Financial Dependence and Growth

Raghuram G. Rajan
and
Luigi Zingales 

University of Chicago & NBER

Abstract
This paper examines whether nancial development facilitates economic growth by scru-
tinizing one rationale for such a relationship; that nancial development reduces the costs of
external nance to rms. Specically, we ask whether industrial sectors that are relatively
more in need of external nance develop disproportionately faster in countries with more
developed nancial markets. We nd this to be true in a large sample of countries over the
1980s. We show this result is unlikely to be driven by omitted variables, outliers, or reverse
causality. (JEL O4, F3, G1)

A large literature, dating at least as far back as Joseph A. Schumpeter (1911), emphasizes
the positive in uence of the development of a country's nancial sector on the level and the rate
of growth of its per capita income. The argument essentially is that the services the nancial
sector provides { of reallocating capital to the highest value use without substantial risk of
loss through moral hazard, adverse selection, or transactions costs { are an essential catalyst of
economic growth. Empirical work seems consistent with this argument. For example, on the
basis of data from 35 countries between 1860 and 1963, Raymond W. Goldsmith (1969, p48)
 Raghuram G. Rajan and Luigi Zingales are both at the University of Chicago, Graduate School of Business,
1101 E 58th street, Chicago IL 60637. We thank George Benston, Marco Da Rin, Eugene Fama, Peter Klenow,
Krishna Kumar, Ross Levine, Jonathan Macy, Colin Mayer, Canice Prendergast, Andres Rodriguez-Clare, David
Scharfstein, Robert Vishny, and two anonymous referees for valuable comments. Jayanta Sen, Dmitrii Kachintsev,
and Alfred Shang provided excellent research assistance. A preliminary study was supported by the World Bank.
We gratefully acknowledge nancial support from NSF grant #SBR-9423645.

1
concludes that \a rough parallelism can be observed between economic and nancial develop-
ment if periods of several decades are considered". Nevertheless, studies such as these simply
suggest correlation. As Goldsmith puts it \There is no possibility, however, of establishing
with condence the direction of the causal mechanism, i.e., of deciding whether nancial factors
were responsible for the acceleration of economic development or whether nancial development
re ected economic growth whose mainsprings must be sought elsewhere." While Goldsmith is
agnostic, other economists have expressed downright scepticism that nancial development is
anything but a sideshow to economic development. Joan Robinson (1952, p86) is representative
of such a viewpoint when she claims \where enterprise leads, nance follows".
In an important recent paper, Robert G. King and Ross Levine (1993a) investigate the
causality problem following a post hoc, ergo propter hoc approach. They show that the pre-
determined component of nancial development is a good predictor of growth over the next 10
to 30 years. However, the sceptic could still oer a number of arguments against attributing
causality.
First, both nancial development and growth could be driven by a common omitted variable
such as the propensity of households in the economy to save. Since endogenous savings (in
certain models of growth) aects the long run growth rate of the economy, it may not be
surprising that growth and initial nancial development are correlated. This argument is also
hard to refute with simple cross-country regressions. In the absence of a well accepted theory
of growth, the list of potential omitted variables that nancial sector development might be a
proxy for is large, and the explanatory variables to include a matter of conjecture.
Second, nancial development { typically measured by the level of credit and the size of the
stock market { may predict economic growth simply because nancial markets anticipate future
growth; the stock market capitalizes the present value of growth opportunities, while nancial
institutions lend more if they think sectors will grow. Thus nancial development may simply
be a leading indicator rather than a causal factor.
One way to make progress on causality is to focus on the details of theoretical mechanisms
through which nancial development aects economic growth, and document their working.
Our paper is an attempt to do this. Specically, theorists argue that nancial markets and
institutions help a rm overcome problems of moral hazard and adverse selection, thus reducing

2
the rm's cost of raising money from outsiders. So nancial development should disproportion-
ately help rms (or industries) typically dependent on external nance for their growth. Such
a nding could be the `smoking gun' in the debate about causality. There are two virtues to
this simple test. First, it looks for evidence of a specic mechanism by which nance aects
growth, thus providing a stronger test of causality. Second, it can correct for xed country
(and industry) eects. Though its contribution depends on how reasonable our micro-economic
assumptions are, it is less dependent on a specic macroeconomic model of growth.
We construct the test as follows. We identify an industry's need for external nance (the
dierence between investments and cash generated from operations) from data on U.S. rms.
Under the assumption that capital markets in the United States, especially for the large listed
rms we analyze, are relatively frictionless, this method allows us to identify an industry's tech-
nological demand for external nancing. Under the further assumption that such a technological
demand carries over to other countries, we examine whether industries that are more depen-
dent on external nancing grow relatively faster in countries that, a priori, are more nancially
developed.
This would imply that, ceteris paribus, an industry such as Drugs and Pharmaceuticals,
which requires a lot of external funding, should develop relatively faster than Tobacco, which
requires little external nance, in countries that are more nancially developed. Consider, for
instance, Malaysia, Korea, and Chile, which are moderate-income, fast-growing, countries, that
dier considerably in their nancial development. Consistent with our hypothesis, in Malaysia,
which was the most nancially developed by our measures, Drugs and Pharmaceuticals grew at a
4 percent higher annual real rate over the 1980s than Tobacco (the growth rate for each industry
is adjusted for the worldwide growth rate of that industry). In Korea, which was moderately
nancially developed, Drugs grew at a 3 percent higher rate than Tobacco. In Chile, which
was in the lowest quartile of nancial development, Drugs grew at a 2.5 percent lower rate than
Tobacco. So nancial development seems to aect relative growth rates of industries in the way
predicted. We establish this result more systematically for a large cross-section of industries
and countries in the body of the paper.
Delving deeper into the components of growth, industry growth can be decomposed into the
growth in the number of establishments and the growth in the average size of existing estab-

3
lishments. New establishments are more likely to be new rms, which depend more on external
nance than established rms. So the growth of the number of establishments in industries
dependent on external nance should be particularly sensitive to nancial development. This
is indeed the case. Our estimates suggest that nancial development has almost twice the eco-
nomic eect on the growth of the number of establishments as it has on the growth of the
average size of establishments. This suggests that an additional indirect channel through which
nancial development could in uence growth is by disproportionately improving the prospects
of young rms. If these are typically innovators, they make possible Schumpeterian \waves of
creative destruction" that would not even get initiated in countries with less developed markets.
Let us be careful about what we nd, and about what we have little to say. Our ndings
suggest that the ex ante development of nancial markets facilitates the ex post growth of sectors
dependent on external nance. This implies that the link between nancial development and
growth identied elsewhere may stem, at least in part, from a channel identied by the theory:
nancial markets and institutions reduce the cost of external nance for rms. Of course, our
analysis suggests only that nancial development liberates rms from the drudgery of generating
funds internally. It is ultimately the availability of protable investment opportunities that
drives growth, and we have little to say about where these come from. In the imagery of Rondo
Cameron (1967, p2), we nd evidence consistent with nance as a lubricant, essential no doubt,
but not a substitute for the machine.
Our paper relates closely to three recent papers that attempt to establish the direction
of causation of the nance-growth correlation. Asli Demirguc-Kunt and Vojislav Maksimovic
(1996) also use micro data to develop a test of the in uence of nancial development on growth.
Using rm-level data, they estimate the proportion of rms whose rate of growth exceeds the
growth that could have been supported only by internal resources. They then run a cross country
regression and nd that this proportion is positively related to the stock market turnover and
to a measure of law enforcement. There are two essential dierences from our paper. First,
their estimate of the internal growth rate of a rm is dependent on the rm's characteristics.
While it is potentially more accurate than our measure of external dependence, it is also more
endogenous. Second, they focus on between-country dierences in the spirit of traditional cross-
country regressions, while our focus is on within-country, between-industry dierences. The

4
latter is an important innovation in this paper.
Jith Jayaratne and Philip E. Strahan (1996) examine the liberalization of the banking sector
in dierent states in the United States in recent years and show that this had a positive in uence
on a state's growth. Our attempt to correct for xed eects is similar to theirs. They use
dierences in growth rates across the temporal shock of liberalization while we use dierences
between industries within a country to do so. Since they focus on a very nice natural experiment
to provide identication, their methodology may be harder to apply to dierent countries or
dierent questions. But the more important dierence is that we focus on providing evidence
for a micro-economic channel through which nance is supposed to work rather than examining,
as they do, the broader correlation between nance and growth.
Finally, Levine and Sarah Zervos (1996) study whether stock markets and banks promote
economic growth. They nd that measures of market liquidity are strongly related to growth,
capital accumulation, and productivity, while surprisingly, more traditional measures of devel-
opment such as stock market size are not as robustly correlated. They also nd that bank
lending to the private sector has a strong independent eect on growth. They focus on a richer
set of measures of nancial development and growth than we do, but their cross-country regres-
sion methodology is also more traditional. The two studies should be viewed as complementary,
theirs providing information on a broader set of correlations, while ours details a mechanism.
The rest of the paper is as follows. We start by describing the theoretical underpinnings of
our work in section 1 and then our measure of external dependence in section 2. In section 3,
we present our data on nancial development, country characteristics, and industry growth.
In section 4 we set up our main test and discuss the results. We explore other tests and the
robustness of our ndings in section 5. Section 6 concludes.

I Theoretical Underpinnings and The Basic Test.

A Theoretical Underpinnings
There has been extensive theoretical work on the relationship between nancial development
and economic growth. Economists have emphasized the role of nancial development in better
identifying investment opportunities, reducing investment in liquid but unproductive assets,

5
mobilizing savings, boosting technological innovation, and improving risk taking.1 All these
activities can lead to greater economic growth. We do not have the space to go into all these
theories (see Levine (1997) for a comprehensive recent survey) so we content ourselves with
outlining the essential theoretical underpinnings for our test.
Jeremy Greenwood and Boyan Jovanovic (1990) develop a model where the extent of nancial
intermediation and economic growth are endogenously determined. In their model, nancial
intermediaries can invest more productively than individuals because of their better ability
to identify investment opportunities. So nancial intermediation promotes growth because it
allows a higher rate of return to be earned on capital, and growth in turn provides the means
to implement costly nancial structures.
Equivalently, the model could be recast to show that nancial development reduces the cost
of raising funds from sources external to the rm relative to the cost of internally generated
cash ows. External funds are generally thought to be costlier because outsiders have less con-
trol over the borrower's actions (see, for example, Michael C. Jensen and William R. Meckling
(1976)) or because they know less about what the borrower will do with the funds (see Joseph
E. Stiglitz and Andrew Weiss (1981) and Stewart C. Myers and Nicholas S. Majluf (1984)). Fi-
nancial development, in the form of better accounting and disclosure rules, and better corporate
governance through institutions, will reduce the wedge between the cost of internal and external
funds and enhance growth, especially for rms that are most reliant on external nancing.2
A second issue is how nancial development takes place. Some economists take the develop-
ment of the nancial market as exogenous to the model arguing that \dierences in the extent
of nancial markets across countries seem to depend primarily on legislation and government
regulation" (Bencivenga and Smith (p 207)). By contrast, Greenwood and Jovanovic (1990)
have a \once-and-for-all" lump sum cost of development and development is endogenous to
their framework. From the perspective of our paper, it really does not matter whether legal and
political or economic forces are responsible for nancial development. Our focus is on whether
the pre-determined level of nancial development aects growth. All we need for the stock of
nancial development to matter even when development is endogenous is that there be a cost to
development (as in Greenwood and Jovanovic) or that nancial development cannot happen in-
stantaneously (as in reputational models of nancial development such as Douglas W. Diamond

6
(1989)). Either assumption seems plausible.
If nancial development cannot take place at low cost and on the y, the above theories would
suggest that the a priori existence of a well-developed nancial market should disproportionately
improve the ex post growth rates of industries that are technologically more dependent on
external funds.

B The basic test

The most disaggregated comprehensive data on growth that we have for countries is at the
industry level (data at the rm level, if available, is typically limited to large listed rms). Our
hypothesis is that industries that are more dependent on external nancing will have relatively
higher growth rates in countries that have more developed nancial markets.
Therefore, the dependent variable is the average annual real growth rate of value added in
industry j in country k over the period 1980-1990. If we can measure industry j's dependence
on external nance and country k's nancial development, then after correcting for country and
industry eects we must nd that the coecient estimate for the interaction between dependence
and development is positive.
The most eective way of correcting for country and industry characteristics is to use indi-
cator variables, one for each country and industry. Only additional explanatory variables that
vary both with industry and country need be included. These are industry j's share in country
k of total value added in manufacturing in 1980 and the primary variable of interest, the inter-
action between industry j's dependence on external nancing and nancial market development
in country k.
The model we want to estimate is then

Growthj;k = Constant + 1::m Country Indicators + m+1::n Industry Indicators+

(1) n+1 (Industry j 0s share of manufacturing in country k in 1980)+

n+2 (External Dependence of industry j Financial Development of country k) + j;k

Of course, in order to estimate the model, we need appropriate measures of nancial development
and external dependence. This is what we will examine shortly.
Before proceeding, we point out that our study has one important advantage over recent

7
cross-country empirical studies of growth.3 That advantage is simply that we make predictions
about within country dierences between industries based on an interaction between a country
and industry characteristic. Therefore, we can correct for country and industry characteristics
in ways that previous studies were unable to correct for, and will be less subject to criticism
about an omitted variable bias or model specication.

II A Measure of Dependence on External Finance

A The proxy for dependence.
Data on the actual use of external nancing is typically not available. But even if it were, it
would not be useable because it would re ect the equilibrium between the demand for external
funds and its supply. Since the latter is precisely what we are attempting to test for, this
information is contaminated. Moreover, we are not aware of systematic studies of the external
nancing needs of dierent industries, either cross-sectionally or over time.4
We, therefore, have to nd some other way of identifying an industry's dependence on ex-
ternal nancing. We assume that there is a technological reason why some industries depend
more on external nance than others. To the extent that the initial project scale, the ges-
tation period, the cash harvest period, and the requirement for continuing investment dier
substantially between industries, this is indeed plausible. Furthermore, we assume that these
technological dierences persist across countries, so that we can use an industry's dependence
on external funds as identied in the U.S. as a measure of its dependence in other countries.
While there are enormous dierences in local conditions between countries, all we really need is
that statements of the following sort hold: If pharmaceuticals require a larger initial scale and
have a higher gestation period before cash ows are harvested than the textile industry in the
U.S., it also requires a larger initial scale and has a higher gestation period in Korea.

B How the proxy is calculated.

We start by computing the external nancing needs of U.S. companies over the 1980s. We use
data from Compustat for this. Compustat does not contain a representative sample of U.S.
rms, because it is limited to publicly traded rms, which are relatively large. Nevertheless, we

8
regard this as an advantage for two reasons. First, in a perfect capital market the supply of
funds to rms is perfectly elastic at the proper risk adjusted rate. In such a market the actual
amount of external funds raised by a rm equals its desired amount. In other words, in such an
idealized setting, the identication problem does not exist. But capital markets in the United
States are among the most advanced in the world, and large publicly traded rms typically
face the least frictions in accessing nance. Thus the amount of external nance used by large
rms in the United States is likely to be a relatively pure measure of their demand for external
nance.5
A second reason for using a database on listed rms is that disclosure requirements imply
that the data on nancing are comprehensive. For most of the paper, we will take the amount
of external nance used by U.S. rms in an industry as a proxy for the desired amount foreign
rms in the same industry would have liked to raise had their nancial markets been more
developed.
Next, we have to dene precisely what we mean by external and internal nance. We are
interested in the amount of desired investment that cannot be nanced through internal cash
ows generated by the same business. Therefore, a rm's dependence on external nance is
dened as the ratio of capital expenditures (Compustat # 128) minus cash ow from operations
divided by capital expenditures. Cash ow from operations is broadly dened as the sum of
Compustat cash ow from operations (Compustat # 110) plus decreases in inventories, decreases
in receivables, and increases in payables.6 Note that this denition includes changes in the non-
nancial components of net working capital as part of funds from operations. In fact, in certain
businesses these represent major sources (or uses) of funds, that help a rm avoid (or force it
to tap) external sources of funds.7
Similarly, the dependence on external equity nance is dened as the ratio of the net amount
of equity issues (Compustat # 108 minus # 115) to capital expenditures. Finally, the investment
intensity is the ratio of capital expenditure to net property plant and equipment (Compustat
# 8).
To make these measures comparable with the industry level data we have for other countries,
we have to choose how to aggregate these ratios over time and across companies. We sum the
rm's use of external nance over the 1980s and then divide by the sum of capital expenditure

9
over the 1980s to get the rm's dependence on external nance in the 1980s. This smooths
temporal uctuations and reduces the eects of outliers. To summarize ratios across rms,
however, we use the industry median. We do this to prevent large rms from swamping the
information from small rms; for instance, we know that IBM's free cash ow does not alleviate
possible cash ow shortages of small computer rms.

C External dependence for dierent industries.

In Table 1, we tabulate by International Standard Classication Code (ISIC) the fraction of
investments U.S. rms nanced externally (rst column) and the level of capital expenditures
divided by net property plant and equipment (second column). We restrict our attention to those
manufacturing industries for which we have value-added data from the United Nations Statistics.
Drugs and Pharmaceuticals emerge as the industry that uses the most external nance, with
Plastics and Computing coming close behind. Tobacco, on the other hand, generates the most
excess cash ow and has negative external funding needs.
It is common wisdom in the corporate nance literature (though we were hard-pressed to nd
formal empirical studies of this phenomenon) that there is a life cycle in the pattern of nancing
for rms; rms are more dependent on external nancing early in their life than later. Figure 1
supports the common wisdom. It plots the median nancing and investment needs across U.S.
rms as a function of the number of years since the initial public oering (IPO). Not surprisingly,
in the year of the IPO, rms raise a substantial amount of external funds (especially equity).
More interestingly, this continues { albeit on a smaller scale { up to approximately the 10th
year. After that period, net equity issues go to zero and the usage of external nance uctuates
around zero. In the third and fourth columns of Table 1, we report the external dependence
and capital expenditures for mature companies (rms that were listed for more than 10 years),
while the fth and sixth columns are for young companies (rms that were listed for less than
10 years).8 This pattern appears to be fairly standard across dierent industries, though there
are exceptions. All this suggests that very young rms are more dependent on external nance
than older rms. This fact will provide an additional test of our hypothesis.

10
D Is the Dependence of U.S. Firms a Good Proxy?
Much of our analysis rests on dependence of U.S. rms on external nance being a good proxy
for the demand for external funds in other countries. We think this is reasonable for four reasons.
First, in a steady state equilibrium there will not be much need for external funds, as
Figure 1 shows. Therefore, much of the demand for external funds is likely to arise as a result
of technological shocks that raise an industry's investment opportunities beyond what internal
funds can support. To the extent these shocks are worldwide, the need for funds of U.S. rms
represents a good proxy.9
Second, even if the new investment opportunities generated by these worldwide shocks dier
across countries, the amount of cash ow produced by existing rms in a certain industry is likely
to be similar across countries. In fact, most of the determinants of ratio of cash ow to capital
are likely to be similar worldwide: the level of demand for a certain product, its stage in the life
cycle, and its cash harvest period. For this reason, we make sure that our results hold even when
we use the amount of internally generated cash, rather than the dierence between investments
and internally generated funds. We also check that the results hold when we use dependence as
measured in Canada, a country which has well developed capital markets but a very dierent
banking system and industry concentration than the United States. Unfortunately, we do not
have access to ow of funds data from any other countries, so we cannot venture further aeld,
but this methodology could, in principle, be used with dependence measured in any country
with well functioning capital markets.
Third, one might argue that the stage of the product life cycle that U.S. rms are in is likely
to be dierent from that of foreign rms. Given that our sample is biased toward developing
countries one might think that the U.S. industry in the 1970s might be a better proxy for the
position of developing countries in a product life cycle. For this reason, we also explore the
robustness of our results to measuring the dependence of U.S. rms in the 1970s rather than in
the 1980s. We also distinguish between dependence as measured for young rms in the United
States (less than 10 years from listing) and dependence for old rms (more than 10 years from
listing).
Last but not least, that we only have a noisy measure of the need for funds creates a bias
against nding any interaction between dependence and nancial development.

11
III Data.
A Data on industries.
Data on value added and gross xed capital formation for each industry in each country are
obtained from the Yearbook of Industrial Statistics (vol 1) database put together by the United
Nations Statistics Division. We checked the data for inconsistencies, changes in classication of
sectors, and changes in units. The U.N. data is classied by International SIC code. In order to
obtain the amount of external nance used by the industry in the U.S., we matched ISIC codes
with SIC codes.10 Typically, the three digit ISIC codes correspond to two digit SIC codes, while
the four digit ISIC code corresponds to three digit SIC codes. In order to reduce the dependence
on country specic factors like natural resources we conne our analysis to manufacturing rms
(U.S. SIC 2000-3999).
We would like data on as many countries as possible. The binding constraint is the avail-
ability of measures of nancial development (specically the availability of data on accounting
standards). Since we also wanted data on equity market capitalization, we started with the 55
countries from the Emerging Stock Markets Factbook. We dropped countries like Kuwait that
did not report a stock market capitalization till the latter half of the 1980s. We could not use
Hong Kong and Taiwan because data on these countries are not present in the International
Financial Statistics volumes. We also dropped countries for which we did not have data from
the Yearbook database that is separated by at least ve years (notably, Switzerland). Finally,
Thailand is dropped because the U.N. notes that data from year to year are not comparable.
The United States is excluded from the analysis because it is our benchmark. This leaves us
with the 43 countries in Table 2.
We want to see if nancially dependent industries are likely to be better o in countries
with well developed nancial sectors. The availability of nance aects not just investment but
also the ability to nance operations and sales through working capital. Therefore, the most
appropriate measure of an industry being \better o" is the growth in value added for that
industry, i.e., the change in the log of real value added in that industry between 1980 and 1990.
Real value added in 1990 is obtained by de ating value added by the Producer Price Index. For
high in ation countries, spurious dierences in value added may be obtained simply because the

12
UN data are measured at a dierent point from the PPI index. So, instead, we determine the
eective de ator by dividing the growth in nominal value added for the entire manufacturing
sector in the UN database by the index of industrial production (which measures the real growth
rate in industrial production) obtained from the IFS statistics.

B Data on countries.
The Gross Domestic Product, the Producer Price Index, the exchange rate, and the Index of
Industrial Production are all obtained from International Financial Statistics (I.F.S.) published
by the International Monetary Fund. Whenever a particular series is not available, we use close
substitutes { for instance, the wholesale price index if the producer price index is not available.
Data on a country's human capital (average years of schooling in population over 25) is obtained
from the Barro-Lee les downloaded from the NBER web site (see Barro and Jong Wha Lee
(1993)).

C Measures of nancial development.

Ideally, nancial development should measure the ease with which borrowers and savers can be
brought together, and once together, the condence they have in one another. Thus nancial
development should be related to the variety of intermediaries and markets available, the e-
ciency with which they perform the evaluation, monitoring, certication, communication and
distribution functions, and the legal and regulatory framework assuring performance. Since
there is little agreement on how these are appropriately measured, and even less data available,
we will have to make do with crude proxies even though they may miss many of the aspects we
think vital to a modern nancial system.
The rst measure of nancial development we use is fairly traditional { the ratio of domestic
credit plus stock market capitalization to GDP. We call this the capitalization ratio. We obtain
stock market capitalization for all countries listed in the Emerging Stock Markets Factbook
published by the International Finance Corporation, which contains data on developed countries
also.11 Domestic credit is obtained from the International Financial Statistics. Specically, it
is the sum of IFS lines 32a through 32f and excluding 32e. Finally, domestic credit allocated to
the private sector is IFS line 32d.

13
 Despite the virtue of tradition, there are concerns with this measure. Unlike domestic
credit, stock market capitalization does not re ect the amount of funding actually obtained by
issuers. Instead, it re ects a composite of retained earnings, the investing public's perception
of the corporate sector's growth prospects, and actual equity issuances. One could argue that
the amount of money raised through initial public oerings and secondary oerings is more
suitable for our purpose. Unfortunately, these data are not widely available. At the same
time, one cannot dismiss the capitalization measure in favor of actual nancing too easily. The
net amount raised from U.S. equity markets by large rms was negative in the 1980s (see, for
example, Rajan and Zingales (1995)). So the actual amount raised may underestimate the
importance of the stock market's role in providing price information and liquidity to investors.
Market capitalization may be a better measure of the importance of the stock market in this
respect. Since we are unsure about whether market capitalization is a reasonable proxy, we will
check that the results are robust to redening the capitalization ratio as the ratio of domestic
credit to the private sector to GDP.
The second proxy for nancial development we use is the accounting standards in a country.
Unlike our rst measure, accounting standards re ect the potential for obtaining nance rather
than the actual nance raised. Specically, the higher the standards of nancial disclosure in
a country, the easier it will be for rms to raise funds from a wider circle of investors. The
Center for International Financial Analysis and Research (CIFAR) creates an index for dierent
countries by rating the annual reports of at least three rms in every country on the inclusion
or omission of 90 items. Thus each country obtains a score out of 90 with a higher number
indicating more disclosure. The Center for International Financial Analysis and Research which
produces this data started analyzing balance sheets from 1983 onwards. However, their rst
comprehensive survey dates from 1990. We will use the accounting standards as measured in
this study in much of the paper. The date of the survey raises concerns about endogeneity, but
we believe such concerns are small to begin with, and can easily be addressed. First, accounting
standards do not change much over time. In 1995, the CIFAR published a study examining
how accounting standards had changed since 1983. This study estimated the standards in
1983 and 1990 based on a subset of annual reports, and for a subset of countries that are in
the comprehensive 1990 survey. The study nds the mean accounting standards for countries

14
followed both in 1983 and 1990 is the same at 65. The Wilcoxon signed rank test for equality
of distributions fails to reject the equality of the distribution of accounting standards across
countries in the two years. Finally, the correlation between the accounting standards in 1983
and 1990 is 0.75.12 Nevertheless, we will instrument accounting standards with variables that
predate the period of growth that we are looking at. Also, we will use the 1983 data to see that
the results hold in the subset of countries for which it is available.
Both our measures of nancial development, accounting standards and the capitalization
ratio, are tabulated for the dierent countries (see Table 2). While more developed countries
have better accounting standards, there are exceptions. For instance, Malaysia scores as high as
Australia or Canada, while Belgium and Germany are in the same league as Korea, Philippines,
or Mexico. Portugal has among the worst accounting standards.
Before we go the the summary statistics, note that for a country's nancial development
to have any eect on industrial growth in that country we have to assume that rms nance
themselves largely in their own country. In other words, only if world capital markets are not
perfectly integrated can domestic nancial development aect a country's growth. There is a
wealth of evidence documenting the existence of frictions in international capital markets: the
extremely high correlation between a country's savings and its investments (Martin Feldstein and
Charles Horioka, 1980), the strong home bias in portfolio investments (Kenneth R. French and
James M. Poterba, 1991), and cross countries dierences in expected returns (Geert Bekaert
and Cambell R. Harvey, 1995). We have little else to say about this assumption other than
noting that its failure would weaken the power of our test but not necessarily bias our ndings.
Summary statistics and correlations are in Table 3. A number of correlations are noteworthy.
First, the nancial sector is more developed in richer countries. The correlation of per capita
income in 1980 with accounting standards and capitalization is 0.56 and 0.26 (signicant at the
1 percent and 10 percent level respectively).
Second, the correlation between our capitalization measure of nancial development and
accounting standards is 0.41 (signicant at the 5 percent level for the 33 countries for which we
have both data). However, the correlations between accounting standards and the components
of capitalization dier. Accounting standards are strongly correlated with equity market cap-
italization (correlation = 0.45, signicant at the 1 percent level) but not with domestic credit

15
(correlation = 0.25, not signicant). Domestic credit is credit oered by depository institutions
and the central bank. One explanation of the low correlation is perhaps that institutions rely on
their own private investigations, and credit from them is little aected by accounting standards.
Another possible explanation is that when accounting standards are low, only institutions oer
credit. But even though institutions benet from improvements in accounting standards, other
sources of nance become available, and rms substitute away from their traditional sources.
We cannot distinguish between these explanations. It will suce for our purpose that the overall
availability of nance, whatever its source, increases with nancial development.

IV Financial dependence and growth

A Results from the basic regression.
A.1 Varying measures of nancial development
Table 4 reports the estimates of our basic specication (1) obtained by using various measures of
nancial development. Since the specication controls for country-specic eects and industry-
specic eects, the only eects that are identied are those relative to variables that vary both
cross countries and cross industries. Thus, Table 4 reports only the coecient of the industry's
share of total value added at the beginning of the sample and the coecient of the interaction
between external dependence and dierent measures of nancial development.13 Since we use
U.S. data to identify the external dependence, we drop the United States in all regressions.
We start with total capitalization as the proxy for development. As can be seen in the rst
column of Table 4, the coecient estimate for the interaction term is positive and statistically
signicant at the 1 percent level (throughout the paper, the reported standard errors are robust
to heteroskedasticity).14
The interaction term is akin to a second derivative. One way to get a sense of its magnitude
is as follows; the industry at the 75th percentile of dependence (high dependence) is Machinery.
The industry at the 25th percentile (low dependence) is Beverages. The country at the 75th
percentile of development as measured by capitalization is Italy, while the country at the 25th
percentile is Philippines. We set the industry's initial share of manufacturing at its overall
mean. The coecient estimate then predicts that Machinery should grow 1.3 percent faster than

16
Beverages annually, and in real terms, in Italy as compared to Philippines. For comparison, the
real annual growth rate is, on average, 3.4 percent per year. So a dierential of 1.3 percent is a
large number.
For each specication, we compute a similar number which is reported as the dierential
in real growth rate in the last row of each table. Of course, the countries at the 75th and
25th percentile vary with the measure of development as do the industries at the 75th and 25th
percentile with the measure of dependence.
The rest of the columns of the table include dierent measures of development. We include
domestic credit to the private sector in the second column, accounting standards in the third
column, and accounting standards from the 1983 subsample in the fourth column (for ease of
presentation, accounting standards have been divided by 100 in the estimation). The coecients
are uniformly signicant at the 1 percent level. The economic magnitudes { as measured by
the dierential in growth rates { are also similar except when development is measured by
accounting standards in 1983. The magnitude in column IV falls to approximately half of its
level otherwise. The explanation for this fall is, perhaps, that the 1983 subsample, being based
on just a few companies for each country, introduces signicant measurement error.15
In the fth column, we include both total capitalization and accounting standards. The
coecient for total capitalization is no longer dierent from zero and its magnitude falls to
one fth of its level in the rst column. Similar results are obtained when we replace total
capitalization by domestic credit to the private sector (coecients not reported). This suggests
that accounting standards capture the information about development that is contained in the
capitalization measures. For this reason, we will use accounting standards as our measure of
development in the rest of the paper. The reader should be assured, however, that the results
are qualitatively similar when capitalization measures of development are used.
Because of potential concerns about endogeneity, we will, however, instrument accounting
standards with predetermined institutional variables. Rafael La Porta et al. (1996) suggest that
the origin of a country's legal system has an eect on the development of a domestic capital
market and on the nature of the accounting system. Countries colonized by the British, in
particular, tend to have sophisticated accounting standards while countries in uenced by the
French tend to have poor standards. This suggests using the colonial origin of a country's legal

17
system (indicators for whether it is British, French, German, or Scandinavian) as reported in La
Porta, et al. as one instrument. Also, countries dier in the extent to which laws are enforced.
So we use an index for the eciency and integrity of the legal system produced by Business
International Corporation (a country-risk rating agency) as another instrument. As the sixth
column of Table 4 shows, the fundamental interaction becomes even stronger in magnitude when
we estimate it using instrumental variables.
Before going further, consider the actual (rather than estimated) eects of development on
the growth of specic industries. In Table 5, we summarize for the three least dependent and
three most dependent industries, the residual growth rate obtained after partialling out industry
and country eects. The pattern is remarkable. For countries below the median in accounting
standards, the residual growth rate of the three least dependent industries is positive, while the
residual growth rate of the three most dependent industries is negative. The pattern reverses
for countries above the median. Clearly, this suggests no single country or industry drives our
results and the realized dierential in growth rates is systematic and large.

A.2 Varying measures of dependence.

We now check that our measure of dependence is, indeed, reasonable. We do this in two ways.
First, we check that past nancing in a country is related to the external dependence of industries
in the country. Second, we check that our result is robust to dierent measures of dependence.
Total capitalization is a (crude) measure of how much nance has been raised in the past
in the country. If external dependence is a proxy for an industry's technological need for exter-
nal nance outside the United States, then countries more specialized in externally dependent
industries should have higher capitalization. We calculate the weighted average dependence
for each country by multiplying an industry's dependence on external nance by the fraction
that the industry contributes to value added in the manufacturing sector in 1980. We then
regress total capitalization against weighted average dependence for the 43 countries in the
sample. Weighted average dependence is strongly positively correlated with capitalization in
1980 ( = 2:89, t=3.06). This suggests that our measure of dependence in the United States is
related to the external nancing used by industry in other countries.16
Next, in Table 6 we check that the results are robust to using the external dependence

18
measured for the sample of young rms. Since Figure 1 suggests that most of the demand for
external funds is expressed early on in the life of a company, it may be legitimate to expect this
to be a better measure of an industry's nancial needs. Regardless of how we measure nancial
development, the interaction eect is positive and statistically signicant at the 10 percent
level or better and at the 5 percent level when we use instrumented accounting standards. The
magnitude of the coecient, however, is smaller (roughly a third of the one estimated in Table 6).
In part, this re ects the higher level of the external nance raised by young companies. But even
when we take this into account (see last row of the table), a dierence, albeit smaller, persist.
One possible explanation for this result is that young rms are not as important as mature rms
in in uencing the growth of the industry. We shall return on this issue in section V.A.
In Table 7, we undertake further robustness checks on our measure of external dependence.
While we vary the measure of external dependence, we maintain as a measure of nancial
development a country's accounting standards, instrumented as above.
In the rst column, external dependence is calculated restricting the sample only to mature
rms (listed for more than 10 years) in the United States. Our interaction variable is positive
and statistically signicant and the estimated dierential growth rate (0.9 percent) is similar to
that for the entire sample.
Next, we check whether there is persistence in dependence. If the pattern of nancing in
the United States in the 1980s is very dierent from the pattern in the 1970s, it would be
unreasonable to expect it to carry any information for other countries (especially developing
countries that may use older technologies). The raw correlation between an industry's demand
for external nancing in the 1980s and its demand in the 1970s is 0.63. The coecient estimate
when dependence is measured by the demand for external nancing in the 1970s is statistically
signicant, and the estimated dierential growth rate is 0.9 percent.
Finally, it may be that our results derive from the peculiarities of the U.S. over the 1980s. Our
method should work so long as we measure dependence in a country where nancial constraints
are thought to be small (so that we measure demand not supply). The only other country we
have detailed data on ow of funds for is Canada. Canada is very dierent from the U.S. along
important dimensions. Its banking system is more concentrated as is corporate ownership, and
the composition of its industries is dierent. Nevertheless, the correlation between dependence

19
measured in the United States and dependence measured in Canada is 0.77. As the third column
of Table 7 shows, the coecient estimate when dependence is measured using Canadian data
is highly signicant. What is especially interesting both in this table and Table 4 is that the
economic magnitude of the interaction eect is generally similar despite variation in the measure
of dependence and development used.

V Other Tests
A Decomposition of sources of growth.
An industry can grow because new establishments are added to the industry or because existing
establishments grow in size. The U.N. database also reports the number of establishments in an
industry.17 In our sample, it turns out that two-thirds of the growth is spurred by an increase
in the average size of establishments, while the remaining third is accounted for by an increase
in the number of establishments. The growth in the number of establishments is the log of
the number of ending-period establishments less the log of the number of establishments in the
beginning of period. The average size of establishments in the industry is obtained by dividing
the value added in the industry by the number of establishments, and the growth in average
size is obtained again as a dierence in logs.
Although the denition of establishments provided by the Yearbook of Industrial Statistics
does not coincide with the legal denition of a rm, there are three reasons why it is interesting
to decompose the eect of nancial development in its eect on the growth in the number of
establishments and growth in the size of the existing establishments. First, since this statistic
is often compiled by a dierent body in a country from the one that produces the value-added
data, this test provides an independent check on our results.18 Second, the creation of new
establishments is more likely to require external funds, while the expansion of existing estab-
lishments can also use internal funds. Thus, the eect of nancial development should be more
pronounced for the rst than for the second. Finally, the growth in the number of establish-
ments is more likely to be generated by new rms than the growth in the size of the existing
establishments. Thus, the growth in the number of establishments should be more sensitive to
the external dependence measured using young rms in the United States.

20
 We, then, estimate the basic regression with growth in number of establishments and growth
in average size as dependent variables. As Table 8 indicates, the interaction variable is statis-
tically signicant only when explaining the growth in the number of establishments. More
important, the dierential in growth rate suggested by the estimate is twice as large in the
second column (the regression with growth in numbers as the dependent variable) as in the rst
column (the regression with growth in average size as the dependent variable).
This nding that the development of nancial markets has a disproportional impact on the
growth of new establishments is suggestive. Financial development could indirectly in uence
growth by allowing new ideas to develop and challenge existing ones, much as Schumpeter
argued.
Recall that in the previous section, we found that the dependence of young rms was of lower
importance (both statistical and economic) than the dependence of mature rms in explaining
the relative growth of industries. One explanation is that the dependence of young rms in
the United States is an accurate measure of the needs of new rms in that industry elsewhere
but only a noisy measure of the dependence of all rms. This seems to be the case. When
dependence is measured for young rms, the interaction coecient has a positive, statistically
signicant, eect on the growth in the number of establishments, but a negative (and statistically
insignicant) eect on the growth of the average size of existing establishments (third and
fourth columns); when dependence is measured for mature rms, the interaction coecient has
a positive a statistically signicant eect on both.
Since most of growth in value added is generated by an increase in the average size of
existing establishments, the most appropriate measure of external dependence seems to be one
that includes both the needs of new rms as well as the needs of existing rms. This is why in
the rest of the paper we shall use external dependence measured across all rms.

B Is the Interaction a Proxy for Other Variables?

Do external dependence or nancial development proxy for something else? In principle, there is
a long list of sources of comparative advantage that may dictate the presence, absence, or growth
of industries in a country. Our results, though, cannot be explained unless the dependence of
industries on this source of comparative advantage is strongly correlated with their dependence

21
on external funding and nancial development is a good proxy for the source of comparative
advantage. We rule out two such possibilities below.
Industries that are highly dependent on external nance { for example, drugs and phar-
maceuticals { could also be dependent on human capital inputs. To the extent that nancial
market development and the availability of human capital are correlated, the observed inter-
action between external dependence and nancial development may proxy for the interaction
between human capital dependence and the availability of trained human capital. To check this,
we include in the basic regression an interaction between the industry's dependence on external
nance and a measure of the country's stock of human capital (average years of schooling in
population over age 25). If the conjecture is true, the coecient of the nancial development
interaction term should fall substantially. As the coecient estimates in the rst column of
Table 9 show, the coecient on the human capital interaction term is small and not statistically
signicant, while the nancial development interaction increases somewhat. This suggests that
nancial dependence is not a proxy for the industry's dependence on human capital.
Another possibility is that lower dependence on external nancing in the United States
simply re ects the greater maturity of the industry. An in uential view of the development pro-
cess is that as technologies mature, industries using those technologies migrate from developed
economies to developing economies (see, for example, Rudiger Dornbusch, Stanley Fischer, and
Paul A. Samuelson (1977)). Since developing countries are more likely to have underdeveloped
nancial markets, the interaction eect we document may simply re ect the stronger growth of
mature technologies in underdeveloped countries.
We already have results suggesting this cannot be the entire explanation. The interaction
eect is present even when dependence is measured only for young rms in the United States.
Furthermore, we can test if nancial development is really a proxy for economic development
in the regression. We include in the basic regression the interaction between the industry's
dependence on external nance and the log per capita GDP for the country, in addition to our
usual interaction term. As seen in the second column of Table 9, the coecient of the interaction
term falls from 0.165 (in the basic regression) to 0.149 but is still statistically and economically
signicant. The interaction between nancial dependence and log per capita income is close
to zero and not signicant. The results do not suggest nancial dependence is a proxy for

22
technological maturity.

C Other explanations: Reverse Causality.

Thus far, we have taken the state of nancial markets as predetermined and exogenous. An al-
ternative explanation of the development of nancial markets is that they arise to accommodate
the nancing needs of nance-hungry industries.
The argument is as follows. Suppose there are some underlying country specic factors
or endowments (such as natural resources) that favor certain industries (such as mining) that
happen to be nance hungry. Then, countries abundant in these factors should experience
higher growth rates in nancially dependent industries and { as a result { should develop a
strong nancial market. If these factors persist, then growth rates in nancially dependent
sectors will persist and we will observe the signicant interaction eect. But here it will result
from omitted factors than any benecial eect of nance.
On the one hand, the lack of persistence in country growth over periods of decades (see
William Easterly, et al. (1993)) and the low correlation of sectoral growth across decades (Peter
Klenow, 1995) suggest that this should not be a major concern. On the other hand, our nding
that capitalization is higher when the weighted average dependence of industries in the country
is high indicates the argument is not implausible.
The results we already have should reduce concerns about reverse causality. By restricting
the sample to manufacturing rms, we have reduced the in uence of availability of natural
resources. More important, the measure of nancial development we use, accounting standards,
is instrumented with pre-determined variables that are unlikely to be correlated with omitted
factors driving the growth of industries dependent on external nance. In fact, it should be less
correlated with past nancing than the capitalization measure, yet it explains future relative
growth rates better.
However, we can also test the argument more directly. If an industry has a substantial
presence in a particular country, it is logical that the country has the necessary resources and
talents for the industry. So by further restricting the sample to industries that are above the
median size in the country in 1980, we reduce the problem of dierences in growth stemming
from dierences in endowment. When we estimate the regression with this smaller sample (third

23
column of Table 9), the interaction coecient is virtually unchanged.
One way to make sense of all our ndings without reverse causality driving the results is that
nancial markets and institutions may develop to meet the needs of one set of industries, but
then facilitate the growth of another younger group of industries. Alfred D. Chandler (1977)
suggests this is, in fact, what happened in the United States. The nancial sector, especially
investment banks and the corporate bond market, developed to meet the nancing needs of
railroads in the mid-nineteenth century. The nancial infrastructure was, therefore, ready to
meet the nancing needs of industrial rms as they started growing in the latter half of the
nineteenth century. Similarly, Goldsmith (1985, p2) based on a study of the balance-sheets of
twenty countries writes \The creation of a modern nancial superstructure, not in its details
but in its essentials, was generally accomplished at a fairly early stage of a country's economic
development".
Again, we can test this possibility more directly. We estimate the eect of nancial develop-
ment only for industries that are small to start out with, and are unlikely to be responsible for
the state of development of the nancial markets. So we estimate the basic regression for indus-
tries that in 1980 were less than the median size in their respective countries. The coecient of
the interaction term is again unchanged (see column four of Table 9) even for these industries,
for whom the economy's nancial development is largely predetermined. We conclude that
reverse causality is unlikely to explain our results.

D Other Explanations: Investment and Cost of Capital.

Investment opportunities in dierent industries may be very dierent. For instance, the tobacco
industry in the United States uses negative external nance (see Table 1) partly because in-
vestment opportunities in the Tobacco industry are small relative to the cash ows the industry
generates. It may be that our measure of dependence on external nance proxies primarily for
the investment intensity of a particular industry. Furthermore, the development of the nancial
sector may proxy for the overall cost of capital in that country (rather than the cost of exter-
nal funds). The interaction eect then indicates that capital intensive rms grow faster in an
environment with a lower cost of capital. Though this is a legitimate channel through which
the nancial sector in uences growth, we are also interested in a dierent channel where the

24
reduction in the incremental cost of external funds facilitates growth.
If investment intensity were all that mattered, and external nance and internal nance were
equally costly, the cash internally generated by industries would be irrelevant in countries that
are more nancially developed. All that mattered would be the size of the required investment
and the cost of capital. By contrast, if there is a wedge between the cost of internal and external
nance which narrows as the nancial sector develops, industries generating lots of internal cash
should grow relatively faster in countries with a poorly developed nancial sector. As indicated
in the rst column of Table 10, they do. This is consistent with nancial development reducing
the cost of external nance. Of course, as is to be expected with both the \cost of capital" and
\cost of external capital" hypotheses, industries that invest a lot also grow faster in countries
with more developed nancial markets (second column). Unfortunately, when both interactions
are introduced in the same regression, the coecients are measured very imprecisely because of
multi-collinearity (cash ow intensity and investment intensity have a correlation of 0.73). So
neither is statistically dierent from zero. However, the coecient on cash ows is still negative
and sizeable (accounting for a real growth rate dierential of about 0.4 percent per year).
Multicollinearity results from our aggregating cash ows and investments over a decade.19
Therefore, we estimate the same regression using a measure of cash ow intensity and investment
intensity measured for just one year (rather than an entire decade). In the fourth column we
report the estimates obtained by using the 1980 measures of cash ow and investment. Both
the cash ow intensity and the investment intensity are statistically signicant at the 5 percent
level. We estimated (but not report) the same regression using a 1985 measure and a 1990
measure. In both cases the results are similar and both coecients are statistically signicant
at the 5 percent level.

VI Conclusion
We develop a new methodology in this paper to investigate whether nancial sector development
has an in uence on industrial growth. In doing so, we partially circumvent some of the problems
with the existing cross-country methodology highlighted by Mankiw (1995). First, it is dicult
to interpret observed correlations in cross-country regressions in a causal sense. Here, we push

25
the causality debate one step further by nding evidence for a channel through which nance
theoretically in uences growth. Also, since we have multiple observations per country, we can
examine situations where the direction of causality is least likely to be reversed. A second
problem with the traditional methodology is that explanatory variables are multi-collinear and
are measured with error. The combination of these two problems may cause a variable to appear
signicant when it is merely a proxy for some other variable measured with error. As a result,
observed correlations can be misleading. By looking at interaction eects (with country and
industry indicators) rather than direct eects, we reduce the number of variables that we rely
on, as well as the range of possible alternative explanations. Third, there is the problem of
limited degrees of freedom { there are fewer than two hundred countries on which the myriad
theories have to be tested. Our approach partially alleviates this problem by exploiting within-
country variation in the data. Our methodology, may have wider applications, such as testing
the existence of channels through which human capital can aect growth.
Apart from its methodological contribution, this paper's ndings may bear on three dierent
areas of current research. First, they suggest that nancial development has a substantial
supportive in uence on the rate of economic growth and this works, at least partly, by reducing
the cost of external nance to nancially dependent rms. We should add that there is no
contradiction when the lack of persistence of economic growth (Easterly, et al. (1993)) is set
against the persistence of nancial development. Other factors may cause (potentially serially
uncorrelated) changes in a country's investment opportunity set. Finance may simply enable
the pursuit of these opportunities, and thereby enhance long run growth. The paper does,
however, suggest that nancial development may play a particularly benecial role in the rise of
new rms. If these rms are disproportionately the source of ideas, nancial development can
enhance innovation, and thus enhance growth in indirect ways.
Second, in the context of the literature on nancial constraints, this paper provides fresh
evidence that nancial market imperfections have an impact on investment and growth.
Finally, in the context of the trade literature, the ndings suggest a potential explanation
for the pattern of industry specialization across countries. To the extent that nancial market
development (or the lack thereof) is determined by historical accident or government regulation,
the existence of a well developed market in a certain country represents a source of comparative

26
advantage for that country in industries that are more dependent on external nance. Simi-
larly, the costs imposed by a lack of nancial development will favor incumbent rms over new
entrants. Therefore, the level of nancial development can also be a factor in determining the
size composition of an industry as well as its concentration. These issues are important areas
for future research.

27
Notes
1
Apart from the papers discussed below, see Valerie R. Bencivenga and Bruce D. Smith (1991), John H. Boyd
and Smith (1996), King and Levine (1993 b), Giles Saint-Paul (1992), and Maurice Obstfeld (1994).
2
In Greenwood and Jovanovic (1990), there are no moral hazard or asymmetric information problems at the
level of the entrepreneur. The intermediary simply provides information about economy wide trends that the
entrepreneur cannot gure out for himself, enabling the entrepreneur to invest his own funds more productively.
An equivalent formulation is to distinguish between savers and entrepreneurs. Absent nancial development,
savers can invest directly only in safe, low return, government-sponsored projects, while nancial development
can reduce adverse selection, enabling savers to invest in risky (but often more productive) entrepreneurs.
3
See, for example, Robert J. Barro (1991), Roger Kormendi and Philip Meguire (1985), King and Levine
(1993a), Levine and David Renelt (1992), N. Grigori Mankiw, David Romer, and David Weil (1992), and
Demirguc-Kunt and Maksimovic (1996).
4
Colin Mayer (1990) does look at external nancing, but largely at the country level.
5
Even if capital markets are imperfect so that the supply is not perfectly elastic, this methodology provides a
reasonable measure of the relative demand for funds provided the elasticity of the supply curve does not change
substantially in the cross-section. By contrast, in a very imperfect capital market, the relative amount of funds
raised may be a function not only of the demand for funds but also of factors that aect supply, such as the
availability of collateral.
6
This item is only dened for cash ow statements with format codes 1, 2, or 3. For format code 7 we construct
it as the sum of items # 123, 125, 126, 106, 213, 217.
7
It could be argued that inter-rm trade credit should be viewed as a component of external nancing. It is
unclear how much of trade credit is used to reduce transactions costs and how much is used for nancing. Much
trade credit is granted routinely and repaid promptly and usually, net trade credit for a rm (accounts receivable
less payables) is small (see Mitchell A. Petersen and Rajan (forthcoming)). This may be why trade credit is
typically treated as part of operations in capital budgeting exercises. We adhere to this tradition.
8
We required that there be more than one observation in the industry for this variable to be computed. Even
with this weak requirement we do not have data for some industries. Most notably there are insucient young
rms in the Tobacco industry.
9
This amounts to saying that if the invention of personal computers increased the demand for external funds in
the U.S. computer industry, it is likely to increase the need for funds in the computer industry in other countries
as well.
10
Not all the ISIC sectors for which the Yearbook of Industrial Statistics report data on value added are mutually
exclusive. For example, drugs (3522) is a subsector of other chemicals (352). In these cases, the values of the
broader sectors are net of the values of the subsectors that are separately reported. We follow this convention
both for the data value added and for the nancial data from Compustat.
11
Stock market capitalization is measured at the end of the year, while Gross Domestic Product may value

28
 ows through the year. This may be a problem in high in ation countries. We therefore measure GDP as the
GDP in constant prices multiplied by the producer price index where the base year for both series is ve years
before the year of interest.
12
The regression estimates are not sensitive to dropping the few countries such as Denmark and Spain that
changed accounting standards substantially.
13
The dependent variable is the average real growth rate over the period 1980-1990. For some countries,
however, data availability limits the period. For no country do we have data separated by less than 5 years. A
potential concern is that we measure growth in value added rather than growth in output. Unfortunately, we do
not have data for the latter. While we may not capture increases in productivity fully, we see no obvious way in
which this should bias our results.
14
We reduce the impact of outliers by constraining growth between -1 and +1. Three observations are af-
fected. The coecient estimates for the interaction coecient are higher and still signicant when we do not do
this, though the explanatory power of the regression is lower. We also re-estimate the same specication after
winsorizing the 1 percent and 5 percent tails of the growth rate distribution obtaining virtually identical results
(except that the explanatory power of the regression is still higher).
15
When we instrument this measure (see next paragraph), the coecient estimate goes up by 50 percent
suggesting the coecient estimate is biased downwards by measurement error.
16
Of course, this raises the possibility of reverse causality which we will address later.
17
An establishment is dened as a \unit which engages, under a single ownership or control, in one, or pre-
dominantly one, kind of activity at a single location." (Industrial Statistics Yearbook p. 4). This denition may
not coincide with the legal boundaries of the rm, but is the only one available for such a large cross section of
countries.
18
The disadvantage is that the industry classication used by the body compiling the number of rms may
dier from the industry classication used by the body compiling value-added data, resulting in an increase in
noise.
19
Early investments will generate later cash ows resulting in the correlation. Aggregating over a decade,
however, will still give a reasonable estimate of the average demand for external funds even though it tells us less
about the components.

29
 References
Barro, Robert J. \Economic Growth in a Cross-Section of Countries." Quarterly Journal of
Economics, May 1991, 106, pp. 407-443.
Barro, Robert J. and Lee, Jong Wha. \International Comparisons of Educational Attainment."
Journal of Monetary Economics, December 1993, 32(3), pp. 363-94.
Bekaert, Geert and Harvey, Campbell R. \Time-varying World Market Integration." Journal
of Finance, June 1995, 50(3), pp. 403-44.
Bencivenga, Valerie and Smith Bruce. \Financial Intermediation and Endogenous growth."
Review of Economic Studies, April 1991, 58(2), pp 195-209.
Boyd, John and Smith Bruce. \The Co-Evolution of the Real and Financial Sectors in the
Growth Process." World Bank Economic Review, 1991, 10(2), pp 371-396.
Cameron, Rondo. Banking in the Early Stages of Industrialization. New York, NY: Oxford
University Press, 1967.
Chandler, Alfred D. Jr. The Visible Hand: The Managerial Revolution in American Business
Cambridge, MA: Harvard University Press, 1977.
Demirguc-Kunt, Asli and Maksimovic, Vojislav. \Financial Constraints, Uses of Funds and
Firm Growth: An International Comparison." World Bank Working Paper, 1996.
Diamond, Douglas W. " Reputation Acquisition in Debt Markets." Journal of Political Economy
August 1989, 97(4), pp. 828-62.
Dornbusch, Rudiger; Fischer, Stanley and Samuelson, Paul A. \Comparative Advantage, Trade
and Payments in A Ricardian Model with a Continuum of Goods." American Economic
Review, December 1977, 67(5), pp. 823-839.
Easterly, William; Kremer, Michael; Pritchett, Lant and Summers, Lawrence H. \Good Pol-
icy or Good Luck? Country Growth Performance and Temporary Shocks." Journal of
Monetary Economics, December 1993, 32(3), pp. 459-83.
Feldstein, Martin and Horioka, Charles. \Domestic Saving and International Capital Flows."
The Economic Journal, December 1980, 90, pp. 314-29.
French, Kenneth R. and Poterba, James M. \Investor Diversication and International Equity
Markets." American Economic Review, May 1991, 81(2), pp. 222-26.
Goldsmith, Raymond, W. Financial Structure and Development, New Haven, CT: Yale Uni-
versity Press, 1969.
Goldsmith, Raymond, W. Comparative National Balance Sheets Chicago, IL: University of
Chicago Press, 1985.

30
Greenwood, Jeremy and Jovanovic, Boyan. \Financial Development, Growth, and the Distri-
bution of Income." Journal of Political Economy, October 1990, 98(5), pp 1076-107.
International Finance Corporation. Emerging Stock Markets, Factbook. Washington, D.C.:
1990.
International Monetary Fund. International nancial statistics. Washington, D.C.: 1993.
Jayaratne, Jith and Strahan, Philip E. \The Finance-Growth Nexus: Evidence from Bank
Branch Deregulation." Quarterly Journal of Economics, August 1996, CXI(3), pp. 639-
671.
Jensen, Michael C. and Meckling, William. \Theory of the Firm: Managerial Behavior, Agency
Costs and Capital Structure." Journal of Financial Economics, October 1976, 3(4), pp.
305-60.
King, Robert G. and Levine, Ross. \Finance and Growth: Schumpeter Might Be Right." The
Quarterly Journal of Economics, August 1993a, CVIII(3), pp. 681-737.
King, Robert G. and Levine, Ross. \Finance, Entrepreneurship, and Growth." Journal of
Monetary Economics, December 1993, 32(3), pp. 513-42.
Klenow, Peter. \Ideas vs. Rival Human Capital: Industry Evidence on Growth Models."
Mimeo, University of Chicago, 1995.
Kormendi, Roger and Meguire, Philip. \Macroeconomic Determinants of Growth." Journal of
Monetary Economics, September 1985, 16, pp. 141-163.
La Porta, Rafael; Lopez de Silanes, Florencio; Shleifer, Andrei and Vishny, Robert. \Law and
Finance," NBER Working Paper 5661, 1996.
Levine, Ross. \Financial Development and Economic Growth: Views and Agenda." Journal
of Economic Literature., June 1997, 35(2), pp. 688-726.
Levine, Ross and Renelt, David. \A Sensitivity Analysis of Cross-Countries Growth Regres-
sions." American Economic Review, September 1992, 82(4), pp. 942-963.
Levine, Ross and Zervos, Sara. \Stock Markets and Economic Growth." Mimeo, World Bank,
1996.
Mankiw, N. Gregory. \The Growth of Nations." Brookings Papers on Economic Activity,
1995, (1), pp. 275-310.
Mankiw, N. Gregory, Romer, David, Weil, David. \A Contribution to the Empirics of Economic
Growth," The Quarterly Journal of Economics, May 1992, 152(2), pp. 407-437.
Mayer, Colin. \Financial Systems, Corporate Finance, and Economic Development," in R.
Glenn Hubbard, ed., Asymmetric Information, Corporate Finance and Investment. Chicago,
IL: The University of Chicago Press, 1990.

31
Myers, Stewart C. and Majluf, Nicholas S. \Corporate Financing and Investment Decisions
when Firms have Information that Investors Do Not Have," Journal of Financial Economics,
June 1984, 13 (2), pp. 187-222.
Obstfeld, Maurice. \Risk Taking, Global Diversication, and Growth." American Economic
Review, December 1994, 84(5), pp 1310-29.
Petersen, Mitchell A. and Rajan, Raghuram R. \Trade Credit: Some Theories and Evidence."
Review of Financial Studies, forthcoming.
Rajan, Raghuram R. and Zingales, Luigi. \What Do We Know About Capital Structure:
Some Evidence From International Data." Journal of Finance, December 1995, 50(5), pp.
1421-60.
Robinson, Joan. \The Generalization of the Generat Theory," in The rate of interest, and
other essays. London: Macmillan, 1952, pp. 67-142.
Saint-Paul, Gilles. \Technological Choice, Financial Markets, and Economic Development."
European Economic Review, May 1992, 36 (4), pp 763-81.
Schumpeter, Joseph, A. A Theory of Economic Development. Cambridge, MA: Harvard Uni-
versity Press, 1911.
Standard & Poor. Compustat Services. New York, NY: McGraw Hill, 1994.
Stiglitz, Joseph, E. and Weiss, Andrew. \Credit Rationing in Markets with Imperfect Infor-
mation." American Economic Review, June 1981, 71(3), pp. 393-410.
United Nations. Industrial Statistics Yearbook. New York, NY: United Nation Publications,
1993.

32
 Table 1:

Pattern of external nancing and investment across

industries in the U.S. during the 1980s
The tablereports the median level of external nancing, equity nancing, and capital expenditure for ISIC
industries during the 1980s. External dependence is the fraction of capital expenditures not nanced with cash
ow from operations. Cash ow from operations is broadly dened as the sum of Compustat funds from operations
(item # 110), decreases in inventories, decreases in receivables, and increases in payables. Equity dependence
is the ratio of the net amount of equity issues to capital expenditures. Capital expenditures are the ratio of
capital expenditures to net property plan and equipment. Mature companies are rms that have been public for
at least 10 years, correspondingly young companies are rms that went public less than 10 years ago. The year
of going public is the rst year in which a company starts to be traded on the NYSE, AMEX, or NASDAQ. All
companies is the union of mature and young rms plus rms for which the year of going public could not be
determined (rms already traded on NASDAQ in 1972). All the information is obtained from the ow of funds
data in Compustat, except for the SIC code which is obtained from CRSP and then matched with the ISIC code.

All companies Mature companies Young companies

ISIC Industrial sectors External Capital External Capital External Capital
code dependence expenditures dependence expenditures dependence expenditures
314 Tobacco -0.45 0.23 -0.38 0.24 . .
361 Pottery -0.15 0.20 0.16 0.41 -0.41 0.13
323 Leather -0.14 0.21 -1.33 0.27 -1.53 0.16
3211 Spinning -0.09 0.16 -0.04 0.19 . .
324 Footwear -0.08 0.25 -0.57 0.23 0.65 0.26
372 Non-ferrous metal 0.01 0.22 0.07 0.21 0.46 0.24
322 Apparel 0.03 0.31 -0.02 0.27 0.27 0.37
353 Petroleum reneries 0.04 0.22 -0.02 0.22 0.85 0.28
369 Non metal products 0.06 0.21 0.15 0.22 -0.03 0.26
313 Beverages 0.08 0.26 -0.15 0.28 0.63 0.26
371 Iron and steel 0.09 0.18 0.09 0.16 0.26 0.19
311 Food products 0.14 0.26 -0.05 0.25 0.66 0.33
3411 Pulp, paper 0.15 0.20 0.13 0.21 0.22 0.20
3513 Synthetic resins 0.16 0.30 -0.23 0.20 0.79 0.45
341 Paper and products 0.18 0.24 0.10 0.23 0.57 0.29
342 Printing & publishing 0.20 0.39 0.14 0.33 0.60 0.41
352 Other chemicals 0.22 0.31 -0.18 0.25 1.35 0.46
355 Rubber products 0.23 0.28 -0.12 0.21 0.50 0.32
332 Furniture 0.24 0.25 0.33 0.17 0.68 0.29
381 Metal products 0.24 0.29 0.04 0.25 0.87 0.34
3511 Basic exclud fert 0.25 0.30 0.08 0.24 0.79 0.29
331 Wood products 0.28 0.26 0.25 0.23 0.34 0.40
384 Transportation equipment 0.31 0.31 0.16 0.28 0.58 0.31
354 Petroleoum and 0.33 0.23 0.16 0.26 -0.26 0.22
coal products

33
 All companies Mature companies Young companies
ISIC Industrial sectors External Capital External Capital External Capital
code dependence expenditures dependence expenditures dependence expenditures
3843 Motor veichle 0.39 0.32 0.11 0.33 0.76 0.32
321 Textile 0.40 0.25 0.14 0.24 0.66 0.26
382 Machinery 0.45 0.29 0.22 0.25 0.75 0.33
3841 Ship 0.46 0.43 0.04 0.34 1.05 0.56
390 Other industries 0.47 0.37 -0.05 0.28 0.80 0.49
362 Glass 0.53 0.28 0.03 0.28 1.52 0.33
383 Electric machinery 0.77 0.38 0.23 0.29 1.22 0.46
385 Professional goods 0.96 0.45 0.19 0.33 1.63 0.52
3832 Radio 1.04 0.42 0.39 0.30 1.35 0.48
3825 Oce &computing 1.06 0.60 0.26 0.38 1.16 0.64
356 Plastic products 1.14 0.44 . . 1.14 0.48
3522 Drugs 1.49 0.44 0.03 0.32 2.06 0.47

34
 Table 2:
Financial development across countries.
Accounting standards is an index developed by the Center for International Financial Analysis & Research
ranking the amount of disclosure in annual company reports in each country. Total Capitalization to GDP is the
ratio of the sum of equity market capitalization (as reported by the IFC) and domestic credit (IFS line 32a-32f
but not 32 e) to GDP. Domestic credit to the private sector is IFS line 32d. Per capita income in 1980 is in
dollars and is from the IFS.

Country Accounting Total capitalization Domestic credit to Per capita

standards over GDP private sector over GDP income (dollars)
Bangladesh . 0.20 0.07 121
Kenya . 0.28 0.20 417
Morocco . 0.41 0.16 807
Sri Lanka . 0.44 0.21 252
Pakistan . 0.53 0.25 290
Costa Rica . 0.53 0.26 2,155
Zimbabwe . 1.01 0.30 441
Jordan . 1.16 0.54 1,109
Egypt 24 0.74 0.21 563
Portugal 36 0.82 0.52 2,301
Peru 38 0.28 0.11 842
Venezuela 40 0.34 0.30 3,975
Colombia 50 0.21 0.14 1,150

35
Country Accounting Total capitalization Domestic credit to Per capita
standards over GDP private sector over GDP income (dollars)
Turkey 51 0.35 0.14 1,081
Chile 52 0.74 0.36 2,531
Brazil 54 0.33 0.23 1,650
Austria 54 1.00 0.77 9,554
Greece 55 0.74 0.44 3,814
India 57 0.50 0.24 240
Mexico 60 0.39 0.16 2,651
Belgium 61 0.65 0.29 11,226
Denmark 62 0.56 0.42 12,188
Germany 62 1.08 0.78 12,345
Italy 62 0.98 0.42 6,460
Korea 62 0.63 0.50 1,407
Netherlands 64 0.91 0.60 11,155
Spain 64 1.02 0.76 5,087
Israel 64 1.18 0.67 3,573
Philippines 65 0.46 0.28 729
Japan 65 1.31 0.86 9,912
France 69 0.70 0.54 11,337
New Zealand 70 0.59 0.19 7,490
South Africa 70 1.51 0.26 2,899
Norway 74 0.63 0.34 13,430
Canada 74 0.98 0.45 10,486
Australia 75 0.82 0.28 9,866
Malaysia 76 1.19 0.48 1,683
Finland 77 0.52 0.48 10,181
UK 78 0.78 0.25 9,600
Singapore 78 1.96 0.57 4,661
Sweden 83 0.79 0.42 14,368

36
 Table 3:

Summary Statistics.
Industry real growth is the annual compounded growth rate in real value added for the period 1980-1990 for
each ISIC industry in each country. The growth in the number of rms is the dierence between the log of number
of ending-period rms and the log of number of beginning-period rms. The average size of rms in the industry
is obtained by dividing the value added in the industry by the number of rms, and the growth in average size
is obtained again as a dierence in logs. The industry's share of total value added is computed dividing the 1980
value added of the industry by the total value added in manufacturing that year. External dependence is the
median fraction of capital expenditures not nanced with cash ow from operations for each industry. Cash ow
from operations is broadly dened as the sum of Compustat funds from operations (item # 110), decreases in
inventories, decreases in receivables, and increases in payables. External dependence has been constructed using
Compustat rms between 1980 and 1990 except for Canada where we use Global Vantage data between 1982 and
1990. Accounting standards is an index developed by the Center for International Financial Analysis & Research
ranking the amount of disclosure of companies' annual reports in each country. In panels B and C the p-value
are reported in brackets.

A: Summary statistics
Variable Mean Median Standard Minimum Maximum Number of
Deviation observations
Industry real growth 0.034 0.029 0.099 -0.447 1.000 1242
Industry growth in number of rms 0.012 0.007 0.071 -0.414 0.759 1073
Industry growth in average rms' size 0.022 0.026 0.094 -0.536 0.410 1070
Industry's share of total value added 0.016 0.009 0.021 0.000 0.224 1217
Log per capita income in 1980 in dollars 7.814 7.883 1.340 4.793 9.573 43
Average years of schooling 5.900 5.442 2.829 1.681 12.141 41
External nance dependence (all rms) 0.319 0.231 0.319 -0.451 1.492 36
External nance dependence (old rms) 0.010 0.075 0.302 -1.330 0.394 35
External nance dependence (young rms) 0.675 0.673 0.643 -1.535 2.058 34
External nance dependence (1970s ) 0.078 0.073 0.188 -0.450 0.542 35
External nance dependence (Canadian rms) 0.427 0.384 0.767 -0.802 3.512 27
Cash ow generated 0.173 0.198 0.112 -0.217 0.331 36
Investment intensity 0.298 0.278 0.095 0.161 0.600 36
Total capitalization over GDP 0.738 0.696 0.375 0.199 1.962 41
Domestic credit to private sector over GDP 0.377 0.302 0.201 0.069 0.856 41
Accounting standards 61.324 62.000 13.238 24.000 83.000 34
Accounting standards (1983) 65.393 68.500 11.426 39.000 81.000 28

37
 B: Correlation between measures of external dependence
All Old Young 1970s Cash ow Investment
External nance dependence (all rms) 1.00
External nance dependence (old rms) 0.46 1.00
( 0.01 )
External nance dependence (young rms) 0.72 0.48 1.00
( 0.00 ) ( 0.00 )
External nance dependence (1970s ) 0.63 0.42 0.48 1.00
( 0.00 ) ( 0.01 ) ( 0.00 )
Cash ow generated -0.91 -0.37 -0.55 -0.50 1.00
( 0.00 ) ( 0.03 ) ( 0.00 ) ( 0.00 )
Investment intensity 0.81 0.28 0.64 0.63 -0.60 1.00
( 0.00 ) ( 0.10 ) ( 0.00 ) ( 0.00 ) ( 0.00 )
External nance dependence (Canadian rms) 0.77 0.36 0.58 0.37 -0.78 0.55
( 0.00 ) ( 0.07 ) ( 0.00 ) ( 0.07 ) ( 0.00 ) ( 0.00 )

C: Correlation between measures of nancial development

Total Market Domestic credit Accounting Accounting
capitalization capitalization to private sector standards standards 1983
Market capitalization over GDP 0.79
( 0.00 )
Domestic credit to private sector 0.67 0.21 1.00
over GDP ( 0.00 ) ( 0.18 )
Accounting standards 0.41 0.45 0.25 1.00
( 0.02 ) ( 0.01 ) ( 0.17 )
Accounting standards (1983) 0.27 0.39 -0.14 0.68 1.00
( 0.17 ) ( 0.05 ) ( 0.50 ) ( 0.00 )
Per capita income 0.26 0.04 0.48 0.56 0.28
( 0.09 ) ( 0.80 ) ( 0.00 ) ( 0.00 ) ( 0.16 )

38
 Table 4:

Industry growth and various measures of

development.
The dependent variable is the annual compounded growth rate in real value added for the period 1980-
1990 for each ISIC industry in each country. External dependence is the fraction of capital expenditures not
nanced with internal funds for U.S. rms in the same industry between 1980-90. The interaction variable is
the product of external dependence and nancial development. Financial development is total capitalization in
the rst column, domestic credit to the private sector over GDP in the second column, accounting standards in
1990 in the third column, accounting standards in 1983 in the fourth column. The sixth column is estimated
with instrumental variables. Both the coecient estimate for the interaction term and the standard error when
accounting standards is the measure of development are multiplied by 100. The dierential in real growth rate
measures (in percentage terms) how much faster an industry at the 75th percentile level of external dependence
grows with respect to an industry and 25th percentile level when it is located in a country at the 75th percentile
of nancial development rather than in one at the a 25th percentile. All regressions include both country and
industry xed eects (coecient estimates not reported). Heteroscedasticity robust standard errors are reported
in brackets.

Financial development measured as

Total Bank Accounting Accounting Accounting Instrumental
capitalization debt standards standards standards and variables
Variable in 1983 Capitalization
Industry's share of total value -0.912 -0.899 -0.643 -0.587 -0.433 -0.648
added in manufacturing in 1980 ( 0.246 ) ( 0.245 ) ( 0.204 ) ( 0.223 ) ( 0.135 ) ( 0.203 )
Interaction (external dependence 0.069 0.012
X total capitalization) ( 0.023 ) ( 0.014 )
Interaction (external dependence 0.118
X domestic credit to private sector) ( 0.037 )
Interaction (external dependence 0.155 0.133 0.165
X accounting standards) ( 0.034 ) ( 0.034 ) ( 0.044 )
Interaction (external dependence 0.099
X accounting standards 1983 ) ( 0.036 )
R-squared 0.290 0.290 0.346 0.239 0.419 0.346
Number of observations 1217 1217 1067 855 1042 1067
Dierential in real growth rate 1.3 1.1 0.9 0.4 1.3 1.0

39
 Table 5:

Eect of Financial Development on Actual Growth

Rates in Dierent Industries.
This tablereports the mean residual growth rate (in percentage term) obtained after regressing the annual
compounded growth rate in real value added for the period 1980-1990 on industry and country dummies.

Countries below Countries above

the median in the median in
accounting standards accounting standards
Least nancially dependent industries
Tobacco 0.53 -0.60
Pottery 0.25 -0.30
Leather 0.77 -0.77
Most nancially dependent industries
Drug -1.11 1.30
Plastics -0.21 0.21
Computers -2.00 1.80

40
 Table 6:

Industry growth and various measures of

development using external dependence measured
for young rms.
The dependent variable is the annual compounded growth rate in real value added for the period 1980-
1990 for each ISIC industry in each country. External dependence is the fraction of capital expenditures not
nanced with internal funds between 1980-90 for U.S. rms who went public in the previous ten years belonging
to the same industry. The interaction variable is the product of external dependence and nancial development.
Financial development is total capitalization in the rst column, domestic credit to the private sector over GDP
in the second column, accounting standards in 1990 in the third column, accounting standards in 1983 in the
fourth column. The sixth column is estimated with instrumental variables. Both the coecient estimate for the
interaction term and the standard error when accounting standards is the measure of development are multiplied
by 100. The dierential in real growth rate measures (in percentage terms) how much faster an industry at the
75th percentile level of external dependence grows with respect to an industry and 25th percentile level when it is
located in a country at the 75th percentile of nancial development rather than in one at the a 25th percentile. All
regressions include both country and industry xed eects (coecient estimates not reported). Heteroscedasticity
robust standard errors are reported in brackets.

Financial development measured as

total bank accounting accounting accounting instrumental
capita- debt standards standards standards and variables
Variable lization in 1983 capitalization
Industry's share of total value -0.911 -0.904 -0.568 -0.616 -0.293 -0.571
added in manufacturing in 1980 ( 0.287 ) ( 0.286 ) ( 0.234 ) ( 0.252 ) ( 0.149 ) ( 0.233 )
Interaction (external dependence 0.021 -0.004
X total capitalization) ( 0.012 ) ( 0.008 )
Interaction (external dependence 0.034
X domestic credit to private sector) ( 0.019 )
Interaction (external dependence 0.046 0.045 0.058
X accounting standards) ( 0.021 ) ( 0.022 ) ( 0.028 )
Interaction (external dependence 0.038
X accounting standards 1983 ) ( 0.019 )
R-squared 0.283 0.283 0.341 0.236 0.415 0.340
Number of observations 1150 1150 1008 808 984 1008
Dierential in real growth rate 0.6 0.5 0.4 0.2 0.1 0.5

41
 Table 7:

Industry growth and various measures of external

dependence
The dependent variable is the annual compounded growth rate in real value added for the period 1980-
1990 for each ISIC industry in each country. External dependence is the fraction of capital expenditures
not nanced with internal funds by rms in the same industry during the 1980s. In the rst column this
ratio is computed only for companies that have been public for at least 10 years, in the second column
the ratio is computed for companies that have gone public in the last 9 years, in the second column it
is computed for U.S. rms during the 1970s. In the third column it is computed for Canadian rms
during the 1980s. Also in the third column, data on U.S. industries are included while data on Canadian
industries are dropped. The dierential in real growth rate measures (in percentage terms) how much
faster an industry at the 75th percentile level of external dependence grows with respect to an industry
and 25th percentile level when it is located in a country at the 75th percentile of nancial development
rather than in one at the a 25th percentile. All regressions are estimated using instrumental variables and
include both country and industry xed eects (coecient estimates not reported). Heteroscedasticity
robust standard errors are reported in brackets.

External dependence measured using

Variable Old 1970s Canadian
rms rms rms
Industry's share of total value -0.648 -0.620 -0.610
added in manufacturing in 1980 ( 0.227 ) ( 0.205 ) ( 0.235 )
Interaction (external depen- 0.255 0.315 0.068
dence X accounting standards) ( 0.064 ) ( 0.127 ) ( 0.023 )
R-squared 0.344 0.3345 0.343
Number of observations 979 1035 802
Dierential in real growth rate 0.9 0.9 0.8

42
 Table 8:

Growth in average size and number of

establishments
The average size of establishments in the industry is obtained by dividing the value added in the
industry by the number of establishments, and the growth in average size is obtained as a dierence in
logs between average size in 1990 and average size in 1980. The growth in the number of establishments
is the log of the number of establishments in 1990 less the log of the number of establishments in 1980.
The dierential in real growth rate measures (in percentage terms) how much faster an industry at the
75th percentile level of external dependence grows with respect to an industry and 25th percentile level
when it is located in a country at the 75th percentile of nancial development rather than in one at the
a 25th percentile. All regressions are estimated using instrumental variables and include both country
and industry xed eects (coecient estimates not reported). Heteroscedasticity robust standard errors
are reported in brackets.

External dependence measured using

All rms Young rms Mature rms
Growth av- Growth Growth av- Growth Growth av- Growth
Variable erage size number erage size number erage size number
Industry's share of total value -0.620 -0.312 -0.662 -0.267 -0.624 -0.282
added in manufacturing in 1980 ( 0.217 ) ( 0.154 ) ( 0.252 ) ( 0.176 ) ( 0.220 ) ( 0.152 )
Interaction (external depen- 0.051 0.115 -0.018 0.078 0.125 0.131
dence X accounting standards) ( 0.043 ) ( 0.037 ) ( 0.028 ) ( 0.023 ) ( 0.055 ) ( 0.041 )
R-squared 0.498 0.347 0.499 0.308 0.492 0.310
Number of observations 951 1011 926 949 923 947
Dierential in real growth rate 0.3 0.7 -0.2 0.6 0.4 0.4

43
 Table 9:

Robustness Checks
The dependent variable is the annual compounded growth rate in real value added for the period 1980-1990
for each ISIC industry in each country. The rst column adds to the basic specication the interaction between
external dependence and a country's human capital. The second column adds to the basic specication the
interaction between external dependence and a country's level of economic development (log per capita income).
The third column estimates the basic specication for industries that in 1980 were above the median industry in
terms of the fraction they accounted for of value added in the manufacturing sector. The fourth column estimates
the basic specication for industries that in 1980 were below the median industry in terms of the fraction they
accounted for of value added in the manufacturing sector. The dierential in real growth rate measures (in
percentage terms) how much faster an industry at the 75th percentile level of external dependence grows with
respect to an industry and 25th percentile level when it is located in a country at the 75th percentile of nancial
development rather than in one at the a 25th percentile. All regressions are estimated using instrumental variables
and include both country and industry xed eects (coecient estimates not reported). Heteroscedasticity robust
standard errors are reported in brackets.

Human Economic Below Above

Variable capital development median median
Industry's share of total value -0.386 -0.422 -0.437 -6.079
added in manufacturing in 1980 ( 0.137 ) ( 0.134 ) ( 0.178 ) ( 1.932 )
Interaction (external dependence 0.191 0.149 0.161 0.161
X accounting standards) ( 0.072 ) ( 0.055 ) ( 0.065 ) ( 0.066 )
Interaction 2 (external dependence -0.002
X average years of schooling) ( 0.003 )
Interaction 3(external dependence 0.000
X log of per capita income in 1980) ( 0.005 )
R-squared 0.413 0.418 0.548 0.390
Number of observations 1006 1042 522 545
Dierential in real growth rate 1.0 0.9 0.9 1.0

44
 Table 10:

Cash Flow and Investments

The dependent variable is the annual compounded growth rate in real value added for the period 1980-
1990 for each ISIC industry in each country. Internal cash ow is the ratio of cash ow from operations
broadly dened (see text) to net property plant and equipment for U.S. rms in the same industry.
Investment intensity is the ratio of capital expenditures to property plant and equipment for U.S. rms in
the same industry. The fourth column uses the cash ow intensity and the investment intensity measured
for the year 1980. The dierential in real growth rate measures (in percentage terms) how much faster an
industry at the 75th percentile level of external dependence grows with respect to an industry and 25th
percentile level when it is located in a country at the 75th percentile of nancial development rather than
in one at the a 25th percentile. All regressions are estimated using instrumental variables and include
both country and industry xed eects (coecient estimates not reported). Heteroscedasticity robust
standard errors are reported in brackets.

Cash ow Investment Both Both

Variable intensiveness intensiveness measured for 1980
Industry's share of total value -0.588 -0.653 -0.639 -0.639
added in manufacturing in 1980 ( 0.201 ) ( 0.205 ) ( 0.205 ) ( 0.207 )
Interaction (internal cash ow -0.482 -0.261 -0.595
X nancial development) ( 0.153 ) ( 0.196 ) ( 0.295 )
Interaction 2 (investment intensiveness 0.623 0.443 0.800
X accounting standards) ( 0.221 ) ( 0.283 ) ( 0.299 )
R-squared 0.343 0.345 0.345 0.344
Number of observations 1067 1067 1067 1035
Dierential in real growth rate -0.7 1.4 0.5 1.6

45
 Figure 1:
Life Cycle of External Financing and Investments

1.5

0.5

-0.5

-1
0 4 8 12 16 20 24 28
# years since the IPO
Equity Financing External Finance Investments

The graph plots the median level of external nancing, equity nancing, and investments in
the U.S. across 3-digit SIC industries as a function of the number of years since the IPO. External
nance is the amount of investments (CAPEX) not nanced with cash ow from operations,
reduction in inventories, or decreases in trade credit. Equity nance is the net amount of funds
raised through equity issues divided by the amount of investments. Investment is the ratio of
CAPEX to net property, plant and equipment. The IPO year is dened as the rst year in
which a company starts to be traded on the NYSE, AMEX, or NASDAQ. All the information
is obtained from the ow of funds data in Compustat, except for the SIC code which is from
CRSP.

46