized that under certain circumstances cooperation
among competitors may generate value for consumers. The European Union’s antitrust laws also
allow firms to cooperate when they can show, in effect, that this is essential for providing benefits
to consumers. Over the years the courts and regulatory authorities have struggled with deciding
whether these associations of banks rivals are restricting competition or engaging in necessary co-
operation that promotes economic efficiency. But any action that MasterCard and Visa take can be
subject to a lawsuit, or a complaint to an antitrust authority, that the banks are violating antitrust
laws that prohibit competitors from acting together. The burden then falls to the associations to
prove that they should get a pass.
The greatest controversy over the years has surrounded the interchange fee. MasterCard
and Visa set the fees that members that acquire merchants have to pay members the acquire
cardholders when a cardholder uses her card to buy something at a merchant. Antitrust and
bank regulators around the world started looking at interchange fees in the late 1990s. Then,
in 2005, a group of retailers in the United States, including many large ones, filed class-action
action cases against MasterCard and Visa accusing them of having engaged in price fixing and
other violations of the antitrust laws. It became apparent early in the decade that the associa-
tions were going to be embroiled in litigation and regulatory proceedings over the setting of
interchange fees for years, that interchange fees in various part of the world could be prohibited
PLATFORM ECONOMICS:
308 Part Four : Payment Cards Essays on Multi-Sided Businesses
or regulated, and that in the United States at least associations (and their members) could face
massive damage awards. MasterCard and Visa agreed to pay about $3.3 billion in 2003 (in
2008 dollars) to settle another merchant class action. The merchants had claimed that the card
networks have restricted competition by “tying” credit and debit cards—merchants had to take
both to get either.
The legal issue at the heart of these lawsuits results almost entirely from the association
model that MasterCard and Visa adopted. The interchange fee largely determines the fee that
merchants pay because it is passed on from the acquiring bank to the merchant. Go-it-alone
systems such as American Express can set the merchant fee for their cards. They are just setting
the price they are charging customers. They can also set the portion of the fee that they share
with banks that agree to issue American Express cards. MasterCard and Visa were vulnerable to
lawsuits largely because members that competed with each other were acting through the asso-
ciation to set the interchange fee.
The banks that owned MasterCard and Visa decided to end their associations. MasterCard
went first in 2006 and Visa next in 2008. In both cases these card networks reorganized themselves
as public equity corporations and did initial public offerings in which a majority of the stock was
sold off. Banks retained some stock, continued to have representation on the boards, but removed
themselves from any decisions concerning interchange fees. These IPOs allowed the banks to cash-
in on their participation in the card associations. By allowing these organizations to raise capital
in the public markets, and eliminating the cumbersome association management, the banks likely
increased the value of MasterCard and Visa and thus their own stakes. And finally, by removing the
banks from a decision making role over interchange fees the banks arguably reduced their exposure
to litigation involving price fixing claims. (The plaintiffs in the US class-action litigation and the
European antitrust regulators, however, continue to claim that the banks are colluding unlawfully
in the setting of interchange fees.)
Another significant change in the role of card networks occurred in the 2000s as a result of a
lawsuit that had been filed by the U.S. Department of Justice. The courts ruled that it was unlawful
for MasterCard and Visa to prohibit member banks from also issuing the cards of competing asso-
ciations. (American Express had already been issuing cards through MasterCard and Visa member
banks outside the United States where the association gave in to the threat of litigation.) American
Express and Discover both have “network” divisions which enter into agreements with banks to
issue their cards. For the United States, these usually go-it-alone systems have not reported how
much of their card transactions occur on cards issued by bank partners although there is no evi-
dence that it is significant as of 2010. (Outside the US American Express reports that 68 percent
of its transactions occur on cards issued by third-parties on its network.)2
2
American Express 2009 10k.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses More than Money 309
3
According to data from the Federal Reserve’s 2007 Electronic Payments Study, signature and pin debit cards ac-
counted for 57% of all general purpose card transactions in 2006 and 35 percent of all general purpose card volume
(this includes excludes store cards and prepaid cards). About two-thirds of both transactions and volume go through
MasterCard or Visa with a signature and one third through one of the PIN-debit networks.]
4
Robin Sidel of the Wall Street Journal
PLATFORM ECONOMICS:
310 Part Four : Payment Cards Essays on Multi-Sided Businesses
through these cards. The employer would use the ACH system to deposit wages into an account
at the prepaid card issuer. The employee could then use their card to either pay for things at mer-
chants that accept the card brand or to take money out of ATMs. Payroll cards are particularly
helpful for employees that don’t have bank accounts.
Prepaid cards were also appealing for the unbanked more generally. Aside from convenience,
by the 2000s it had become difficult for a consumer to carry on the ordinary business of life with-
out a plastic card. They were essential for buying things online, a few places had stopped accepting
cash, and some transactions required a card for a security deposit. The unbanked overlapped sig-
nificantly with another group of people who bound prepaid cards useful: immigrants who wanted
to transfer funds to family in friends in other countries. They could do this through money transfer
services such as Western Union but these services, which rely on labor-intensive networks of of-
fices throughout the world, were expensive. Prepaid card issuers introduced cheaper card-based
alternatives.
Prepaid card solutions were also developed for governments and non-profits. Many states use
prepaid cards to distribute benefits such as food stamps. Visa alone ran 70 state prepaid pro-
grams in 38 states in 2009. They load child support, workers comp and unemployment insurance
payments on these cards. The American Red Cross provided massive assistance to the Hurricane
Katrina victims on prepaid debit cards from MasterCard and JPMorgan Chase and is using these
cards as part of its general relief efforts.
Although there is a consensus across many data sources that prepaid cards have grown over the
2000s there are no reliable estimates of the amount of spending that takes place on these cards.
(That is because the main data collection schemes cannot readily distinguish between prepaid
and debit cards.) A detailed study sponsored by the Board of Governors of the Federal Reserve
estimated that, in 2006, consumers charged $58.3 billion on retailer and general purpose prepaid
cards.5 Most of the prepaid spending, 73.3 percent was on retailer cards. Prepaid cards accounted
for a small portion of overall payment card spending mid decade. Retailer and general purpose
prepaid cards amounted to 1.6 percent of total spending on all retailer and general purpose pay-
ment cards in 2006. General purpose prepaid cards accounted for only 0.5 percent of spending on
general purpose payment cards.
5
Federal Reserve Board, Electronic Payments Study, 2007.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses More than Money 311
Consumers were also irritated by banks increasing interest rates on their cards without notifying
them and applying these higher rates to existing balances. The press highlighted extreme cases of
bad behavior by issuers and consumer advocates agitated for relief.
Meanwhile, merchants, the other half of the two-sided payments business, became increasingly
vociferous about the fees they had to pay when consumers paid with plastic. Their bills had increased
over the decade in part because interchange fees had increased and because consumers were paying
more with cards (in particular with debit cards). In addition to pursuing an antitrust lawsuit they
lobbied Washington for relief.
We will never know whether Congress would have adopted legislation to deal with either of
these issues if the financial crisis hadn’t hit in 2008. The banking industry was blamed for having
caused the Great Recession and forcing taxpayers to fund massive bailouts to stem financial col-
lapse. Calls for financial reform soon followed. Although the crisis was largely caused by the burst
of a housing bubble the payment card industry was swept along in efforts to devise stricter regula-
tions for banks and curtail what some thought as excessive credit.
The Credit Card Accountability and Disclosure Act (CARD Act), signed by President Obama
in May 2009 after having won bipartisan support in Congress, regulated many aspects of how is-
suers could charge consumers. The Act prohibited some practices such as raising interest rates in
the first year after a consumer has opened a credit card account, charging for exceeding limits, and
setting early morning deadlines for payments. It also required issuers to give consumers at least
45 days notice of increase in interest rates and to mail bills at least 21 days in advance of the due
date of the payment. Credit card companies complained that the legislation limited their ability to
charge more to riskier borrowers. Although systematic data on the results of the legislation are not
yet available there were reports that banks raised card fees and interest rates to cover their lending
risks and reduced lending to riskier borrowers. They were able to respond that way because the
CARD Act did not limit their fees—only the timing of them.
Slightly more than a year later Congress, divided along party lines, passed the Wall Street
Reform and Consumer Protection Act which was signed by President Obama on July 21, 2010.
Two parts of this legislation, which was the major effort to reform financial regulation in light of
the crisis, affect the payment card industry.
The most direct is a section which calls for “reasonable fees and rules for payment card trans-
actions.” The Federal Reserve Board is supposed to make sure the debit card interchange fees are
reasonable and proportional to the costs incurred in making the transaction. That applies to signa-
ture and PIN debit as well as most general purpose prepaid cards. Since banks typically earn most
of their debit card revenue from merchants rather than cardholders these rules will likely reduce
interchange fees, especially for signature debit, substantially. The Federal Reserve is supposed to
have its regulations in place by the end of March 2011.
The legislation also creates the Consumer Financial Protection Board (CFPB), within the
Federal Reserve Board, which is responsible for regulating almost all consumer financial service
PLATFORM ECONOMICS:
312 Part Four : Payment Cards Essays on Multi-Sided Businesses
products including all payment cards. In part, the CFPB takes over existing responsibilities under
past legislation from various federal agencies. But it also has new powers to prevent financial ser-
vice companies from engaging in “abusive” practices. The legislation furthermore allows states to
adopt and enforce more stringent rules and laws than those adopted by the federal government. It
remains to be seen how this new agency will affect the card industry. It will take time for it to be
established and much of its powers are vested in its director who will be appointed by the President
and confirmed by Congress.
6
http://www.ctia.org/advocacy/research/index.cfm/AID/10323
7
http://www.pewinternet.org/~/media//Files/Reports/2010/PIP_Mobile_Access_2010.pdf
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses More than Money 313
Merchants also started seeing the possibility of using internet technology to accept payments
at the physical point of sale. They had relied on private networks and devices that came bundled
with software. The company Square introduced a new payment system that enabled merchants
to accept payments using internet-connected mobile devices such as the iPad or Android phones.
They provided an attachment to these devices through which merchants could swipe cards. The
transaction was then sent to the merchant processor over the internet.
For most of the previous century Americans used some form of “card”, whether metal, paper,
cardboard, or plastic, to identify themselves to merchants. The mag stripe plastic card was the pre-
dominant method for paying at merchants in 2010—despite a well-funded effort by the card net-
works to persuade merchants to install equipment to take contactless cards that consumers could
wave at terminal. Swipe transactions took just a few seconds to complete. Yet many were betting
that the mobile phone would topple the card. They just weren’t sure how or when.
As cards entered roughly their second century and the general purpose payment card turned
sixty, paper was still not vanquished. Almost half of consumer expenditures—even excluding
mortgages and rent—were paid with cash or check or some other paper method of payment. That
means that the electronic payments industry, and the general purpose card portion of it, continues
to have significant opportunities for growth.
CHAPTER
Interchange Fees and
12 Their Regulation
David S. Evans and Richard Schmalensee
ABSTRACT:
This chapter surveys the economic literature on interchange fees and the debate over whether
interchange should be regulated and, if so, how. We consider, first, the operation of unitary pay-
ment systems, like American Express, in the context of the recent economic literature on two-
sided markets, in which businesses cater to two interdependent groups of customers. The main
focus is on the determination of price structure.
We then discuss the basic economics of multi-party payment systems and the role of inter-
change in the operation of such systems under some standard, though unrealistic, simplifying
assumptions. The key point of this discussion is that the interchange fee is not an ordinary price;
its most direct effect is on price structure, not price level. We then examine the implications for
privately determined interchange fees of some of the relevant market imperfections that have
been discussed in the economic literature. While some studies suggest that privately determined
interchange fees are inefficiently high, others point to fees being inefficiently low. Moreover,
there is a consensus among economists that, as a matter of theory, it is not possible to arrive,
except by happenstance, at the socially optimal interchange fee through any regulatory system
that considers only costs. This distinguishes the market imperfections at issue here for multi-
party systems from the more familiar area of public utility regulation, where setting price equal
to marginal cost is theoretically ideal.
Next, we consider the issues facing policy makers. Since there is so much uncertainty about
the relation between privately and socially optimal interchange fees, the outcome of a policy
debate can depend critically on who bears the burden of proof under whatever set of institutions
and laws the deliberation takes place. There is no apparent basis in today’s economics - at a the-
oretical or empirical level - for concluding that it is generally possible to improve social welfare
by a noticeable reduction in privately set interchange fees. Thus, if antitrust or other regulators
had to show that such intervention would improve welfare, they could not do so. This, again, is
quite unlike public utility regulation or many areas of antitrust including, in particular, ordinary
cartels. By the same token, there is no basis in economics for concluding that the privately set
interchange fee is just right. Thus, if card associations had to bear the burden of proof -for ex-
ample, to obtain a comfort or clearance letter from authorities for engaging in presumptively
illegal coordinated behavior - it would be difficult for them to demonstrate that they set socially
optimal fees.
314
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 315
We take a pragmatic approach by suggesting two fact-based inquiries that we believe policy-
makers should undertake before intervening to affect interchange. First, policymakers should es-
tablish that there is a significant market failure that needs to be addressed. Second, policymakers
should establish that it is possible to correct a serious market imperfection, assuming one exists,
by whatever intervention they are considering (such as cost-based regulation of interchange fee
levels) and thereby to increase social welfare significantly after taking into account other distor-
tions that the intervention may create. We illustrate both of these points by examining the recent
Australian experience.
I. INTRODUCTION
In 1958 the Bank of America began operating the BankAmericard credit card system, the predeces-
sor of Visa, as a unitary system.1 It performed both the issuing function (dealing with cardholders)
and the acquiring function (dealing with merchants) itself. Similarly, it set the fees charged to both
these customer classes—the annual fee, interest rate, late fees, and other fees charged to cardhold-
ers; and the per-transaction fee to merchants known as the merchant discount. It was therefore
able to determine both the overall level of fees (which might be measured as total fees per dollar
of transactions) and their structure (which might be measured by the shares of total fees paid by
merchants and cardholders).
In 1966 the Bank of America began to bring other banks into the system as franchisees.
Individual banks within the system were free then, as now, to determine the fees they charged
merchants and cardholders. When a consumer holding a card issued by bank A made a purchase at
a merchant that had bank A as its acquirer, bank A could, if it wished, have the same fee structure
as Bank of America. But what if this same consumer made a purchase from a merchant acquired
by bank B? Bank of America required the acquiring bank to pass the full merchant discount to the
issuing bank. Acquiring banks had incentives to lie about their merchant discounts under this rule,
as issuing banks were well aware. More importantly, this rule meant that acquiring banks received
zero revenue for transactions for which they provided the merchant but had not issued the card
being used. The rule therefore blunted the incentives for all banks to sign up merchants, to the
obvious detriment of the system as a whole.
In 1970 the BankAmericard system was converted into a membership corporation, a multi-
party system. This cooperative association established an interchange fee in 1971 to deal with trans-
actions in which issuing and acquiring banks were different. This fee was paid by the acquiring
bank to the issuing bank and initially set at 1.95 percent. It was not linked to any individual bank’s
1
This discussion in this paragraph and the next two follows Evans and Schmalensee 2005, pp. 153-156.
PLATFORM ECONOMICS:
316 Part Four : Payment Cards Essays on Multi-Sided Businesses
merchant discount. The interchange fee thus became a revenue source on the issuing side of the
credit card business and a cost element on the acquiring side. Acquiring banks had to charge a
merchant discount that was greater than the interchange fee to recover this cost. The interchange
fee was an element of a standard contract that the multi-party system established for its members;
other terms of the contract defined who bore the risk of fraud or nonpayment, as well as how
disputes would be resolved.
This chapter surveys the economic literature on interchange fees and the debate over whether
interchange should be regulated and, if so, how.
Another justification for evaluating the [interchange fee] under the rule of reason is because
it is a potentially efficiency creating agreement among members of a joint enterprise. There
are two possible sources of revenue in the VISA system: the cardholders and the merchants.
As a practical matter, the card-issuing and merchant-signing members have a mutually de-
pendent relationship. If the revenue produced by the cardholders is insufficient to cover the
card-issuers’ costs, the service will be cut back or eliminated. The result would be a decline
in card use and a concomitant reduction in merchant-signing banks’ revenues. In short, the
cardholder cannot use his card unless the merchant accepts it and the merchant cannot
accept the card unless the cardholder uses one. Hence, the [interchange fee] accompanies
“the coordination of other productive or distributive efforts of the parties” that is “capable
of increasing the integration’s efficiency and no broader than required for that purpose.”3
In 1983, William Baxter, a leading antitrust scholar who had worked for Visa on the case,
published an important paper on the economic rationale for interchange fees in 1983.4 But after
the NaBanco decision, interchange fees faded from view in academic and policy circles and was a
topic of interest mainly to industry insiders.
A few academic papers in the 1990s mentioned interchange fees,5 but for the most part, this
topic languished in obscurity until around the turn of this century. And then, as the dates on
2
National Bancard Corporation v. Visa U.S.A., Inc. 1984, 1986a, and 1986b.
3
National Bancard Corp. v. Visa U.S.A., Inc. 1986a, p. 602.
4
Baxter 1983. We discuss this paper below.
5
These include Carlton and Frankel 1995a and 1995b, Evans and Schmalensee 1995, and Frankel 1998.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 317
most of the entries in this chapter's list of references indicate, interest in interchange fees in-
creased dramatically among academics, banking regulators and competition authorities around
the world.
Two developments caught the attention of policymakers. Cards had become an increasingly
important part of the payment system in many countries. The share of consumer expenditures in
the United States paid for with cards had increased from about 3 percent in 1986, the year of the
NaBanco decision, to 25 percent in 2000.6 Similar increases occurred in other countries. For exam-
ple, in 2000, cards accounted for 30 percent of consumer expenditures in Australia and 35 percent
in the United Kingdom.7 While American Express played a significant role in the United States,
globally most cards were associated with multi-party systems that had interchange fees. In the
United Kingdom, for example, an influential report on the banking industry issued in 2000—the
“Cruickshank Report”—addressed interchange fees in multi-party systems, concluding that “[t]
here is a strong case for reform of the interchange fee system.”8
For retailers, merchant discounts (which included interchange fees in the bank card systems)
had become a growing portion of their costs, as more people paid with cards. Some retailers had
periodically complained about merchant discounts; hotels had gone so far as to create their own
card system in the United States in the mid 1950s to avoid the merchant discount of Diners Club
(a unitary system). However, with increases in interchange fees and perhaps other legal and politi-
cal developments, various organizations of retailers around the world sought regulatory relief from
the fees. For example, EuroCommerce, a retailer association, filed a complaint with the European
Commission in 1997. This led to an investigation of Visa Europe and ultimately a settlement in
which Visa Europe agreed to lower the interchange fees. Interchange fee levels have also been under
active attack and/or regulated in a number of other countries, including Australia and the United
Kingdom.9
Why has interest in interchange fees surged among economists in recent years? Some, includ-
ing us, were exposed to the topic through their involvement in litigation and/or regulatory pro-
ceedings. But the same could be said for many other legal and regulatory issues that have spawned
much smaller literatures over longer periods of time. The large volume of theoretical literature
on interchange fees has arisen for the simplest of reasons: understanding their determination and
effect is intellectually challenging. As the discussion below indicates, this is not necessarily good
news for policy-makers.
6
HSN Consultants 1987, 1991, 1992, 2001a, and 2001b, and United States Department of Commerce2003.
7
Visa International 2001. Measures of consumer expenditures may not be entirely consistent across countries but are
sufficiently comparable to provide a rough sense of importance. See also Weiner and Wright 2005.
8
Cruickshank 2000, p. 272.
9
See Reserve Bank of Australia 2002, European Commission 2002 and 2003, and Office of Fair Trading 2003.
Weiner and Wright (2005) provide a useful overview of government involvement.
PLATFORM ECONOMICS:
318 Part Four : Payment Cards Essays on Multi-Sided Businesses
and costs of alternative payment systems as they vary among transactions; we suggest that there
is highly incomplete information available on these benefits and costs, so that any inference from
the current data is at best problematic. Second, policymakers should establish that it is possible
to correct a serious market imperfection, assuming one exists, by whatever intervention they are
considering (such as cost-based regulation of interchange fee levels) and thereby to increase social
welfare significantly after taking into account other distortions that the intervention may create.
We illustrate both of these points by examining the recent Australian experience.
The final section summarizes our conclusions. Many of the results of the economic literature
necessarily depend on various simplifying assumptions. In this last section we highlight those that
we believe are robust, in the sense that they are likely to hold generally. Our main focus throughout
is on conceptual issues; the companion paper by Weiner and Wright provides a good deal of useful
factual material.10
10
Weiner and Wright 2005.
11
Van Raalte and Webers 1998. While he used different terminology, Baxter 1983 did the first two-sided analysis, as
we discuss below.
12
See Ackerberg and Gowrisankaran 2003 for an attempt to quantify these externalities in a payments context.
PLATFORM ECONOMICS:
320 Part Four : Payment Cards Essays on Multi-Sided Businesses
numbers, it will fail. If it needs to serve men for free to accomplish this, it is rational for it to do so.
Giving the service away to men, even though serving them involves positive marginal costs, can be
profitable if it attracts sufficient fee-paying women.
The recent general literature on two-sided platform markets began around 2002 with early
versions of a seminal contribution by Jean-Charles Rochet and Jean Tirole.13 Early versions of an
important related paper by Mark Armstrong appeared shortly thereafter.14 These papers pointed
out that many businesses or markets can usefully be thought of as two-sided, and they developed
some general implications of the importance of balanced participation from the involved customer
groups (or, as it is more commonly put, “getting both sides on board”). Since then the literature on
two-sided platform markets has grown explosively.15 This literature generally considers one or more
vendors dealing directly with both (or all) involved customer groups and it is thus most directly
relevant in our context to the analysis of unitary payment systems.
13
Rochet and Tirole 2003b.
14
Armstrong 2004.
15
Contributions that look beyond payment systems include Armstrong 2005, Armstrong and Wright 2004, Bolt and
Tieman 2004a and 2004b, Caillaud and Jullien 2003, Gabszewicz and Wauthy 2004, Hagiu 2004a and 2004b,
Jullien 2004, Rochet and Tirole 2004a and 2004b, and Schiff 2003. Evans 2003a and 2003b and Wright 2004b
provide accessible discussions of the implications of some of this work as well as empirical support; Rochet and
Tirole 2004b provide a more technical overview.
16
As we note below, the payment card industry has followed this pricing structure since its birth, as a two-sided plat-
form, in 1950.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 321
price and quality of available substitutes and other factors, including, in some models, the pres-
ence of two-sided competitors.
The profit-maximizing price structure—the ratio of the price paid by men to the price paid by
women, for instance—also depends in general on costs, price sensitivities, and the way participa-
tion by the members of each group affects the demand of the other group.17 In addition, all general
models of two-sided platform markets imply that profits may be maximized by highly asymmetric
pricing in which one group is served at a price close to or even below marginal cost, and most or
all gross margin is earned by serving the other group.18
It is important to note that many, if not most two-sided markets exhibit this sort of asymmetry
in pricing and gross margin generation.19 Shopping malls, for instance, often provide free parking
to consumers, sometimes in expensive parking structures, and make all their money by charging
rent to merchants. Yellow Pages and competing telephone directories of merchants in the US are
given away to consumers; all the revenue is provided by merchants.20 Similarly, Microsoft and
Apple do not charge applications software developers anything for the highly valuable software
services (sometime called APIs) included in their software platforms. Both these firms make almost
all of their money from end-users of computer systems. (In Microsoft’s case, Windows is usually
licensed to computer makers that in turn license it to end-users.) On the other hand, makers of
video game consoles sell them to end users at or below cost and make most or all of their gross
margin from license fees paid by game developers.21
17
In the Rochet-Tirole model, which is based on particular assumptions about the structure of demand, the price
structure is completely independent of the relative costs of serving the two groups (there are no fixed costs on either
side in their formulation). See Rochet and Tirole 2003b, pp. 996-997.
18
We define gross margin here and below as revenues minus direct, side-specific variable costs, such as the manufactur-
ing costs of video game consoles. Attributing profits to one side or another is strictly incorrect, since both sides must
be on board for profits to be earned, and many costs in platform businesses, like the cost of developing the Windows
operating system, are properly thought of as joint between the two sides.
19
Bolt and Tieman 2003 offer an explanation based on the fact that some demand functions that are well-behaved in
an ordinary one-sided context yield non-concave profit functions in some two-sided models, thus making corner
solutions (e.g., prices of zero) optimal.
20
In the payment card context, Vickers 2005 has argued that one might find “a distortion of competition” when one
side of the market is charged “more than all the costs” of providing services to both sides. There is no theoretical
basis for this test of which we are aware, and this sort of pricing is not uncommon. It occurs whenever, as in the two
examples in the text, one side of a profitable two-sided business pays nothing, so the other side must be charged
“more than all the costs” if the enterprise is to be profitable.
21
See Evans et al. 2004.
PLATFORM ECONOMICS:
322 Part Four : Payment Cards Essays on Multi-Sided Businesses
s ystem today (ignoring recent franchise-like bank deals), or the store-specific cards offered by such
merchants as Neiman-Marcus. Neglecting for the moment the distinction between access and vari-
able prices, the price structure here can be described by the ratio of fees paid by merchants, typi-
cally in the form of the merchant discount, to transaction-related fees paid by consumers, in the
form of annual and other fees.22 Using this measure, the available evidence indicates that unitary
systems have generally adopted asymmetric pricing structures and earned the bulk of their revenue
from merchants, rather than consumers.
Before 1950, payment cards were issued by retailers for use in their stores. Then, as now, card-
holders who paid their bills within a specified amount of time (usually a bit less than a month)
did not pay any fee for charging and in fact benefited from the float. Those who financed their
store card charges paid interest, of course. But to our knowledge, then as now, retailers did not
cancel cards held by customers who chronically did not finance. Although we would not want to
push this point too far, given that store cards are one-sided and are bundled with finance services,
store card transaction services have what might be thought of as a slight negative variable price
and a zero access price.23 Merchants with retail cards presumably find that this is the optimal
pricing scheme.
Diners Club introduced the first two-sided payment platform—that is, a general purpose
payment card that could be used by cardholders at many retailers—in 1950. After initially offer-
ing the cards for no fee to consumers, Diners Club settled on a business model in which card-
holders paid an annual fee of $3 (over $18 in 2004 dollars) and a slightly negative transaction
fee in the form of float, while merchants paid a variable fee of seven percent of each transaction.
During the 1950s and into the early 1960s, based on available data, Diners Club earned about
70 percent of its revenues—and probably most of its gross margin—from the merchant side of
the business.24
American Express entered the card business in 1958. It adopted a similar business model, though
it eventually settled on slightly lower merchant fees and slightly higher cardholder fees than Diners
Club. But available data indicate that for the last four plus decades American Express has earned
upwards of 65 percent of its transaction-related revenues from merchant fees (in later years some
American Express cards began to bundle borrowing with transaction services, which makes clean
22
By “transaction-related,” we mean that we are here following almost the entire literature and focusing on the pay-
ment function of credit, charge, and debit cards, putting to one side the credit function that only credit cards per-
form, and the associated revenue flow from consumers.
23
Store cards may increase purchases by relaxing short-run liquidity constraints. As we note below, this argument has
been advanced as a motivation for retailer acceptance of general purpose payment cards (particularly credit cards)
by Chakravorti and To 2003. See also Chakravorti and Emmons 2003 and Wright 2000. However, again note that
we would expect that retailers would take efforts to weed out transactors if that were the entire motivation.
24
Evans and Schmalensee 2005, pp. 54-55. Annual fee converted to real 2004 dollars using the GDP implicit price
deflator from United States Department of Commerce 2004.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 323
comparisons more difficult).25 Bank of America also entered in 1958, with a card that did not charge
consumers a fee, although there were finance charges (and associated costs of funds and default risk)
from consumers who chose to revolve. Their merchant fee was 5 percent of each transaction.26
Even though transactions-related costs on the acquiring side of the business seem to belower
than those on the issuing side,27 these unitary systems apparently concluded that the profit maxi-
mizing price structure for the system—the price structure that gets and keeps both sets of custom-
ers on board and permits both issuers and acquirers to be profitable—is one that obtains the bulk
of the revenue from the merchant side of the business.
These unitary systems instituted these pricing schemes in their early years, when they would
appear to have had little market power. We think it is fair to conclude that the“competitive”—cer-
tainly in the sense of non-collusive and non-monopolistic—pricing structure for payment cards is
(or at least was for some time) one in which merchants pay a relatively high transaction price and
cardholders pay zero or possibly slightly negative transaction prices plus modest fixed fees, and in
which the bulk of the profits, loosely speaking, thus flow from the merchant side. Being the com-
petitive pricing structure does not necessarily mean that it is socially optimal, however, and it is to
that issue we turn next.
ward. In an ordinary one-sided market under the usual assumptions, buyers’ willingness to pay for
incremental units of output provides a measure of the social value of that output. Thus the social
optimum in such markets occurs at the output level at which price is equal to marginal cost, since
at lower levels of output buyers are willing to pay more than marginal cost for incremental output,
while at higher levels of output they are willing to pay less than marginal cost. In a two-sided mar-
ket, however, increases in output on side A of the market provide positive benefits to buyers on side
B that are not reflected in the side A demand curve. Thus if price equals marginal cost to customer
group A, it is nonetheless socially beneficial to increase output to A because of the (externality)
benefits that would thereby be conferred on members of group B.
What about price structure? In general single-firm models of two-sided markets in otherwise
perfectly competitive economies, one can compare the conditions defining profit-maximizing pric-
ing, welfare-maximizing pricing (which involves the seller losing money and, presumably, being
subsidized by the government), and Ramsey pricing (which involves maximizing social welfare
subject to the constraint that the firm does not lose money). Comparing these conditions in vari-
ous models yields two general observations that appear to be robust to modeling assumptions.
First, neither welfare-maximizing pricing nor Ramsey pricing is ever purely cost-based. In both
cases, the optimal price structure also depends on price sensitivities and externalities on both sides
of the market.29 Second, there is no simple, general description of the relations among the profit-
maximizing and Ramey price structures. They are rarely identical, but the sign and magnitude of
the difference between them depends on essentially all the demand, cost, and externality param-
eters in the model.
29
In fact, in the Rochet-Tirole model, which assumes a particular demand structure, the Ramsey price structure, like
the profit-maximizing price structure, is completely independent of group-specific marginal costs. See Rochet and
Tirole 2003b, pp. 997-998.
30
This discussion in this section is based on Evans and Schmalensee 2005, pp. 153-156.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 325
non-delivery of goods by the merchant, bankruptcy of the merchant or, for that matter, of the mer-
chant’s bank. Another issue was how much the issuer and acquirer were compensated for executing
a transaction that could not take place without participation by both of them.
As noted above, when Bank of America began franchising, its first response to this problem was
a rule stipulating that whenever acquiring bank A and issuing bank B were different, A was to send
the full amount of its merchant discount to B (and, of course, B was to reciprocate when one of
its merchants dealt with one of A’s cardholders). Presumably the idea was that if all banks adopted
similar price structures, then to a first approximation typical bank A’s merchant/consumer revenue
mix would not depend on whether its cardholders dealt with its merchants or with those acquired
by other banks. If this had worked, the system could have maintained a merchant-centered (or
any other) price structure. But, as we observed above, this device failed—both because it reduced
everyone’s incentives to sign up merchants and because it invited deception.
In the early cooperatives, some banks entered into bilateral agreements with each other; this
was possible when the systems had few members. But ultimately the cooperatives decided to de-
velop a default set of rules, or contracts, that defined the allocation of risks and payments. In
NBI—the cooperative that evolved from the BankAmericard franchise system and was the pre-
decessor of Visa—as long as an acquirer’s merchant met certain terms, such as properly author-
izing transactions and checking card numbers against a list of known fraudulent accounts, it was
guaranteed payment. If a transaction turned out to be fraudulent or a consumer failed to pay, the
issuer was responsible. Various procedures were also set up to resolve disputes between merchants
and cardholders as to the validity of particular charges. MasterCharge—MasterCard’s predeces-
sor—developed a similar contract with a similar interchange fee.
Both contracts were presumably the result of a bargain struck within the cooperative organiza-
tions between members with different stakes in acquiring and issuing (although back then there
was less specialization on this dimension than there later came to be). Given the payment card
industry at the time, the price structure that resulted from these interchange fees was similar to
that of the unitary systems, although the price levels were lower. Since the interchange fee was a
cost to the acquirer it was passed on to merchants as part of merchant discounts. The resulting
merchant discounts were lower, however, than those charged by American Express at the time.
The issuers then chose to issue cards with modest access fees and slightly negative transaction fees
for cardholders.31 Unlike the charge-card systems, of course, the credit cards came bundled with
longer-term financing. For transactions, the resulting percentages of revenue that the multi-party
systems earned from the merchant side were similar to those of American Express and Diners
Club, if not higher.
31
At least through most of the 1970s, annual fees for Visa and MasterCard cards were relatively uncommon. National
Bancard Corporation v. Visa U.S.A., Inc., p. 27.
PLATFORM ECONOMICS:
326 Part Four : Payment Cards Essays on Multi-Sided Businesses
32
E.g., this was one of the remedies proposed by NaBanco. National Bancard Corporation v. Visa U.S.A., Inc. 1984,
p. 1241.
33
A recent analysis (Small and Wright 2000) concludes that, at least under certain somewhat special assumptions,
even if transactions costs were low enough to make bilateral negotiations practical in large systems, banks’ strategic
behavior would undermine the system’s viability. It is worth noting, though, that the terms of two-bank transactions
in the Australian EFTPOS system (a PIN-based debit card system) are set exclusively by bilateral negotiation; see
Reserve Bank of Australia 2004. There are only 11 acquirers and 150 issuers in the EFTPOS system; these numbers,
while nontrivial, are orders of magnitude lower than the thousands of issuers and (at least) hundreds of acquirers in,
e.g., the Visa system in the United States. See Reserve Bank of Australia 2005a, pp. 4, 40-45. Wright and Weiner
2005 report that interchange fees are also set by bilateral negotiation in Sweden. Banking concentration is quite
high in Sweden, with the top four banks accounting for over 80 percent of total assets in 2003. See Swedish Bankers’
Association 2004. By contrast, the top four banks in the United States accounted for only a 24 percent share in
2002. See United States Census Bureau 2004.
34
See, e.g., Frankel 1998 and Balto 2000 and, for a rebuttal, Ahlborn et al. 2001.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 327
no economic basis for concluding that an interchange fee of “0” is better or worse for society than
any randomly chosen positive or negative percentage.35
Second, setting the interchange fee at zero imposes a particular price structure on the system,
one in which side-specific prices are tightly linked to side-specific costs. For example, had the pre-
decessors of MasterCard and Visa had a zero interchange fee in the early 1970s, they would have
had to raise card fees by $4.88 per account (or about $18 in 2004 dollars) to compensate for the
loss of interchange fee revenues.36 Imposing less pricing flexibility on the emerging cooperative
card systems in the early 1970s would have necessarily placed them at a competitive disadvantage
relative to the more established unitary systems, which could choose their price structures with-
out constraints. This distortion in competition between unitary and multi-party systems would
need to be weighed against whatever benefits policymakers believe would arise from mandating
exchange at par.
One also has to consider what other changes might result from a zero interchange fee. It is
possible that a multi-party system is not even viable with the pricing structure that would result
from a zero interchange fee.37 In their early days, the viability of these systems was very much an
open question. Even today, large numbers of issuers might move to other card systems (possibly
new ones) that are organized so as to be able to replace interchange revenue without regulatory or
antitrust scrutiny. (Consider, for instance, franchise arrangements with for-profit unitary systems
in which merchant discounts and payments to issuers are set unilaterally by the system.) It is not
clear that the pricing structure from such systems would be more favorable to merchants. Another
possible effect of a zero interchange fee would be to change the other terms of the contract among
member banks. For example, rules governing disputes among issuers and acquirers might be made
more favorable toward issuers in order to avoid issuer defections. Or the circumstances under
which payment is guaranteed to acquirers might become more limited.
35
Interchange payments in the Australian EFTPOS system flow from issuers to acquirers; see Reserve Bank of Australia
2004. Negative interchange fees of this sort existed in some online debit card systems in the United States until 1997
and remain the norm in U.S. ATM systems; see Hayashi et al. 2003.
36
Calculated from data provided by Visa U.S.A. Card fee converted to real 2004 dollars using the GDP implicit price
deflator from United States Department of Commerce 2004.
37
We illustrate this point in a simple model immediately below.
38
Baxter 1983.
PLATFORM ECONOMICS:
328 Part Four : Payment Cards Essays on Multi-Sided Businesses
Baxter assumed perfect competition among issuers, among acquirers, and everywhere else in
the economy, and, like almost all the subsequent literature, he assumed away all fixed costs and
access prices. Under these conditions, collective determination of the interchange fee cannot be an
exercise of market power, since there is no market power anywhere in the economy. It is simply a
payment from one set of perfectly competitive firms, which will have to raise their (variable) prices
to cover it, to another set of perfectly competitive firms, which will lower their (variable) prices so
as to compete it away completely. The interchange fee thus can only affect the price structure, not
the price level. Baxter assumed that consumers and merchants would use a particular payment card
if and only if the per-transaction price charged to them was less than the per-transaction benefit
from using the card rather than cash or check.
To see the role of the interchange fee in ensuring system viability in this setup, suppose for
simplicity that all consumers have the same per-transaction benefits and that so do all merchants.
Let the per-transaction prices charged by the card system to consumers and merchants be Pc and
Pm, respectively; let the corresponding per-transaction benefits relative to cash or check be Bc and
Bm; and suppose the constant per-transaction marginal cost of serving a consumer is Cc and of
serving a merchant is Cm. Under these assumptions, it is efficient to use the card for all transactions
if and only if
With a zero interchange fee and perfect competition, consumers will agree to use the card and is-
suers will break even if
Bc Pc 5 Cc, (2)
And merchants will agree to accept the card and acquirers will break even if
Bm Pm 5 Cm. (3)
For a zero-interchange system to be viable, both (2) and (3) must be satisfied. It is easy to find
numerical examples in which (1) is satisfied, but either (2) or (3) is not. Suppose, for instance:39
Bc 5 1, Bm 5 8, Cc 5 3, and Cm 5 2. (4)
Here total per-transaction benefits from using the card are almost double the corresponding cost,
but there is no price to consumers that satisfies (2). In this example retailers receive the most ben-
efits if the card is used, and it would be logical for them to cover most of the system’s costs. But
there is no way to accomplish this with a zero interchange fee. An interchange fee of 2 solves the
39
This example is from Rochet 2003, where this model is instructively explored in more detail.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 329
problem. This raises the acquirer’s cost and (because of perfect competition among acquirers) the
merchant discount, Pm, to 4, and it lowers the issuer’s cost and (because of perfect competition
among issuers) the consumers’ fee to 1. Merchants contribute 80 percent of the system’s revenue
but are better off by 4 per transaction, while consumers are just indifferent to the card’s existence
even though the positive interchange fee has reduced their fees substantially.
By assumption, the costs incurred by both sides are necessary to execute a transaction. If this
assumption is correct, it makes no sense to think of either side as providing particular services to
the other; both must incur all the costs stated for either to benefit. Moreover, there is no way for
regulators to look only at cost conditions and conclude that an interchange fee of 2 is appropriate.
Even in this simplest possible example, demand conditions must be considered. It could happen,
of course, that some cost-based formula produced an interchange fee of 2. But this could only
happen by chance.
In Baxter’s model, the interchange fee is not set to maximize profits, since there are no profits
earned anywhere in the payment system. He argued that its level was determined uniquely by the
need to balance the supply and demand of card transactions. This seems somewhat artificial, since
as long as some merchants have agreed to accept a card, the volume of transactions is determined
unilaterally by card-carrying consumers’ deciding whether or not to use the card for particular
transactions. Nonetheless, the proposition that at some interchange fee levels the system would
not be viable is likely to hold in more general models. The important—and robust—insight from
Baxter’s analysis, although it is not framed precisely this way, is that the interchange fee helps inter-
nalize an externality between the two customer groups and in so doing has the potential of making
both customer groups better off.
40
Schmalensee 2002.
41
Much of the subsequent literature has persuasively called that standard assumption into question in this context, as
we discuss in the next section. Nonetheless, it is instructive to see its implications before considering how it might
be altered in the direction of greater realism.
PLATFORM ECONOMICS:
330 Part Four : Payment Cards Essays on Multi-Sided Businesses
assumed to operate like the Visa and MasterCard systems operate in fact, on a break-even basis,
and fixed costs and access prices were again assumed away. Thus, as in the numerical example
above, the interchange fee simply shifts costs between issuers and acquirers, raising costs on one
side of the market by exactly as much as it lowers costs on the other side.
In this model the level of the interchange fee can affect the profits earned by issuing and acquir-
ing banks whenever competition among them is imperfect, making it possible to compare profit-
maximizing and welfare-maximizing fees. Under some special assumptions the comparison is sim-
ple: when there is a single issuer and a single acquirer and demand curves are linear, for instance,
Schmalensee shows that the profit-maximizing interchange fee also maximizes system output and
economic welfare. In this case regulation could only reduce overall performance. In general, how-
ever, these three fees may be different even under the particular demand structure Schmalensee
assumed, and even when the further assumption of linear demand is imposed. Thus, even under
strong assumptions about demands and costs, the relations among these quantities is complex and
depends on demand parameters, cost conditions, and, an element not present in analysis of unitary
two-sided platforms, the nature of competition among issuers and among acquirers.
In the Schmalensee model the interchange fee is not an ordinary market price: it is a balancing
device for shifting costs between issuers and acquirers and thus shifting charges between consum-
ers and merchants.42 Fixing the interchange fee is quite unlike fixing a price in a typical one-sided
market. The first-order effect of ordinary price-fixing is to harm consumers by restricting output.
The first-order effect of collective determination of interchange fees in this model is generally to
enhance the value of the system by balancing participation of the two customer groups, thus in-
ternalizing indirect externalities. This is illustrated most clearly by the existence of a special case,
noted above, in which collective determination of the interchange fee in order to maximize profit
also maximizes output and economic welfare. And even in Schmalensee’s simple model, the so-
cially optimal interchange fee depends on costs, demand conditions, competition among issuers
and among acquirers, and externalities between merchants and consumers. Thus as a practical
matter, there is no rule for regulatory determination of the interchange fee that could be relied on
to improve overall system performance and thus enhance economic efficiency.
42
In addition, the interchange fee can be used to shift profits between issuers and acquirers, though when it is used for
that purpose total system profit is necessarily reduced.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 331
the relation between the profit-maximizing and welfare-maximizing interchange fees. Almost all
assume away fixed costs and access pricing and assume that all revenues flow from variable prices.
When used to consider regulatory policy, these models become exercises in the economics of the
second best.43 That is, they consider policy in the presence of multiple, interacting departures from
the competitive ideal. In general, such exercises rarely yield tractable rules that can be used to de-
sign practical policies. The interchange literature, which points to many additional distortions but
typically considers them one at a time, provides no exceptions.44
43
Lipsey and Lancaster 1956.
44
For surveys of some of the literature discussed here, see Chakravorti 2003, Hunt 2003, Rochet 2003, Rochet and
Tirole 2003a, and Schmalensee 2003.
45
In an influential review of the literature undertaken for the Reserve Bank of Australia, Katz 2001 stressed the effects
of such competition. See also Wright 2003c and Hayashi 2004.
46
It is worth noting that De Grauwe and Rinaldi 2002 have developed a model involving rent-seeking by merchants
in which the level of card usage is below the welfare-maximizing level. They treat merchants’ decisions to accept or
not accept cards as exogenous, however.
PLATFORM ECONOMICS:
332 Part Four : Payment Cards Essays on Multi-Sided Businesses
mizing interchange fee may be above or below the welfare-maximizing level, and there may thus
be too many card transactions or too few.47 The relation between profit-maximizing and welfare-
maximizing fee levels in this more general model is complex and depends on details of demand and
competitive conditions as well as costs.
47
Wright 2004a.
48
Chakravorti and To 2003. See also Chakravorti and Emmons 2003 and Wright 2000.
49
Chang and Evans 2000; for an alternative view, see Frankel 1998.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 333
(a payment system in which a bank or a post office transfers money from one account to another
when it receives authorization to do so) and related transactional systems.
There is no rigorous analysis of which we are aware of the effects of government-determined
pricing in these competing systems on pricing in payment card systems. It is worth noting, for in-
stance, that a large fraction of consumers in the United States and Western Europe pay no variable
fees for writing checks, primarily as a result of the decision by banks to bundle this service (and
ATM/debit cards) with the general banking relationship. This might explain why credit and charge
card systems have generally not imposed transaction-specific variable fees on consumers. Similarly,
there has been no analysis of which we are aware that considers the impact of this sort of competi-
tion on the relationship between profit-maximizing and welfare-optimizing interchange fee levels.
The existing literature does contain a number of (generally complex) analyses of competition
between payment systems or between two-sided platforms in general.50 It is fair to say that this
work is at an early stage. It seems clear that the nature of consumer and merchant behavior shapes
the competitive price structure, and that there is no general tendency for competition between
platforms to make the price structure closer or farther from the social optimum, though compe-
tition will generally tend to lower the price level. Suppose, for instance, that consumers all have
multiple cards (this is termed “multi-homing” in the literature) and are indifferent among them,
and that merchants can effectively persuade consumers which card to use. In this case competing
systems will have an extra incentive to compete for merchants’ favor, and this will cause a tilt in the
price structure against consumers that may or may not improve performance. Similarly, if consum-
ers tend to use only one card (to “single-home”) while merchants find it easy to accept all cards,
competing systems will have an extra incentive to attract consumers, and the price structure will
be tilted in their favor accordingly. These aspects of consumer and retailer behavior have received
little empirical study, however.51
50
See Armstrong 2004 and 2005, Armstrong and Wright 2004, Chakravorti and Rosson 2004, Gabszewicz and
Wauthy 2004, Guthrie and Wright 2003, Manenti and Somma 2003, Rochet and Tirole 2002 and 2003b,
51
Rysman 2004 finds that most U.S. consumers carry multiple cards but tend to concentrate their purchases on only one,
suggesting a weak form of single-homing. It is unclear what combination of preferences, costs, and merchant behavior
this finding reflects, however, or how much a typical consumer prefers her primary card over others she carries.
PLATFORM ECONOMICS:
334 Part Four : Payment Cards Essays on Multi-Sided Businesses
use, the interchange fee will be irrelevant and card usage will be efficient.52 If acquirers are perfectly
competitive, an increase in the interchange fee is passed along dollar for dollar to the merchant
discount, and perfect competition among merchants means that the merchant discount is passed
along dollar for dollar to card-using consumers (and not at all to those who pay with cash). On
the other side of the market, competition among issuers means that the increase in interchange is
passed to card-users, dollar for dollar, in reduced fees. Thus card-using consumers pay the full cost
of the system regardless of the interchange fee, and, if they also bear the full costs of all other pay-
ment systems, they will use payment cards if and only if they are socially less costly. Accordingly,
Carlton and Frankel advocate abolishing NSRs.
It turns out that the Carlton-Frankel assumptions are stronger than necessary for the inter-
change fee to be irrelevant in the absence of an NSR. What is required is only costless surcharg-
ing—i.e., the ability of merchants at no cost to charge different prices depending on the means of
payment used.53 But without perfect competition everywhere, abolishing an NSR does not gener-
ally lead to an efficient outcome. In particular, imperfect competition among issuers then tends to
lead to under-provision of card services, and merchants could use surcharges as a mechanism for
price discrimination. Economic welfare may be lower than at the profit-maximizing equilibrium
with an NSR—even if card usage is excessive in the latter case. In the murky realm of the second-
best, this sort of ambiguity is not uncommon.
Moreover, it is clear that the assumption of costless surcharging is unrealistically strong; most
merchants do not discriminate among people using different means of payment even when they
are not prevented from doing so. For instance, it has generally been permissible for U.S. merchants
to give a discount for cash purchases and, though this was done for a time at gasoline stations, it is
now extremely rare.54 In the Netherlands about 10 percent of merchants imposed surcharges when
they were allowed.55 In the United Kingdom, surcharges are permitted for credit and charge cards
but surcharging is uncommon.56 And when NSRs were abolished in Sweden, only about 5 percent
of merchants imposed surcharges.57 Although we are not aware of any concrete data yet from the
elimination of NSRs in Australia,58 our understanding from colleagues there is that the prevalence
of surcharging, at least to date, is likely in line with these other experiences. The one instance we
are aware of with a somewhat higher incidence of surcharging, although still far from pervasive,
52
Carlton and Frankel 1995a and 1995b; see also Evans and Schmalensee 1995.
53
See Rochet and Tirole 2002, Wright 2003a, and, especially, Gans and King 2003b. Schwartz and Vincent 2004
provide a related analysis for a unitary system.
54
Chakravorti 2003, p. 55.
55
ITM Research 2000.
56
Office of Fair Trading 2003b, pp. 114-115.
57
IMA Market Development AB 2000, p. 18; but see Katz 2001, p. 44.
58
For example, the Reserve Bank of Australia notes that “some merchants” have started surcharging but that “there are
no comprehensive data.” Reserve Bank of Australia 2004, p. 11.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 335
was in Denmark, where earlier this year, 19 percent of merchants (primarily grocery retailers)
passed a new 0.55 Kroner debit card surcharge through to consumers.59
We suspect merchants are reluctant to impose surcharges for two reasons. The first is that there
are transaction costs of imposing different prices based on payment methods. The second is that
consumers may prefer to patronize stores that do not surcharge. No analysis of which we are aware
has considered or attempted to measure these costs and preferences. There is nothing unusual here,
of course: there are many things that merchants could surcharge for—because they entail specific
costs that are caused by particular customers—and do not.60 Parking in shopping malls is an obvi-
ous example.
E. Implications
The quantity of recent theoretical literature discussed in this section makes it clear that econo-
mists find interchange fees fascinating. Almost all of these papers find that profit-maximizing
interchange fees are unlikely to be socially optimal, but none yields workable rules for welfare-
improving regulatory intervention. (In particular, none points to the optimality of any rule that is
purely cost-based.) Moreover, these models are highly stylized. A variety of market imperfections
are considered one at a time; not only are some visible imperfections not considered at all (e.g.,
government-determined zero interchange in U.S. checking), no analyst has even attempted to
consider them all together. Most papers assume consumers are faced with variable (i.e., transac-
tion-specific) charges, even though most systems do not impose such charges and rely instead on
annual fees and other access charges. Finally, there has been essentially no empirical work devoted
to testing any of these models or to measuring the importance of any of the effects they predict.
This literature, in short, is not very useful for either rationalizing or designing a system of inter-
change fee regulation. Of course, for exactly the same reasons, it is not capable of proving that the
interchange fees determined by the card associations are exactly or approximately socially optimal.
It is useful to compare the results for interchange fees for payment cards with some other busi-
ness practices that economists have analyzed. In many ways the theoretical results surveyed here
are similar to those for advertising, research and development, product design, product variety,
location decisions, firm entry in the presence of fixed costs, bundling, and price discrimination.61
In all these cases economic models show that the profit-maximizing result under imperfect com-
petition may deviate from the social welfare-maximizing result. However, in most cases the bias
can go in either direction, and in all cases determining the socially optimal result depends on com-
plex factors that cannot be measured in practice. Based on our review of the theoretical literature
59
The Danish competition authority had announced that it was going to investigate whether there had been any col-
lective action by grocery stores. Konkurrencestyrelsen 2005.
60
See, for example, Evans and Schmalensee 2005, p. 131.
61
See, generally, Tirole 1988.
PLATFORM ECONOMICS:
336 Part Four : Payment Cards Essays on Multi-Sided Businesses
on interchange fees to date there does not seem to be any basis for concluding that the potential
distortions caused by collective determination of interchange fees are any more—or any less—sig-
nificant than the potential distortions caused by these other deviations from the ideal model of
perfect competition.
Since regulation is a blunt policy instrument in practice, unless there is a substantial market
failure there is scant chance that regulation will reliably improve matters. And unless there is a
known, practical regulatory rule that, if followed, is reasonably certain to improve performance, it
is likely that regulation will be on balance harmful. At least in theory, there is no economic point
to interfering with even imperfect markets unless those imperfections are serious and capable of
correction by known methods.
62
See, for instance, Armstrong et al. 1994, Megginson and Netter 2001, and Littlechild 2003. In the United States,
price cap regulation has been used as an alternative to rate of return regulation by the FCC and some states in
regulating telephone companies starting in the 1980s. Viscusi et al. 2000, Chapter 12. This approach has been very
widely used outside the United States. Winston and Morrison 1986 document some of the adverse effects of airline
regulation in the US.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 337
…we saw that credit cards were growing faster than the other means of payment. This was
initially somewhat surprising as credit card transactions are more expensive than most other
means of payment—that is, they involve a larger payment from the users of the payments
system to the providers of the payments system. … Why was this possible?
Governor Macfarlane poses almost the right question: Are payment system usage patterns sig-
nificantly inconsistent with system costs—and, it is essential to add, benefits? There is nothing
unusual about a high-cost product driving out cheaper competition if the high-cost product is
much better. U.S. drivers generally prefer automatic to manual transmissions in their automobiles,
for instance, even though automatic transmissions cost more and are more expensive to maintain.
Drivers seem to believe the difference in benefits outweighs the difference in cost.65
In our context, U.S. banks typically charge consumers the same variable price for handling
a check as for handling a signature-based debit card payment: zero. Debit cards are nonetheless
rapidly replacing checks because consumers find them more convenient.
63
It is worth noting one difference explicitly. Both profit- and welfare-maximizing interchange fees depend in general
on the nature of competition among issuers and among acquirers, and this additional complexity is not present in
the analysis of ordinary, unitary two-sided platform businesses.
64
Macfarlane 2005.
65
Similarly, as Joanna Stavins pointed out in a paper on electronic check truncation and presentment, bicycles may be
cheaper than cars but that does not mean that society would necessarily be better off if bicycles replaced cars. Stavins
1997, p. 28.
PLATFORM ECONOMICS:
338 Part Four : Payment Cards Essays on Multi-Sided Businesses
An approach that is based on careful measurement of costs and benefits has the potential to
distinguish what is important from what is only a theoretical possibility. Dealing with quantita-
tive evidence rather than qualitative possibilities can also inform regulatory policies if regulation
is deemed appropriate. Since the objective of interchange-fee regulation, where it is warranted,
should be to correct the effects of a distortion of price signals, it is important to get a quantitative
sense of the importance of those effects. Let us, accordingly, turn to the available evidence.
Some studies have argued that payment cards are used too much, based on the observation
that they are more expensive on average for merchants than cash and checks but are nevertheless
increasing in use at the expense of cash and checks.66 While this is intuitively appealing, it ignores
both cardholders and the role of benefits in determining the social optimum.
To see this, consider a simple economy with only cash and cards and with a fixed set of trans-
actions to be executed. For each transaction, one can in principle compute the social marginal
cost of executing the transaction using cash—the sum of the marginal costs to the merchant and
consumer involved, as well as the net costs to all other involved parties, including governments and
commercial banks. One can similarly compute the social marginal cost of executing each transac-
tion using cards, along with the social marginal benefit of using cards as opposed to cash—the
sum of the marginal benefits to consumers, merchants, governments, and commercial banks.67
The marginal benefit in any of these cases may be negative, of course; consumers may find cards
less convenient than cash for very small transactions, for instance. In this simple case the efficient
outcome is clearly for cash to be used for transactions for which the social marginal cost of using
cash is less than the net social marginal cost of using cards—the social marginal cost of using cards
minus the social marginal benefit of using cards rather than cash—and for cards to be used for
the others. Transactions-specific marginal costs and benefits are in principle necessary to assess the
importance (i.e., the net social cost) of any deviation from this ideal. Considering only costs and
only one of the parties involved cannot be very informative.
To our knowledge, there are no empirical studies in the literature that consider the marginal
social costs and benefits for merchants and cardholders, and thus there are no comparisons of
actual versus optimal use of alternative payment systems that have even approximate economic
validity. There is some evidence on marginal costs for merchants, some highly incomplete evidence
on marginal benefits for merchants, and essentially nothing on marginal costs or benefits for card-
holders or other parties.68
Garcia Swartz, Hahn, and Layne-Farrar provide a useful overview of much of the available evi-
dence, examine some of the issues one would need to consider in evaluating whether an economy
66
See, for example, Balto 2000 and Reserve Bank of Australia 2002.
67
For consumers, benefits would be measured by changes in consumers’ surplus; for the other entities the measure
would generally be changes in rents.
68
Compare, for instance, Humphrey et al. 2003, which argues that electronic payments systems are inherently socially
more efficient, and Food Marketing Institute 1998, which argues that payment cards impose excessive costs on grocers.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 339
has roughly the socially optimal use of payment cards, and present some rough calculations based
on available data and some plausible assumptions about consumer benefits for a few transactions
sizes and types.69 They take the merchant-based cost surveys as a point of departure but then pro-
ceed to make three types of adjustments. First, merchant-based surveys of costs usually compare
payment instruments at different transaction sizes, typically the average transaction size for each
payment instrument. The authors depart from this norm and compare payment instruments at
a set of fixed transaction sizes, small and large, to assess the sensitivity of their results. Second, by
combining plausible assumptions with available data they attempt to incorporate all other parties
to the transaction—consumers, the government, and commercial banks—into the calculation of
cost and benefits. Finally, they consider the benefits received by the cardholder and merchants; as
a result they compare the “net” cost of payment methods. Garcia Swartz, Hahn, and Layne-Farrar
present rough estimates of these magnitudes and find that the optimal payment method (consider-
ing quantifiable costs and benefits) varies by transaction size, store type, and other circumstances.
For many transactions it appears that credit and debit cards do not have higher “net costs” than
cash or checks. While this work is instructive, however, it falls well short of a rigorous overall as-
sessment of payment system performance.
one-for-one—or, indeed, at all. Issuers can respond to the loss of interchange fee revenue by vary-
ing either fixed or variable fees, or both. The extent to which they vary each will depend on a va-
riety of factors including the elasticity of demand for access to cards and the elasticity of demand
for transactions. In addition, there are likely to be one-time costs of various sorts (including costs
of changing accounting and billing systems) caused by moving from the traditional regime of zero
variable charges to a regime with positive variable charges. The one-time costs of making reward
programs less generous are likely to be less significant, if they exist at all.
Suppose, for instance, that issuers increased annual fees but did not reduce variable fees at all.
In that extreme case the regulation of the interchange fee would not alter consumer incentives to
use cards, although it might reduce the average number of cards that people carry. (Since it is hard
to function in modern economies without at least one payment card, we doubt that the fraction
of households with zero cards would rise noticeably.) Preliminary data from Australia suggests
that even though interchange fees were reduced by nearly half in late 2003, the marginal price to
cardholders of using credit cards has not changed much. While some reward programs were made
somewhat less generous, these cover only a fraction of consumers and card transactions,70 and we
have not seen widespread evidence of surcharging by merchants or the imposition of fees by issu-
ers that increase with card usage.71 So while available data suggest that interchange fee reductions
were passed through more or less completely to reductions in merchant discounts, it does not seem
that the stated objective of the RBA to make consumers face the “right” variable prices for different
payment methods was realized to any appreciable extent.72 Consistent with this, the data do not
reveal much, if any, impact of the reforms on the use of credit cards.73
70
Industry estimates suggest that reward cards constitute about one third to one half of all credit cards in the United
States; we lack comparable data for Australia. Bayot 2003.
71
Making some reward programs less generous moved some negative variable prices toward zero. Many of these
changes were in the form of caps on the total number of rewards points that can be earned each year. These limits
have been set at relatively high levels, over AUS$40,000 (approximately US$31,000 at the time of this writing) a
year, and likely have not been a binding constraint for many cardholders. See Reserve Bank of Australia 2004, p. 11.
There were also some reductions in the marginal reward per dollar spent in some cases. There is no reason to think
that this would have a big effect on card volume. In fact, if one is accumulating points in a rewards program to take
a trip, and the rewards program is made less generous so that more purchases are required to earn the trip, one might
use the card more rather than less intensively. There would be costs of various sorts associated with imposing positive
transaction fees on non-reward cards, so it is not surprising that this does not seem to have been done.
72
The RBA stated that its reforms were in line with the principle that “consumers should face prices that take into account
the relative costs of producing goods and services, as well as demand conditions.” Reserve Bank of Australia 2002, p. 34.
73
Credit card dollar volume grew significantly in Australia in the late 1990s. The rate of growth started declining fol-
lowing 1999. The rate of decline in the growth rate leveled off in 2003 and 2004, in contrast to an acceleration in
the rate of decline that would be expected were the reforms to have had a substantial impact. Nonetheless, the real
dollar volume of credit card purchases rose 20.0 percent between 2002 and 2004, about four times as rapidly as real
household consumption. Interestingly, the number of credit card accounts grew by 11.1 percent over this period,
compared to a growth of only 2.6 percent in population. See Reserve Bank of Australia 2005b.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 341
All this is not to say that dramatic reductions in the interchange fee had no economic ef-
fects. Because retailing in Australia is relatively concentrated and thus competition is likely to
be imperfect,74 it is reasonable to expect that only a fraction of the fall in merchant discounts
was passed on to consumers in the form of lower prices, and the remainder went to increasing
retailers’ profits. When consumers use cash or credit cards without rewards they are better off as
a result; when they use cards with reward schemes they are likely worse off. We have seen no evi-
dence that these effects can be associated with distinct groups of consumers – different income
quintiles, for instance.
On the other side of the market, the best evidence we have been able to obtain indicates that is-
suers have recovered between a third and a half of the fall in interchange revenue through increased
fees to consumers.75 Thus issuers’ profits have been reduced, and consumers with credit cards of all
sorts have been directly harmed, particularly if they use reward cards. In order to mitigate reduc-
tions in MasterCard and Visa interchange revenues, three of the top four Australian banks have
signed agreements to issue American Express or Diners Club cards, which can provide greater,
unregulated transaction-related revenues to issuing banks.76 In part as a consequence, the shares of
these unitary systems have risen, though from initially low levels.77
All in all, it seems that in the short time since interchange fee reductions were imposed retail-
ers have been made better off, issuers have been made worse off, and some consumers have been
made better off (particularly those who tend to use cash a lot) and others have been make worse
off (particularly those who use credit cards with reward schemes). In the long run some of these
effects are likely to be undone as consumers and issuing banks move toward American Express and
74
Government statistics on firm concentration are not available. We have identified what data we could find from
a range of sources. Many merchant categories appear to have significant levels of concentration. For example,
within their respective categories, the top department store had a 71 percent share (and the top two had an 83
percent share) in 2003; the top two supermarket and grocery stores (excluding convenience, specialty and miscel-
laneous food stores) had a 75 percent share in 2003; the top two mobile telephone operators had a 78 percent
share in 2005; the top land-line telephone operator had a 75 percent share in 2005; and the top two airlines had
an 83 percent share in 2005. In contrast, the four-firm concentration index for these categories is significantly
lower in the United States based on 2002 U.S. Census data: 65 percent for department stores, 33 percent for su-
permarkets and grocery stores, 61 percent for mobile telephone operators, and 60 percent for land-line telephone
operators. The U.S. Bureau of Transportation Statistics reports the top two firms accounting for 32 percent of
enplaned passengers among major U.S. airlines in 2000. The one merchant category that we found with lower
concentration was for warehouse clubs and superstores, for which the four-firm concentration index in Australia
was 41 percent, compared to 92 percent in the United States. See Euromonitor 2005a, Euromonitor 2005b,
Euromonitor 2005c, United States Census Bureau 2004, Bureau of Transportation Statistics 2000, Maxwell
2005, and McFarland 2005.
75
See Chang, Evans and Garcia Swartz 2005.
76
See, e.g., Cornell 2004 and 2005.
77
Reserve Bank of Australia 2004, p. 13.
PLATFORM ECONOMICS:
342 Part Four : Payment Cards Essays on Multi-Sided Businesses
Diners Club, as these schemes have substantially higher average merchant discounts than the bank
card systems.78
The apparent ineffectiveness of the fairly dramatic Australian regulatory intervention in terms
of its stated goals may result from the particular competitive environment in that country and
the types of cards used by consumers. However, it would appear to be generally the case that the
interchange fee is a highly imprecise instrument for affecting the volume of transactions on cards
and thus for correcting any perceived market distortion. That is because there is only a loose con-
nection between interchange fees and transactions prices to cardholders.
Let us suppose for the sake of argument that regulators went further than any have gone in fact
and banned the use of consumer access fees. Interchange fee regulations would then necessarily
affect the transaction price paid by consumers, as they are assumed to do in the theoretical litera-
ture. (This ban would, of course, likely impose a new set of costs and distortions on the system.)
There would still be the question whether it is possible to estimate the optimal interchange fee with
sufficient precision that policymakers could expect to increase rather than decrease social welfare.
A robust conclusion from the theoretical literature is that an estimate of the optimal interchange
fee would depend on a host of factors: estimates of the price responsiveness of cardholders and
merchants, indirect network effects between cardholders and merchants, competition in issuing
and acquiring and among merchants, price distortions in competing payment systems, transac-
tions costs and liquidity constraints, and marginal costs of serving cardholders and merchants. It
would also depend on how competing systems—some of which may be unitary—would respond
to changes in prices to cardholders and merchants. Because of the difficulty of the task, there are
no serious attempts of which we are aware to estimate the socially optimal interchange fee for any
real payment system. Given currently available data and estimation methods, we believe that any
such attempt could at best yield highly imprecise estimates.
This task would require far more empirical information than classic public utility regulation.
Most public utilities have historically been monopolies, so the strategic interaction with competi-
tors could be ignored, unregulated prices can be safely presumed to be too high, and reducing
78
Some have argued that reducing the interchange fees of the multi-party systems will put pressure on unitary systems
to follow suit, but this is not persuasive. To the extent that interchange regulation disadvantages the multi-party sys-
tems, it reduces competitive pressure on the unitary systems. (We are indebted to Michael Katz for this point.) The
Reserve Bank of Australia suggests that there has been some downward pressure on merchant discounts for American
Express and Diners Club in Australia, with a decline of 13 basis points over the 12 months ending June 2004 (al-
though it does not discuss whether other factors might have been responsible for this), but expresses concern that the
gap between the merchant discount for the unitary versus multi-party systems had widened by 30 basis points. See
Reserve Bank of Australia 2004, pp. 12-13. Given that the merchant discounts for the unitary systems were, accord-
ing to one account, about 100 basis points higher than for the multi-party systems (Reserve Bank of Australia 2001,
Figure 2.2), even if there is a modest decrease in the merchant discounts for multi-party systems attributable to the re-
forms, the post-reform unitary merchant discount would still exceed the pre-reform multi-party merchant discount.
And the merchant discount paid on volume shifted from the multi-party to the unitary systems would increase.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 343
prices until the utility just breaks even will generally increase economic welfare—at least as long as
impacts on the utility’s incentives for efficiency are ignored. Although sometimes present, network
effects rarely played an important role in the analysis, and the calculation of optimal (Ramsey)
prices required only estimates of marginal costs and demand elasticities. However, even in the
public utility context it has proved difficult to calculate precise estimates of the relevant param-
eters, and the determination of optimal prices has often led to considerable controversy among
economists and policymakers.
A robust conclusion of the economic literature on interchange fees and two-sided markets is
that cost-based interchange fees are generally not socially optimal.79 Even if one were convinced,
as some regulators seem to be, that current interchange fees are too high, unlike the public utility
case, there is no guarantee that lowering them toward any particular target will improve welfare.
In particular, there is no basis for believing that any particular cost-based formula for determining
interchange fees would even provide a first approximation to the socially optimal interchange fee.
Nor is there any basis at the moment for believing that moving from the collectively determined
interchange fee to a fee based on any formula that considers only costs would be likely to improve
social welfare. Unlike the public utility situation, therefore, there is no basis in economic theory or
fact for cost-based regulation of interchange fees such as the regime adopted in Australia or by the
European Commission.80
In both Australia and Europe, the regulators (or at least their economic consultants) recog-
nized that socially optimal interchange fees also depend on demand factors or network effects
but, presumably, decided not to incorporate those factors because of the difficulty of doing so,
citing instead the “objectivity” and “transparency” benefits of a cost-based measure.81 Objectivity
and transparency may have benefits, of course, but by themselves they do not, necessarily lead to
greater economic efficiency.
Although cost-based interchange fee regulation could by happenstance improve the efficiency
of the payment system, there are two fundamental reasons to doubt that it would regularly do so
in practice. The first is that regulating the interchange fee will not necessarily have a significant
effect on the variable prices paid by cardholders and therefore will not necessarily have a significant
effect on the volume of transactions—generally the putative target of the intervention. The second
79
The economic literature appears to be unanimous on this point; examples include Bergman 2005, Gans and King
2003a and 2003c (who accept for the purpose of argument the assumption that externalities between merchants and
consumers are unimportant in mature card systems) and Wright 2003b.
80
These regulatory schemes have the further peculiarity from an economic standpoint that the merchant discount
charged by unitary systems is left unregulated while the merchant discount for cooperative systems is regulated
through cost-based interchange fees. This favors the unitary systems and thus leads to a further economic distortion
whose effects would need to be considered in evaluating the net social benefits from moving to cost-based inter-
change fees.
81
Reserve Bank of Australia 2002, p. 12; Katz 2001, p. 29; European Commission 2002.
PLATFORM ECONOMICS:
344 Part Four : Payment Cards Essays on Multi-Sided Businesses
is that even if regulating the interchange fee could affect the variable prices paid by both merchants
and cardholders, cost-based regulation is not capable of achieving the optimal prices except by hap-
penstance. Whether cost-based regulation in practice would increase efficiency is unknown given
the current state of theoretical and empirical knowledge.
VI. CONCLUSIONS
Economists have only scratched the surface of the theoretical and empirical work that will be need-
ed to understand pricing in two-sided markets in general and the determination of interchange
fees in particular. Like much work in economics, many of the existing theoretical models are
based on highly simplified representations of the industries in question and employ highly special
82
That is the remedy that was sought by the plaintiff in NaBanco.
83
As we noted above, Small and Wright 2000 conclude bilateral negotiation would threaten system viability, but it is
unclear how robust this result is.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Interchange Fees and Their Regulation 345
1. The socially optimal prices for customer groups in multi-sided industries depend on price
elasticities of demand, indirect network effects between the customer groups, marginal
costs for providing goods or services to each group, and other factors.
2. Although socially optimal prices in the payment card industry depend on the same set of
factors, the socially optimal interchange fee also depends on other characteristics of these in-
dustries that affect the relationship between the interchange fee and final prices. Those factors
include the use of fixed and variable fees, competitive conditions among merchants, issuers,
acquirers, and the nature of competition from cash, checks, and unitary payment systems.
3. Thus the socially optimal interchange fee is not in general equal to any interchange fee
based on cost considerations alone.
4. One cannot presume on the basis of theory alone that the interchange fee set collectively by
an association is greater than, less than, or equal to the socially optimal interchange fee.
5. One cannot presume on the basis of theory alone that movements from the collectively set
interchange fee to any particular cost-based interchange fee will increase or decrease social
welfare.
6. One cannot presume on the basis of theory alone that the collectively set interchange fee
is greater than, less than, or equal to the interchange fee that would be set by bilateral
negotiations.84
If regulators have concluded that multi-party payment systems have exhibited significantly
sub-optimal economic performance, the current state of theoretical and empirical research leaves
them with three serious challenges if they try to increase payment system performance overall and
thereby raise economic welfare:
1. There is no empirical research that reliably addresses whether payment cards or any other
payment mechanism is used too much or too little. Such research would need to consider
the social costs and benefits of alternative payment systems and consider the effect of other
market distortions.
84
As we noted above, in the one example of bilateral interchange negotiation on which we have seen data, the Australian
EFTPOS debit card system, the fee is negative, flowing from issuers to acquirers. And the Reserve Bank of Australia
is considering raising EFTPOS interchange, shifting costs from consumers to merchants. See Macfarlane 2005.
PLATFORM ECONOMICS:
346 Part Four : Payment Cards Essays on Multi-Sided Businesses
2. Although it is possible that economists will be able to estimate the quantities necessary for
determining optimal interchange fees, very little empirical work has been done thus far on
most of the relevant quantities.
3. It is not clear that interchange fee regulation is the appropriate intervention for correcting
distortions in payment systems. The interchange fee is a blunt instrument for affecting the
prices faced by consumers if issuers assess fixed fees as well as variable fees.
CHAPTER
The Effect of Regulatory
13 Intervention in Two-Sided Markets
Howard Chang, David S. Evans, And Daniel D. Garcia Swartz
ABSTRACT
The Reserve Bank of Australia reduced interchange fees by almost half thereby eliminating a sig-
nificant source of revenue to issuers of credit cards. The purpose of this intervention was to align
the prices of using various payment instruments with their social costs and thus reduce the use of
cards, which the RBA viewed as a socially less efficient payment method than cash, checks, and
PIN debit cards. The short-run result of this regulatory intervention has been the following: (1)
Bank issuers have increased the fixed prices for cards and thereby recovered between 30 and 40
percent of the loss of interchange fee revenue; this fraction is likely to increase over time as cards
renew and new solicitations go out. Bank issuers have not changed the per-transaction fees for
cards much. (2) Merchants experienced a very small reduction in their costs. Both theory and
limited empirical evidence suggest that the highly concentrated merchant sector in Australia has
captured the reduction in interchange fees as profits and has not passed it on in the form of lower
consumer prices. (3) The per-transaction price at the point of sale has not changed significantly.
Merchants have not generally availed themselves of their right to surcharge card transactions
and the per-transaction price faced by consumers from their card issuers has not changed much.
Holding the number of cards fixed, the regulatory intervention has not altered prices in a way
that could achieve the intent of the intervention. (4) There is relatively little evidence thus far that
the intervention has in fact affected the volume of card transactions in Australia as intended by
the regulation. (5) In the short-run, the effect of the regulation has been to transfer significant
profits to the Australian merchant sector with that transfer being borne partly by bank issuers
and partly by cardholders. (6) Since proprietary systems such as American Express were not sub-
ject to the pricing regulations and since American Express can enter into deals with banks to is-
sue cards, banks have shifted volume from the regulated association systems to the unregulated
proprietary systems.
347
PLATFORM ECONOMICS:
348 Part Four : Payment Cards Essays on Multi-Sided Businesses
I. INTRODUCTION
This chapter examines the effect of a significant regulatory-mandated alteration in pricing policy in
a two-sided industry.1 In doing so it provides empirical evidence that is helpful for understanding
how two-sided industries work. It also sheds light on the interaction between the design of regula-
tory interventions and the pricing policies in two-sided industries.
In 2003 the Reserve Bank of Australia (RBA) mandated a reduction in the “interchange fee.”
In the context of a credit card transaction, this is the fee that the bank that acquires the receivable
from the merchant pays to the bank that issued the card to the consumer. Three associations of
banks in Australia had centrally set the interchange fee at around 0.95 percent of the transaction
value. The RBA imposed cost-based regulation that resulted in a reduction of the interchange fee
to around 0.55 percent. Absent any other adjustment, this 0.40 percentage-point reduction in
interchange fees eliminated roughly AU$490 million in revenues that banks would have received
from these fees in 2004.2
The credit-card industry is often cited as a classic two-sided product: it intermediates the
transactions between merchants and cardholders.3 This massive regulatory intervention there-
fore provides a natural experiment, almost, for assessing how an alteration in the price on one
side of a two-sided industry affects the other interdependent parts of the system. In addition,
more so than many interventions, the RBA’s mandated reduction in interchange fees is so sub-
stantial that it provides the hope that one can determine empirically whether the intervention
achieved its objectives.
After a number of studies, the RBA concluded that interchange fees that were “too high”
helped sustain card transaction prices (to consumers) that were “too low” from a social perspective.
This resulted in the overuse of cards relative to other, allegedly cheaper, payment instruments. The
main objective of the regulation, therefore, was to raise the price of credit transactions to cardhold-
ers and thereby reduce the use of cards. Of course, at the time of this writing this experiment has
only lasted for less than two years—it is possible at this time to assess only short-run effects.
In Section 2 we provide some background on the Australian credit-card industry. Then we
turn in Section 3 to a summary of the RBA’s theory for regulatory intervention. Section 4 provides
a brief discussion of the possible effects of this intervention based on economic theory. Section 5
presents our empirical analysis and Section 6 lays out its implications for the RBA’s intervention.
We end with conclusions and suggestions for further research in Section 7.
1
See Rochet and Tirole (2003) for the seminal work on two-sided markets, and Evans and Schmalensee (2005c) for a
review of the industrial organization of two-sided markets.
2
The reduction was 0.40 percent. According to the RBA, total nominal purchase volume on credit and charge cards
was about $147 billion in 2004. Furthermore, Bankcard, Visa and MC accounted for about 83.8 percent of total
volume in that year, i.e. they accounted for about $123 billion. See RBA (2005c).
3
See Baxter (1983), Rochet and Tirole (2002, 2003) and Schmalensee (2002).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 349
4
See Australian Bureau of Statistics (2001).
5
See Australian Bureau of Statistics (2003).
6
See Reserve Bank of Australia (RBA) and Australian Competition and Consumer Commission (ACCC) (2000), p.
15. For a general introduction to payment cards see Evans and Schmalensee (2005a). American Express reached an
agreement with AMP in 1998 to issue cards in Australia. More prominently, as we discuss below, major Australian
banks have signed up with American Express and Diners Club since the RBA’s interchange fee regulation.
7
Excluding subsidiaries. Source: RBA, at http://www.rba.gov.au/FinancialSystemStability/FinancialInstitutionsIn
Australia/the_main_types_of_financi al_institutions_in_aus.html
8
See Australian Prudential Regulation Authority (2005). The APRA calculates total deposits as the sum of transaction
deposit accounts, non-transaction deposit accounts, and certificates of deposit. It excludes intra-group deposits.
9
See RBA and ACCC (2000), p. 17.
PLATFORM ECONOMICS:
350 Part Four : Payment Cards Essays on Multi-Sided Businesses
other contract terms that are set centrally by the association or regulators, or an industry standard
that substitutes for an agreement.10
In Australia, Bankcard, MasterCard, and Visa adopted a multilaterally set fee.11 Members could
negotiate bilaterally if they wanted or rely on the multilateral fee. Almost all relied on the centrally
set fee. As mentioned, that fee was slightly less than one percent of the transaction amount before
the regulatory intervention.12
The interchange fee is a cost to the acquirer and is passed on to the merchant in whole or in
part. The merchant discount is the percent of the transaction that the merchant pays to the acquir-
er. The merchant service fees for the four card brands were the following just before regulation: for
Amex, 2.57 percent; for Diners Club, 2.35 percent; and for Visa and MasterCard, 1.41 percent.13
As of 2001, the number of merchants accepting MasterCard, Bankcard, and Visa cards was
about twice as large as the number of merchant accepting American Express. The merchant base for
Diners Club was apparently smaller than for American Express.14 At this time there were roughly
13 million credit cards in use by consumers in Australia.15 Table 1 shows the shares of each brand.
10
See Evans and Schmalensee (2005b) for further discussion on interchange fee setting.
11
See, for example, RBA (2001), p. 14.
12
According to the Joint Study, the average interchange fee that issuers received in 1999 was 0.95 percent. Strictly
speaking, interchange fees differed for Visa and MasterCard depending on whether the transaction was carried out
with the “card present” or not. As of the end of 2001, the electronic (card-present) rate for Visa and MasterCard was
0.8 percent and the rate for all other transactions was 1.2 percent. The Bankcard banks charged 1.2 percent on all
transactions, although apparently they had agreed to introduce an electronic rate of 0.8 percent toward the end of
2001. See RBA (2001), pp. 14-15.
13
See RBA, Merchant Fees for Credit and Charge Cards, available at http://www.rba.gov.au/Statistics/Bulletin/
C03hist.xls
14
See RBA (2001), p. 119.
15
See RBA and ACCC (2000), p. 15.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 351
16
See RBA and ACCC (2000).
17
The debit cards the RBA views as less costly are the EFTPOS PIN debit cards, not the Visa debit cards that are also
offered. MasterCard does not offer a debit card in Australia.
18
See RBA (2001), p. vi.
19
ibid, p. 116.
20
See RBA (2004), p. 7.
21
ibid, p. 7.
PLATFORM ECONOMICS:
352 Part Four : Payment Cards Essays on Multi-Sided Businesses
issuer, in turn, receives interchange fees as revenues. To increase this source of revenue it has incen-
tives to encourage consumers to use their cards and it does so by providing rewards and other in-
ducements. As a result, cardholders face negative prices for using cards and the price of using cards
at the point of sale is lower than their cost. This results in a misallocation of resources (the overuse
of cards and under-use of allegedly more efficient forms of payment) and a perverse distributional
effect (people who do not use cards at the point of sale subsidize cardholders).22
In September 2002, MasterCard International and Visa International challenged the RBA’s
decision in the Federal Court on procedural and jurisdictional grounds. The Court rejected the
challenge in September 2003, finding against MasterCard and Visa. The no-surcharge standard
came into effect on January 1, 2003. The standard on interchange fees came into effect on July
1, 2003. The Bank required the credit card schemes to publish and put in force the new inter-
change fees by October 31, 2003. For all practical purposes, in November 2003 interchange
fees declined from an average of around 0.95 percent to around 0.55 percent.23 The timeline is
shown in Table 2.
22
We do not dispute this analysis for the purpose of this article; see Evans and Schmalensee (2005b) and Rochet
(2005) for some comments. Our focus here is only on whether the RBA intervention achieved its goals.
23
See RBA (2004), pp. 8-9.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 353
from merchants will get passed on to cardholders depends on the degree of competition among
card issuers. Given that card issuing in Australia is relatively concentrated we would not expect full
pass through, at least in the short run.24
It is well known that in perfectly competitive markets there is full pass through of cost changes.
A $1 decrease in marginal costs will lead to a $1 decrease in price. With imperfectly competitive
markets the extent of pass through depends as a theoretical matter on the shape of the demand
curve.25 With linear demand curves there is less than a 100 percent pass through of costs to final
consumers. For the very small cost decreases involved here it is reasonable to assume linear demand
(since the curvature of demand can be ignored for such small changes) and therefore to expect less
than 100 percent pass-through as a matter of theory. Empirical studies tend to find less than 100
percent pass through more frequently than greater than 100 percent pass through; the greater-
than-100-percent pass through rate appears to happen in the empirical tax incidence literature and
the empirical effects are confounded with the sticky-price issue discussed below.26 Thus, we would
expect less than 100 percent pass-through as an empirical matter.
24
See below for a discussion of the theoretical and empirical evidence on the degree of pass through.
25
Cotterill (1998) provides a summary of the relationship between market structure and cost pass through. If the industry is
perfectly competitive and firms maximize profits, then the rate of pass through is 100 percent, no matter what the value of the
market elasticity of demand is. If the industry is a monopoly, demand for its product is linear, and the monopolist maximizes
profits, the pass through rate is less than 100 percent. More specifically, for a monopolist that faces a linear demand curve,
the pass through rate is 50 percent. See, for example, Bulow and Pfleiderer (1983). If, on the other hand, a monopolist faces
a constant-elasticity demand curve over the relevant range it is possible that pass through could be greater than 100 percent.
Assuming a linear demand curve is reasonable for small changes in costs but not necessarily for larger changes in cost.
26.
The empirical literature on pass through is fairly vast and covers a number of areas, including exchange rate and tax rate pass
through to prices. The literature on pass through of exchange rate changes to domestic prices of traded goods finds that the
median rate of pass through is roughly 50 percent for shipments to the US. See Goldberg and Knetter (1997). In the context of
the proposed merger between Staples and Office Depot, Ashenfelter et al (1998) made a distinction between the firm-specific
cost pass through rate and the industry-wide pass through rate—i.e., the reaction of the firm’s prices to changes in its own costs
and to changes in costs common to all firms in the industry. They found that the firm-specific pass through rate for Staples
was roughly 15 percent, whereas the industry wide pass through rate was around 57 percent. Furthermore, there is a literature
focused on pass through of trade promotions to retail prices, which is a form of firm-specific pass through. There are a number
of relevant studies in this area, including Chevalier and Curhan (1976), Walters (1989), Armstrong (1991), and Besanko, Dube,
and Gupta (2005). Besanko et al (2005) analyze the degree of pass through of trade promotions using scanner data for eleven
product categories at a Chicago supermarket chain. 8 out of 11 categories show pass through rates of less than 100 percent. At a
more disaggregated level, only 165 products out of 1164 (about 14 percent) show pass through rates significantly larger than 100
percent. Besanko et al (2005) cite a study of trade promotions according to which retailers themselves said that they pass through
roughly 62 percent of the promotion, use 24 percent to cover promotion costs, and keep the rest as profits. In the context of
this literature, Blattberg et al (1985) suggest that, as an empirical matter, most products display pass through rates much smaller
than 100 percent. Furthermore, Tyagi (1999) suggests that pass through rates of less than 100 percent should occur much more
frequently than pass through rates of more than 100 percent, since the set of demand functional forms that imply less than
100 percent pass through is quite large. It seems to us that pass through rates of more than 100 percent are less frequent as an
empirical matter than rates of less than 100 percent, and they tend to appear most often in the tax incidence literature. Poterba
(1996), for example, is unable to reject the hypothesis of 100 percent pass through for the commodities he examines, but Besley
and Rosen (1999) find greater than 100 percent pass through for several of the commodities in their sample. In line with our
discussion of sticky prices below, note that empirical estimates of pass through mix pass through effects and sticky price effects.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 355
There is a further consideration, from a theoretical perspective, on how the regulatory in-
tervention would affect final good prices. It is well known that many prices are sticky in the
short run (Stigler and Kindahl, 1970; Carlton, 1986). Of particular interest in our case is the
finding that prices tend to rise faster than they fall. A number of studies in the 1990s found that
retail prices respond faster to input price increases than to input price decreases (Karrenbrock,
1991; Newmark and Sharpe, 1992; Borenstein, Cameron, and Gilbert, 1997; Jackson, 1997).
More recently, Peltzman (2000) confirmed this finding with a large sample of consumer and
producer goods. This is especially important for considering short-run versus long-run effects of
the intervention.
The credit-card acquiring business in Australia is highly concentrated. The four major banks
accounted for 93 percent of credit card transactions acquired in 1999.27 In contrast, the share of
the four major acquirers in the United States is around 41 percent.28 As we note below, the RBA
has reported that the reductions in the interchange fee appear to have been fully passed through
to merchants. This is not surprising because, despite the relative concentration of acquirers, many
merchants are large customers and can bargain effectively with acquirers. In addition, the acquir-
ers are aware that their actions have been monitored closely by the RBA. As noted above, credit-
issuing is also highly concentrated. We would therefore not anticipate full pass-through and, at
least for the short-run, that is what we find.
Comprehensive government statistics on merchant concentration are not available for
Australia. We have identified what data we could find from a range of sources. Many merchant
categories appear to have significant levels of concentration. For example, within their respective
categories, the top department store had a 71 percent share (and the top two had an 83 percent
share) in 2003;29 the top three supermarket and grocery stores had a 75.4 percent share in the late
1990s;30 the top two mobile telephone operators had a 78 percent share in 2005; the top land-line
telephone operator had a 75 percent share in 2005;31 and the top two airlines had an 83 percent
share in 2005.32
27
See RBA and ACCC (2000), pp. 16-17.
28
These calculations are based on data from HSN Consultants Inc. (2003).
29
See Euromonitor International (2005a).
30
See http://www.aph.gov.au/senate/committee/retail_ctte/report/c04.htm. The Australian Bureau of Statistics re-
ported the 75.4-percent figure, which was based on sales of all supermarkets and grocery stores, including the non-
petrol sales of convenience stores at petrol stations. Furthermore, Euromonitor International (2005b) estimated
that, in 2003, the size of the total grocery stores/food retailers/supermarkets sector was AU$ 72.5 billion and the
supermarket sub-sector represented 60.5 percent of that. The top two supermarkets had a combined market share
(in the total sector) of 37.9 percent. Therefore, they had a combined market share in the supermarket sub-sector of
62.6 percent.
31
See Maxwell (2005).
32
See Bureau of Transportation Statistics (2000).
PLATFORM ECONOMICS:
356 Part Four : Payment Cards Essays on Multi-Sided Businesses
There are significant differences between the cardholder and merchant sides of the business
that likely affect relative pass through rates to consumers on each side. The regulatory intervention
caused issuers to experience a significant reduction in revenue (which could also be viewed as an
increase in costs). That makes menu costs and other sources of sticky prices less binding. Moreover,
it appears that it is quite easy for issuers to adjust prices by varying service and other card fees. For
example, it seems that average annual fees on standard and gold cards changed (in a non-trivial
manner) every year between 2000 and 2004.33 Lastly, linear demand is a more reasonable assump-
tion for the merchants who experience a small relative cost decrease than for issuers who experience
a relatively large cost increase (decreased interchange fees).
The main source of friction for issuers concerns the annual fee and other prices that issuers
commit to in trying to persuade people to take their cards or to switch from another card. Cards
are replaced about every three years; consequently that is the opportunity to institute or increase
annual fees for current cardholders. For new solicitations it is easy to change fees whenever the
solicitation goes out.
Merchants on the other hand experienced a relatively small reduction in cost. If fully passed
through by acquirers, the interchange fee reduction amounts to less than half a percent of their
selling price (and only on those transactions that take place on credit cards). The evidence on price
rigidities, and particularly the one on asymmetric price responses cited above, makes one doubt
that such a small cost reduction would affect final goods prices very quickly, even if there were
extensive retail competition. We return to this point below.
33
See RBA (2005a).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 357
Our best estimate on timing is that there were at least some changes in bank behavior in
early 2003 in anticipation of the regulatory intervention.34 The annual report for one of the ma-
jor banks, ANZ, noted that in 2003 it had reshaped its “product set across the Australian Cards
Issuing portfolio to address the impact of the Reserve Bank interchange reforms.”35 There are a
number of newspaper articles along the same lines in the period prior to October 2003.36 The
Sunday Telegraph (Sydney), for example, reported in its September 21, 2003 edition that the five
major banks had increased credit card fees by up to 50 percent in the previous 12 months. The
article goes on to quote a bank executive who expressed that the rise in fees had the goal of making
up for the loss in interchange income that would happen after the implementation of the regula-
tory scheme.37 In our analyses below, we consider whether we see any effect of regulation starting
in 2003 and, alternatively, starting in 2004.
We analyse two questions. (1) How did the intervention affect prices to issuers, cardholders,
merchants, and consumers? (2) What was the effect of the intervention on card use? Appendix A
provides information on our data sources and detailed statistical results for interested readers.
34
Some changes in fees may have been happened already in 2002.
35
See ANZ (2003), p. 31.
36
See, for example, McKinnon (2001), Hanna (2002), Brammall (2002, 2003), Horan (2003), and Graeme (2003).
37
See Horan (2003).
38
At the time of this writing the information on other service charges and finance charges was available only through
the second quarter of 2004.
39
As noted above, all figures are reported in real 2004 Australian dollars.
PLATFORM ECONOMICS:
358 Part Four : Payment Cards Essays on Multi-Sided Businesses
Interchange-fee regulation was implemented in the fourth quarter of 2003. If we compare the
first two quarters of 2003 with the first two of 2004, we find that issuers lost, on average and in real
terms, about AU$9.35 per card in interchange income per quarter, a loss of about 40 percent on
total interchange income per card of about AU$23.52 per quarter in the first two quarters of 2003.
A more interesting way to look at this loss is to ask how much the issuers would have made
in interchange income in the absence of regulation. From this perspective, we calculate a “but
for” interchange income per card under the assumption that the regulatory scheme was not im-
plemented and the interchange rate remained at an average of 0.95 percent. We then subtract the
actual interchange income per card from the “but for” interchange income per card and we find
that, on average, in the first two quarters of 2004 issuers lost about AU$10.31 per quarter per card.
This represents a loss of about 42 percent with respect to the interchange income they would have
obtained in the “but for” world. Figure 1 shows these facts.
The aevidence
calculate “but for” thus suggests that
interchange issuersper
income started
card losing,
under inthereal terms, roughly
assumption between
that the AU$9.30
regulatory
and AU$10.30 per card per quarter in interchange income as a consequence of the RBA interven-
scheme
tion. Inwas not implemented
the months that followedandthethe interchange
introduction rateregulation
of the remained(andat an average
likely of 0.95 that
in the months
preceded
percent. We thethen
regulation as well)
subtract the they
actualrecovered between
interchange 30 andper
income 40 percent
card from of that
theloss through
“but for” the
imposition of higher fees.
interchange
Appendix income per card
A provides theand we find
support for that, on average,Ainregression
this conclusion. the first two
modelquarters of 2004 the
that compares
level of
issuers real
lost other
about service charges
AU$10.31 per card
per quarter perafter
card.regulation with the
This represents levelofbefore,
a loss about 42controlling
percent for
seasonal effects, changes in the real purchase volume per card, and changes in unobserved fac-
with respect to the interchange income they would have obtained in the “but for” world.
tors over time, suggests that the regulation (or its anticipation) was accompanied by a rise in
Figure 1 shows
real other these
service facts. of roughly 30 to 40 percent of the interchange loss amount. (As we
charges
Figure
Figure1 Real
1. Real Interchange
Interchange Revenue
Revenue per Card
per Card
30
AU$)
(2004AU$)
25
Card (2004
20
Interchange per Card
15
10
Interchange
Actual
5
But-For
0
92 93 94 95 96 97 98 99 00 01 02 03 04
92
93
94
95
96
97
98
99
00
01
02
03
04
19 19 19 19 19 19 19 19 20 20 20 20 20
19
19
19
19
19
19
19
19
20
20
20
20
20
T he evidence thus suggests that issuers started losing, in real terms, roughly between
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 359
discuss further in the appendix, the quarterly Visa data we used track closely with annual RBA
data on fees. The increase in issuer annual and service fee revenue per card from 2001 to 2003
was AU$20.84 from the RBA data and AU$22.78 from the Visa data.) A similar model for real
finance charges per card produces positive coefficients for the post-regulation period. The coef-
ficients, however, are for the most part not statistically significant. Furthermore, we found no
evidence of a structural break (associated with the RBA regulation) in the interest rate that issu-
ers charge on outstanding balances.
The components of the “other service fees” variable from the Visa data are annual fees and
service fees. Annual fees are fixed with respect to transaction volume. Service fees (such as late
payment fees and over-limit fees) are also primarily fixed fees. A late payment fee, for example, is
independent of the amount charged.40
In addition to the increases estimated above for these fixed fees, there has also been an impact
on the level of rewards offered by issuers. The RBA has reported that the average reward decreased
from 0.8 to 0.6 percent for most of the bank schemes.41 While rewards (miles or points accumu-
lated as a function of purchase volume) affect marginal incentives to use cards, the way in which
issuers have implemented changes has likely limited the impact on purchase volume. First, some
of the changes have been in the form of caps on the number of reward points accumulated. For
example, National Australia Bank (NAB) capped rewards on its Visa rewards card. It offers 1 point
for each dollar spent up to AU$3,000 per month. Above the $3,000 a month spending thresh-
old, cardholders receive 1 point for every two dollars spent, up to a maximum of 13,000 points a
month. For cardholders below the AU$3,000 threshold, the marginal incentive to use cards has
not changed as a result of the imposition of the cap.42
For customers who are likely to exceed the monthly cap on a regular basis, the marginal incen-
tives have changed on the NAB Visa rewards card. However, one of NAB’s other responses to the
regulation was to partner with American Express, in offering a rewards card without a cap (in fact
offering 1.5 points for each dollar in the first year). Similarly, Westpac now offers an American
Express rewards card and ANZ offers a Diners Club rewards card, both without rewards caps and
40
The over-limit fee may be partially variable, in the sense that a consumer may be more likely to exceed the credit
limit at higher purchase volumes. But it is not clear that higher over-limit fees would significantly affect consumer
purchase volume on credit cards. Among other things, over-limit fees would appear to be more directly related to
accumulated revolving balances as opposed to purchase volume (although the two are related).
41
See Testimony of Dr. Philip Lowe, House of Representatives Standing Committee on Economics, Finance and
Public Administration (EFPA) (2005), at p. 26-27. There is insufficient detail on this data point for us to conduct
a detailed analysis. The RBA does not report, for example, how this estimate is constructed, nor does it provide a
data series over time (or even which years are the beginning and end points). It is also unclear whether this includes
rewards for cards issued on the American Express and Diners Club systems.
42
The cost of having rewards cards has increased, which may result in fewer consumers deciding to hold them, but con-
tingent on having a card, the marginal incentive to use a card has not changed for consumers below the threshold.
PLATFORM ECONOMICS:
360 Part Four : Payment Cards Essays on Multi-Sided Businesses
both more generous than the MasterCard/Visa rewards cards from the respective banks.43 As we
note below, while the interchange fee regulation appears to have had relatively little effect to date
on overall card volume, it does appear to have switched cards from MasterCard and Visa to the
American Express and Diners Club systems.44
What we have estimated is the short-run impact. Since cardholders are valuable assets—issuers
spend considerable resources to acquire new cardholders and pay a premium on portfolio pur-
chases—one would expect issuers to be cautious in raising prices quickly to their existing mature
cardholder portfolio. Over the longer run, as cardholders switch from one issuer to another, one
would expect price effects to be more fully realized.
43
See, Westpac Bank, at http://www.westpac.com.au/internet/publish.nsf/Content/PBCCCSCR+Altitude; ANZ
Bank, at http://www.anz.com/australia/support/library/MediaRelease/MR20030912.pdf
44
See EFPA (2005), p. 25. See also, Diners Club of Australia website, at http://www.dinersclub.com.au/s06_media/
p62_view.asp?id=144
45
See RBA (2004), p. 9.
46
We have not made any attempts to verify this independently. In any case, according to the RBA, total nominal purchase
volume on credit and charge cards was about $147 billion in 2004. According to the RBA, Bankcard, Visa and MC
accounted for about 83.8 percent of total volume in that year, i.e. they accounted for about $123 billion. The reduction
in the merchant discount was 0.0042 percentage points, which gives savings to merchants of roughly $517 million.
47
The RBA claimed in testimony before the Australian House of Representatives in 2005 that these cost savings would
be passed on almost entirely to final consumers. It reasoned that the retail sector overall was not concentrated even
though particular segments such as supermarkets were. This analysis ignores the point, of course, that the extent of
pass through depends on the structure of the relevant market, which is not overall retail but the individual markets
that comprise it. Testimony of Mr. Ian MacFarlane, House of Representatives Standing Committee on Economics,
Finance and Public Administration (EFPA) (2005), at pp. 23-24.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 361
servative, we take the card share of retail transactions to be 50 percent, which will likely overstate
the card share for many merchants, especially those in the service sector.48 Taking this 50 percent
estimate, then the average reduction in overall merchant costs as a result of the interchange fee
reduction was 50 percent of 0.42 percent or 0.21 percent.
There are three reasons, mentioned earlier, why we would expect that prices to consumers
would have fallen by less than 0.21 percent. First, this decrease in costs amounts to roughly 8 cents
on an AU$40 transaction. The empirical literature on price rigidities makes it doubtful that a de-
crease in cost of this small magnitude would be passed on to consumers quickly. Second, many of
the significant retail markets in Australia are highly concentrated. Given the small decrease in cost
it is reasonable to approximate the demand curve facing these merchants with a linear demand
curve. In this case, pass through rates will be less than 100 percent. Third, it seems that the empiri-
cal evidence on pass through—while not specific to Australia—finds rates of less than 100 percent
more often than rates of more than 100 percent.49 With a 50 percent pass through rate, the reduc-
tion in prices to consumers from the RBA’s interchange fee reduction would be 0.105 percent.
That amounts to roughly 4 cents on a AU$40 transaction. It would not require much in the way
of price rigidities for merchants to decide not to adjust prices in the short run.
The very little empirical evidence there is suggests that, in fact, merchants have tended not to
pass through the reduction in the merchant discount to consumers in the form of lower prices.
Cannex, an independent research group, surveyed merchants in Australia regarding the impact
of the interchange fee regulation on their regular business practices.50 Among merchants who
reported a change in the merchant discount during the previous year, less than 5 percent declared
that they had reduced prices to consumers. On the other hand, more than 20 percent reported
that their profits had increased and almost 60 percent reported that they had not experienced any
changes in their regular operations. Surcharging for credit card transactions, following the repeal
of non-surcharge rules, is still the exception in Australia, as it has been in other countries.51 One
survey of Australian merchants in November 2004 found that only 2.3 percent of all merchants
surcharged, with larger merchants slightly more likely to surcharge, at slightly over 5 percent.52 A
48
The 25 percent figure of share of cards of total consumption may be an underestimate of the percent of retail transac-
tion dollars on cards since the consumption figure includes some non-retail merchants where cards are not commonly
used. Unfortunately, more precise data are not available for Australia. In the United States, where the card share of total
consumption is only slightly higher than Australia, the portion of transactions at merchants paid for with cards is slightly
under 50 percent for retail and travel and entertainment merchants, and less than 10 percent for service merchants.
49
See supra notes 25-26.
50
See Cannex Australia (2004).
51
Only about 10 percent of merchants imposed surcharges in the Netherlands, and only about 5 percent of merchants
imposed surcharges in Sweden. Surcharging is also uncommon in the United Kingdom. See Evans & Schmalensee
(2005b), p.27.
52
See Network Economics Consulting Group (NECG) (2005), p. 46, citing East & Partners 2004, “Australian mer-
chant acquiring and cards markets: Multiclient market analysis report,” December.
PLATFORM ECONOMICS:
362 Part Four : Payment Cards Essays on Multi-Sided Businesses
different survey found that 7 percent of all merchants surcharged regularly.53 It is also worth not-
ing that the average surcharge was 1.8 percent, which is higher than the merchant discount fee
on credit card transactions, and also almost surely higher than any differential between the costs
to merchants of processing credit cards versus other forms of payment. This suggests that at least
some of the surcharging that takes place may be opportunistic and does not increase the efficiency
of relative prices for payment instruments facing consumers.
53
The survey reported that 19 percent of merchants “sometimes” surcharged, but it is unclear how commonly and on
what criteria they did so. Two other surveys that asked incidental questions on surcharging reported that 12 percent
and 2.9 percent, respectively, of respondents surcharged (although the frequency of surcharging was unknown). See
NECG (2005), at p.44.
54
In this calculation we are excluding revenues derived from the so-called “interest margin.”
55
According to the RBA data, the regulated systems accounted for 86.3 percent of the total value of credit card
purchases in September 2003. In June 2005 they accounted for 83 percent. See the market-share data available at
http://www.rba.gov.au/Statistics/Bulletin/C02hist.xls
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 363
35.0
35.0 $140
$140
10/00:Publication
10/00:Publication
30.0 of Joint
of Joint Study
Study $120
30.5 $120
25.0
25.0 $100
$100
(%)
Billions
08/02:
Rate (%)
08/02:
AU$, Billions
20.0
20.0 Publication of no-
Publication of $80
$80
Growth Rate
surcharge
no-surcharge
and
and interchange
interchange
AU$,
Growth
15.0
15.0 standards
standards $60
10.0
10.0 12/01: $40
12/01:
Publication of
Publication of
Consultation
Consultation 10/03: New IFIF
10/03: New
5.0 $20
Document implemented
implemented
0.0 $0
$0
1995
1995 1996 1997
1997 1998
1998 1999
1999 2000
2000 2001 2002
2002 2003 2004
2004
Year
Year
Growth rate
Growth rate Real Purchase
Real Volume
Purchase Volume
Figure 2 shows the level and growth rate of real purchase volume on credit cards along with
Table 4. Annual Growth Rates of Indicators of Credit Card Use, All Credit Cards, 1996-2004
key points in the timeline for the regulatory intervention. Despite the massive degree of the RBA’s
intervention, there appears to be no evidence Purchase Real
Purchaseof any effect of the intervention
Real on the use of cards.
Transaction Purchase
Table 4 goes into more detail by showing the trends over
Accounts Transaction time in several
s per measures
Purchase of card
Volume peruse.
s
Other than the number of accounts, the indicators of credit-cardVolume
Account use grew at an increasing rate
Account
through 1999 and then grew at a declining rate through 2002. Assessing the impact of the RBA’s regula-
1996 7.7 14.2 6.1 13.7 5.6
tion depends on our assumptions about card industry performance in the absence of regulation. For ex-
ample, if the1997
question is whether
5.6 the regulations
19.7 stopped the 13.3growth of card
23.3ownership and16.7 usage, then
the answer is1998
“no”—all card 6.1
metrics continued
30.8
to grow post-regulation.
23.2
Our
31.7
main focus is on two differ-
24.2
ent questions, namely (1) whether growth rates were lower post-regulation than pre-regulation and (2)
whether the 1999 5.9
decline in the growth rates that32.5 25.1pre-regulation
was taking place 33.0accelerated or
25.6
decelerated.
The aggregate
2000 data suggest7.0 that (1) 24.0industry growth 15.9was lower 23.5
after regulation than before,
15.5
with the exception of the number of accounts, and (2) the decline in the rates of growth that
20011999 has continued,
started around 2.5 16.9
again 14.0
with the exception of the 16.4 13.5
number of accounts, but has
not accelerated.
2002 These basic1.5empirical regularities
13.9 are confirmed
12.2 by our
12.4detailed analysis
10.7 discussed
below. The first comparison would say that regulation has lowered the growth rates, while the sec-
2003 4.7 8.9 4.0
ond would say that the decline in the growth rates was already taking 9.9 4.9
place and the regulation had
2004 6.0 8.7 2.5 9.2 3.0
Source: all the growth rates have been calculated on the basis of RBA (2005c).
PLATFORM ECONOMICS:
364 Part Four : Payment Cards Essays on Multi-Sided Businesses
Table 4 Annual Growth Rates of Indicators of Credit Card Use, All Credit Cards, 1996-2004
Purchase Real Real Purchase
Purchase Transactions Purchase Volume per
Accounts Transactions per Account Volume Account
1996 7.7 14.2 6.1 13.7 5.6
1997 5.6 19.7 13.3 23.3 16.7
1998 6.1 30.8 23.2 31.7 24.2
1999 5.9 32.5 25.1 33.0 25.6
2000 7.0 24.0 15.9 23.5 15.5
2001 2.5 16.9 14.0 16.4 13.5
2002 1.5 13.9 12.2 12.4 10.7
2003 4.7 8.9 4.0 9.9 4.9
2004 6.0 8.7 2.5 9.2 3.0
Source: all the growth rates have been calculated on the basis of RBA (2005c).
relatively little impact. Our prior is that the second comparison is more appropriate, as we observe
a significant slowdown in the rate of growth in the pre-regulation period. We acknowledge, how-
ever, that our analysis does not come close to explaining the dramatic shifts in industry output that
have taken place— either the accelerating growth leading up to 1999 or the decelerating growth
following 1999. Therefore, we believe it is at least possible that absent regulation, growth rates of
the different card metrics might have levelled off, or even increased.56
In order to address the first question—i.e., whether growth rates were lower post-regulation—
we estimated a number of simple regression models of the quarterly growth rates of the credit
card use variables on a set of quarterly binary variables, a binary variable that takes on the value 1
between 2000 and 2002 and 0 otherwise,57 and a binary variable that takes on the value 1 starting
in 2003. The results are reported in Table 4a.58
56
For example, the entry of Virgin as an issuer in 2004 could have increased output, all else equal. To the extent that
is true, our comparisons would underestimate the impact of the RBA’s regulation. We do not have detailed and
systematic data on industry concentration, or on entry and exit, to attempt to control for these factors.
57
In the accounts model this binary variable takes on the value 1 in 2001-2002 rather than in 2000-2002.
58
All the regressions include binary variables for the second, third, and fourth quarters on the right hand side. When
the growth rate in the number of accounts is the dependent variable, we control for changes in population and
income per capita. In the case of the number of purchases and real purchase volume, we control for changes in real
consumption. In the case of the number of purchases per account and real purchase volume per account, we control
for changes in consumption per capita.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 365
The interpretation of these results is straightforward. In the case of the number of accounts,
growth rates in the regulated period have been no different from what they were through 2000 and
higher than they were in 2001-2002. In the case of all other variables, however, growth rates in
the regulated period have been lower than they were through 1999 and also lower than they were
in 2000-2002. From this perspective, the regulated period has been associated mostly with lower
growth rates in the indicators of credit card use.
PLATFORM ECONOMICS:
366 Part Four : Payment Cards Essays on Multi-Sided Businesses
We then addressed the second question—i.e., has the decline in the growth rates that started in the
late 1990s accelerated or decelerated with regulation? We used regression methods to examine more
carefully whether credit card activity grew more slowly than it would have in a world without the RBA
regulation. We studied the growth patterns in greater detail by regressing the growth rates of the relevant
variables on a linear spline function.59 We performed the analysis with annual data and quarterly data.
With quarterly data, we calculated proportional growth rates between quarter q and quarter (q-1).60
We describe the model on the basis of quarterly data. We first created a linear trend t that
starts at 1 in the first quarter for which growth rates can be calculated and grows by 1 every quarter
through the first quarter of 2005. We then defined the spline time dimension variables as follows:
X(t) 5 t, t 5 1, 2, …, T;
Y(t) 5 max (0, t 2 a);
Y(t)’ 5 max (0, t 2 b) ; and
Z(t) 5 max (0, t 2 c).
In this model, a is the number that corresponds to the last quarter of 1999 in the trend se-
quence, b is the number that corresponds to the last quarter of 2000, and c is the number that
corresponds to the last quarter of 2002.
Think of a world where the number of credit card accounts, the number of credit card pur-
chases, and the real purchase volume on credit cards are functions of a set of demand shifters. For
example, the number of credit card accounts is a function of population and income per capita,
and the number of credit card purchases is a function of real consumption. Then in order to assess
whether there have been significant changes in the trend of the growth rates of the credit card use
variables, we can estimate models of the following form:
GCC(t) 5 a1 1 d11 X(t) 1 d12 Y(t) 1 d13 Z(t) 1 1 C(t) 1 u1(t), or
GCC(t) 5 a2 1 d21 X(t) 1 d22 Y(t)’ 1 d23 Z(t) 1 g2 C(t) 1 u2(t).
In these models, GCC(t) stands for the growth rate of the credit-card-use variable in question
and C(t) stands for the growth rate of the relevant demand shifter. These models focus on the
percentage-point change in growth rates over time.61
59
On linear spline functions see, for example, Poirier (1976), ch. 2; Poirier and Garber (1974); Johnston (1983), p.
392ff; and Greene (1993), p. 235ff. For a study that uses linear splines in the context of testing for the existence of
unit roots in economic time series, see Perron (1989). For a discussion of the various Perron models, see, for exam-
ple, Enders (2004), pp. 200-207.
60
We also performed the analysis with the growth rates calculated between quarters q and (q-4). Although the magni-
tude of the coefficients was, of course, different, the substance of the conclusions we draw did not change much.
61
The results we obtain do not change in any substantive manner if we exclude the demand shifters from the estimated
regressions.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 367
The di1 parameters (for i 5 1, 2) capture the trend in the growth rate between the starting point
and 1999 (or between the starting point and 2000). The di2 parameters capture the change in the
trend of the growth rate over 1999-2002 relative to the previous period (or 2000-2002 relative to
the previous period). The di3 parameters, finally, capture the change in the trend of the growth rate
over 2003-2004 (the regulatory period) relative to 1999–2002 (or 2000–2002).
The estimated parameters with the p-values calculated on the basis of the Newey-West standard
errors are reported below in Table 5.62
The results, based on quarterly growth rates, show the following. All of the models estimated
show a positive point estimate for the trend that starts in 2003. In three of the models, the relevant
coefficients are statistically significant—the exceptions are purchases per account and real purchase
volume per account. These models convey the idea either that the trend of growth of the relevant
card-use variable accelerated during the regulated period (e.g., number of accounts) or that the
decline in the trend of growth that had started in the late 1990s decelerated during the regulated
period (e.g., number of purchases).
The two models that raise some questions are the ones estimated in the growth rate for
the number of purchases per account and real purchase volume per account. They both show
a positive coefficient for the trend that starts in 2003 but the coefficients are not statistically
significant. (A simple examination of Table 4, furthermore, shows that these two series seem to
have exhibited considerably lower rates of growth in the post-regulation period than in the pre-
regulation period as compared to other indicators of credit card use.) Whether the regulation
had any impact in terms of curtailing purchases per account and purchase volume per account
is thus an open issue.
Overall, taking into account existing trends in the growth rates of card variables, there is lit-
tle evidence that the regulatory intervention has affected overall card use in the admittedly short
period of time examined here. Figure 3 below, however, reveals that there have been important
compositional changes—volume has shifted from the regulated systems to the unregulated ones.63
More specifically, between October 2003 and June 2005 the share of American Express and Diners
62
All the regressions include binary variables for the second, third, and fourth quarters on the right hand side. When
the growth rate in the number of accounts is the dependent variable, we control for changes in population and
income per capita. In the case of the number of purchases and real purchase volume, we control for changes in
real consumption. In the case of the number of purchases per account and real purchase volume per account, we
control for changes in consumption per capita. We also tried estimating the models via instrumental variables—we
instrumented the growth in real consumption and the growth in real consumption per capita. We calculated the
Newey-West standard errors for the IV-estimated coefficients. Overall, as far as the trend coefficients are concerned,
the results were not substantially different from the ones obtained via OLS with Newey-West standard errors. We
also calculated feasible-generalized-least-squares (Prais-Winsten) estimates. The conclusions we draw did not change
under the FGLS approach.
63
Analysis of Visa data on card and volume also indicates that Visa output declined relative to total industry output
following regulation.
PLATFORM ECONOMICS:
368 Part Four : Payment Cards Essays on Multi-Sided Businesses
Table 5 Regression of the growth rate of the credit-card-use variables on a linear spline
function on the basis of quarterly data, 1994-2005 (P-values calculated on the
basis of Newey-West standard errors)
Coefficient P-value
Growth in Number of Accounts
d(1) 0.0001 0.592
d(2) 20.0018 0.001
d(3) 0.0039 0.000
Growth in Number of Purchases
d(1) 0.0021 0.000
d(2) 20.0066 0.000
d(3) 0.0044 0.003
Growth in Number of Purchases per Account
d(1) 0.0018 0.001
d(2) 20.0049 0.000
d(3) 0.0010 0.499
Growth in Real Purchase Volume
d(1) 0.0020 0.002
d(2) 20.0063 0.000
d(3) 0.0044 0.005
Growth in Real Purchase Volume per Account
d(1) 0.0016 0.014
d(2) 20.0047 0.001
d(3) 0.0010 0.509
Club increased by roughly 21 percent in terms of the number of purchases and by roughly 19
percent in terms of the value of purchases. This has happened because, as a consequence of regula-
tion, relative prices seem to have changed. First, the relative price that issuers receive from a Visa or
MasterCard transaction has declined relative to, say, an American Express transaction, since inter-
change has been capped for Visa and Master Card but American Express is allowed to sign issuing
deals with banks under no interchange ceiling. Therefore, issuers have an incentive to shift volume
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 369
18.0
16.0
14.0
Diners Club
12.0
10.0
8.0
6.0
4.0
2.0
0.0
Jan-02
Mar-02
May-02
Jul-02
Sep-02
Nov-02
Jan-03
Mar-03
May-03
Jul-03
Sep-03
Nov-03
Jan-04
Mar-04
May-04
Jul-04
Sep-04
Nov-04
Jan-05
Mar-05
May-05
Quater
Number of Purchases Value of Purchases Volume
to American Express, and they have. Secondly, the price of a card (or account) that cardholders
face may have increased for the Visa and MasterCard cards relative to the proprietary ones, since
Visa and MC issuers have attempted to recover some of the lost interchange income via an increase
in other service charges, most of which are fixed fees. (Of course, in light of the fact that Visa and
MasterCard issuers have raised other service charges, American Express and Diners Club may have
done the same, so it is not altogether clear whether this relative price has changed much or not.)
Note, by the way, that the shift in volume from the regulated systems to the unregulated ones has
a “perverse” effect on prices. In other words, the merchant discount on the regulated systems has come
down significantly and the merchant discount on the unregulated systems has declined slightly, but it
is considerably higher than the associations’ pre-regulation merchant discount. Although the pure price
effect leads to a lower average merchant discount, the compositional-change effect leads to a higher
average merchant discount. (Of course, it will take an extremely large compositional change effect to
produce an average merchant discount rate that is higher than the average pre-regulation rate.)
One predictable result is that a massive reduction in revenue from one side of a two- sided
market had consequences on the other side. Banks lost roughly $490 million in interchange fee
revenue but appear to have regained between 30 and 40 percent of that through increased mem-
bership fees for cardholders. We believe this is a lower bound on the portion passed on to cardhold-
ers because of the cycle of replacing cards.
Another predictable result is the absence of evidence that consumer prices have fallen as a result of
lower merchant discounts. This is not surprising because the cost savings are too small to be measurable
with any degree of confidence. However, based on the economics literature on pass-through effects, we
believe that it is highly unlikely that consumers have received any significant benefit over the period
of time considered given the likely sticky prices and high concentration in the Australian retail sector.
Two results are, viewed prospectively, surprising. It appears that issuers have chosen thus far to
adjust their pricing structures mainly through fixed fees—i.e., fees that are independent of transac-
tion volume. That is so despite a reduction in the variable revenue they received from the merchant
side that had translated into negative variable prices on the cardholder side. Related to this is the
finding that the near halving of interchange fees has not thus far led to a substantial reduction in
card use.64 That is to be expected for now given that marginal prices for completing transactions
with cards have not changed significantly as a result of the regulation.
Over time we would expect a reduction in total credit-card use on the regulated side as a result of
the decrease in the number of cardholders. However, it is difficult to predict the magnitude of that re-
duction without knowing the responsiveness of cardholders to increased membership fees. As noted,
offsetting this decrease will be the increased use of cards from the unregulated proprietary systems. It
appears that especially for high-spenders on rewards cards, issuers have made a serious, and predict-
able, effort to move them from the regulated associations to the unregulated proprietary systems.
The RBA’s hope that regulating the associations’ interchange fees would exert substantial downward
pressure on American Express and Diners Club’s respective merchant discount rates is particularly
unlikely to be realized as high-spend rewards cardholders move to the unregulated systems.
VII. CONCLUSIONS
Did the RBA regulatory intervention achieve its goals? One answer is that it is too soon to tell. It
takes time for markets to adjust and it may be that banks and merchants will make further adjust-
ment over time. But to the extent that the first couple of years provide information, the evidence
indicates that the intervention has not achieved its goals for a reason that was not apparent to
either the RBA or the card associations at the time the intervention was being debated.
64
As we note above, growth rates of card output are lower post-regulation than pre-regulation. Although our prior is
that this is a continuation of pre-existing declines in the growth rate, we do not rule out the possibility that growth
rates might have levelled off absent regulation, and that the regulation has therefore lowered card growth.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 371
The purpose of the intervention was to raise the cost of transacting with credit cards, to bring
social costs and benefits in alignment, and thereby to reduce the use of what was thought to be
an inefficient instrument. The RBA saw the interchange fee as the source of the problem because
banks subsidized card use to get this revenue source and merchants could not make cardholders
bear the differential costs of card use. If the intervention had resulted in banks raising usage fees
in lockstep with the reduction in interchange fees or if merchants had raised surcharges to account
for the interchange fees, the RBA would have achieved its objectives.
Neither expectation was fulfilled. For the most part, it seems that banks raised fixed fees and
left the per-transaction incentives alone. In addition, the banks started to switch volume from the
regulated to the unregulated systems. By and large, merchants have not chosen to impose sur-
charges. Thus far the incentives that cardholders face to use cards at the point of sale do not seem
to have changed radically. If these patterns persist, the effects of the RBA regulation, if any, will
likely take place through a reduction in total cards on the regulated systems.
This result does not just provide further support for the law of unintended consequences. It
also raises interesting research questions. For the study of two-sided markets, it emphasizes that
further work is needed to understand the role of two-part tariffs in guiding membership and us-
age decisions. This is an important topic for many two-sided industries since these two-part tariffs
are quite common. For the study of regulatory interventions, it emphasizes that we need to know
more about how two-sided businesses set prices to design interventions in those industries that can
accomplish specified goals.
VIII. APPENDIX A
In order to determine whether issuers have made an attempt at recovering from cardholders some
of the income lost due to the RBA regulation, we posed the question: In the period associated with
the RBA regulation, is there evidence of a jump in the level of the card fees that issuers charge
cardholders?
An examination of the Visa Australia series for other service charges per card and finance
charges per card revealed that a few data points dramatically break the series trends. In the case of
other service charges per card, the outliers are located in the second and third quarters of 1999 and
the second quarter of 2000. In the case of finance charges per card, the outlier is located in the first
quarter of 2004. To our knowledge, these are data problems rather than reflections of true change
in the economic time series.
We took two different approaches in order to preclude our estimation from being driven by
a few “influential” observations. First, we replaced the outliers with observations obtained via
linear interpolation (Newton method). The results we report below were obtained on the basis of
the interpolated data. Second, we kept the outliers in the dataset but ran robust regressions a la
Huber (1964). In this iterative approach, observations receive weights that are a function of the
PLATFORM ECONOMICS:
372 Part Four : Payment Cards Essays on Multi-Sided Businesses
magnitude of the associated residuals. For any given specification of the relevant model, the results
obtained via robust regression were not substantially different from the ones obtained on the basis
of the interpolated data.
In equation (A1), Y(t) is the level of the fees that issuers charge cardholders in each quarter, t
is a trend variable that takes on the value 1 at the starting point and grows by 1 each quarter, D(t)
is a binary variable associated with the regulated period, and X(t) is a vector of other variables that
may have had an impact on the level of Y(t).
More specifically, we want to isolate the behaviour of issuers—therefore, we have to control for
those elements of the behaviour of cardholders that could have had an impact on Y(t). For exam-
ple, in a model where Y(t) is real other service charges per card (including annual fees and other
usage fees) we may want to control for real purchase volume per card (in case there are elements of
Y(t) that vary with the level of purchase activity). In a model where Y(t) is real finance charges per
card we will certainly want to control for the level of real outstanding balances per card. The trend
variable picks up the effect of unobservable factors that may have changed over time and may have
exerted an influence upon Y(t). We also include binary variables for the second, third, and fourth
quarters in order to control for potential seasonality in Y(t).
We estimated a number of different models to probe how robust our results were to various
specifications. In some of these models we took into account the fact that real purchase volume per
card and real outstanding balances per card may be endogenous—in other words, cardholders may
choose the optimal level of these variables by taking into account the issuers’ choice of fees and
interest rates. In other words, we checked whether we obtained considerably different results with
OLS vis-à-vis an instrumental-variable approach. For space reasons, we do not report the results
of all the estimated models here. The results not reported were not substantially different from the
ones reported and they are available from the authors upon request.
In estimating models like (A1), researchers usually take two potential issues into account—
the presence of serial correlation and the presence of stochastic trends. We dealt with serial cor-
relation by estimating the models with robust standard errors—more specifically, Newey-West
standard errors with four lags. We also used feasible generalized least squares (Prais-Winsten)
estimation but obtained results not substantially different from those reported here. We checked
for the presence of stochastic trends by running Dickey-Fuller tests and co-integration tests on
the relevant v ariables.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses The Effect of Regulatory Intervention in Two-Sided Markets 373
65
Incorporating the real 30-day interest rate on the right hand side generated an estimated coefficient of 3.18 (p-value
5 0.002) for the reform binary variable.
66
Incorporating the real 30-day interest rate on the right hand side generated an estimated coefficient of 3.90 (p-value
5 0.001) for the reform binary variable.
67
See RBA (2005a)
PLATFORM ECONOMICS:
374 Part Four : Payment Cards Essays on Multi-Sided Businesses
cards in this survey, fees increased by $24 a year for standard cards and $30 for gold cards. If we
include one late fee and one over-limit fee penalty per year, total fees increased by $48 a year for
standard cards and $54 a year for gold cards.
There are at least two reasons why these survey data on rewards cards indicate higher fee in-
creases than the calculations we have reported based on the aggregate data (leaving aside the fact
that the limited annual data series do no allow for full controls). First, cardholder spending is
higher on rewards cards, so the loss in interchange fee income is greater. Second, replacing existing
cardholders with new ones is costly and banks may be reluctant to raise fees as much to existing
cardholders as to new cardholders.
14 Invisible Engines
David S. Evans
ABSTRACT
The use of software platforms to drive innovation and transform industries has exploded in the
four years since the publication of the English-language edition of Invisible Engines in 2006.
Around the globe, invisible engines are ushering in a new era of software-based technological
change. The Apple iPhone has shaken the mobile phone industry worldwide in part by creating
a massive applications business built on the phone’s operating system. Firefox has revolution-
ized the browser industry by encouraging web developers to write add-ons and in doing so top-
pled Microsoft’s Internet Explorer from dominance in many countries. Facebook has created a
powerful social networking platform by opening itself up to developers. Amazon has released
a cloud-computing platform that enables entrepreneurs to access its vast software, hardware
and global communication systems over the Internet. A less well known company, IP Commerce,
is starting to transform the payments business in the United States by helping developers build
applications that work with the diverse hardware and software than handle the various types of
payments.
In this preface to the 2010 Chinese edition of Invisible Engines we will survey what has hap-
pened since we finished the book and explain why software platform-based business models
continue to create enormous social value, while often producing great profits for the companies
and entrepreneurs behind them.
Invisible engines are based on software code. Software programs are written in various languages.
Working by themselves or with other programs they tell hardware – from the pixels on your iPhone
to the chip on your desktop to the communication devices providing your internet connection –
what to do. They are the brains behind everything from the card reader in which you swipe your
debit card to your social network page to the complex trading decisions at hedge funds to your
favorite word processing package.
Software programs can become platforms that support other software programs when code
that is written to perform a particular task could be made available to other software programs so
the writer does not have to write that code again. That is what Windows does. It contains many
little programs that help developers write their own programs. Microsoft provides a link to that
376
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invisible Engines 377
1
David S. Evans and Richard Schmalensee, Catalyst Code: The Strategies behind the World’s Most Dynamic Companies
(Boston: Harvard Business School Press, 2007). Multi-sided platforms are sometimes referred to as “two-sided mar-
kets” even though they are usually businesses rather than markets.
PLATFORM ECONOMICS:
378 Part Five : Software Platforms Essays on Multi-Sided Businesses
that could support third-party applications. That meant it had to create APIs which would enable
outsiders to link into the iPhone’s operating system. These APIs give developers access to blocks of
code that manipulate various aspects of the iPhone. Apple then provided a “software developer kit”
(SDK) to help programmers develop applications using the iPhone’s APIs and put together a web
site to provide other resources for developers.
An important set of business dynamics starts when such an operating system is opened up and
thereby becomes a software platform that can support developers. Apple initially provided more
than enough applications and features on the iPhone to persuade many people to get one. By
opening up its operating system it then started a virtuous circle. Entrepreneurs and hobbyists saw
the opportunity to write interesting applications that could use the features of the iPhone, includ-
ing its large touch screen and connectivity. The iPhone users soon found that they could download
applications that made their phones even more useful than they were when they first bought them.
More people started buying iPhones because of the increasingly valuable applications. As the num-
ber of iPhone users increased, developers had even more incentives to write applications.
Apple has not taken this process for granted. Like most other successful platforms it has stoked
the fires. It developed a marketplace, an online shopping mall, to help developers make their appli-
cations available to iPhone users and earn revenues from them. Apple is responsible for collecting
money from iPhone users and then gives the developers about 70% share of the revenue. Users
have benefitted from having effectively a single location to search for applications. Developers have
benefited because Apple has reduced the transactions costs of selling and has assembled an audi-
ence of users.
Apple has obviously made enormous profits and seen the market value of the firm increase be-
cause of the success of the iPhone and its application store. iPhone users have done very well also as
a result of the tens of thousands of applications they can choose from, some of which enable them
to do things that were not possible before. New markets have opened for entrepreneurs that use the
iPhone, and some of them are earning significant profits or obtaining fame. By bringing developers
and users together the iPhone has served as a catalyst which has created value—and profit—out of
thin air. Others have followed suit. The most significant follower is Google, which has introduced
the Android software platform for mobile phones, has stimulated hardware makers to introduce
phones using this platform, and has worked hard at persuading developers to write applications for
its store. China Mobile set up its own application store in the summer of 2009.
The invisible engine model followed by Apple’s iPhone has helped power the information tech-
nology industry for about three decades. Apple itself was one of the pioneers in encouraging de-
velopers to write applications for its personal computer operating system. It invented the “software
evangelist”. Microsoft, though, was the maestro catalyst. Its Windows software platform attracted
thousands of software developers and hundreds of millions of users. As a result it has been the du-
rable center of a vast computer based ecosystem since at least the launch of Windows 3.0 in 1990.
Invisible Engines tells the story of how software platforms have transformed industries including
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invisible Engines 379
computers, video games, and handheld devices and then accelerated innovation. It then examines
the forces behind the various business models that have been adopted in these industries. Many
software platforms have decided not to charge developers while others have. Some have decided to
vertically integrate into hardware, while others have remained pure software vendors.
The development of the internet and the spread of high-speed broadband throughout the
world have generated a new wave of software platforms. Several of the major web businesses have
turned themselves into platforms. The heart and soul of any web business does not reside in a serv-
er farm somewhere or in its buildings—it lies in the thousands and thousands of lines of software
code that enable people to see and interact with web pages. Once written these software programs
can be opened up to others by exposing APIs that enable developers to use portions of that code to
interact with the web properties.
Facebook has allowed developers to write applications that work inside its social network. As
a result developers can write games - such as Farmville – which Facebook users can play, or online
shopping and advertising programs - such as those developed by FreeCause for raising money for
charities. By the end of 2009 there were more than 500,000 applications running on Facebook.
This five-year old site has also developed a program called Facebook Connect. Developers can write
applications that through APIs made available by Facebook enable users of those applications to
pull their Facebook identity, friends and privacy to other websites. As of the end of 2009 there
were more than 6,000 developers working on Facebook Connect applications. These applications
make Facebook more valuable to its users, and the greater number of users makes Facebook more
valuable to developers. Facebook makes money in large part by selling access to the eyeballs drawn
to Facebook and these applications to advertisers.
Amazon is one of several companies that are competing to create a general purpose software
platform that can be accessed over the internet. Such platforms are said to reside “in the cloud”
because they are off in the distance as opposed to on the user’s desktop or other client computer.
The idea is to develop a set of services that application developers can use. These include software
services that allow developers to use the code rather than writing their own. But they also involve
access to vast server farms and communication networks around the globe. Amazon’s Web Services
plays to this company’s strength in selling products over the internet. It provides special services to
developers who are writing programs to help merchants operate virtual stores and receive payment.
Invisible engines are beginning to disrupt the payments business both online and offline. This
industry is at the heart of commerce and makes it possible for businesses and people to exchange
value around the world. By definition it involves essentially all the money in the world. If invisible
engines start driving innovation in this industry they could have an enormous impact on consum-
ers and businesses who transact with each other.
An intricate set of systems and an enormous group of businesses are involved in payments.
They range from the companies that print paper checks, to ones that make the point-of-sale ter-
minals where you swipe your debit card, to banks that operate deposit accounts, to clearing and
PLATFORM ECONOMICS:
380 Part Five : Software Platforms Essays on Multi-Sided Businesses
settlement networks. At the beginning of 2010 this industry involves many different computer sys-
tems, each with a software program, that have been made to interoperate with each other. Consider
what happens when you swipe your credit card. The terminal sends your card information along
with details of the transaction to a switch. That switch, which may be operated by a variety of busi-
nesses, has a software program that decides what to do with the transaction. It will send it on to a
merchant processor which keeps track of the details of the transaction. This processor will also act
as an intermediary with a clearing and settlement system, which will contact the bank processor
that acts on behalf of the bank that issued the card (or possibly the bank directly). Many software
programs are involved in the various steps of this process.
What we have just described is one “rail” in the payments system—the one for credit cards.
While this may vary by country, in the United States there are other rails for debit cards, paper checks,
electronic funds transfer, and other “tender” types. Now suppose that you are an entrepreneur who
has come up with a great idea that involves incorporating payments into your application. This could
be something as simple as a software program that helps small businesses accept payment in multiple
ways and integrate these payments into basic accounting software. To succeed you would have to
integrate your application into the many software programs that are used for the many tender types
taken by businesses. That is such a daunting task—one involving substantial money and time not
to mention the cooperation of the other businesses that handle payments—that you would give up.
This is a perfect problem for an invisible engine to solve. Massive transactions costs make it
hard for payments entrepreneurs and payments users to get together. A software platform can
lower those costs by investing in linking to the multitude of software programs that handle various
elements of payments. By exposing APIs, this software platform then makes it possible for entre-
preneurs to quickly integrate into most relevant aspects of the payments business. Such a software
platform is analogous to Windows which, among other things, enables software applications to
work with the multitude of device drivers controlling various hardware peripherals such as com-
puter hardware, printers, and cameras.
In the United States several companies are working to create software platforms for payments.
The most ambitious effort has been that of Denver-based IP Commerce. They have developed a
Windows-like software platform enabling developers to link into software programs such as mer-
chant processing platforms and point-of-sale equipment that account for the bulk of transactions
in the United States. Developers have already used IP Commerce’s invisible engine to create new
products. PaySimple created the payments software for small businesses mentioned above. A small
firm can install their web-based application and quickly accept payments offline and online with
multiple tender types. Of course, IP Commerce is a new player and it remains to be seen whether
it will succeed in generating the kind of virtuous circle that has made the fortunes of other software
platforms and their creators.
There is nonetheless a great deal of evidence that the software platform model has caught on in
payments. Amazon and PayPal are competing to offer payments platforms to merchants for online
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invisible Engines 381
transactions. They have made software services available to developers through APIs. Application
developers for merchants can use these APIs to make their applications capable of accepting pay-
ments using the PayPal or Amazon payments system. It remains to be seen whether the card net-
works will consider opening up their powerful software programs to outsiders and trying to build
an ecosystem of applications around themselves.
Invisible engines are the unsung heroes of the economic progress that we have seen in the last
three decades. It is true that these engines could not accomplish much without the massive im-
provements in microprocessor speeds, computer storage, and communication capabilities. But all
those accomplishments would not have amounted to much without the development of software
languages; the creation of software platforms that eliminate the duplication of effort in writing
code; and the catalyst business model that enables the virtuous circle between users and applica-
tions.
Based on what has happened in the last four years it appears that we are entering a new age in
the development of invisible engines. We suspect that historians will look back and define 1980-
2004 as the first age of invisible engines. That is the period when software platforms helped create
vast ecosystems around “thick clients” in which most of the work was done on a local computing
device. The second age began after 2004 with the spread of Web 2.0, the rapid growth of smart
mobile phones, and the global increase in broadband penetration. How long this second age will
last we do not know, and it is even harder to guess what will come after it. With confidence,
though, we can say that invisible engines will create new industries, destroy old ones, and drive
innovation in China and elsewhere in the world.
David S. Evans
Andrei Hagiu
Richard Schmalensee
Boston, 23 January 2010
CHAPTER
ABSTRACT
Two features of the technology we described in the last chapter shape the economics of soft-
ware platforms. Software platforms are a written product of the mind. They are in effect docu-
ments, usually written in a high-level computer language. The code involved is malleable. It can
be moved, altered, added to, and subtracted from with great ease. It is created almost entirely
by people—“almost” because, like composers and writers, most programmers use computers
for help. Software platforms are inherently multisided. They usually serve distinct groups of cus-
tomers, who benefit from having each other on the same platform. Application Programming
Interfaces (APIs) forge the crucial relationship between application developers and end users.
Two features of the technology we described in the last chapter shape the economics of software
platforms.
Software platforms are a written product of the mind. They are in effect documents, usually
written in a high-level computer language. The code involved is malleable. It can be moved, al-
tered, added to, and subtracted from with great ease. It is created almost entirely by people— “al-
most” because, like composers and writers, most programmers use computers for help.
1
Eric Raymond, The Cathedral and the Bazaar (Sebastapol, Calif.: O’Reily Press, 1999).
382
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 383
Software platforms are inherently multisided. They usually serve distinct groups of customers,
who benefit from having each other on the same platform. Application Programming Interfaces
(APIs) forge the crucial relationship between application developers and end users. The developer
can benefit from using APIs when she can sell the resulting software to users who have those APIs
on their computing devices.
Although we examine both of these features in this chapter, the multisided nature of software
platforms is a main focus of the remainder of the book and the economic aspect of these invisible
engines from which we will glean many insights. We conclude this chapter with a discussion of an-
other remarkable aspect of the software business: people working collaboratively over the Internet,
often without pay, produce software, including software platforms, that compete with software
produced by for-profit firms.
degree, in computer science. Software programmers and related professionals who worked in the
U.S. software industry had an average of 15.3 years of education as of 2000. That compares with
13.8 years for the workforce on average and 14.7 years for professional service industries (including
law, medicine, accounting, and architecture).2
There were 1,194 degree programs in computer science in American colleges and universities
in 2006.3 These programs usually offer courses in the design of operating systems. All of the top
ten programs as ranked by U.S. News and World Report did.4 There are a number of textbooks on
the design of computer operating systems and related topics.
Microsoft is notable for screening people for intelligence and problem-solving skills. A 1995
study reported that Microsoft recruited from the top fifty colleges and universities and hired less
than 3 percent of the people it initially interviewed. Microsoft is famous for asking job candidates
to solve problems on the spot, such as estimating the number of gas stations in the United States.
As of 2004, over 95 percent of the architects, designers, and programmers working on Windows
had a college degree and 40 percent had computer science degrees. Ten years later Google has de-
veloped a reputation as a company where only brainiacs need apply.5 It advertises mainly in tech-
nical magazines and puts people through numerous interviews that test intellectual skills before
hiring one out of the 200 candidates who send in a résumé.
Made of Malleable Code We have already seen that software programs, including platforms, are
a series of instructions usually written in a language such as C++. Managing the creation of mil-
lions of lines of code that work together as planned is no mean feat. But one reason these programs
have grown so large is that they have been designed to make it easy to add to them. In some re-
spects, doing so is like adding a paragraph to a chapter of a book or another riff when playing jazz.
The key difference is that since the ultimate product is digital, users do not experience the addition
in the way that adding a chapter makes a book thicker.
Like the contents of a newspaper, the contours of a software program can be changed readily.
Just as newspapers have suburban or regional editions that add coverage specific to a particular
geographic area, modern software programs may have versions targeted to particular groups, such
as Java for small devices. And just as newspapers can add sections to bring in more readers, so soft-
ware programs can add features or functionality.
2
http://www.bls.gov/oco/ocos110.htm.
3
http://www.usnews.com/usnews/edu/college/tools/brief/cosearch_advanced_ brief.php
4
“America’s Best Graduate Schools 2005 Edition,” U.S. News and World Report, December 31, 2004.
5
Michael A. Cusumano and Richard W. Selby, Microsoft Secrets (London: HarperCollins, 1996), pp. 92–93;
Internal Microsoft information; John Battelle, The Search: How Google and Its Rivals Rewrote the Rules of Business and
Transformed Our Culture (New York: Portfolio Press 2005); http://www.cbsnews.com/stories/2004/12/30/60minutes/
main664063.shtml.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 385
Many operating systems added features in the mid-1990s that helped users communicate over
networks such as the Internet. Apple’s Macintosh included AppleTalk proprietary networking pro-
tocols in 1985 and added protocols for communicating over the Internet in 1995.6
Easy to Reproduce All information goods are easy to copy. But software programs, including
platforms, are especially so because they are necessarily digital. The easiest way to see this is with
the open-source operating system Linux. You can download this 5.7 million-line operating system
over the Internet from numerous Web sites.7 With a cable modem connection it takes a couple
of minutes. Not surprisingly, piracy is a major problem for software firms that sell their products.
Software platforms are often installed on computer hardware before it is sold. The manufac-
turer does this itself when it makes both the software and the hardware. Apple zaps its software
right onto its iPods and Apple computers. Or the software manufacturer may license the software
to other manufacturers that install it, often from a single master CD. Distribution costs are higher
when the software platform is sold directly to users. Thus, Microsoft incurs some costs in reproduc-
ing its Windows software on CDs and distributing it through retail channels. The same is true for
Linux distributors such as Red Hat. But even in these cases the per-unit costs are relatively low, as
with other information goods such as music CDs.
Complementarities and Network Effects System components are generally complements: add-
ing another component to a system or improving an existing component generally increases the
value of the other components. Moreover, in many cases, systems have what economists call in-
direct network effects linked to the presence of components.8 That is, an increase in the number of
users of one component often makes that component more valuable as a complement to the other
components. As Sony’s Internet–based game center for the PlayStation 2 draws more users, for
instance, more PlayStation 2 owners will want to buy games supported by the Internet service. As
its games become more popular, more consumers will prefer PlayStation 2 consoles to competitors’
6
Jim Carlton, Apple: The Inside Story of Intrigue, Egomania, and Business Blunders (New York: HarperCollins,
1997), p. 59; “Apple goes to the core; Apple introduces Power Macintosh 9500 that uses TCP/IP and PCI bus archi
tecture; Product Announcement; Brief Article,” LAN Magazine, October 1, 1995.
7
http://zdnet.com.com/2100-1104-864256.html.
8
Michael Katz and Carl Shapiro, “Systems Competition and Network Effects,” Journal of Economic Perspectives 8
(Spring 1994): 99. For a general discussion of network effects and their business implications, see Carl Shapiro and
Hal Varian, Information Rules (Boston, Mass.: Harvard Business School Press, 1998).
PLATFORM ECONOMICS:
386 Part Five : Software Platforms Essays on Multi-Sided Businesses
models. Likewise, an increase in the variety of components (printers for PCs, for example) often
increases the value of other components to end users. In recent years economists have tended to
apply the multisided platform framework to these situations, as we discuss in the next section.
There also may be direct network effects. These arise when an increase in the number of users
of an application or platform directly makes that application or platform more valuable to each
user. Its value increases because users can share information and work together more efficiently.
When WordStar was the leading PC word-processing program, many people bought it at least in
part because they could share documents with friends and co-workers.
Intellectual Property Protection If people could get the source code for any software product,
they could reproduce it for next to nothing. The price would fall to almost zero, and the original
writer would derive no financial benefit.
Software companies rely on all three major forms of intellectual property protection to guard
their investments against this fate.9 First, they keep the source code secret as much as possible.
Before they distribute the software they turn it into the 1s and 0s of machine code. In principle, an
able, dedicated, and patient programmer could translate machine code back into a high-level lan-
guage. But this sort of decompiling is forbidden by almost all commercial software licenses and all
but impossible in practice for multi-million-line software platforms. In addition, “trade secret” law
protects software developers from the unscrupulous employee or agent who might try to release the
source code without authorization.
Software companies also copyright their code. As with a book, you cannot reproduce copies of
software programs without violating copyright law. Of course, as with other information goods, pi-
racy is nonetheless rampant, especially in countries with weak intellectual property laws. In India,
an estimated 80 to 85 percent of the copies of Macromedia Flash and Dreamweaver in use are not
legal. Even in the United States, almost 30 percent of Macromedia software is pirated.10
Finally, software companies get patents on algorithms and other features. The United States
had granted about 127,000 software patents through 2004. (It takes about 3.5 years on average for
the grant of a patent application.11) Apple’s iTunes software, for example, allows users to import
9
Suzanne Scotchmer, Innovation and Incentives (Cambridge, Mass.: MIT Press, 2004), chap. 3.
10
Venkatesh Ganesh, “The continuing story of software piracy,” The Financial Express (August 8, 2005), http://www.
macromedia.com/devnet/logged_in/ swozniak_piracy.html.
11
It is difficult to define “software patents,” but in 2004, 127,098 patents were granted under the G06F classification,
which covers electronic data processing. Data set available from www.uspto.gov.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 387
Average Costs
Total Costs
an unlimited number of audio tracks and encode them into the popular MP3 format, as well as
listen to MP3s, audio CDs, or hundreds of Internet radio stations. A patented system for accessing
digital media across networks was important for the success of iTunes.
Economies and Diseconomies of Scale Although no one has ever quantified it, it is gener-
ally understood that there are diminishing returns to scale in writing software platforms. That is,
doubling the size of a platform by adding more features more than doubles the cost (holding the
quality of the code constant—one can always write inefficient code). Increases in size create more
interdependencies (with N objects, there are N2 possible pairwise interactions, for instance), thus
raising the likelihood of bugs, and thereby raising development, debugging, and test costs more
than proportionately.12 (Object-oriented programming and the use of modules are designed to
temper these diseconomies.) But once created, cheap reproduction means that additional copies
cost little. The production technology is therefore as shown in Figure 1.
These economic features suggest some caution in characterizing the marginal cost of producing
software. It is true that once the costs of creating software have been sunk, the marginal costs of
reproducing and distributing it are very low. That is an ex post perspective on cost. But it is also
true that the likely adoption of a software program is not independent of the costs that are incurred
in creating it or revising it. Software designers add features in part to bring in more consumers; in
the case of software platforms, those consumers include both users and developers of applications
that run on top of the platforms. Ex ante, the marginal cost of acquiring additional customers by
improving the platform, is likely to be much higher than the marginal cost of reproducing and
distributing the software.
12
“Linux: Fewer Bugs Than Rivals,” Wired News, December 14, 2004.
PLATFORM ECONOMICS:
388 Part Five : Software Platforms Essays on Multi-Sided Businesses
This distinction is also relevant for other information goods. Movies are a good example. The
cost of making a completed movie available to an additional viewer is close to zero. However, the
number of viewers a movie garners is partly dependent on the investment in the actors, special
effects, and other features that make a movie popular. Movies made with low budgets are often
aimed for a narrow audience, while movies with blockbuster potential typically have extravagant
budgets. The marginal cost of garnering a viewer, viewed ex ante, is positive.
Pricing and the Recovery of Investments As with all information goods, software poses some
challenges related to recovering investments and earning profits. Pricing at ex post marginal cost or
anything close to it would lead to bankruptcy. Software pricing thus depends primarily on demand
(particularly the responsiveness of demand to changes in prices) rather than on cost and has as its
main goal at least recovering fixed and sunk development costs. The pricing of software platforms
is considerably more complex because of their multisided nature, as we discuss later.
Why have software prices not declined at the same pace as hardware prices? Basically because
software development costs have not declined as rapidly as hardware costs, for two related reasons.
The first is that educated labor, which is not becoming cheaper, accounts for most of the cost of
producing software. In 2001, U.S. software firms paid about 33 percent of revenues to their em-
ployees, while semiconductor companies paid less than half that percentage.13 The second reason is
that software products are becoming more complex: with advances in hardware, software programs
typically grow over time through the accretion of features. A typical PC game program in 1994 was
20 megabytes; a typical PC game program in 2004 was 2,200 megabytes.14
Bundling Features Most goods are bundles of features, many of which could be provided sepa-
rately but are not. Cars come with spark plugs and tires even though you could buy your own.
Moreover, many goods are improved over time through the addition of features. Few cars come
these days without air conditioners and rear window defrosters.15 Many cereals add fruit and flavors
over time, leading to many variations.
The same is true for computer systems. Microprocessors, memory, and other components are
typically combined to create a hardware platform such as a Nokia mobile phone handset or an
Xbox game console. With time, many peripherals come to be integrated into the hardware plat
form. Consider the case of the math coprocessor, which facilitates number crunching. Before the
release of Intel’s 486 chip, Intel’s microprocessors did not include a math coprocessor. Customers
13
2002 Economic Census, Industry Series Reports, available at http://www. census.gov.
14
“Doom II Game Has 500,000 Pre-Orders,” Newsbytes News Network, October 13, 1994; http://www.amazon.com/
exec/obidos/tg/stores/detail/videogames/ B00006C2HA/tech-data/103-8949918-3544622.
15
David Evans and Michael Salinger, “Why do Firms Bundle and Tie? Evidence from Competitive Markets and
Implications for Tying Law,” Yale Journal on Regulation (January 2005): 37–89.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 389
who wanted one purchased it separately from one of several vendors at substantial cost. Today, one
cannot buy an Intel x86 processor without a built-in math coprocessor.
Creating products through feature addition is particularly easy with information goods. That
is the beauty of the pliability of music and language. Pop music was mainly distributed as singles
on seven-inch records until the early 1960s. The success of the Beatles’ Sergeant Pepper’s Lonely
Hearts Club Band album made clear the value of compiling songs and helped make the market
for long-playing albums. Bundling multiple songs into albums became standard practice, and the
distribution of songs as singles became less common. Newspapers have added various features such
as style and living sections over time.
Similar forces apply to software in general and platforms in particular. Where exactly the tasks
performed by software are accomplished is a matter of business and design decisions. Many tasks
that used to be performed by stand-alone applications have become integrated into other applica-
tions (such as spell checkers, which originally were sold separately from word processing programs)
or into the software platform itself. Early operating systems, for example, did not include com-
munications functionality.
The malleability of code reinforces several economic forces that encourage the inclusion and
accretion of features in products.
Bundling and integration. Combining features in a single product reduces transaction costs for
consumers. Rather than having to buy two products, they can buy just one. Moreover, the manu-
facturer can create additional value by creating connections between the features. An example is
making the features of a spreadsheet program available to a word-processing program.
Economies of scope. When there are fixed costs in offering separate products, firms may find
it profitable to bundle those products if demand for the separate components is not particularly
strong. Several major automobile makers, for example, have decreased the number of different cars
people can purchase.16 They have done this by developing bundles of options that “most people”
want, even though some people would not value some of those options separately. For software
there are cost savings from combining several features into a single package, as well as savings in
distribution and product support. (There may be diseconomies, of course, if making the program
larger and thus more complex results in disproportionately large increases in the costs of writing,
debugging, testing, and supporting the package.)
Demand aggregation. When there are fixed costs of producing and distributing products but
low marginal costs of adding components, it may be possible to lower average costs and reduce
variation in what people are willing to pay by combining components that appeal to different
groups of customers.17 Hardware and software typically include many features that most con-
16
Ibid.
17
Yannis Bakos and Eric Brynjolfsson, “Bundling and Competition on the Internet,” Marketing Science, 1 (Winter
2000): 63–82.
PLATFORM ECONOMICS:
390 Part Five : Software Platforms Essays on Multi-Sided Businesses
sumers never use. However, by including these features vendors expand the number of consum-
ers who find the product valuable at the offered price. This is why many word processors include
equation editors, newspapers have horoscopes, and cable companies include channels that most
of us never watch.
payment. Standards sometimes give rise to two-sided platforms. Fax machines facilitate commu-
nication between senders and receivers. Cooperatives of firms also operate two-sided platforms—
Visa is the most significant example. For-profit businesses operate two-sided platforms in a wide
variety of industries and in many economically significant ones. Highly visible examples include
American Express (travelers checks and charge cards), Google (search engine–based portal), and
News Corporation (advertising-supported media).
William Baxter presented one of the first formal analyses of a two-sided business in 1983.18 He
was a law professor who was self-taught in economics. He observed that payment cards provided
a service only if both cardholders and merchants jointly agreed to use a card for a transaction. He
derived some of the fundamental economic consequences of this joint demand. (Baxter went on to
become a highly innovative antitrust chief in the United States.)
The notion, however, that diverse industries are based on two-sided platforms and are gov-
erned by the same basic economic principles is due to a pathbreaking paper by Jean Tirole and
Jean-Charles Rochet that began circulating in 2001.19 They showed that businesses such as com
puter operating systems, dating clubs, exchanges, shopping malls, and video game consoles were
two-sided.
Economists now recognize that many industries, including the manufacture of software plat-
forms, are guided by economic principles that differ in important ways from those that govern tra-
ditional industries. Many of these two-sided or multisided industries are subject to network effects,
which were studied extensively by economists during the 1980s and 1990s.20 Network effects are
also central to the economics of multisided platforms, and more recent analysis provides additional
insight into their business implications.
Internalizing Externalities Multisided platform businesses tend to arise in markets that have
three characteristics:
18
William Baxter, “Bank Interchange of Transactional Paper: Legal and Economic Perspectives,” The Journal of Law
and Economics 26 (October 1983): 541–588.
19
Jean Tirole and Jean-Charles Rochet, “Platform Competition in Two-Sided Markets,” Journal of the European
Economic Association 1, no. 4 (2003): 990–1029.
20
Shapiro and Varian, Information Rules. See the articles in Symposium on Network Externalities, Journal of Economic
Perspectives, 8 (Spring 1994): 93–150.
PLATFORM ECONOMICS:
392 Part Five : Software Platforms Essays on Multi-Sided Businesses
As a practical matter, multisided platforms tend to arise when a stronger version of condition 2
applies: most platform businesses exhibit indirect network externalities. Consumers, for example,
get more value from their credit cards when more merchants take them, and merchants get more
value from accepting credit cards when more consumers use them. This has not been lost on the
card systems. The current advertising slogans highlight merchant acceptance: “Visa. Everywhere
you want to be.” “MasterCard: No card is more accepted.” American Express, MasterCard, and
Visa persuade merchants to pay for taking their cards by emphasizing the millions of consumers
that have these cards and want to use them to pay. The sales pitch for the merchants is similar: then
the card systems tout the number of cardholders they have who could transact at the merchant if
it accepted the card for payment.
Customer groups can sometimes get together without a platform. Men and women have found
each other without matchmakers. Buyers and sellers figured out ways to transact before there was
money. Merchants can advertise their wares without the media. Successful multisided platforms,
however, generally reduce the transactions costs that members of different customer groups would
incur in trying to reap the benefits of getting together.
The fact that a platform could exist does not mean that it necessarily will or that it will provide
the only method for providing benefits to customers. Apple has thus far operated its iPod/iTunes
platform as a single-sided business.. It buys music by paying publishers royalties and distributes
this music to customers who want it. Similarly, many consumers have “store cards”—payment
cards issued by stores such as Bloomingdale’s. In fact, the payment card industry was based entirely
on this single-sided model until Diners Club introduced a card in 1950 that put multiple mer-
chants and consumers on the same platform.
Similarly, many businesses deal with multiple diverse groups without being platforms. There
is a sense in which auto companies bring tire manufacturers and consumers together, but they
do not do so in a way that makes Toyota, for example, a multisided platform business. In this
case there is no direct interaction between the two sides. Toyota substitutes itself for consumers
when dealing with tire producers, just as Apple does before sending music to consumers through
iTunes. By contrast, two-sided platform businesses provide support for direct interaction be-
tween the two sides. Thus, game developers sell directly to PlayStation users, for instance, not
through Sony.
Multisided businesses can generate profits for themselves and benefits for their customers if
they can figure out ways to increase and then capture indirect network externalities. There are three
major ways in which they do this.
First, they serve as matchmakers. Financial exchanges such as NASDAQ and on-line auction
sites such as eBay match buyers and sellers. The Yahoo! Personals and 8MinuteDating match men
and women.
Second, they build audiences. Advertising-supported media do mainly that: they use content
to attract eyeballs and then sell access to those eyeballs to advertisers. Many platforms engage in
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 393
less overt audience building. Auction houses such as Sotheby’s try to build an audience of buyers
for the art they sell on consignment. Nightclubs sometimes try to build an audience of women for
men, and vice versa. We saw that payment card systems try to build an audience of cardholders for
merchants and an audience of merchants for cardholders.
Third, they reduce costs by providing shared facilities for the customers on each side. That’s the
shopping mall case with which we began. But other platforms also do this to some degree. Buyers
and sellers have shared auction institutions and auction sites from the Roman forum to eBay.
Readers and advertisers share the pages of Vogue. Payment card systems provide a shared network
for conducting transactions between merchants and consumers.
Software platforms provide value through matchmaking and building audiences, as well as
through reducing duplication. Apple, for example, helped bring commercial artists and developers
of design software together. It did this by including services in the Mac OS that developers could
use to develop programs such as Adobe Photoshop for commercial artists. Sony PlayStation has
developed an audience of console users that it can make available to game developers. The main
economic value of software platforms, however, is in economizing on the amount of code that
developers must write to serve the needs of consumers.
The Pricing Balancing Act In single-sided markets, price usually tracks costs and demand for
the product pretty closely. Firms figure out what their marginal cost will be and then mark it up—a
little if customers are price-sensitive because there is a lot of competition, more if there is little
competition. Particularly in stable markets, this is not rocket science. That is why how-to books
on starting your own small business can offer reliable advice, such as “charge X times cost in sector
Y.” For example, one guide advises that the markup is generally 40 percent of the retail price in
hardware stores and that for jewelry it ranges between 400 and 800 percent.21
In multisided markets, pricing is more complicated because of indirect network effects be-
tween the distinct customer groups. If you charge women the same price as men to enter your
singles club, you may not get enough women. If this happens, men will not come, and suddenly
you will have an empty club. Many Internet publications discovered that viewers deserted in droves
when they attempted to charge them, although some did make the successful transition to paid
subscriptions plus advertising.22
21
Stephen C. Harper, The McGraw-Hill Guide to Starting Your Own Business, 2nd ed. (New York: McGraw-Hill, 2003),
pp. 100–101; Jan Kingaard, Start Your Own Successful Retail Business (Irvine, Calif.: Entrepreneur, 2003), pp. 152–153.
22
Michael Liedtke, “Online Subscriptions Herald the End of Web Freedom,” Associated Press newswires, March 18,
2002; Thomas E. Weber, “Web Users May Balk at New Fee Services That Deliver Little Value,” Wall Street Journal,
April 8, 2002; Timothy J. Mullaney, “Sites Worth Paying For? The Paid Web Is a Work in Progress, But Some Are
Already Getting It Right,” Business Week, May 14, 2001.
PLATFORM ECONOMICS:
394 Part Five : Software Platforms Essays on Multi-Sided Businesses
Multisided platform economics shows that it may make sense for firms to charge very low
prices to one or more groups or even to pay them to take the product. And that is what multisided
businesses do. Magazines, newspapers, and television broadcasters typically earn the preponder
ance of their revenues from advertisers.23 Charge card companies such as American Express earn
the bulk of their revenue from merchants.
Businesses in multisided markets often subsidize one side of the market to get the other side on
board—sometimes explicitly by charging low or negative prices. At other times subsidies are less
apparent, such as when the platform makes significant investments in one side and does not charge
for it. Table 1 shows some examples. We will see that all software platforms make services available
to at least one side for free. Most make free services available to developers through the APIs.
The economics of pricing for multisided platform businesses has another key implication. In
single-sided businesses, the principle that the one who causes the cost should pay the cost is good
23
Lisa George and Joel Waldfogel, “Who Benefits Whom in Daily Newspaper Markets?” NBER Working Paper no.
7944 (October 2000), p. 9.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 395
advice, for businesses as well as for policymakers. For example, a car buyer “causes” the cost of
manufacturing the car, and thus pays the full cost. That principle usually does not make any sense
in multisided markets, however. Often a product cannot exist unless several different customers
participate simultaneously. They all “cause” costs and “cause” benefits. That is true even if it is pos-
sible to identify costs that increase as a result of an additional user on one side—for example, the
cost of printing another copy of the Yellow Pages. Economists have shown that the best prices—
either from the standpoint of the business maximizing profits or from the standpoint of policy-
makers maximizing social welfare—involve complex relationships between the price sensitivity of
each side, interdependencies between these demands, and marginal costs.
Is There Anything New Here? Multisided platforms have a number of features that econo-
mists have examined before. Yet traditional learning does not deal with the role of intermediaries
in internalizing network externalities. Most businesses have distinct consumer types: workers or
retirees, households or corporate entities, men or women. But multisided platforms differ in that
they must serve two or more distinct types of consumers to generate demand from any of them.
Hair salons may cater to men, women, or both. Heterosexual dating clubs must cater to men and
women. For hair salons the ratio of men to women does not matter much; for heterosexual dating
enterprises it is absolutely critical.
Most businesses in single-sided and multisided markets engage in price discrimination (charg-
ing different prices that aren’t proportional to the corresponding marginal costs) because it is pos-
sible to increase revenue by doing so and because, in the case of businesses with extensive scale
economies, it may be the only way to cover fixed costs. A dating club may charge men a higher
price just because they have more inelastic demand and because it is easy to identify that group of
consumers. But businesses in multisided markets have an additional reason to price discriminate:
by charging one group a lower price the business can charge another group a higher price; and un-
less prices are low enough to attract sufficient numbers of the former group, the business cannot
obtain any sales at all.
Like firms in multisided markets, many firms in single-sided markets sell multiple products,
and there is extensive economic literature explaining why they do so. The standard explanations
for why firms produce multiple products probably apply to many of the platforms discussed here.
But firms that make multiple products for several one-sided markets (for example, General Electric
makes light bulbs and turbine engines) or several complementary products for the same set of con-
sumers (for example, IBM sells computer hardware and computer services) do not secure profit
opportunities from internalizing indirect network effects.
Finally, it is important to ask how the business implications of the recent work on multisided
markets differ from those of the older economic literature on network effects. This is not as simple
as it might seem, since popular discussions of network effects often missed important subtleties in
the academic literature.
PLATFORM ECONOMICS:
396 Part Five : Software Platforms Essays on Multi-Sided Businesses
Take the case of single-sided markets with direct network effects. Because of those effects it fol-
lows that there is an advantage to size, all else equal. But it does not follow that this advantage, if
present, is large, and it certainly does not follow that the firm with the biggest market share always
wins in the end, let alone that the first entrant always wins. Nonetheless, in the frenetic days of
the Internet bubble, lots of businesses were founded on the assumption that network effects were
present and important in their markets and that the key to success was to get in fast, price low,
and build share at any cost. Proponents of this simplistic view emphasized tipping—you build up
critical mass, and then the whole market flocks to you. And they emphasized an extreme sort of
lock-in— once you get most of the customers, nobody can enter against you, even with a better
product.
As Brian Arthur, an author of several influential papers in network economics, put it, “You
want to build up market share, you want to build up user base. If you do you can lock in that
market.”24 This is a nice, simple theory—much simpler than the economic literature from which
it claimed to be derived. But it is hard to find many businesses that succeeded by following it.
Unfortunately, many dot-com entrepreneurs and investors thought that “build share fast” was
the path to great riches. Only a few made it very far down that path before reality closed it off
and supposedly locked-in buyers left en masse. It turns out that only rarely are direct network
effects strong enough to prevent buyers from switching to a better product, as the massive defec-
tions of buyers from once dominant word-processing programs illustrates.
Those who believed that riches could be made quickly and easily by harnessing network
effects tended not to distinguish sharply between direct and indirect network effects. In both
cases the managerial prescription was to build share rapidly; indirect network effects, like di-
rect network effects, would kick in automatically and both fuel and protect further growth.
Work on two-sided markets makes it clear that this is dangerously simplistic in two important
respects.
First, even though at least two distinct groups must be involved or there to be indirect net-
work effects, the network enthusiasts assumed both that it is obvious that one should pay and
the other should not, and that it is obvious which group should pay.
Second, they generally assumed that the group that did not pay could be ignored in setting
business strategy because it would automatically fall into line and generate valuable network
effects. In contrast, economic analyses of multisided platforms, show that successful multisided
platform businesses must pay careful attention to all relevant groups, and typically must worry
more about balance among them than about building share with one of them. The multi-sided
approach is consistent with asymmetric treatment of customer groups, but getting it right re-
quires great luck or careful analysis.
24
Joel Kurtzman, “An Interview with W. Brian Arthur,” Strategy+Business 11 (1998): 100.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 397
The popular network economics literature also suggested that markets with direct or indi-
rect network effects would tend to tip toward a single provider.25 That does not happen much in
practice, though. Sometimes congestion costs outweigh network effects—that is the case with
nightclubs, trading pits, and shopping malls. Platforms also differentiate themselves, and thereby
counter the network effects of their rivals, by trying to appeal to different consumer preferences.
That is one of the reasons for the proliferation of magazines.
Consider some markets that seem to display important indirect network externalities: PC
operating systems, real estate agencies, payment cards, auction houses, local and national newspa-
pers, broadcast networks, parcel delivery services, banks, dating services, standards for encoding
DVDs, financial information services, music publishers, and recorded music manufacturers. Of
these many markets, the only ones in which a single large player accounts for the preponderance
of sales are PC operating systems (i.e., Windows) and some local newspaper markets (such as the
Los Angeles Times).
Most software platform categories are competitive as a result of providers differentiating them-
selves to appeal to different types of customers on either side of the market.
25.
Some of the network effects models allowed for differentiated tastes and the coexistence of multiple networks. See
Jeffrey Church and Neil Gandal, “Network Effects, Software Provision and Standardization,” Journal of Industrial
Economics 40, no. 1 (March 1992): 85–104. S.J. Liebowitz and S.E. Margolis, “Network Externality: An Uncommon
Tragedy” Journal of Economic Perspectives 8 (1994): 133–150. However, this literature was often taken to suggest
overall that network effects naturally lead to a single firm dominating a category. See Brian Arthur, “Increasing
Returns and the New World of Business,” Harvard Business Review 74 (July–August 1996): 100–109.
26
eBay was not begun as a B2B Web site, but as more and more businesses began to do business on it, it became one.
In 2003, eBay officially launched a separate B2B site. “Prior to the B2B site, eBay listed more than 500,000 business
items for sale every week on its consumer site, with business buyers representing more than $1 billion in annual-
ized gross merchandise sales, officials said.” Renee Boucher Ferguson, “eBay Launches B2B Site,” eWeek, January
28, 2003 (http://www.eweek.com/print_article/0,3048,a=36363,00.asp); eBay Press Release, “eBay Launches
eBay Business to Serve Its Growing Community of Business Buyers,” January 28, 2002 (http://investor.ebay.com/
news/20030128100772.cfm); Mark Berniker, “SAP, eBay Setup Industrial B2B Marketplace,” internet News.com,
June 16, 2003 (http://www.internetnews.com/xSP/article. php/2222371).
PLATFORM ECONOMICS:
398 Part Five : Software Platforms Essays on Multi-Sided Businesses
entrants into the payment card industry, Diners Club and Carte Blanche, barely exist today. The
second entrant, American Express, had a 14 percent share of credit and debit card purchase volume
in 2003.27
Getting All Sides on Board An important characteristic of multisided markets is that the de-
mand on each side vanishes if there is no demand on the others, regardless of what the price is.
Merchants will not accept a payment card if no customer carries it because no transactions will
materialize. Computer users will not use an operating system for which the applications they need
to run have not been written (except those rare users who plan to write their own applications).
The businesses that participate in such industries have to figure out ways to get both sides on board.
One way to do this is to obtain a critical mass of users on one side of the market by giving
them the service for free or even paying them to take it. Especially at the entry phase of firms in
multisided markets, it is not uncommon to see precisely this strategy. Diners Club gave away
its charge card to cardholders at first—there was no annual fee, and users got the benefit of the
float.28 Netscape gave away its browser to many users, particularly students, to get a critical mass
on the end-user side of its business.29 (Initially the other side was providers of Web sites, to whom
Netscape sold its server software.)
Another way to solve the problem of getting the two sides on board simultaneously is to
invest to lower the costs of consumers on one side of the market. As we saw earlier, for instance,
Microsoft invests in the creation of software tools that make it easier for application developers
to write application software for Microsoft operating systems and provides other assistance that
makes developers’ jobs easier. In some cases, firms may initially take over one side of the business
in order to get the market going. Palm would never have succeeded in creating the vibrant Palm
economy, with thousands of software applications and hardware add-on developers and millions
of users, had it not provided the first applications itself (especially Graffiti, the handwriting
recognition system).30
Providing low prices or transfers to one side of the market may help the platform solve the
simultaneity problem by encouraging the benefited group’s participation—which in turn, o wing
to network effects, encourages the nonbenefited group’s participation. In addition, providing
benefits to one side can discourage its use of competing multisided platforms. For example, when
Palm provides free tools and support to PDA applications software developers, it encourages those
27
Nilson Report 805 (February 2004).
28
“Credit Cards for Diners,” New York Times, March 30, 1950, p. 37; Diners Club display advertisement, New York
Times, March 30, 1950, p. 42.
29
David Plotnikoff, “Internet Born with Netscape,” Mercury News, February 28, 2003 (http://www.tc.umn.
edu/~jbshank/7_NetscapeIPO.html).
30
Annabelle Gawer and Michael Cusumano, Platform Leadership: How Intel, Microsoft and Cisco Drive Industry
Innovation (Boston: Harvard Business School Press, 2003).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 399
evelopers to write programs that work on the Palm OS platform and automatically induces those
d
developers to spend less time writing programs for other operating systems.31
Pricing Strategies and Balancing Interests Firms in mature multisided markets—those that
have already gone through the entry phase, in which the focus is on getting the various sides on
board—still have to devise and maintain an effective pricing structure. In most observed multisid-
ed markets, companies seem to settle on pricing structures that are heavily skewed toward one side
of the market, as Table 3.1 shows. Google earns the preponderance of its revenue from advertisers,
for instance, and real estate brokers usually earn most or all of their revenues from sellers.
Sometimes all competing platforms converge on the same pricing strategy. In principle,
Microsoft, Apple, IBM, Palm, and other operating system companies could probably have charged
higher fees to applications developers and lower fees to hardware makers or end users. Most discov-
ered that it made sense to charge developers relatively modest fees for developer kits and, especially
in the case of Microsoft, to give away a lot for free.
Getting the pricing balance right, however, requires considerable care. For example, in 2000,
Yahoo!’s Internet auction site was second only to eBay in terms of the number of listings. Sellers
found the site appealing because unlike eBay, Yahoo! did not charge sellers a fee for listing their
products. In 2001, Yahoo! changed its pricing strategy and began charging a fee. Yahoo!’s listings
dropped by 90 percent as sellers presumably moved to the larger venue, eBay.32 The price change
affected Yahoo!’s buyer-side market as well, since buyers were now left with little to bid on.
Two important factors influence multisided pricing structures. There may be certain customers
on one side of the market—Rochet and Tirole refer to them as “marquee buyers”33—who are ex-
tremely valuable to customers on the other side of the market. The existence of marquee customers
who create strong network effects tends to reduce the price to all customers on the same side of the
market and increase it to customers on the other side. A similar phenomenon occurs when certain
customers are extremely loyal (or captive) to the multisided platform firm, perhaps because of
long-term contracts or sunk cost investments. The effect is then opposite: the presence of captive
customers leads to an increase in the price charged to those on the same side and a decrease in the
price charged to the other side.
For example, American Express has been able to charge a relatively high merchant discount as
compared to other card brands, especially for its corporate card, because merchants have viewed
the American Express business clientele as extremely attractive.34 Corporate executives on expense
31
Jean-Charles Rochet and Jean Tirole, “Platform Competition in Two-Sided Markets,” working paper, December 13,
2002; http://www.palmsource.com/ developers/why_develop.html.
32
Saul Hansell, “Red Face for the Internet’s Blue Chip,” New York Times, March 11, 2001, section 3, p. 1.
33
Rochet and Tirole, “Platform Competition in Two-Sided Markets,” pp. 23–24.
34
Jon Friedman and John Meehan, House of Cards: Inside the Troubled Empire of American Express (New York:
Kensington Publishing, 1992), pp. 13, 56.
PLATFORM ECONOMICS:
400 Part Five : Software Platforms Essays on Multi-Sided Businesses
accounts were “marquee” customers, who allowed American Express to raise its prices to the other
side of the market, merchants. Similarly, marquee customers—in the guise of popular stores, often
called anchor tenants—are important for shopping malls as well: by attracting customers they
make a mall more attractive to other stores. The decline of a marquee store can sound the death
knell for an entire mall.
In the software world, marquee customers are usually businesses on the user side and “killer
applications”—an application so innovative and popular that people and businesses buy the com-
puter system mainly because they want the app—on the developer side. VisiCalc was the killer app
for the Apple II computer. It was one of the most important reasons behind the initial popularity
of this platform. Likewise, Mario Bros. was largely responsible for Nintendo’s millions of sales of
its NES video game console in the United States, and Sonic the Hedgehog was the main reason for
its displacement by Sega’s Genesis as the dominant console several years later.
Multihoming As Table 2 illustrates, customers on at least one side of a multisided market often
belong to several different networks. This is known as multihoming. Take payment cards. Most
merchants accept charge, credit, and debit cards associated with several systems; consider how
many card symbols there are at the next gasoline pump you use. On the other side of the market,
the average consumer has 3.6 payment cards.35 Advertisers typically place advertisements in several
different magazines, and consumers read various magazines.
In general, multihoming by one side of the market relaxes platform competition for that side
and intensifies it on the other. For instance, if game developers suddenly become more prone to
porting their games to both Sony PlayStation 3 and Microsoft Xbox 360, there would be less rea-
son for Sony and Microsoft to hold royalties down to attract developers. Moreover, in this case the
two consoles would become closer substitutes from the users’ perspective, since they would have
more games in common, so one might expect the battle for the end users (many of whom buy only
one console) to become fiercer, resulting in lower console prices.
Sometimes unrelated platforms evolve into intersecting ones, which target one or more groups
of customers in common; we will see this for digital media platforms. Platform competition can
be fierce when either group of customers is price-sensitive because they have other alternatives.
The Houston Chronicle may have 89 percent of the newspaper readers in Houston, but that does
not mean that it can exercise a great deal of pricing power.36 Advertisers have many other ways of
getting messages to readers, so they are sensitive to prices. And while readers may not have many
newspaper alternatives, they do have other ways of getting the news, and having a lot of readers is
what makes advertisers pay.
35.
http://www.cardweb.com/cardtrak/pastissues/april2004.html.
36.
This number is the total daily circulation of the Houston Chronicle divided by the total daily circulation of all daily
newspapers in the Houston area. Circulation 2003, SRDS (2002), p. 67.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 401
Scaling Many successful multisided firms seem to have adopted a fairly gradual entry strategy in
which they scale up their platform over time.37 Many payment card systems, for example, started
in one city or region before expanding nationally. It is often difficult to predict just what the right
technology and operations infrastructure will be. Therefore, the multisided firm may find it advan-
tageous to establish efficient buyer-seller transactions and balanced pricing first, and make large
investments only after the platform has been tested. Platforms such as eBay, Palm, and Yahoo! have
expanded gradually and methodically, building up customers on both sides of their markets.
Strategy Markets hardly ever cooperate with professors by following simple textbook rules ex-
actly. But in traditional markets there are classic truisms that can at least serve as a benchmark,
a starting point for more nuanced analysis. By contrast, multisided platforms, especially those in
new markets, all too often require clean-sheet planning. With multiple yet interdependent busi-
ness constituencies to serve, costs provide little guidance for pricing strategies. By the same token,
early entry may yield first-mover advantages or provide an instructive failure that simplifies the
search for successful strategies by businesses that follow. And, in light of the interdependence be-
tween different stakeholders, changes in the business environment may have multisided effects that
are very difficult to anticipate.
open-source projects. The source code of the resulting programs is made available for free; hence
the term open source. Users must sign a license that requires that if they redistribute the pro-
gram, even in modified form, they must make the source code available. As a result, it is hard to
make money directly from open-source programs or anything derived from open-source programs.
Open source began as an ideology—“free software is a matter of liberty,” according to Richard
Stallman—but has evolved into a multi-billion-dollar business based on selling hardware, software,
and services that are complementary to open-source programs.39
39
http://www.fsf.org/philosophy/free-sw.html.
40
http://httpd.apache.org/ABOUT_APACHE.html.
41
http://en.wikipedia.org/wiki/Apache_web_server.
PLATFORM ECONOMICS:
404 Part Five : Software Platforms Essays on Multi-Sided Businesses
42
http://www.idei.fr/doc/conf/sic/papers_2005/pdavid_slides.pdf.
43
Joe Wilcox, “IBM to Spend $1 Billion on Linux in 2001,” CNETNews.com, December 12, 2000 (http://news.com.
com/2100-1001-249750.html).
44
James Evans, “IBM to Invest Almost $1 Billion on Linux Development,” InfoWorld, December 12, 2000 (http://
www.infoworld.com/articles/hn/xml/00/ 12/12/001212hnibmlin.html).
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 405
offices.45 A year after its initial $1 billion investment, the company announced that it had already
recouped that amount and more.
You must cause any work that you distribute or publish, that in whole or in part contains or
is derived from the Program or any part thereof, to be licensed as a whole at no charge to
all third parties under the terms of this License.46
(Despite the copyleft name, the GPL is enforced by copyright law. Copyright is the source of
the property protection that enables those who release software under a GPL to impose conditions
on others who obtain that code.) The copyleft provision means that if people choose to distribute
software that is based in part on other software covered by the GPL, they must distribute their
new software under the GPL. GPL software thereby propagates itself. Copyleft makes it difficult
for anyone to earn significant profits from selling software code subject to the GPL. As Richard
Stallman observed,
Proprietary programs can use or communicate with GPL programs in some limited ways with-
out themselves becoming subject to the viral license condition, but the FSF recognizes that the di-
viding line can be murky. The terms of the GPL apply only to the distribution of software licensed
45
http://www.infoworld.com/articles/hn/xml/03/01/23/030123hnibmlinux. html?0124fram.
46
http://www.fsf.org/licensing/licenses/gpl.html.
47.
http://www.rons.net.cn/RMS/ms_oss.html.
PLATFORM ECONOMICS:
406 Part Five : Software Platforms Essays on Multi-Sided Businesses
under the GPL, although what “distribution” means in this context is not entirely clear either. It
may be possible for an enterprise to modify a GPL program and use it internally without being
legally bound to make the source code for its modified version available to others. On the other
hand, if the same enterprise distributed its modified GPL program to a subsidiary, the terms of the
GPL might well require it to make the source code available to all comers.
Most open-source projects are subject to the GPL. However, several commercial ventures
have chosen to use modified licenses. The two most prominent examples are the Common
Development and Distribution License (CDDL) that covered Sun’s Solaris as it went open source
and the Mozilla Public License (MPL) that governs the Firefox browser, among other products.
Both contain provisions that GPL does not, and thus code cannot be freely moved between
GPL and projects covered under these other licenses. Opponents of this balkanization of open-
source licenses contend that it leads to islands of legally incompatible code. For example, owing
to different licenses, no cross-pollination between Linux (GPL) and Solaris (CDDL) is possible.
Proponents argue that companies have varying needs and catering to these differences is neces-
sary for open-source software to flourish. In addition to relying on more restrictive licenses some
open-source software companies are using intellectual property rights to help protect their invest-
ments and guard their profits. Red Hat, for example, has used trademark law to help protect its
compilation of Linux from others.
C. Incentives
The incentives for writing open-source software are different from those for writing commercial
software. Many people write open-source code without being paid directly for it. These are volun-
teers who write code in their spare time because it interests them. Others write code because their
companies have asked them to. This may sound traditional but is not, since their employers cannot
sell the resulting code or obtain intellectual property rights over it.
Why Individuals Work on Open-Source Software Why programmers donate time to open-
source software projects is a subject that has generated considerable discussion.48 Open-source
advocates have suggested several motives, four of which involve nonfinancial rewards:
48.
For a more detailed discussion of open-source software development, see Joshua Lerner and Jean Tirole, “Some
Simple Economics of Open Source,” Journal of Industrial Economics, 2002; Joshua Lerner and Jean Tirole, “The
Open Source Movement: Key Research Questions,” European Economic Review 45, nos. 4–6 (2001): 819–826.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 407
• It is prestigious. Success at open-source development rates highly among those whose opin-
ions most programmers most value—other programmers.
• It “scratches an itch.” Programmers attack problems that they personally face or because they
are intrigued by the intellectual challenge.
• It meets an ideological urge—the desire for free software and the “liberty” it entails.
The “scratches an itch” motive has been considered by some analysts as leading to something
like a cooperative of users. A number of developers all consider a particular type of software poten-
tially useful, so they pool their talents to develop the software. With this type of motivation, the
GPL has sometimes been considered beneficial as an enforcement mechanism: it ensures that no
one can take the collective intellectual property, add some private intellectual property, and treat
the whole as a private good.
Business Models Based on the Open-Source Concept Businesses have incentives to “donate”
employees to the development of open-source projects that stimulate the demand for other prod-
ucts or services sold by the firm. This has become an increasingly large source of labor for open-
source projects.49 IBM and Red Hat illustrate the motivations. As discussed earlier, IBM’s model
is built on driving the sales of its key products: supporting Linux software increases IBM’s sales of
hardware, proprietary software, and services. Linux offered a way for IBM to integrate its entire
line of servers without having to develop a software platform of its own, and without having to
shoulder the continued support and development of that system on its own.
Red Hat is a somewhat different story. That company began as a pure open-source vendor of-
fering a distribution of Linux. Over time, it has gradually moved toward more traditional software
licensing, presumably because it is difficult to support a for-profit company with a pure open-
source business model. Red Hat is focused on solving a problem inherent in the Linux develop-
ment model. For major proprietary operating systems such as Windows, the components of the
software are integrated by the distributor and sold as a single program. Since no one developer
exists for Linux, bits and pieces of the operating system tend to float around—some in forms
unusable by nonprofessionals. Specific Linux distributions consolidate these bits and pieces into a
convenient package.
Red Hat is arguably the premiere Linux distribution, with more than 46 percent of Linux
server distribution shipments in 2004.50 The company was founded in 1995 and subsequently
49
“OpenOffice Team Wants IBM Contribution,” VNUNet, April 25, 2005; Timothy Prickett Morgan, “Novell
Creates Hula Open Source Collaboration Server,” ComputerWire, February 16, 2005 (http://www.computerwire.
com/ industries/research/?pid=23554158-E49D-4ED2-9482-72B4B5D4119F&type= CW%20News).
50
Al Gillen, Milla Kantcheva, and Dan Kusnetzky, “Worldwide Linux Operating Environments 2005–2009 Forecast
and Analysis: Product Transitions Continue” (IDC report no. 34390), table 2.
PLATFORM ECONOMICS:
408 Part Five : Software Platforms Essays on Multi-Sided Businesses
enjoyed significant growth, topped off with an IPO in 1999 that generated the eighth biggest first-
day percentage gain in Wall Street history.51 Like many other high-tech companies, Red Hat lost
quite a bit of value in the dot-com crash, but it has since rebounded successfully.
Red Hat really does three things. First, it integrates components of Linux into a cohesive
distribution, including commonly used open-source products along with the core operating sys-
tem. Second, it adds its own software to provide a better user experience and to make installation
and updating easier. Third, it sells support packages and certifies that external administrators are
qualified to work on Red Hat products. The company includes only open-source software, and the
code it writes is licensed mainly under the GPL.
Red Hat changed its business model drastically in 2003 by splitting its distribution into two
products—the Fedora Project, a more traditional open-source project, and Red Hat Enterprise
Linux (RHEL), the flagship product. Fedora is the place for experiments to run and outside devel-
opers to submit code, while RHEL is a stable version of Linux for paying customers. Along with
the split came a new licensing agreement. RHEL source code is available for free from Red Hat,
but the code computers need to run the operating system is available only with the purchase of a
support subscription.52 And support subscriptions must be purchased for each computer, just like
traditional proprietary software licenses.
One fundamental problem with generating revenue from GPL software is that anyone can take
the source code, “compile” it for computers to read, and resell it without incurring the original
creator’s development costs. Red Hat has tried to sidestep this problem. Another company could
rebuild RHEL from freely available source code, but it would have to strip out all references to Red
Hat to comply with trademark law. Purchasers could not be certain that the distribution really con-
tained all of the pieces in Red Hat’s, or that the installation would work as seamlessly. Thus, Red
Hat has used its reputation in combination with trademark law to limit the potential for another
company to undercut its profits.
Open source seems so New Age. Yet when one looks over the history of the computer industry,
it turns out that the business of selling software—including software platforms—really didn’t take
hold until the late 1970s. Microprocessors created a mass market for software that attracted entre-
preneurs. Many of these pioneers wrote applications that would run on the new microprocessor-
based personal computers. A few focused on refining software platforms whose shared code could
be used by many developers and customers at the same time.
51
W. G. Rohm, “Inside The Red Hat IPO,” Linux Magazine, November 15, 1999 (http://www.linux-mag.com/1999-
11/redhatipo_01.html).
52
Advanced users could compile the source code themselves. As discussed later in the chapter, this has many disadvan-
tages for corporate customers.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses Invinsible Engines: Both Sides Now 409
INSIGHTS
• Like other information goods, software platforms are produced by educated workers, are
malleable and easily changed, and are reproducible at virtually no cost.
• Bundling features into the software platform is often efficient for the platform producer
and for end users, as it is for most information goods, because it lowers distribution costs
and expands demand.
• Software platforms create value by reducing the costs for their multiple customer groups
to come together and thereby enhance the value that each customer group delivers to
the other. They do this mainly through providing shared services—made available through
APIs—that reduce costs for developers and users.
• Multisided platforms must consider marginal costs and price sensitivity in pricing, like sin-
gle-sided businesses, but they must also consider which side values the other side more.
Software platforms generally charge low prices on one side in order to attract customers
who can then be made available to the other side. Getting the balance right among all
sides is more important than building market share.
• Commercial and open-source production methods have both proved viable models for
producing software platforms. Commercial methods seem better suited for managing the
multisided aspects of platforms, while open-source methods have produced reliable plat-
forms and applications.
CHAPTER
What’s Next in Payments:
16 Invisible Engines
David S. Evans
ABSTRACT
The use of software platforms to drive innovation and transform industries has exploded since
the 2006 publication of my book Invisible Engines with MIT Professor and former Sloan School
Dean Richard Schmalensee and Harvard Business School Professor Andrei Hagiu. Around the
globe, invisible engines are ushering in a new era of technological change based on software.
The Apple iPhone has shaken the mobile phone industry worldwide in part by creating a mas-
sive applications business built on the phone’s operating system. Firefox has revolutionized
the browser industry by encouraging developers to write add-ons and, in doing so, toppled
Microsoft’s Internet Explorer from dominance in many countries. Facebook has created a pow-
erful social networking platform by opening itself up to developers. Amazon started a cloud-
computing platform that enables entrepreneurs to access its vast software code, hardware and
global communication systems over the Internet.
This chapter explain how software platforms are going ready to transform the payments in-
dustry and why you will be touched by this new revolution and perhaps even want to be one of
the entrepreneurs or investors who pioneer this new area. I’m first going to provide background
on the technology and business models behind invisible engines. I’m then going to show why this
is important for payments and describe some of the main players that are leading the revolution
at this point.
410
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 411
writer does not have to write that code again. That is what Windows does. It contains many
little bundles of code that help developers write their own programs. Microsoft provides a link
to that code—called Application Programming Interface (API)—that allows developers to link
into the Windows code. Some businesses make their service available by providing an API. For
example, YouTube provided an easy-to-use API for people to include access to videos on their
social network pages.
By making APIs available, the owners of software programs are letting others use their intellec-
tual property, the results of their hard work writing code. They can benefit from doing this if others
decide to write valuable applications that work with their software platform. More people are likely
to use a platform if there are more attractive applications that work with it. For example, many us-
ers decided to switch from Internet Explorer to Firefox because there were add-ons for Firefox that
did things that were not available with Internet Explorer such as support for XHTML – a popular
programming language for webpage developers. Software platform providers do not have to make
their code available for free though. Developers that sell their applications through the iPhone app
store give a portion of their revenue (about 30 percent as of the end of 2009) to Apple.
Invisible engines are catalysts that use a multi-sided platform business model. They ultimately
create value by making it easier—lowering the transactions cost—to get several different groups,
whose members value each other, together. In the physical world, a shopping mall is a two-sided
platform that helps bring shoppers and merchants together.
Software platforms generate value by reducing the cost to developers of writing applications for
consumers and providing a common place for users and developers to get together. Facebook for
example provides a place, not only for friends to get together, but also a place for friends to find ap-
plications that can help them and applications to find users that can benefit from them. Facebook
helps developers do this and thereby creates value for its community and ultimately itself.
to itself and focus on making the iPhone a self-contained device that only did things that Apple
wanted it to do; it would have had a “closed software platform”. Instead, in a smart move, it
decided to make that operating system a software platform that could support third-party appli-
cations. That meant it had to create APIs which would enable outsiders to link into the iPhone’s
operating system. These APIs give developers access to blocks of code that manipulate various
aspects of the iPhone. Apple then provided a “software developer kit” (SDK) to help program-
mers develop applications using the iPhone’s APIs and put together a web site to provide other
resources for developers.
An important set of business dynamics starts when such an operating system is opened up
and thereby becomes an “open software platform” that can support developers. Apple initially
provided more than enough applications and features on the iPhone to persuade many people
to get one. By opening up its operating system it then started a virtuous circle. Entrepreneurs
and hobbyists saw the opportunity to write interesting applications that could use the features of
the iPhone, including its large touch screen and connectivity. The iPhone users soon found that
they could download applications that made their phones even more useful than they were when
they first bought them. More people started buying iPhones because of the increasingly valuable
applications. As the number of iPhone users increased, developers had even more incentives to
write applications.
Apple has not taken this process for granted. Like most other successful platforms it has stoked
the fires. It developed a marketplace--an online shopping mall--to help developers make their ap-
plications available to iPhone users and earn revenues from them. Apple is responsible for collect-
ing money from iPhone users and then gives the developers about 70% share of the revenue. Users
have benefitted from having effectively a single location to search for applications. Developers have
benefited because Apple has reduced the transactions costs of selling and has assembled an audi-
ence of users.
Apple has obviously made enormous profits and seen the market value of the firm increase be-
cause of the success of the iPhone and its application store. iPhone users have done very well also as
a result of the tens of thousands of applications they can choose from, some of which enable them
to do things that were not possible before. New markets have opened for entrepreneurs that use the
iPhone, and some of them are earning significant profits or obtaining fame. By bringing developers
and users together the iPhone has served as a catalyst which has created value—and profit—out of
thin air. Others have followed suit. The most significant follower is Google, which has introduced
the Android software platform for mobile phones, has stimulated hardware makers to introduce
phones using this platform, and has worked hard at persuading developers to write applications
for its store. (Microsoft has been trying to be the open software platform for mobile for many years
but has thus far failed to catch on.)
Apple’s experience with the iPhone shows the power of the invisible engine model. As we will
see software platforms can drive payments innovation and revolutionize the payments business.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 413
But they also involve access to the vast server farms and communication networks around the
globe. Amazon’s Web Services plays to this company’s strength in selling products over the inter-
net. It provides special services to developers who are writing programs to help merchants operate
virtual stores and receive payment.
Invisible engines are beginning to disrupt the payments business both online and offline. This
industry is at the heart of commerce and makes it possible for businesses and people to exchange
value around the world. By definition it involves essentially all the money in the world. If invisible
engines start driving innovation in this industry they could have an enormous impact on consum-
ers and businesses who transact with each other.
1. Accessing things over the internet has become much faster and more reliable in many parts
of the world as a result of the spread of broadband.
2. Software development has moved from local devices to the internet so there are more lan-
guages, tools, and code to leverage.
3. There’s been a lot of progress in developing inexpensive, efficient, and massively large com-
puter systems. Of necessity Google and other companies had to figure these kinds of things
out.
4. The companies that have invested in massive and scalable server farms have realized that
they can sell others access to it. Many of the obvious suspects have gotten into offering
cloud-based services including Amazon, Google, IBM, and Microsoft.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 415
Cloud computing is also associated with a change in the traditional business model. Consumers
and enterprises typically bought (technically they licensed, but usually in perpetuity) software
packages which they installed on their computers. Expanding information-technology resources
meant front capital expenditures for hardware and software. In contrast, cloud computing is usu-
ally sold in bite sizes under what is known as the “software as a service” model. In fact it is even
better than that. In a typical situation consumers and enterprises are paying monthly or annual fees
to obtain access to hardware including storage as well as software.
Amazon Merchant Services provides a good example of what cloud computing is all about and
one that is relevant for the payments industry. Suppose you are a retailer who wants to start selling
over the internet. If you wanted to stay on earth, you would hire someone to build an e-commerce
site for you, line up someone to host it, and get a processor to enable you to take payment cards
online. If you wanted to go to the clouds, you could rely on the software that Amazon has created
for quickly building a store on line, have them host it at one of their server farms, and use their
payments system.
Amazon Merchant Services also illustrates how cloud computing has resulting in innovation
moving to the cloud. Amazon has developed a payment services product that relies on the payment
systems back down on the ground. They connect to one or more processors that in turn connect
to the rest of the payments system. Amazon’s innovative product stays in the cloud where everyone
down on earth can connect to it. PayPal, Apple, and other companies are also engaging in this sort
of innovation in the cloud as we will see.
PLATFORM ECONOMICS:
416 Part Five : Software Platforms Essays on Multi-Sided Businesses
Cloud computing also unleashes innovation from hardware. For most of its history the pay-
ments business has been tied to devices that pretty much have the software hard-wired in. Maybe
the supplier of the hardware could modify the software but the merchant really couldn’t nor could
other players in the ecosystem. And since these devices were closed systems a clever entrepreneur
couldn’t develop applications that could provide extra value to the merchant or to members of the
payments value-chain. That can and will change. Software will come down to the devices from the
cloud and APIs will enable entrepreneur/developers to add value.
There are other opportunities for moving innovation off the ground in payments that don’t
necessarily involve using internet connections to software residing on remote hardware. At the
moment the payments system consists of a number of highly secure private networks that compa-
nies can plug into. A point of sale terminal at a retailer plugs into several private networks. It does
that by connecting to a switch which then routes transactions across the debit or credit rails to the
relevant clearing and settlement system and so forth. One doesn’t have to rely on the internet nec-
essarily to move innovation off the ground. Companies could provide remote services over private
networks.
One way to do this would involve developing a software platform that enables third parties
to write applications that would work with the payments system. That could happen in a couple
of ways.
1. The major processors currently operate software platforms that provide services to banks.
They could potentially open up portions of these platforms to developers to build applica-
tions that use some of the features and functionality of these platforms. The same is true for
networks. Obviously, there are security issues that would have to be dealt with to ensure the
safety and integrity of the platforms.
2. A third party could build an uber-software platform that sits on top of all of the various
software programs that lie on the payment rails. That uber-software platform would con-
nect into these software programs. Then developers could plug into the uber-software plat-
form to write applications. In both of these cases it becomes possible for innovation to take
place outside of the software and hardware that currently comprises the payments systems.
The resulting innovative applications then also sit outside.
When we talk about payments in the cloud for the rest of this chapter we will use this term to
refer to the development of applications that sit outside of the traditional payments system and
link to the traditional payments system through APIs on software programs that sit somewhere on
the rails of the traditional system.
One thing is for sure in my view: innovation in payments will move to the cloud and anything
who wants to be a player will need to be there as well or work with someone or is.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 417
making it easier to ignite their businesses. With that they will be able to obtain scale more quickly and
lower costs. It will be possible to roll out innovations to many channels simultaneously by porting apps
between different platforms or using APIs for platforms to connect to multiple channels.
In the next section I will describe a platform that sits on top on the rails I’ve just described and
helps developers of new payments businesses access the messy plumbing of the payments system
right down to the point-of-sale devices at retailers. Then I will describe how PayPal has created a
software development platform for online transactions. These platforms help developers of applica-
tions, say for the iPhone, easily integrate payments. And finally I will explain how the traditional
card networks will play in this new world in which innovation, and value-added services, are driven
by invisible engines Should the card networks create their own open software platforms or focus
on just running the rails?
Innovation in payments will, in my view, be driven over the next decade by software platforms
that facilitate a vast ecosystem of applications. And there’s going to be competition over who provides
the best APIs (and underlying code) to support applications. One of the things we’ve learned about
software platforms is that they have huge network effects. The platform that gets more applications and
users will get even more applications and users. That doesn’t mean that only one software platform will
survive because each could differentiate itself by catering to different needs of developers. But it does
mean that it is unlikely that more than a handful of payment platforms will survive. The race is on.
plumbing. Through the miracle of the division of labor, PaySimple got to focus on what it could
do best—develop an innovative solution for small business—and offload a lot of dirty work to a
specialist. As a result, American Express got access to a technology that would have taken it a long
time and a ton of money to develop on its own—so much so that the AcceptPay innovation prob-
ably never would have launched without something like IP Commerce.
The APIs that IP Commerce makes available are enabling many entrepreneurs to develop
payments-related applications far more quickly and cheaply than they could have done on their
own or through other solutions. Square—the new smart-phone enabled payments system for small
merchants and their consumers—is powered in part by the IP Commerce invisible engine. So is
BrainTree Payment Solutions, a gateway and value add provider of online payment solutions.
A new e-commerce application that relies on IP Commerce’s APIs, getta , that just launched on
MySpace. Getta uses the power of social networks to encourage group buying—the buyers get
volume discounts and the merchants get sales.
IP Commerce has made these applications—and more like them—possible by spending years
doing the dirty work of creating code that links into the multitude of discrete and non-interoper-
able software programs that ultimately run the payment systems. This tedious work is what many
successful software platforms do. There’s nothing glamorous about Windows 7 or Snow Leopard.
These software platform—that we don’t really think much about, except when they crash—are
based on millions of lines of code that do all the nasty business of computing—from telling the
chip to move 1s and 0s around, to working the pixels on your screen, to connecting to your home
wireless system. IP Commerce has spent the last 4 years and many millions of dollars writing the
code that’s needed to get a lot of the mundane but complicated tasks that need to be done to work
with each of the several tender types that are used in the US.
IP Commerce’s work isn’t done. While it covers much of the payments business through which
most of the dollars flow it doesn’t have 100% coverage yet. And it is going to need to keep up
with ever changing software systems. Moreover, like all software platforms, its value will ultimately
come from spawning numerous applications that rely on it. But it has done a lot of the tedious but
important work of developing a cloud based platform that sits on top of the payments system and
it has a rapidly growing stable of applications that have shown its power.
IP Commerce has the prospect of creating the same sort of powerful network effects that
Windows has (which isn’t to say they are necessarily the next Microsoft—that franchise is tough to
beat!). These come in two ways. First, IP Commerce’s platform will become more valuable to the
participants in the payments ecosystem (such as processors and device makers) the more applica-
tions there are there are that drive volume through these points. Likewise it will become more valu-
able to developers of payments applications as even more of the payments system standardizes on
this platform. Second, some of the applications that are being built on IP Commerce—like some
of those built on Windows—have there own APIs and are creating there own ecosystems. These
applications can work more easily together as a result of relying on the same platform.
PLATFORM ECONOMICS:
420 Part Five : Software Platforms Essays on Multi-Sided Businesses
Don’t look, of course, for tens of thousands of applications coming out of garages over the next
year. Creating “applications” in payments is a lot harder than it sounds. The success of the iPhone
has made us think about applications as software programs that do cool things like help you find
friends nearby or learn Mandarin. In payments, the software application is usually just one piece
of a business model. Developing successful business models in payments—no matter how good
the software application is—is extremely difficult. There’s everything from glacial inertia to chicken
and egg problems to solve.
Despite these challenges it is likely that cloud based platforms such as that provided by IP
Commerce will become the one of the main drivers of innovation in payments in the coming
years. These platforms drastically lower the cost of creating businesses that need to connect to
the payments plumbing to create value. They are likely to therefore unleash a vast amount of
entrepreneurial energy devoted to discovering ways to provide value to businesses, consumers,
and governments that transact with each other. PaySimple, Square, Getta, BrainTree Payment
Solutions, TIO Networks, SoftTouch, and over 100 other applications, are early indications of
what’s to come.
s tandard “software developer kit” (sdk). Then it has provided a set of sample applications that give
developers some guidance on what to do. PayPal X also have a “sandbox” where developers can test
out their applications to debug them and make sure they work.
Taking a cue from the Invisible Engine strategy that Andrei Hagiu, Dick Schmalensee and I
described in our book, PayPal put on a develop conference as soon as it opened its new platform.
Over 1500 came to the November 3-4 conference in San Francisco. PayPal also offered prizes for
the best applications. The winner of the $50k prize was Rentalic which helps people rent their stuff
to other people. It has been described as a mashup of eBay and Craigslist. Here’s how it works.
Joe lists her lawnmower on www.rentalic.com. Sue offers to rent it for a fee they agree on. Joe uses
PayPal to verify they can pay the deposit for it. Sue gets a secret code. After meeting Joe and mak-
ing sure that she’s getting a working lawnmower she gives Joe the secret code.
PayPal has been trying to get people excited about the iPad. On April 16th, 2010, it held a
developer bootcamp in San Jose that was focused just on using PayPal X for iPad apps. It got 300
participants. Naveed Anwar, who is the head of Pay Pal X and seemingly its chief evangelists, talks
about PayPal X and the IPad, and shows some clips of the developers beavering away on apps, in
the Pay Pal X IPad Developer Camp video on YouTube.
PayPal is also rolling out its own applications but even here it is allowing developers to build
value-added services on top of them. Most significantly, PayPal just rolled out the PayPal iPhone
Bump which makes it possible for people to transfer money to each other using their iPhones.
(Now, personally, on this one I’m a bit skeptical. The video provides a great example of overhyped
P2P. Maybe there’s a market for P2P, especially for small B to P transactions, but it isn’t likely to
be splitting meal bills which seems to be the favorite example of the P2P optimists, or charging a
friend for a slice of my pizza as featured on the video.)
With X.com PayPal has ignited a new global payments platform very quickly. Here’s the
virtuous circle:
• They’ve started with rolling out a series of services that are available through APIs.
• They’ve poured money into encouraging developers to write applications by providing prizes
and putting together a VC network.
• They’ve provided these developers with tools and soon a place where develops, like the iP-
hone Store, will be able to sell their applications.
• This has led to the rapid growth of applications, the growth of a developer community, and
more buzz that will drive more developers to PayPalX.
• They’ve started building alliances with companies like Facebook, Microsoft, IBM, and
Salesforce. That’s the start of building out a potentially huge ecosystem.
It is remarkable how quickly PayPal has ignited its open platform strategy.
PLATFORM ECONOMICS:
422 Part Five : Software Platforms Essays on Multi-Sided Businesses
What does this mean for the traditional payment networks? For one it puts further distance be-
tween PayPal’s online dominance and the traditional networks. PayPal conquered eBay, has made
great strides in moving its payment services off eBay, and is now creating a potentially enormous
ecosystem of online applications. This is coming at a time when the distinction between online
and offline will soon end. For example, the iPad with wiFi at the point-of-sale is online and can
compete with traditional hardware and private connections. PayPal and the traditional networks
have had an uneasy relationship for the last decade. PayPal drives an enormous volume of card
transactions but everyone knows that they have strong financial incentives to move transactions
to ACH and that in any event PayPal—not a card network and not a bank—owns the online cus-
tomer relationship.
For another, it means that the traditional networks are far behind their alternative payment
system rival in seizing the most important opportunity of the coming decade: creating an open
platform for development payments applications. First movers such as PayPal aren’t necessarily
the category winner—and in fact often get leapfrogged by those who bide their times. That said,
PayPal—like Apple in the smart mobile space—is rapidly generating network effects and creating
an increasing number of applications that will make it hard to beat.
Finally, PayPal’s seemingly flawless execution of its open platform strategy makes one wonder
whether the traditional card networks have the genes to compete in this very Web 3.0 kind of
world. PayPal has its roots as a software application that lived within one of the great internet com-
panies. It wasn’t that long ago that the traditional card networks mainly used mainframes running
Cobol. If the card networks can’t pull it off though they could be to payments what the mobile
carriers are to mobile communications.
It is unclear at this point, however, whether the card networks understand how open software
payments platforms will radically change the payments business in the next decade. They may have
lots of plans in the works, may be behaving stealthily, just waiting to pounce on the first movers
in their own sweet time. Or they may be unsure of what’s happening, what to do, and how to do
it—they may be deer in the headlights of nimble rivals who aren’t saddled with old code and pre-
web says of doing things.
In fact, the card networks have three stark choices: spur innovation with an open platform,
innovate with a closed platform, or just stick to running the rails.
• Open software platform strategy. They could use an invisible-engine strategy by building
an open software platform on top of their existing networks and promoting the creation of
applications that use services based on their rails. They would expose APIs to third party
developers that would provide services. They would thereby compete for the attention of
application developers with PayPal X, Amazon Merchant Services, IP Commerce, and who-
ever else tries to build a platform layer.
• Closed software platform strategy. They could try to develop value added services them-
selves in competition with third-party application developers writing for competing platforms,
or partner with other firms to develop value-added services that rely on their network. With a
closed strategy they would not provide APIs that third-party developers generally could access.
• Just run the rails strategy. They could focus on making money from transactions that go
over their rails and focus on running those rails as efficiently as possible. They could leave
operating software platforms and fostering third-party applications to other companies.
They would benefit from the success of these companies who would drive more transactions
over their rails.
Each of these strategies poses risks. To evaluate them it is important to agree on three premises con-
cerning how the payments world will evolve over the next decade regardless of what the networks do.
First, over time the distinction between the online and offline world in payments will vanish.
Fast internet connections will become ubiquitous. Almost every device we interact with will have
a connection. Those devices will access web-based software which will provide an increasing array
of services.
We have already seen this happening. When the commercial internet started only a few com-
puters were connected. Over time virtually all computers had internet connectivity. Then mobile
phones started being connected. It is widely expected that most phones in industrialized countries
will be internet-enabled in not too long. The third screen—the television—is next. Other devices
are in the queue. Automobile manufacturers, for example, are incorporating internet capabilities in
their cars and providing software platforms for applications. Importantly, for our purposes, point-
PLATFORM ECONOMICS:
424 Part Five : Software Platforms Essays on Multi-Sided Businesses
of-sale devices will become internet-enabled and rely on services provided by web-based software.
That will include everything from card terminals to cash registers.
This means that the neat distinction of today between online transactions that you do with
your computer (or increasingly mobile phone) over the internet, and offline transactions where
you walk into a brick-and-mortar store without touching the internet, will disappear. The internet
will become like the electric power grid—something that almost every device is plugged into. The
corollary is that it will be meaningless to talk about internet-payment providers, such as PayPal,
as if they were an interesting group of foreigners. The alternative payment providers will be at the
physical point of sale because the physical point of sale will be connected to the web.
Second, most devices even at the point of sale will rely on software, probably residing in the
cloud, that provide valuable services. At the moment most point-of-sale devices rely on software
that either resides on the device or on servers that are accessed over a corporate intranet. Devices
usually rely on a dedicated software solution. Once devices can obtain easy access to the internet
it will prove more efficient to move software to the internet as well. Device makers may do this on
their own to obtain greater control over software—including their choice of vendor and the ability
to provide easy updates over the lifecycle of their equipment. Merchants may demand this as well
because cloud-based software will provide them greater flexibility.
Innovation for the point-of-sale will move out to the cloud or will at least become untethered
with the particular point-of-sale device. Internet-connected computers made it extremely easy for
software developers to distribute products to businesses and people via download. That of course
led to an increasingly vibrant application community. Internet-connected point of sale devices
will do the same. The mere availability of the direct link to the point-of-sale devices will ignite a
developer community.
Third, there’s going to be limited room for software payments platforms. Several software
payments platforms will succeed in becoming the go-to places for developers who want to incor-
porate payments and related functionality into their products and services. But it is doubtful that
there’s going to be room for more than two or three.
Software platforms are driven by indirect network effects. The more applications that are writ-
ten for them the more customers (businesses and consumers in this case) will want to use them;
and the more customers there are for the applications for a platform the more other developers will
want to write more applications. The main reason this virtuous circle doesn’t lead to a monopoly
platform is that different platforms can specialize in providing different services—product differ-
entiation can offset indirect network effects. The other limiting factor is that software developers
will port their applications to a few platforms but not many. They will go for the two or three
platforms that have the largest reach assuming they are technically on par.
For payments, it is doubtful that the four card networks plus several alternative payments pro-
viders plus others will all succeed in operating software payments platforms. This point has, as we
will see, has huge implications for the risks faced by the networks in choosing a strategy.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 425
With these premises in place it is easy to see the risks and opportunities that the card networks
face during the new age of invisible engines.
Let’s start with the stick-to-your knitting strategy of focusing on just running the rails. That
would involve future value-added services moving to third-parties. The software payments plat-
forms would support ecosystems of application developers that would incorporate payments func-
tionality into their applications. (Just to be clear, these applications could be software programs
that support substantial businesses. That’s the vision for Square for example.) The applications
businesses would earn revenue from their customers and the software platforms would earn license
fees (and perhaps revenue cuts) from the applications business.
The card networks could charge for access to their rails. They would benefit from more applica-
tions generating more transactions over their networks. The risk is two-fold. On the one hand the
software platforms and applications suck up a lot of the value that’s available from using the rails. The
card rails could try to get a piece of the action but that brings us to the other hand: the software plat-
forms, and perhaps other players, could try to disintermediate the rails altogether. In fact, one would
expect that PayPal will squeeze card fees down. They are already moving transactions over to cheaper-
for-them ACH. Their ability to do so will eventually put downward pressure on transaction fees for
the card networks (not to mention interchange fees if those survive the current regulatory and legal
onslaughts). The card networks could try to be like the cable companies and squeeze a lot from eve-
ryone who tries to use their rails. That has worked, although it is getting ugly, in cable because there
isn’t much competition. It is unlikely to work in payments where there are multiple sets of payments
rails and lot of opportunities (see PayPal above) to shift transactions in response to cost differences.
The card networks could also take the middle way. They could embrace the idea of creating
applications that use their network services but do that themselves, but more aggressively than
they have in the past and with a clear eye on what’s being built on the open platforms, or enlist
business partners to that with them. This has several advantages. For one, it allows them to keep
tight control over what’s built on their networks, make sure it is consistent with their overall busi-
ness objectives, tightly integrate the applications, and prevent applications that could degrade the
performance or reputation of their networks. For another, it allows them to make money from the
applications directly rather than limiting themselves to access and transactions fees. They will own
customer relationships and data as well. The chief disadvantage of the closed-platform approach is
that it does not generate the sorts of indirect network effects that have powered the growth of the
iPhone and seem to be working well so far for PayPal. The networks would have to hope that they
make up in quality of apps what they lose in quantities.
Finally, the card networks could adopt an invisible engine strategy. That would generally in-
volve creating a software layer above their networks, developing features for that layer that use
their rails and provide other services as well, and making these features available through APIs.
Following the PayPal X approach they would need to evangelize their platforms and provide de-
velopers with the tools and encouragement to create successful applications. There is a huge upside
PLATFORM ECONOMICS:
426 Part Five : Software Platforms Essays on Multi-Sided Businesses
potential here. They could, as I’ve described throughout this series, become the centre of a ecosys-
tem of applications that would drive value back to the center and in doing create barriers for rivals.
See Windows.
Like most things in life that have major payoffs, there are at least two related and significant
risks: First, it is highly unlikely that all four of the card networks could succeed in establishing
successful software payments platforms. One could easily imagine that only 2-3 platforms emerge
and that at least one of those is one of the alternative payments providers. That would mean that
if all four networks tried 2-3 of them would probably fail. Second, there can be a wide gulf be-
tween aspiration and ability as every little boy who wants to play pro-ball soon finds out. The card
networks have much experience developing sophisticated software systems and communications
platforms. So it isn’t like a car company wanting to get into the software platform business. But the
card networks have evolved slowly from mainframe based companies and don’t, at least obviously,
have demonstrated skills in doing internet-related development work. They may lack the genetic
matter to go up against PayPal or other companies that were born in the internet world. Thus, the
card networks could make significant investments of financial and reputational capital in trying to
become one of the software payments platforms and fail in succeeding.
The rise of invisible engines in payments therefore presents significant opportunities and risks for
the traditional card networks. One or more of them could distance themselves from their rivals by
becoming one of the surviving software payments platforms with a robust set of applications driving
traffic. Those that don’t develop a successful platform could find themselves as service providers to
much more interesting and robust businesses. It is beginning to look like this is the way the mobile
communications business could evolve—with the iPhone, the Android, and perhaps another software
development platform driving applications and the carriers settling for charging for carrying traffic.
The traditional card networks are clearly trying to navigate what to do. Visa’s acquisition of
CyberSource for $2 billion may be an important step in its developing an open software platform.
CyberSource is the e-commerce engine for about 300,000 merchants. It mainly helps these firms with
key back office tasks—risk and fraud most importantly, call centers, and then processing. Combined
with other technology it could help Visa challenge PayPal X. American Express’s acquisition of
Revolution Money for $300 million is another example. Revolution Money has a payment platform
based on internet technology. It could also provide some useful building blocks for a larger efforts.
The decisions these networks make on how to proceed, and how to meet the Pay Pal X chal-
lenge, will determine the course of payments over the next decade.
evidence that consumers want to use the card. But there hasn’t been a groundswell of demand
from consumers because there are few places to use the cards or really compelling reasons to use
it. It is hard to ignite innovations in payments because it is costly to change current systems and
because it is difficult to solve the chicken-and-egg problem of getting multiple stakeholders in
the ecosystem on board.
Cloud-based software platforms may change that very quickly. Innovation will take place
through the creation of software-enabled businesses that access the payments rails through soft-
ware development platforms that live in the cloud. These innovations will be pushed through the
broader payments ecosystem through the internet.
Whether Square succeeds or not as a business it points the way to how this will happen.
It is based on a software application that relies on two software development platforms—
IP Commerce’s payments software development platform and Apple’s iPhone software develop-
ment platform. It connects to merchants through an internet-connected device—the iPhone, iPad,
or other smart mobile technology. It connects to processors through an internet connection that
takes it to IP Commerce’s software platform in the cloud. And it connects to consumers through
email that traverses the internet.
Rentalic—the winner of PayPal X’s first application competition—may also succeed or fail but
again points the way. An entrepreneur comes up with an idea for bringing buyers and sellers together
(in this case renters and rentees) and needs a way to deal with complex payments (in this case han-
dling the security deposit). She goes to a payments software platform in the cloud for these services.
PayPal X enables Rentalic to connect buyers and sellers and ultimately to multiple payment rails.
Rentalic and Square demonstrate why the pace of innovation is likely to increase rapidly in
payments and highlight the opportunities. Innovation will pick up because the cost of getting
new ideas that require payments solutions is declining rapidly as a result of these software plat-
forms. The opportunities are enormous—one only has to look at what Apple has accomplished
with its Applications Store and glance at how quickly PayPal X has attracted developers and new
applications.
The “applications” also highlight the risks facing the traditional card networks if they fall be-
hind. Both Rentalic and Square involve physical transactions—the bread and butter of the card
networks. Rentalic is mainly about renting physical items where the buyer and seller will actually
meet in person probably. Square is about transactions at physical points of sale. That’s the future.
One or more of the card networks may figure out how to create an open software development
platform and get a virtuous circle going around an “applications store”. Some card networks may
find themselves on the sidelines and watch a huge amount of value being sucked towards the cloud.
Invisible engines will drive innovation and transform the payments industry. Some incum-
bents will win, and some will lose, as this disruption unfolds over the next decade. But there’s no
doubt in my mind the entrepreneurs, consumers, retailers, and other stakeholders in the payments
ecosystem will benefit tremendously
428
Chapter 3: “How Catalysts Ignite: The Economics of Platform-Based Startups,” Platforms, Markets
and Innovation, A. Gawer, ed., Cheltenham, UK and Northampton, MA, US: Edward Elgar, 2009.
Chapter 4: “The Antitrust Economics of Multi-Sided Platforms,” Yale Journal on Regulation, Vol.
20, Issue 2, Summer 2003.
Chapter 5: “Two-Sided Market Definition,” ABA Section of Antitrust Law, Market Definition in
Antitrust: Theory and Case Studies, forthcoming 2012.
Chapter 6: “The Analysis of Mergers that Involve Multi-Sided Platforms”, Journal of Competition
Law and Economics 2008:4, pp. 663-695, with Michael D. Noel.
Chapter 7: “The Economics of the Online Advertising Industry,” Review of Network Economics,
Vol. 7, No. 3, September 2008.
Chapter 8: “The Online Advertising Industry: Economics, Evolution, and Privacy,” Journal of
Economic Perspectives, Summer 2009.
Chapter 9: “Web Economy, Two-Sided Markets, and Competition,” Concurrences, May 2010
Chapter 10: “Antitrust Issues Raised by Global Internet Economy,” Northwestern University Law
Review, Vol. 102, No. 4, p. 285, Colloquy Essay, 2008.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 429
Chapter 11: “More Than Money,” Paying with Plastic (Massachusetts: MIT Press, first edition
1999, second edition 2005; third edition 2012), with R. Schmalensee.
Chapter 12: “Interchange Fees and Their Regulation,” Interchange Fees in Credit and Debit Card
Industries: What Role for Public Authorities?, Federal Reserve Bank of Kansas City, Santa Fe, New
Mexico, May 2005
Chapter 13: “The Effect of Regulatory Intervention in Two-Sided Markets,” Review of Network
Economics, Vol. 4, Issue 4, December 2005, with Howard H. Chang and Daniel D. Garcia Swartz.
Chapter 14: Invisible Engines: Introduction to the Chinese Edition Invisible Engines: How Software
Platforms Drive Innovation and Transform Industries, (Massachusetts: MIT Press, 2010) with A
Hagiu and R. Schmalensee.
Chapter 15: “Invisible Engines: Both Sides Now,” Invisible Engines: How Software Platforms Drive
Innovation and Transform Industries, (Massachusetts: MIT Press, 2006) with A Hagiu and R.
Schmalensee.
Chapter 16: “What’s Next in Payments: Invisible Engines,” PYMNTS.com, October 24, 2009
430
BIBLIOGRAPHY
Ackerberg, Daniel A., and Gautam Gowrisankaran. 2003. ”Quantifying Equilibrium Network
Externalities in the ACH Banking Industry,“ mimeo, University of Arizona, October.
Ahlborn, Christian, Howard H. Chang, and David S. Evans. 2001. “The Problem of Interchange
Fee Analysis: Case without a Cause?” European Competition Law Review, vol. 22, no. 8 (August),
pp. 304-312.
American Express. 2005. American Express Company Annual Report 2004. February.
Anderson, Simon P., Jean J. Gabszewicz. 2005. “The Media and Advertising: A Tale of Two-Sided
Markets.” CEPR Discussion Paper No. 5223.
ANZ (2003) “Annual Report.”
Arango, Tim. 2008. “Cable Firms Join Forces to Attract Focused Ads,” New York Times (March10).
Arcavision. “Exchange Volume Summary-1/1/2008-12/31/2008.” Accessed on April 2, 2009.
Argentesi, E. and L. Filistrucchi (2005). “Estimating Market Power in a Two-Sided Market: The
Case of Newspapers,” Journal of Applied Econometrics, forthcoming.
Argentesi, E. and M. Ivaldi (2005). “Market Definition in Printed Media Industry: Theory
and Practice,” CEPR Discussion Paper No. 5096, Available at SSRN: http://ssrn.com/ab-
stract=779107.
Armstrong, M. and J. Wright (2007). “Two-Sided Markets, Competitive Bottlenecks and Exclusive
Contracts,” Economic Theory, Vol. 32, No. 2, pp. 353-380.
Armstrong, Marcia (1991) “Retail Response to Trade Promotion: An Incremental Analysis
of Forward Buying and Retail Promotion,” unpublished doctoral dissertation, School of
Management, University of Texas at Dallas
Armstrong, Mark (2002) “Competition in Two-Sided Markets,” Nuffield College, Oxford,
Working Paper.
Armstrong, Mark. 2004. “Competition in Two-Sided Markets,” mimeo, University College London.
__________. 2005. “Two-Sided Markets: Economic Theory and Policy Implications.” mimeo,
University College, London, April.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 431
Becker, David (2000) “Revenue from Game Consoles will Plunge, Report Predicts,” CNET News.
com, July 31, http://techrepublic-cnet.com.com/2100-1040-43841.html?legacy=cnet.
Becker, David (2002) “Xbox Drags on Microsoft Profit,” CNET News.com, Jan. 18, http://news.
com.com/2100-1040-818798.html.
Bergin, Richard (1998) “Interview with Yahoo! Senior Vice President Tim Brady,” Harvard
Business School, doctoral research.
Bergman, Mats A. 2005. “Two-Sided Network Effects, Bank Interchange Fees, and the Allocation
of Fixed Costs.” Mimeo, Swedish Competition Authority, January.
Berndt, Ernst, Mohammad Arzaghi, James C. Davis, Alvin J. Silk. 2008. “Economic Factors
Underlying the Unbundling of Advertising Agency Services.” NBER Working Paper No.
W14345.
Besley, Timothy and Harvey Rosen (1999) “Sales Taxes and Prices: An Empirical Analysis,”
National Tax Journal, 52(2): 157-178.
BIGresearch, Simultaneous Media Usage Study (SIMM12), Media Influence to Purchase & New
Media Usage –Voters, June 2008.
Blodget, Henry. 2008. “Newspapers’ Fab 2007: $11 Billion of Market Value Vaporized.” Silicon
Alley Insider. (January 2), at <http://www.alleyinsider.com/2008/1/newspapers_11_billion_
of_market_value_vaporized_in_2007>.
Bolt, Wilko, and Alexander F. Tieman. 2004a. “Skewed Pricing in Two-Sided Markets: An IO
Approach,” Working Paper No. 13, De Nederlandsche Bank, Amsterdam, October.
__________. 2004b. “A Note on Social Welfare and Cost Recovery in Two-Sided Markets,”
Working Paper No. 24, De Nederlandsche Bank, Amsterdam, December.
Borenstein, Severin, C. Cameron, and R. Gilbert (1997) “Do Gasoline Prices Respond
Asymmetrically to Crude Oil Price Changes?,” Quarterly Journal of Economics 112 (February).
Brammall, Bruce (2002). “Watchdog Fears ANZ Card Hikes,” Herald Sun, November 5.
Brammall, Bruce (2003) “Bank May Ditch Reward Scheme,” Herald Sun, February 20.
Bulow, Jeremy and Paul Pfleiderer (1983) “A Note on the Effect of Cost Changes on Prices,”
Journal of Political Economy 91(1): 182-185.
Bureau of Transportation Statistics (2000) “Airport Activity Statistics of Certificated Air Carriers:
Summary Tables 2000,” Table 1, online: http://www.bts.gov/publications/airport_activity_sta-
tistics_of_certificated_air_carri ers/2000/tables/table01.html.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 433
Bureau of Transportation Statistics. 2000. “Airport Activity Statistics of Certificated Air Carriers:
Summary Tables 2000,” Table 1, online: http://www.bts.gov/publications/airport_activity_sta-
tistics_of_certificated_air_carriers /2000/tables/table01.html.
Caillaud, Bernard and Bruno Jullien (2001) “Chicken & Egg: Competing Matchmakers,” Center
For Economic Policy Research, Working Paper No. 2885.
Caillaud, Bernard, and Bruno Jullien. 2003. “Chicken and Egg: Competition among Intermediation
Service Providers,” Rand Journal of Economics, vol. 34, no. 2 (Summer), pp. 521-552.
Cannex Australia (2004) Data on merchant service fees. August/September.
Carlton, Dennis (2007). “Market Definition: Use and Abuse,” Competition Policy International,
Vol. 3, No 1.
Carlton, Dennis W. (1986) “The Rigidity of Prices,” American Economic Review, 76(4): 637-58
Carlton, Dennis W., and Alan S. Frankel. 1995a. “The Antitrust Economics of Credit Card
Networks,” Antitrust Law Journal, vol. 63, no. 2 (Winter), pp. 643-668.
__________, and __________. 1995b. “The Antitrust Economics of Credit Card Networks:
Reply to Evans and Schmalensee,” Antitrust Law Journal, vol. 63, no. 3 (Spring), pp. 903-915.
Chakravorti, Sujit (2003) “Theory of Credit Card Networks: A Survey of the Literature,” Review
of Network Economics, 2(2): 50-68.
Chakravorti, Sujit. 2003. “Theory of Credit Card Networks: A Survey of the Literature,” Review of
Network Economics, vol. 2, no. 2 (June), pp. 50-68.
Chakravorti, Sujit, and Alpha Shah. 2003. “Underlying Incentives in Credit Card Networks,”
Antitrust Bulletin, vol. 48, no. 1 (Spring), pp. 53-75.
Chakravorti, Sujit, and Roberto Roson. 2004. “Platform Competition in Two-Sided Markets: The
Case of Payment Networks,” Federal Reserve Bank of Chicago, Working Paper 2004-09, July.
Chakravorti, Sujit, and Ted To. 2003. “A Theory of Credit Cards,” mimeo, Federal Reserve Bank
of Chicago, July.
Chakravorti, Sujit, and William R. Emmons. 2003. “Who Pays for Credit Cards?” Journal of
Consumer Affairs, vol. 37, no. 2 (Winter), pp. 208-230.
Chang, Howard H., and David S. Evans. 2000. “The Competitive Effects of the Collective Setting
of Interchange by Payment Card Systems,” Antitrust Bulletin, vol. 45, no. 3 (Fall), pp. 641-677.
Chang, Howard H., David S. Evans, and Daniel Garcia Swartz. 2005. “An Assessment of the
Reserve Bank of Australia’s Interchange Fee Reforms,” Working Paper, May.
PLATFORM ECONOMICS:
434 Part Five : Software Platforms Essays on Multi-Sided Businesses
Chevalier, Michel and R. Curhan (1976) “Retail Promotions as a Function of Trade Promotions:
A Descriptive Analysis,” Sloan Management Review 18(3): 19-32.
Chiaravalle, Bill and Barbara Schenck. 2006. Branding for Dummies. John Wiley & Sons.
Cisneros, Oscar S (2000) “Ebay Accused of Monopolization,” Wired News, July 31, http://www.
wired.com/news/print/0,1294,37871,00.html.
ComScore Press Release. 2008. “U.S. Search Engines Rankings.” (November 26), at <http://
www.marketwatch.com/news/story/comScore-Releases-October-2008-US/story.
aspx?guid={FE165F0B-907D-4869-BDC9-379F061AEB6E}>.
Cornell, Andrew. 2004. “Deal Highlights Change in Card Market,” Australian Financial Review,
July 12.
Cornell, Andrew. 2005. “American Express Aims for Expansion,” Australian Financial Review,
April 18.
Cotterill, Ronald (1998) “Estimation of Cost Pass Through to Michigan Consumers in the ADM
Price Fixing Case,” Food Marketing Policy Center, University of Connecticut, Research Report
No. 39.
Cotterill, Ronald (2000) “Dynamic Explanations of Industry Structure and Performance,” presen-
ted at the USDA Conference “The American Consumer and the Changing Structure of the
Food System,” May 3-5, 2000.
Cruickshank, Don. 2000. “Annex D3: Interchange Fees for Credit and Debit Cards,” UK Banking:
A Report to the Chancellor of the Exchequer, UK: HM Treasury (March), pp. 247-272.
De Grauwe, Paul, and Laura Rinaldi. 2002. “A Model of the Card Payment System and The
Interchange Fee,” in Clemens Esser and Michael H. Stierle, eds., Current Issues in Competition
Theory and Policy, Berlin: VWF.
DOJ and FTC (1997). “U.S. Horizontal Merger Guidelines.” Available at: http://www.usdoj.gov/
atr/public/guidelines/horiz_book/hmg1.html.
Duboff, Robert S. 2007. ROI for Marketing: Balancing Accountability with Long-Term Needs. New
York: Association of National Advertisers.
Dubois, P., A. Hernandez -Perez and M. Ivaldi (2007). “The Market of Academic Journals: Empirical
Evidence from Data on French Libraries,” Journal of the European Economic Association, Vol.
5, No. 2, April-May 2007, pp. 390-399
Dye, Jessica. 2009. “Consumer Privacy Advocates Seek Search Engine Solution.” EContent (March
Issue), at <http://www.econtentmag.com/Articles/News/News-Feature/Consumer-Privacy-
Advocates—Seek-Search-Engine-Solution-52679.htm>.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 435
Evans, D. and R. Schmalensee (2007b). Catalyst Code: The Strategies behind the World’s Most
Dynamic Companies. Harvard Business School Press, 2007.
Evans, David and Richard Schmalensee (2005a). Paying with Plastic: The Digital Revolution in
Buying and Borrowing, 2d Ed. MIT Press, Cambridge, MA.
Evans, David and Richard Schmalensee. 2005. Paying with Plastic, 2nd Edition: The Digital
Revolution in Buying and Borrowing. Cambridge: The MIT Press.
Evans, David and Richard Schmalensee. 2009. “Platform Failure.” Available at SSRN, http://ssrn.
com/abstract=1353502.
Evans, David S. 2003a. “Some Empirical Aspects of Multi-Sided Platform Industries,” Review of
Network Economics, vol. 2, issue 3 (September), 191-209.
__________. 2003b. “The Antitrust Economics of Multi-Sided Platform Markets,” Yale Journal on
Regulation, vol. 20, no. 2 (Summer), pp. 325-381.
__________. 2003b. “The Antitrust Economics of Multi-Sided Platform Markets,” Yale Journal on
Regulation, vol. 20, no. 2 (Summer), pp. 325-381.
Evans, David S. (2003) “The Antitrust Economics of Multi-Sided Platform Markets,” Yale Journal
on Regulation (forthcoming).
Evans, David S. 2008. “The Economics of the Online Advertising Industry,” Review of Network
Economics, 7(3), pp 359 – 391.
Evans, David S. 2009. “How Catalysts Ignite: The Economics of Platform-Based Start-Ups.”
Platforms, Markets and Innovation, A. Gawer, ed., Cheltenham, UK and Northampton, MA,
US: Edward Elgar, 2009. Available at SSRN: http://ssrn.com/abstract=1279631.
Evans, David S., Andrei Hagiu, and Richard Schmalensee. 2004. “A Survey of the Economic Role of
Software Platforms in Computer-Based Industries,” CESifo Working Paper No. 1314, October.
Evans, David S., and Richard Schmalensee. 1995. “Economic Aspects of Payment Card Systems and
Antitrust Policy toward Joint Ventures,” Antitrust Law Journal, vol. 63, no. 3 (Spring), pp. 861-901.
Evans, David S. and Richard Schmalensee (1999) Paying with Plastic: The Digital Revolution in
Buying and Borrowing, The MIT Press, Cambridge, MA.
Evans, David S., and Richard Schmalensee (2005a) Paying with Plastic: The Digital Revolution in
Buying and Borrowing 2d Ed. Cambridge, MA: MIT Press
Evans, David S., and Richard Schmalensee (2005b) “The Economics of Interchange Fees and
Their Regulation: An Overview,” paper presented at a Conference Interchange Fees in Credit
and Debit Card Industries: What Role for Public Authorities? Federal Reserve Bank of Kansas
City, Santa Fe, New Mexico, May 4-6.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 437
Evans, David S., and Richard Schmalensee (2005c) “The Industrial Organization of Markets with
Two-Sided Platforms,” NBER Working Paper Series, No. 11603, September.
Fahey, Rob (2002) “MS to Lose J525M on Xbox This Year,” GameIndustry.biz, June 26, http://
www.gamesindustry.biz/content_page.php?section_name=pub&aid=210.
Farrell, J and C. Shapiro (1990). “Horizontal Mergers: An Equilibrium Analysis,” The American
Economic Review, Vol. 80, No. 1, pp. 107-126.
Federal Trade Commission Hearing. Behavioral Advertising: Tracking, Targeting, and Technology.
November 2006.
Federal Trade Commission Release, Online Behavioral Advertising: Moving the Discussion Forward
to Possible Self-Regulatory Principles. December 2007.
Feldman, R. (1994) “The Welfare Economics of a Health Plan Merger,” Journal of Regulatory
Economics, 6(1), pp. 67-86.
Feldstein, Janet and Christopher Flanagan (2001) “Handspring – Partnerships,” Harvard Business
School.
Ferguson, James M. (1983) “Daily Newspaper Advertising Rates, Local Media Cross-Ownership,
Newspaper Chains, and Media Competition,” Journal of Law and Economics, 26: 635-654.
Festa, Paul (1996) “Looking for Love Online,” CNET News.com, December 17, 1996 http://news.
com.com/2100-1023-255523.html?tag=rn.
Fisher, James (1997) “An Ad Blitz for the 21st Century,” Sprint Press Release, September 24, http://
www3.sprint.com/PR/CDA/PR_CDA_Press_Releases_Detail/0,3245,,00.html?ID=1294.
Food Marketing Institute. 1998. A Retailer’s Guide to Electronic Payment System Costs, Washington,
DC: Food Marketing Institute.
Frankel, Alan S., 1998. “Monopoly and Competition in the Supply and Exchange of Money,”
Antitrust Law Journal, vol. 66, no. 2 (Winter), pp. 313-361.
Frew, J. and G. Jud (1987) “Who Pays the Real Estate Broker’s Commission?” Research in Law and
Economics: The Economics of Urban Property Rights, 10: 177-188.
Friedman, Jon and John Meehan (1992) House of Cards: Inside the Troubled Empire of American
Express, Putnam, New York.
Gabszewicz, Jean J. and Xavier Y. Wauthy. 2004. “Two-Sided Markets and Price Competition with
Multi-homing,” mimeo, CORE, Catholic University of Louvain, Belgium.
Gans, Joshua S., and Stephen P. King. 2003a. “Approaches to Regulating Interchange Fees in
Payment Systems,” Review of Network Economics, vol. 2, no. 2 (June), pp. 125-145.
PLATFORM ECONOMICS:
438 Part Five : Software Platforms Essays on Multi-Sided Businesses
__________, and __________. 2003b. “The Neutrality of Interchange Fees in Payment Systems,”
Topics in Economic Analysis and Policy, vol. 3, no. 1, online: http://www.bepress.com/bejeap/
topics/vol3/iss1/art1/.
__________, and __________. 2003c. “A Theoretical Analysis of Credit Card Reform in
Australia,” Economic Record, v. 79, issue 247 (December), pp. 462-472.
Garcia Swartz, Daniel D., Robert W. Hahn, and Anne Layne-Farrar. 2004. “The Economics of
a Cashless Society: An Analysis of the Costs and Benefits of Payment Instruments,” AEI-
Brookings Joint Center for Regulatory Studies, Related Publication 04-24, September.
Gawer, Annabelle and Michael A. Cusumano (2002) Platform Leadership, Harvard Business School
Press, Boston, MA.
George, Lisa and Joel Waldfogel (2000) “Who Benefits Whom in Daily Newspaper Markets?”
NBER Working Paper #7944.
Goettler, Ronald. (1999) “Advertising Rates, Audience Composition, and Competition in the
Network Television Industry,” Carnegie Mellon University GSIA Working Paper #1999-E28.
Goldberger, Arthur S. (1991) A Course in Econometrics, Cambridge, Massachusetts: Harvard
University Press.
Goldberg, Pinelopi and Michael Knetter (1997) “Goods Prices and Exchange Rates: What Have
We Learned?” Journal of Economic Literature 35(3): 1243-1272.
“Google’s growth raises privacy concerns.” 2008. CNN.com. November 4.
“GoTo to Overture to YSM-Timeline,” 2005. September 21. bg Theory website, at <http://www.
bgtheory.com/blog/goto-to-overture-to-ysm-timeline/>.
Graeme, James (2003) “Loyalty is No Longer Paying Off,” Sunday Telegraph, May 4.
Greene, William H (1993) Econometric Analysis. New York: Macmillan, 1993.
Grossman, Peter (1987) American Express: The Unofficial History of the People Who Built the Great
Empire, Crown Publishers, New York.
Guerrera, Francesco (2002) “Brussels Set to Let Visa off the Hook,” Financial Times, May 13.
Guthrie, Graeme, and Julian Wright. 2003. “Competing Payment Schemes,” mimeo, National
University of Singapore, September.
Hacker, Scot (2001) “He Who Controls the Bootloader,” Byte, August, http://www.byte.com/
documents/s=1115/byt20010824s0001/0827_hacker.html.
Hagel, John and Marc Singer. 1999. Net Worth. Cambridge: Harvard Business School Press.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 439
Hagiu, Andrei. 2004a. “Two-Sided Platforms: Pricing and Social Efficiency,” RIETI Discussion
Paper Series 04-E-035, Research Institute for the Economy, Trade and Industry, Tokyo,
December.
__________. 2004b. “Two-Sided Platforms: Pricing and Social Efficiency – Extensions,” RIETI
Discussion Paper Series 04-E-036, Research Institute for the Economy, Trade and Industry,
Tokyo, December.
Hallerman, David. 2008. “US Online Advertising: Resilient in a Rough Economy: Summary.” March,
available at eMarketer.com, <http://www.emarketer.com/Reports/All/Emarketer_2000488.aspx>.
Hamilton, James (1994) Time Series Analysis. Princeton, NJ: Princeton University Press.
Hammer, Alexander R. (1962) “American Express Raises Profit: Credit Card Unit in the Black,”
The New York Times, April 25.
Hanna, Jim (2002) “Fed: Consumers to Save $500m-Plus Under Credit Card Shake-Up,” AAP
Newsfeed, August 27.
Hansen, Bruce (2001) “The New Econometrics of Structural Change: Dating Breaks in U.S. Labor
Productivity,” Journal of Economic Perspectives, 15(4).
Harris, B. (2003). “Recent Observations about Critical Loss Analysis.” Available at: http://www.
usdoj.gov/atr/public/workshops/docs/202599.htm
Harris, B. and C. Veljanovski (2003). “Critical Loss: Its Growing Use in Competition Law,”
European Competition Law Review, Vol. 5, pp. 213-218.
Harris, B. and J.J. Simons (1989). “Focusing Market Definition: How Much Substitution is
Necessary?,” Research in Law and Economics, Vol. 207, pp. 207-226.
Harris, Larry. 2002. Trading and Exchanges: Market Microstructure for Practitioners. Oxford
University Press.
Hayashi, Fumiko. 2004. “A Puzzle of Payment Card Pricing: Why Are Merchants Still Accepting
Card Payments?” mimeo, Federal Reserve Bank of Kansas City, December.
Hayashi, Fumiko, Richard Sullivan, and Stuart E. Weiner. 2003. A Guide to the ATM and Debit
Card Industry, Federal Reserve Bank of Kansas City.
Helft, Miguel. 2008.”Google’s New Tool Is Meant for Marketers.” The New York Times. (August 6),
at http://www.nytimes.com/2008/08/06/business/media/06adco.html.
“History of Pay Per Click Advertising and Yahoo! Sponsored Search,” 2007. November 1.
Semvironment.com website at <http://www.semvironment.com/the-history-of-pay-per-click-
advertising-and-yahoo-sponsored-search>.
PLATFORM ECONOMICS:
440 Part Five : Software Platforms Essays on Multi-Sided Businesses
Horan, Matthew (2003) “Banks Up Card Fees Yet Again,” Sunday Telegraph, September 21.
House Of Representatives Standing Committee On Economic Economics, Finance And Public
Administration (2005) Official Committee Hansard. August 12, online: http://www.aph.gov.
au/hansard/reps/commttee/R8516.pdf
HSN Consultants. 1987. The Nilson Report, no. 406 (June).
__________. 1991. The Nilson Report, no. 510 (October).
__________. 1992. The Nilson Report, no. 522 (April).
__________. 2001a. The Nilson Report, no. 737 (April).
__________. 2001b. The Nilson Report, no. 738 (April).
HSN Consultants (2003) The Nilson Report, no. 783, March.
Huber, Peter J. (19640 “Robust Estimation of a Location Parameter,” Annals of Mathematical
Statistics, 35(1): 73-101.
Humphrey, David, Magnus Willesson, Ted Lindblom, and Göran Bergendahl. 2003. “What Does
it Cost to Make a Payment?” Review of Network Economics, vol. 2, no. 2 (June), pp. 159-174.
Hunt, Robert M. 2003. “An Introduction to the Economics of Payment Card Networks,” Review
of Network Economics, vol. 2, no. 2 (June), 80-96.
IMA Market Development AB. 2000. “Study Regarding the Effects of the Abolition of the Non-
discrimination Rule in Sweden: Final Report, Results and Conclusions,” mimeo, Lerum,
Sweden, February.
Interactive Advertising Bureau. 2000-2007. Press Releases.
ITM Research. 2000. “The Abolition of the No-discrimination Rule.” mimeo, Amsterdam, March.
Ivaldi, M. and F. Verboven (2005). “Quantifying the Effects from Horizontal Mergers in European
Competition Policy,” International Journal of Industrial Organization, Vol. 23, pp. 669-691.
Jackson, William, III (1997) “Market Structure and the Speed of Price Adjustments: Evidence of
Non-monotonicity,” Review of Industrial Organization, 12.
Jeon, Doh-Shin, Jean-Jacques Laffont, and Jean Tirole (2001) “On the Receiver Pays Principle,”
mimeo.
Johnston, Jack (1983) Econometric Methods. New York: McGraw-Hill.
Jullien, Bruno (2001) “Competing in Network Industries: Divide and Conquer,” Institut
D’Economie Industrielle, Working Paper No. 9.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 441
Jullien, Bruno. 2004. “Two-Sided Markets and Electronic Intermediaries,” CESifo Working Paper
No. 1345, November.
Kaiser, U. and J. Wright (2006). “Price Structure in Two-Sided Markets: Evidence from the Magazine
Industry,” International Journal of Industrial Organization, Vol. 24, Issue 1, pp. 1-28.
Karrenbrok, Jeffrey D. (1991) “The Behavior of Retail Gasoline Prices: Symmetric or Not?,” Federal
Reserve Bank of St. Louis Review, 73.
Katz, M. and C. Shapiro (2003). “Critical Loss: Let’s Tell the Whole Story,” Antitrust Magazine,
Vol. 17, No. 2, pp. 49-56.
Katz, Michael L. (2001) “Network Effects, Interchange Fees, and No-Surcharge Rules in the
Australian Credit and Charge Card Industry,” Commissioned Report in Reform of Credit Card
Schemes in Australia, Reserve Bank of Australia, August.
Katz, Michael L. 2001. Reform of Credit Card Schemes in Australia II, Commissioned Report:
Network Effects, Interchange Fees, and No-Surcharge Rules in the Australian Credit and Charge
Card Industry, Reserve Bank of Australia, August.
Katz, Michael L., and Harvey S. Rosen (1998) Microeconomics. Boston, Massachusetts: Irwin/
McGraw-Hill.
Kaye, Barabara K and Norman Medoff. 2001. Just A Click Away: Advertising on the Internet.
Massachusetts: Allyn and Bacon.
Klein, B., Andres V. Lerner, Kevin M. Murphy and Lacey L. Plache (2006). “Competition in Two-
Sided Markets: The Antitrust Economics of Payment Card Interchange Fees,” Antitrust Law
Journal, 73(3), pp. 571-626.
Konkurrencestyrelsen. 2005. “Konkurrencestyrelsens Undersøgelse af Dankortgebyret,” ISA/1107-
8900-003/bjs/fab/lkf, Copenhagen, February.
Lafayette, Jon. 2008. “Nielsen Gains Access to Data From Charter’s Set-Top Boxes,” TVWeek.com.
(March 12), at <http://www.tvweek.com/news/2008/03/nielsen_gains_access_to_data_f.php>.
Lerner, Abba (1934) “The Concept of Monopoly and the Measurement of Monopoly Power,”
Review of Economic Studies, 1: 157-175.
Lerner, Josh (2002) “Did Microsoft Deter Software Innovation?” Working Paper, http://papers.
ssrn.com/sol3/delivery.cfm/SSRN_ID269498_code010528670.pdf?abstractid=2 69498.
Lipsey, R.G., and Kelvin Lancaster. 1956. “The General Theory of the Second Best,” Review of
Economic Studies, vol. 24, no. 1 (October), pp. 11-32.
Littlechild, Stephen. 2003. “Reflections on Incentive Regulation,” Review of Network Economics,
vol. 2, no. 4 (December), pp. 289-315.
PLATFORM ECONOMICS:
442 Part Five : Software Platforms Essays on Multi-Sided Businesses
Network Economics Consulting Group (2005) “Early evidence of the impact of Reserve Bank of
Australia regulation of open credit card schemes,” May.
Nevo, A. (2000). “Mergers with Differentiated Products: The Case of the Ready-to-Eat Cereal
Industry,” RAND Journal of Economics, 31(3), pp. 395-421.
Newmark, David, and Steven Sharpe (1992) “Market Structure and the Nature of Price Rigidity:
Evidence from the Market for Consumer Deposits,” Quarterly Journal of Economics, 107.
Nocera, Joseph. (1994) A Piece of the Action: How the Middle Class Joined the Money Class, Simon
& Schuster, New York.
O’Brien, D. and A.Wickelgren. (2003). “A Critical Analysis of Critical Loss,” FTC Working
Paper 254.
Office of Fair Trading. 2003a. “MasterCard Interchange Fees: Preliminary Conclusions,” London,
OFT634, February.
Office of Fair Trading. 2003b. “UK Payment Systems: An OFT Market Study of Clearing Systems
and Review of Plastic Card Networks,” London, OFT658, May.
O’Hara, Maureen. 1998. Market Microstructure Theory. Oxford: Blackwell Publishing Inc.
Pacelle, Mitchell, Jathon Sapsford and Kara Scannell (2003) “Visa to Settle With Retailers for $2
Billion,” The Wall Street Journal, May 1.
Parker, Geoffrey and Marshall W. Van Alstyne (2002) “Unbundling the Presence of Network
Externalities,” Working Paper, http://www.idei.asso.fr/Commun/Conferences/Internet/
Janvier2003/Papiers/VanAl.pdf.
Pashigian, B. Peter and Eric D. Gould (1998) “Internalizing Externalities: The Pricing of Space in
Shopping Malls,” Journal of Law and Economics, 41: 115-142.
Paul, Ryan. 2008. “EU calls phoul over ad company Phorm’s invasive snooping,” Ars Technica
(August 15), available at http://arstechnica.com/old/content/2008/08/eu-calls-phoul-over-ad-
company-phorms-invasive-snooping.ars.
Peltzman, Sam (2000) “Prices Rise Faster than they Fall,” Journal of Political Economy, 108(3).
Perron, Pierre (1989) “The Great Crash, the Oil Price Shock, and the Unit Root Hypothesis,”
Econometrica, 57(6): 1361-1401.
Poirier, Dale J. (1976) The Econometrics of Structural Change. Amsterdam: North Holland.
Poirier, Dale J. and Steven G. Garber (1974) “The Determinants of Aerospace Profit Rates 1951-
1971,” Southern Economic Journal, 41(2): 228-238.
PLATFORM ECONOMICS:
444 Part Five : Software Platforms Essays on Multi-Sided Businesses
Pope, Daniel. 1983. Making of Modern Advertising. New York, NY: Basic Books.
Poterba, James (1999) “Retail Price Reactions to Changes in State and Local Sales Taxes,” National
Tax Journal, 49(2).
Reserve Bank of Australia (2001) “Reform of Credit Card Schemes in Australia I: A Consultation
Document,” December.
Reserve Bank of Australia. 2001. “Reform of Credit Card Schemes in Australia I: A Consultation
Document,” December.
__________. 2002. “Reform of Credit Card Schemes in Australia IV: Final Reforms and Regulation
Impact Statement,” August.
__________. 2004. “Payment System Board Annual Report 2004.”
__________. 2005a. “Reform of the EFTPOS and Visa Debit Systems in Australia: A Consultation
Document,” February.
__________. 2005b. “Additional Credit Card Statistics,” May, online: http://www.rba.gov.au/
PaymentsSystem/AustralianPaymentsSystem/ExcelFiles/RPS.xls.
Reserve Bank of Australia (2002) “Reform of Credit Card Schemes in Australia IV: Final Reforms
and Regulation Impact Statement,” August.
Reserve Bank of Australia (2004) “Payment System Board Annual Report 2004.”
Reserve Bank of Australia (2005a) “Banking Fees in Australia.”
Reserve Bank of Australia (2005b) “Table C01: Credit and Charge Card Statistics.” online: http://
www.rba.gov.au/Statistics/Bulletin/C01hist.xls.
Reserve Bank of Australia (2005c) “Additional Credit Card Statistics.” Online: http://www.rba.
gov.au/PaymentsSystem/PaymentsStatistics/ExcelFiles/RPS.xls
Reserve Bank of Australia (2005d) “Market Shares Of Credit And Charge Card Schemes.” Online:
http://www.rba.gov.au/Statistics/Bulletin/C02hist.xls.
Reserve Bank of Australia and Australian Competition and Consumer Commission (2000) “Debit
and Credit Card Schemes in Australia: A Study of Interchange Fees and Access,” October.
Rochet, J-C. and J. Tirole (2003). “Platform Competition in Two-Sided Markets,” Journal of the
European Economic Association, 1:4, pp. 990-1029.
Rochet, J-C. and J. Tirole (2006). “Two-Sided Markets: A Progress Report,” RAND Journal of
Economics, Vol. 37, Issue. 3, pp. 645-667.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 445
Rochet, Jean-Charles (2003) “The Theory of Interchange Fees: A Synthesis of Recent Contributions,”
Review of Network Economics, 2(2): 97-124.
Rochet, Jean-Charles. 2003. “The Theory of Interchange Fees: A Synthesis of Recent Contributions,”
Review of Network Economics, vol. 2, no. 2 (June), pp. 97-124.
Rochet, Jean-Charles (2003) “The Theory of Interchange Fees: A Synthesis of Recent Contributions,”
Working Paper.
Rochet, Jean-Charles (2005) “The Interchange Fee Mysteries—Discussion of ‘Economic Rationale
for Interchange Fees’ by D. Evans and R. Schmalensee,” presented at a Conference Interchange
Fees in Credit and Debit Card Industries: What Role for Public Authorities? Federal Reserve Bank
of Kansas City, Santa Fe, New Mexico, May 4-6.
Rochet, Jean-Charles and Jean Tirole (2002) “Cooperation Among Competitors: Some
Rochet, Jean-Charles, and Jean Tirole. 2002. “Cooperation among Competitors: Some Economics
of Payment Card Associations,” Rand Journal of Economics, vol. 33, no. 4 (Winter), pp. 549-570.
__________, and __________. 2003a. “An Economic Analysis of the Determination of Interchange
Fees in Payment Card Systems,” Review of Network Economics, vol. 2, no. 2 (June), pp. 69-79.
__________, and __________. 2003b. “Platform Competition in Two-Sided Markets,” Journal
of the European Economics Association, vol. 1, no 4 (June), pp. 990-1209.
__________, and __________. 2004a. “Defining Two-Sided Markets,” mimeo, IDEI, Toulouse,
France, January.
__________. and __________. 2004b. “Two-Sided Markets: An Overview.” Mimeo, IDEI,
Toulouse, France, March.
Rochet, Jean-Charles and Jean Tirole (2002) “Cooperation among Competitors: Some Economics
of Payment Card Associations,” The RAND Journal of Economics, 33(4): 1-22.
Rochet, Jean-Charles and Jean Tirole (2003) “Platform Competition in Two-Sided Markets,”
Journal of the European Economic Association, 1(4): 990-1029.
Rochet, Jean-Charles and Jean. Tirole. 2003. “Platform Competition in Two-Sided Markets.”
Journal of the European Economic Association, 1(4), 990-1029.
Rochet, Jean-Charles and Jean Tirole (2003) “Platform Competition in Two-Sided Markets,”
Journal of the European Economics Association (forthcoming).
Rochet, Jean-Charles and Jean. Tirole. 2006. “Externalities and Regulation in Card Payment
Systems.” Review of Network Economics 5(1), 1-14.
PLATFORM ECONOMICS:
446 Part Five : Software Platforms Essays on Multi-Sided Businesses
Ronan, Courtney (1998) “Apartment Locaters: How Do They Make Their Money?” Realty Times,
June 30, http://realtytimes.com/rtnews/rtcpages/19980630_aptlocator.htm.
Rysman, M. (2004). “Competition Between Networks: A Study of the Market for Yellow Pages,”
Review of Economic Studies, 71, pp. 483-512.
Rysman, M. (2007). “An Empirical Analysis of Payment Card Usage,” Journal of Industrial
Economics, 55, pp. 1-36.
Rysman, Marc (2002) “Competition Between Networks: A Study of the Market for Yellow Pages
1-2,” Boston University Industrial Studies Project, Working Paper No. 104.
Rysman, Marc. 2004. “An Empirical Analysis of Payment Card Usage,” mimeo, Boston University,
January.
Scheffman, D. and J. Simons (2003). “The State of Critical Loss: Let’s Make Sure We Understand
the Whole Story,” The Antitrust Source, Vol. 3, No. 2, pp. 1-9.
Scherreik, Susan (2002) “Is Your Broker Leaving You Out in the Cold?” Business Week, http://
www.businessweek.com/magazine/content/02_07/b3770110.htm.
Schiff, Aaron (2003) “Open and Closed Systems of Two-Sided Networks,” Information Economics
and Policy, forthcoming.
Schiff, Aaron. 2003. “Open and Closed Systems of Two-Sided Networks,” Information Economics
and Policy, vol. 15, no. 4 (December), pp. 425-442.
Schmalensee, Richard (2002) “Payment Systems and Interchange Fees,” Journal of Industrial
Economics, 50(2): 103-122.
Schmalensee, Richard. 2002. “Payment Systems and Interchange Fees,” Journal of Industrial
Economics, vol. 50, no. 2 (June), pp. 103-122.
__________. 2003. “Interchange Fees: A Review of the Literature,” Payment Card Economics
Review, vol. 1, no. 1 (Winter), pp. 25-44.
Schmalensee, Richard (2003) “Interchange Fees: A Review of the Literature,” Payment Card
Economics Review, 1: 25-44.
Schumpeter, Joseph. 1942. Capitalism, Socialism and Democracy. New York: Harper and Brothers.
Schwartz, Marius, and Daniel R. Vincent. 2004. “The No Surcharge Rule and Card User Rebates:
Vertical Control by a Payment Network,” mimeo, Georgetown University.
Sears, Jay. 2005. “The Evolution of Contextual Advertising,” Upgrade Magazine. (April/May), at
<http://www.contextweb.com/pdf/UpgradeMagazine.pdf>.
Shapiro, Carl (1996). “Mergers with Differentiated Products,” Antitrust, 10, pp. 23-30.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 447
Shapiro, Carl and Hal R. Varian (1999) Information Rules, Harvard Business School Press,
Boston, MA.
Silk, Alvin J., Lisa R. Klein and Ernst R. Berndt. 2001. “The Emerging Position of the Internet as
an Advertising Medium.” Netnomics 3, 129:148.
Silk, Alvin J., Lisa R. Klein and Ernst R. Berndt. 2002. “Intermedia Substitutability and Market
Demand by National Advertisers.” Review of Industrial Organization 20: 323-348, 2002.
__________, Simon Cowan, and John Vickers. 1994. Regulatory Reform—Economic Analysis and
British Experience. Cambridge, Massachusetts: MIT Press.
Small, John, and Julian Wright. 2000. “Decentralized Interchange Fees in Open Payment Networks:
An Economic Analysis,” mimeo, University of Aukland, December.
Stavins, Joanna. 1997. “A Comparison of Social Costs and Benefits of Paper Check Presentment
and ECP Truncation,” New England Economic Review, (July/August), pp. 27-44.
Stigler, George J., and James K. Kindahl (1970) The Behavior of Industrial Prices. New York:
National Bureau of Economic Research.
Stiglitz, Joseph and Carl Walsh. 2002. Principles of Microeconomics. New York: W. W. Norton.
Swedish Bankers’ Association. 2004. Banks in Sweden—Facts about the Swedish banking market.
September.
Thaler, Richard H. and Cass R. Sunstein. 2008. Nudge: Improving Decisions about Health, Wealth,
and Happiness. Yale University Press.
Tirole, Jean. 1988. The Theory of Industrial Organization, Cambridge, Massachusetts: MIT Press.
Tyagi, Rajeev K. (1999) “A Characterization of Retailer Response to Manufacturer Trade Deals,”
Journal of Marketing Research, 36: 510-516.
United States Census Bureau. 2004. “2002 Economic Census: Industry Series,” online: http://
www.census.gov/econ/census02/guide/INDSUMM.HTM.
United States Department of Commerce: Bureau of Economic Analysis. 2003 “Imputations in the
National Income and Product Accounts,” September 29.
__________. 2004 “Gross Domestic Product: Implicit Price Deflator,” November 30. van Raalte,
Chris, and Harry Webers. 1998. “Spatial Competition with Intermediated Matching,” Journal
of Economic Behavior and Organization, vol. 34, no. 3 (March), pp. 477-488.
U.S. Department of Commerce. 2008. Quarterly Retail E-Commerce Sales 3rd Quarter 2008, Press
Release (November 19).
PLATFORM ECONOMICS:
448 Part Five : Software Platforms Essays on Multi-Sided Businesses
Varian, Hal R. 2007. “Position Auctions.” International Journal of Industrial Organization 25(6)
Winter: 1163-1178.
“Verizon posts in-line 1Q adj eps; wireless biz adds 1.5M customers.” 2008. Forbes.com. (April 28),
at <http://www.forbes.com/feeds/afx/2008/04/28/afx4939948.html>.
Vickers, John. 2005. “Public Policy and the Invisible Price: Competition Law, Regulation, and the
Interchange Fee, paper presented at a Conference Interchange Fees in Credit and Debit Card
Industries: What Role for Public Authorities? Federal Reserve Bank of Kansas City, Santa Fe, New
Mexico, May 4-6, 2005
Visa International. 2001. International Country Overviews, October.
Visa USA (various years) “Payment Systems Panel Study.”
Viscusi, W. Kip, John M. Vernon, and Joseph E. Harrington, Jr. 2000. Economics of Regulation and
Antitrust, Cambridge, Massachusetts: MIT Press.
Walters, Rockney G. (1989) “An Empirical Investigation into Retailer Response to Manufacturer
Trade Promotions,” Journal of Retailing, 65(2): 253-272.
Weiner, Stuart E., and Julian Wright. 2005. “Interchange Fees in Various Countries: Developments
and Determinants,” paper presented at a Conference Interchange Fees in Credit and Debit Card
Industries: What Role for Public Authorities? Federal Reserve Bank of Kansas City, Santa Fe, New
Mexico, May 4-6, 2005.
Weiner, Stuart E. and Julian Wright (2005) “Interchange Fees in Various Countries: Developments
and Determinants,” Working Paper, April 25.
Werden, G.J. and L. M. Froeb (1994). “The Effects of Mergers in Differentiated Products
Industries: Logit Demand and Merger Policy,” Journal of Law, Economics, & Organization,
194, pp. 407-426.
Willig, R. (1991). “Merger Analysis, Industrial Organization Theory, and Merger Guidelines,”
Brookings Papers on Economic Activity: Microeconomics, 281, pp. 292-293.
Winston, Clifford, and Steven A. Morrison. 1986. The Economic Effects of Airline Deregulation,
Washington, DC: The Brookings Institution.
Wolters, Timothy (2000) “Carry Your Credit Card in Your Pocket: The Early History of the Credit
Card at Bank of America and Chase Manhattan,” Enterprise and Society, 1: 315-354.
Wolters, Timothy. 2000. ““Carry Your Credit in Your Pocket”: The Early History of the Credit
Card at Bank of America and Chase Manhattan,” Enterprise and Society, vol. 1, no. 2 (June),
pp. 315-354.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses What’s Next in Payments: Invisible Engines 449
Wong, Wylie (1998) “Netscape Applauds Microsoft Suit,” TechWeb, May 20, http://www.techweb.
com/wire/story/msftdoj/TWB19980519S0007.
Wright, J. (2004). “The Determinants of Optimal Interchange Fees in Payment Systems,” Journal
of Industrial Economics, Vol. 52, No. 1, March, pp. 1-26.
Wright, Julian. 2000. “An Economic Analysis of a Card Payment Network.” Mimeo, NECG and
University of Auckland, December.
__________. 2003a. “Optimal Card Payment Systems,” European Economic Review, vol. 47, no.
4 (August), pp. 587-612.
__________. 2003b. “Pricing in Debit and Credit Card Schemes,” Economics Letters, vol. 80, no.
3 (September), pp. 305-309.
__________. 2003c. “Why Do Firms Accept Credit Cards?” mimeo, University of Auckland.
__________. 2004a. “The Determinants of Optimal Interchange Fees in Payment Systems,”
Journal of Industrial Economics, vol. 52, no. 1 (March), pp. 1-26.
__________. 2004b. “One-sided Logic in Two-sided Markets,” Review of Network Economics, vol.
3, no. 1 (March), pp. 44-64.
Yahoo Press Release. 2007. “Yahoo!’s New “SmartAds” Meld Brand and Direct Response Advertising.” (July
2), at <http://yahoo.client.shareholder.com/PRESS/ReleaseDetail.cfm?ReleaseID=252034>.
PLATFORM ECONOMICS:
Essays on Multi-Sided Businesses
David S. Evans, is an economist, business advisor and Founder of Market
Platform Dynamics. He is a recognized global authority on the design
and implementation of platform business strategies, a body of work that
he pioneered and that was captured in Invisible Engines: How Software
Platforms Drive Innovation and Transform Industries, a 2006 American
Publishers Association best business book award winner. David has
consulted with many of the leading companies in the world and has served
as an advisor to their CEOs and Boards of Directors. A frequent keynote speaker, David is often
called upon to lead thought provoking industry discussions and to share his insights on the
success and failure of platform businesses worldwide. David is also an advisor to and investor in
many emerging ventures.
CPI provides comprehensive resources and continuing education for the global antitrust and
competition policy community. Created and managed by leaders in the competition policy
community, CPI delivers timely com-mentary and analysis on antitrust and global competition
policy matters through a variety of media and applications.
Visit: competitionpolicyinternational.com