European Economic Review 56 (2012) 1730–1745

Contents lists available at SciVerse ScienceDirect

European Econom ic Review

jo u rn al h o m ep ag e: w w w .elsevier.co m /locate/eer

Infrast ruct ure and inequalit y $

Sant anu Chat t erjee a,1, St ephen J. Turnovsky b,n
a
Department of Economics, Terry Col ege of Business, University of Georgia, Athens, GA 30602, USA
b
Department of Economics, University of Washington, Seattle, WA 98195, USA

ar t i c l e i n f o abst r ac t

Article history: W e develop a m odel in w hich public capital is both an engine of grow th and a
Received 30 April 2012 determ inant of the distributions of w ealth, incom e, and w elfare. Governm ent invest-
Accepted 10 August 2012 m ent increases w ealth inequality over tim e, regardless of its fi nancing. The tim e path
Available online 27 September 2012
of incom e inequality is, how ever, highly sensitive to fi nancing policies, and is often
JEL classification: characterized by sharp intertem poral tradeoffs, w ith incom e inequality declining in
O1 the short run but increasing in the long run. Public investm ent generates a positive
O2 correlation betw een grow th and incom e inequality along the transition path, but their
O3 short-run and long-run relationship depends critically on (i) how externalities im pinge
on allocation decisions, (ii) fi nancing policies, and (iii) the tim e period of consideration.
Keywords: Finally, these policies also generate sharp trade-offs betw een average w elfare and its
Public capital
distribution, w ith governm ent investm ent im proving average w elfare, but also increas-
Growth
ing its dispersion. Our results are obtained num erically but extensive sensitivity
Inequality
analysis confi rm s their robustness across key param eter values.
& 2012 Elsevier B.V. All rights reserved.

1. In t r oduct ion

The expressway network (in China) hasy helped to promote a sharp increase in private car ownershipy roads are
sometimes built expressly for the purpose of converting countryside into revenue-generating urban landy For Beijing’s
airport expansion, 15 villages were flattened and their more than 10,000 residents resettledy buty former farm-
ersy (were) barred from unemployment benefits and other welfare privileges.
The Econom ist (February 14, 2008).

Governm ent provision of public goods such as infrastructure represents an im portant m echanism through w hich
w ealth can be redistributed across society. M any em erging-market countries such as India, China, and Brazil have
em barked on am bitious expansions of public investm ent in roads, ports, com m unication and transportation netw orks,
pow er generation and w ater services, m ainly as a m eans to sustain their high grow th rates of the last tw o decades. Am ong

$
This paper has benefited from presentations to the Annual Meetings of the Society for Computational Economics in Paris, the NEUDC conference at
MIT, the International Development Conference at the Federal Reserve Bank of Atlanta, the Sixth International Growth and Development Conference at
the Indian Statistical Institute, New Delhi, the Castor Macroeconomics Workshop at the University of Washington, and seminars at the International
Monetary Fund, University of Florida, Florida State University, University of Georgia, Georgia State University, and the University of Hamburg. Comments
received from Dilip Mookherjee, Martin Evans, Mark Rider, Felix Rioja, Clinton Shiells, an anonymous referee, and the Editor, Theo Eicher, have improved
the paper significantly. Chatterjee acknowledges support from the Terry-Sanford Research Award at the University of Georgia. Turnovsky’s research was
supported in part by the Castor endowment at the University of Washington. This support is gratefully acknowledged. An earlier version of this paper w as
previously circulated under the title, ‘ The Distributional Consequences of Government Spending’ .
n
Corresponding author. Tel.: þ 1 206 685 8028; fax: þ 1 206 685 7477.
E-mail addresses: [email protected] (S. Chatterjee), [email protected] (S.J. Turnovsky).
1
Tel.: þ 1 706 542 3696.

0014-2921/$- see front matter & 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.euroecorev.2012.08.003
 S. Chatterjee, S.J. Turnovsky / European Economic Review 56 (2012) 1730–1745 1731

developed countries, a signifi cant fraction of the increase in governm ent stim ulus spending in the afterm ath of the
Financial Crisis of 2007–08 w as targeted tow ards infrastructure. At the sam e tim e, incom e inequality has also risen
steadily across the w orld, both in em erging m arkets and m ost OECD countries. An im portant question that arises in this
context is the effect of pro-grow th policies on the dynam ics of inequality. Indeed, as Anand and Segal (2008) point out,
reducing inequality m ay be an im portant social objective for a governm ent. The issue then is the extent to w hich
investm ent in public infrastructure, w ith its accom panying grow th and productivity benefi ts, is com patible w ith this
objective.
Beginning w ith Arrow and Kurz (1970) and later Barro (1990), the relationship betw een public investm ent and grow th
has been w idely studied, w ith general agreem ent that governm ent spending on infrastructure can yield signifi cant
productivity and grow th benefi ts.2 At the sam e tim e, by affecting factor productivity and therefore relative factor returns,
public investm ent m ay also play a critical role in the evolution of w ealth and incom e distributions as the econom y grow s
over tim e. How ever, a priori, it is unclear w hat the nature of such a relationship w ill be.
In contrast to the public investm ent–grow th relationship, em pirical evidence on the relationship betw een infra-
structure investm ent and inequality is sparse, inconclusive, and largely anecdotal. For instance, Ferranti et al. (2004), Fan
and Zhang (2004), Lopez (2004), and Calderon and Serven (2004) fi nd that public investm ent has prom oted grow th and
contributed tow ards the alleviation of inequality. W olff and Zacharias (2007) docum ent an inverse short-run relationship
betw een governm ent expenditure and inequality for the United States, though they do not distinguish betw een public
consum ption and investm ent. In contrast, Brakm an et al. (2002) fi nd that governm ent spending on infrastructure has
increased regional disparities w ithin Europe, and Artadi and Sala-i-M artin (2003) point to excessive public investm ent as a
contributing factor to rising incom e inequality in Africa. Banerjee and Som anathan (2007) report that in India, access to
critical infrastructure services and public goods is in general positively correlated w ith social status, w hile a W orld Bank
(2006) report also fi nds that the quality and perform ance of state-provided infrastructure services tend to be the w orst
in India’s poorest states. Further, Khandker and Koolw al (2007) fi nd that access to paved roads has had a lim ited
distributional im pact in rural Bangladesh. The diversity of these em pirical fi ndings underscores the need for a w ell-
specifi ed analytical fram ew ork w ithin w hich the link betw een infrastructure spending, econom ic grow th and inequality
can be system atically studied.
This paper seeks to synthesize these tw o extensive, but independent, strands of literature into a unifi ed fram ew ork. On
the one hand, the theoretical literature on grow th and inequality has not dealt w ith issues related to public investm ent and
its fi nancing. 3 On the other hand, the literature on public investm ent and grow th has generally ignored distributional
issues.4 Studying the public investm ent–inequality relationship in the context of a dynam ic grow th m odel therefore
represents an im portant synthesis of previous w ork. In doing so, w e address the follow ing issues:

(i). The m echanism through w hich governm ent spending on infrastructure and accom panying taxation policies affects
the distributions of w ealth, incom e, and w elfare over tim e.
(ii). The dynam ics of the grow th–inequality relationship along the transitional path.
(iii). Trade-offs betw een average w elfare and its dispersion resulting from fi scal shocks.

The m odel w e em ploy has several key elem ents. First, the underlying source of heterogeneity arises through agents’
differential initial endow m ents of private capital. 5 Com bined w ith an endogenous labor–leisure choice, this yields an
endogenous distribution of incom e. Second, w e introduce a grow ing stock of a governm ent-provided good (public capital)
that is non-rival and non-excludable. This interacts w ith the aggregate stock of private capital to generate com posite
externalities for both labor productivity in production and the labor–leisure allocation in utility. The governm ent has a
range of fi scal instrum ents available to fi nance its investm ent, nam ely distortionary taxes on capital incom e, labor incom e,
and consum ption, and a non-distortionary lum p-sum tax (equivalent to governm ent debt). The accum ulation of public
capital and the spillovers it generates serves both as an engine of sustained grow th, and also as a driver of relative returns
to capital and labor, w ith consequences for the evolution of w ealth and incom e inequality. In equilibrium , both the
econom y’s grow th rate and inequality are endogenously determ ined. 6

2
Agenor (2011) provides an exhaustive survey of the theoretical literature on this issue. On the empirical side, the consensus remains that
infrastructure contributes positively and significantly to output, though its exact magnitude is a subject of debate. See Bom and Ligthart (2010) for a
review of the empirical literature.
3
This literature has explored issues such as capital market imperfections, the role of human capital, and technological progress, among others; see,
for example, Banerjee and Newman (1993), Galor and Zeira (1993) and Acemoglu (1998).
4
In a recent contribution, Garcı´a-Pen˜ alosa and Turnovsky (2011) consider the case where the tax revenues are allocated to government
consumption. Their work, however, is in context of the neoclassical Ramsey model and does not focus on the growth and distributional effects of
infrastructure investment and its financing.
5
Recent empirical evidence points to the importance of the return to private capital as one of the determinants of inequality; see, for example,
Atkinson (2003), Piketty (2011) and Checchi and Garcı´a-Pen˜ alosa (2010).
6
In this context, our work is related to Getachew (2010), who uses an OLG set-up with public investment and initial differences in skills as the source
of heterogeneity. While Getachew (2010) focuses on the role of credit markets, our focus is on the interaction between the endogenous allocation of time
between labor and leisure, the initial distribution of private wealth, the stock of public infrastructure, and various underlying tax policies used for
financing public investment.
1732 S. Chatterjee, S.J. Turnovsky / European Economic Review 56 (2012) 1730–1745

Given the com plexity of the theoretical fram ew ork, the m odel is analyzed num erically. Specifi cally, w e com pare an
increase in the rate of governm ent investm ent on public capital, fi nanced by the use of alternative fi scal instrum ents. A
num ber of interesting results em erge:

(i) Governm ent spending on public capital leads to a persistent increase in w ealth inequality over tim e, regardless of how
it is fi nanced. In contrast, the tim e path of incom e inequality is sensitive to the fi nancing policy adopted, and in m any
cases is characterized by sharp intertem poral tradeoffs. For exam ple, w hile governm ent investm ent fi nanced by a
lum p-sum or consum ption tax leads to a short-run decline in incom e inequality, this is com pletely reversed over tim e,
leading to an increase in the long-run dispersion of incom e. This is som ew hat surprising, since lum p-sum taxes are a
non-distortionary source of fi nancing and governm ent spending creates a larger stock of a non-excludable and non-
rival public good. W e also fi nd that m ore than tw o-thirds of the long-run increase in incom e inequality can be
attributed to an increase in labor incom e inequality, consistent w ith the recent em pirical fi ndings of Atkinson et al.
(2011).
(ii) The grow th–income inequality relationship generated by government spending depends critically on (a) how externalities
impinge on allocation decisions, (b) the underlying financing policies, and (c) the time period of consideration—i.e. short
run, transition path, or the long run. These results underscore the ambiguity in the grow th–inequality relationship that is
characteristic in the empirical literature. 7
(iii) Public investm ent generates sharp trade-offs betw een average w elfare and its distribution: w hile governm ent
expenditure on infrastructure im proves average w elfare, it also increases its dispersion. How ever, spending fi nanced
by taxing consum ption or labor incom e is associated w ith less adverse tradeoffs.

The rest of the paper is organized as follow s. Section 2 lays dow n the analytical fram ew ork and Section 3 derives the
m acroeconom ic equilibrium for the aggregate econom y. Section 4 derives the distributional dynam ics and characterizes
the evolution of the different m easures of inequality. Section 5 conducts num erical policy experim ents and discusses their
im plications. Section 6 conducts an extensive robustness check of the benchm ark results. Finally, Section 7 concludes.

2. An alyt ical fr am ew ork

The analytical fram ew ork is that of a closed-econom y w ith heterogeneous agents in w hich both private and public
capital are accum ulated, w ith the evolution of the econom y being characterized by transitional dynam ics and endogenous
grow th, as in Futagam i et al. (1993) and Turnovsky (1997).8

2.1. Firms and technology

All fi rm s are identical and are indexed by j. The representative fi rm produces output in accordance w ith the CES
production function:
Y
0r 0r
j ¼ P LjÞ þ ð10 aÞKj 80 1=r ð1aÞ
A½aðX

w here K and L represent the individual fi rm ’s capital stock and em ploym ent of labor, respectively, and s7 1/(1 þ r )
j j

represents the elasticity of substitution in production betw een capital and effective units of labor. In addition, production
is infl uenced by an aggregate com posite externality, XP, (infrastructure) w hich w e take to be a geom etric w eighted average
of the econom y’s aggregate stocks of private and public capital (K and KG, respectively):

XP ¼ K eK 1 G 0 e, 0 r er 1 ð1bÞ
that is, ‘ raw ’ labor interacts w ith the composite production externality to create labor efficiency units, w hich in turn interact w ith
private capital to produce output. The production function has constant returns to scale in both the private factors and in the
accumulating factors, and accordingly, sustains an equilibrium of endogenous grow th. The composite externality represents a
combination of the role of private capital, (proxying know ledge), as in Romer (1986), together w ith public capital as in Futagami
et al. (1993) and subsequent authors, and can be justified in tw o w ays. First, as w ill become evident below , it helps provide a
plausible calibration of the aggregate economy, something that is generically problematic in the conventional one-sector

7
Empirical studies that have explored the causality between growth and income inequality have generally yielded conflicting results. For example,
while Alesina and Rodrik (1994), Persson and Tabellini (1994), and Perotti (1996) find an inverse relationship, Li and Zou (1998) and Forbes (2000) have
documented a positive link, while Barro (2000) finds a positive relationship for developing countries and a negative one for developed economies. The
diversity of these results is unsurprising given that both growth and income inequality are endogenous outcomes and their co-movement will depend
upon the structural changes to which they are responding.
8
The solution procedure for this model follows Turnovsky and Garcı´a-Pen˜ alosa (2008), where it is discussed at length, and therefore details are
omitted here insofar as possible. We should, however, emphasize that their analysis is very different in that it employs a Ramsey model, rather than an
endogenous growth model. It is also addresses very different issues, being concerned with structural changes, such as changes in technology, and indeed
abstracts from fiscal issues that we are addressing here.
 S. Chatterjee, S.J. Turnovsky / European Economic Review 56 (2012) 1730–1745 1733

endogenous grow th model. Second, the notion that an economy’s infrastructure contributing to labor efficiency comprises a
combination both public and private components is itself a plausible representation of reality.9
All fi rm s are also assum ed to face identical com petitive production conditions, and hence w ill choose exactly the sam e
levels of em ploym ent of labor and private capital, i.e., K ¼K, and L ¼L, for all j, w here K and L denote the average econom y-
j j

w ide levels of private capital and labor em ploym ent, respectively. Letting z7 K /K denote the ratio of the econom y-w ide
G

stock of public capital to private capital, w e can w rite y 7 Y/K, the average product of aggregate private capital as
0r
y 7 yðz,lÞ¼ A½10 a þ afð10 lÞz10 e g
80 1=r ð2Þ

w here l ¼1 0 L denotes the average allocation of tim e to leisure in the econom y. W ith both factors being paid their
respective private m arginal products, the econom y-w ide returns to capital and labor, determ ined in com petitive factor
m arkets, m ay be expressed as

r ¼ rðz,lÞ7 ð10 aÞA0 r yðz,lÞ1 þ r ð3aÞ

w ¼ o ðz,lÞK; o ðz,lÞ7 aA0 r yðz,lÞ1 þ r z0 r ð10 eÞð10 lÞ0 ð1 þ rÞ

ð3bÞ

Thus, as long as eo 1, the real w age rate and the return to private capital depend on the ratio of public to private capital
and the average allocation of tim e to w ork (or leisure). 10

2.2. Consumers

There is a continuum of infi nitely-lived consum ers, indexed by i, w ho are identical in all respects except for their initial
endow m ents of private capital, Ki,0.11 Each consum er is also endow ed w ith one unit of tim e that can be allocated to either
leisure, l , or w ork, L ¼1 0 l . Consum er
i i i i m axim izes utility over an infi nite horizon from his fl ow of consum ption, C , and i

leisure, using the follow ing CES utility function:

Z 1 1
Ui ¼ ½C0i u þ yðXU l Þ0 u 80 g=ue0 bt dt ð4aÞ
0 g

w here q 7 1/(1 þ u) denotes the intra-tem poral elasticity of substitution betw een consum ption and leisure in the utility
function, and e7 1/(1 0 g) represents the inter-tem poral elasticity of substitution. Each consum er’s utility is also affected by
an aggregate com posite externality, XU, w hich is a geom etric w eighted average of the econom y’s aggregate stocks of public
and private capital:

X ¼ K j K1 G 0 j ,
U 0r j r 1 ð4bÞ

this com posite externality in Eq. (4b) interacts w ith the tim e allocated to leisure by consum er i to generate utility benefi ts,
w hich in turn are w eighted by y in yielding overall utility.
Several reasons m otivate specifying the preferences as in Eqs. (4a) and (4b). The fi rst is that the conventional Cobb–
Douglas form ulation of utility has the undesirable im plication that for plausible values of the intertem poral elasticity of
substitution (0 o eo 1), consum ption and leisure are Edgew orth ‘ substitutes’ . Generalizing the utility function to CES
allow s them to be com plem ents or substitutes depending upon w hether e 4 q.12 But for the CES function to have the
o
hom ogeneity properties required to sustain endogenous grow th, the externality m ust interact w ith leisure in the form w e
have specifi ed. Second, the notion that the utility derived from leisure depends upon am enities due to the provision of
both public and private capital is in fact a plausible one. As originally em phasized by Arrow and Kurz (1970), and m ore
recently by Agenor (2008); Econom ides et al. (2011) and Chatterjee and Ghosh (2011), m ost public goods, including
infrastructure, education, healthcare, law and order, etc., play a dual role in private allocation decisions by sim ultaneously
affecting both productivity and utility. Furtherm ore, as w e w ill dem onstrate below , the endogenous labor–leisure choice
plays an im portant role in infl uencing the dynam ics of inequality. Therefore, it is im portant to ensure that any public and
private externalities that im pinge on this choice – w hether they arise from the supply side or the dem and side – are
accounted for. 13

9
Note that the production function Eq. (1a) and the composite externality in Eq. (1b) nest both the traditional linear ‘ AK’ endogenous growth model
as in Romer (1986) ( e¼1) and the stock version of the Barro (1990) model ( e¼0), as in Futagami et al. (1993).
10
If e¼1 the factor returns depend only on leisure, as public capital does not affect production.
11
Private capital can be viewed as an amalgam of physical and human capital, as in Romer (1986).
12
The conventional Cobb–Douglas utility function is of the form: ðCi l yi Þg=g. Two goods are said to be Edgeworth complements
or substitutes
according to whether their cross partial derivatives are positive or negative. Estimates of the (intratemporal) elasticity of substitution between
consumption and leisure are sparse. However, in a well known study Stern (1976) estimates a value of 0.4, well below the value of 1 implicit in the
conventional Cobb–Douglas specification. His finding that optimal tax policy is sensitive to this elasticity suggests that generalizing the utility function to
the CES form is potentially important.
13
Chatterjee and Ghosh (2011) provide several examples of the dual role played by public goods in affecting both utility and productivity. They
discuss the productivity and utility-enhancing role of roads and highways, schools, power and water services, among others.
1734 S. Chatterjee, S.J. Turnovsky / European Economic Review 56 (2012) 1730–1745

Each agent chooses Ci , l i, and his rate of capital accum ulation, K i to m axim ize Eq. (4a) subject to Eq. (4b), their initial
_
endow m ent of capital, Ki,0, and the follow ing fl ow budget constraint

K_ i¼ ð10 t k ÞrK i þ ð10 t w Þw ð10 l i Þ0 ð1 þ t c ÞCi 0 T ð5Þ

w here t k, t w , and t c are the tax rates on the agent’s capital incom e, labor incom e, and consum ption expenditures,
respectively, and T represents a lum p-sum tax levied by the governm ent. In m aking these decisions the agent takes the
real w age rate and the return on private capital, determ ined in com petitive factor m arkets, as given, and treats all tax and
policy variables as exogenous.
Optim izing w ith respect to Ci , l i , and Ki yields the follow ing standard fi rst-order conditions
½C0i u 0u
þ yðXU li Þ 80 ðg=uÞ0 1Ci 0 u0 1¼ l ð1 þ t Þ ð6aÞ

yX0Uu½C0i u
þ yðXU l i Þ0 u80 ðg=uÞ0 1 l i0
u0 1
¼ l i ð10 t w Þo
ðz,lÞK ð6bÞ

l_i
ð10 t k Þrðz,lÞ¼ b0 l i ð6cÞ

Li l i Ki e0 bt
m ¼0 ð6dÞ
t- 1

w here l i is agent i’s shadow value of private capital. Dividing Eq. (6b) by Eq. (6a) yields the m arginal rate of substitution
betw een consum ption and leisure, w hich turns out to be identical for all agents:
Ci
l i ¼ Oðz,lÞK,Oðz,lÞ7 ½ð10 w Þo ðz,lÞzuð10 j c Þ81=ð1 þ uÞ ð7Þ
t Þ= yð1 þ t

2.3. Government

The governm ent provides the stock of public capital, w hich is assum ed to be non-rival and non-excludable, and evolves
according to

K_ G ¼ G ¼ gY, 0o go 1 ð8Þ
w here G is the fl ow of new public investm ent, w hich is tied to aggregate output Y. Therefore, g represents the fraction of
aggregate output allocated to public investm ent by the governm ent, and is the key policy param eter in the m odel.14
The governm ent fi nances its investm ent by tax revenues and m aintains a balanced budget at all points of tim e:
G¼ t k rK þ t w wð10 lÞþ t c Cþ T ð9Þ
dividing Eq. (9) by K, w hile noting Eq. (7), w e can w rite this in the form
gyðz,lÞ¼ t k r ðz,lÞþ t w o ðz,lÞð10 lÞþ t c Oðz,lÞl þ t yðz,lÞ ð10Þ
w here lum p-sum tax revenues are expressed as a proportion t (0 o t o 1) of aggregate output, nam ely T¼ t Y. It is clear
from Eq. (10) that if the tax and expenditure rates, t k, t w , t c, and g are m aintained constant, then as z and l progress along
the transitional path the fraction of output levied as lum p-sum taxes,t , w ill continually vary in order for the governm ent
budget to rem ain in balance.

3. M acr oecon om ic equilibr ium

In general, the econom y-w ide average of a variable, Xi is represented by ð1=NÞP Ni

Xi 7 X. Because of the hom ogeneity of
the utility function and perfect factor m arkets, w e can show that all individuals choose the sam e grow th rates for
consum ption and leisure, im plying that average consum ption, C, and leisure, l, w ill also grow at the sam e rates; i.e.,
_
Ci
C_ _l _l
Ci ¼ , ¼ , for each i ð11Þ
C i l
li
as a result, the system can be aggregated perfectly over agents.15 Each individual, how ever, w ill choose different levels of
consum ption and leisure, depending upon his resources; in particular,
1 XN
li ¼ p i l and p ¼1 ð12Þ
N i
 14
For simplicity we abstract from depreciation of either form of capital.
15
The relationship Eq. (11) is critical in facilitating the aggregation, and is due to Gorman (1953); See also Caselli and Ventura (2000). It is obtained
by taking the time derivative of Eqs. (6a) and (7), and noting (6c); see Turnovsky and Garcı´a-Pen˜ alosa (2008) for more details.
 S. Chatterjee, S.J. Turnovsky / European Economic Review 56 (2012) 1730–1745 1735

w here pi is relative leisure chosen by agent i, and is to be determ ined; see Eq. (20) below . Sum m ing Eqs. (7) and (8) over all
agents, w hile noting Eq. (10), yields the grow th rate of aggregate private capital
K
_ ¼ ð10 t k Þr þ 2ð10 t w Þo ðz,lÞð10 lÞ0 ð1 þ t c ÞOðz,lÞl0 t yðz,lÞ3 ð13Þ
K
com bining Eq. (13) w ith Eq. (10) yields the aggregate goods m arket clearing condition
K_
K ¼ ð10 gÞyðz,lÞ0 Oðz,lÞl ð13 0Þ

Given the hom ogeneity of the underlying utility and production functions in the capital stocks, the long-run
equilibrium of this econom y is a balanced grow th path along w hich all aggregate variables grow at a com m on rate and
average leisure is constant. The transitional dynam ics of the aggregate econom y are driven by the evolution of the ratio of
public to private capital, z, and leisure, l.
z
_
¼g ð14aÞ
z yðzz,lÞ0 2ð10 gÞyðz,lÞ0 Oðz,lÞl3

_
l ¼ Hðz,lÞ ð14bÞ
l Jðz,lÞ

w here the functions H(z,l) and J(z,l) are described in Appendix A.

Eq. (14a) asserts that the grow th of the public to private capital ratio equals the differential grow th rates of its
com ponents. Eq. (14b) is m ore involved and is obtained by com bining Eq. (11), w ith the tim e derivatives of Eqs. (6a)
and (7). It describes the adjustm ent in leisure that is necessary to equalize returns on consum ption and capital along the
transition path. 16

3.1. Steady state and aggregate dynamics

Assum ing that the system is stable, the aggregate econom y w ill converge to a balanced grow th path characterized by a
constant public to private capital ratio, , and leisure, ~ l . Setting ~ ¼ ~
l ¼ 0 in Eqs. (14a) and (14b) determ ines ~ and ~
l , such
that public capital, private capital, and consum ption, all grow at a com m on rate, given by

c~ ~ ~
ð10 t k Þr
~, ðzl Þ0 b y z~,
lÞ
¼ 7 g z~ ð15Þ
10 g
given ~
and l , Eq. (7) then determ ines the steady-state consum ption-private capital ratio, : Finally, the transversality
condition Eq. (6d) together w ith Eq. (6c) im plies c~þ b0 ~
,l Þð10 t k Þ0 b o 0, i:e:, c~o ð10 t k Þ, w hich com bined w ith Eq. (13)
in steady state, yields
~ ~ ~
ð10 t w Þo ðz~,l Þð10 l Þ0 t yðz~,l Þ
c~4
1þ tc ð16Þ

for the long-run grow th rate to be sustainable, consum ption expenditure (inclusive of tax) m ust exceed after-tax labor
incom e (inclusive of lum p-sum taxes), so that som e (net) capital incom e is allocated to consum ption. This viability
condition im poses a restriction on leisure that is necessary to constrain the grow th rate and is im portant in characterizing
the distributional dynam ics.
The aggregate transitional dynam ics for the econom y are obtained by linearizing (14) around the steady state values of
~ and ~l . The stable transition path of the aggregate econom y can be described by

zðt Þ¼
zð0Þ0 emt ð17aÞ

lðt Þ ~ a21
lþ ðt Þ0 8 ð17bÞ
¼ ðm0 a22Þ
w here mis the stable (negative) eigenvalue corresponding to the linearized dynam ic system , and aij are the corresponding
coeffi cients of the linearized m atrix. 17 Our num erical sim ulations reveal that the slope of the saddle path is negative,
so that along the transition path the evolution of leisure is inversely related to that of the public–private capital ratio.
Intuitively, an increase in public to private capital raises the productivity of private capital, raising the w age rate and
inducing agents to increase their labor supply and to reduce their leisure. 18 Finally, the consum ption-private capital ratio

16
Details of these calculations are available from the authors on request.
17
Given the analytical complexity of the model, we have conducted extensive numerical simulations over the plausible ranges for all of the model’s
deep structural parameters to ensure saddle-point stability of the system.
18
The exception is if e¼ j ¼1 (externality is fully private), so that a ¼0 and l(t) immediately jumps to l ~ .
21
1736 S. Chatterjee, S.J. Turnovsky / European Economic Review 56 (2012) 1730–1745

evolves according to
0 8 a 91
~~ ~~ ~ 21
cðt Þ ¼ Oz l l þ f Ol ,l l þ O , l Þg m0 a zðt Þ0 8 ð17cÞ
22
0
the dynamic time paths described in Eqs. (17a–17c) represent the average (mean) behavior of this heterogeneous agent
economy. Since both infrastructure and private capital represent stocks that are being accumulated, w e rule out instantaneous
jumps in z. How ever, leisure, the consumption-capital ratio, and the various grow th rates can respond instantaneously to new
information.

4. Dist r ibut ion al dyn am ics

The fact that the aggregate economy’s behavior characterized in Section 3 is independent of any distributional aspects is a
consequence of the homogeneity of the utility function and the perfect aggregation that this permits. The next step is to
characterize the behavior of a cross-section of agents, and to determine the evolution of that cross-section relative
to that of the average (mean) agent. Specifi cally, w e focus on the distributions of private capital (w ealth), income, and w elfare.

4.1. Distribution of private capital (wealth)

To derive the dynam ics of the relative capital stock of individual i, k 7 K /K (the agent’s relative w ealth) w e com bine
i i

Eqs. (5) and (13). To facilitate this, it is convenient to defi ne:

Dðz,lÞ7 ð10 t Þo ðz,lÞ0 t yðz~,lÞ; Gðz,lÞ7
w ½ð1 þ t ÞOðz,lÞþ ð10 t Þo ðz,lÞ84
c w 0
this enables us to express the evolution of relative w ealth (capital) in the convenient form
k
ðt Þ¼ 0 lÞþ ½Gðz,lÞl0
i0 i ðt Þ0 1Þ ð18Þ
_i
Gðz,lÞðl Dðz,lÞ8ðk
using this notation, the viability condition Eq. (16) can be expressed as G ~~ ~
, l l 4 D , l Þim plying that the dynam ic Eq. (18) is
locally unstable near the steady state. A key elem ent of a stable (bounded) solution includes the steady-state to Eq. (18),
w hich im plies a positive relationship betw een the agent’s steady state share of the private capital stock and leisure:

~
li "
D ~#
~ Þ ðz~,l k~
0l¼ ~l0 0 1Þ ð19Þ
G ~
Þ ðz~,l i

thus, the transversality condition im plies that an individual w ho in the long run has above-average private capital, given
by k ~ ~ 19
0 1, also enjoys above-average leisure, i.e., l i 0 l 4 0. Using Eq. (12), this equation also yields agent i’s (constant)
~i
allocation of leisure tim e:
~ !Þ
Dðz~,l
p i0 1 ¼ 10 ~
~~Þlðki 0 1Þ
Gðz~,l
ð20Þ

~
linearizing Eq. (20) around the steady-state levels , l , and k~, w hile noting Eqs.(17–19) im ply
i

k ~ ~
¼ d1 ,l ~i0 1 zðt Þ0 8þ d2 , l k ðt Þ0 k~8i ð21Þ
_i
w here d1 ~ ~
, l Þand d 2 , l Þ4 0 are constants defi ned in Appendix A. Eq. (21) highlights how the evolution of the econom y-
w ide ratio of public to private capital affects the evolution of relative w ealth, both directly, and indirectly through l(t).
The bounded solution to Eq. (21) is of the form
0 d1 1
ki ðt Þ0 1 ¼ ~i0 1Þ 1 þ ½zðt Þ0 ¼ k~ ð22Þ
z~8 0 1Þ01 þ m0d 1d2 ðz0 0 z~Þemt 1
m0 d2 i

setting t ¼0 in (22) gives

ki ð0Þ0 1 7 ki,0 0 1 ¼ k~i ð220Þ

0 1Þ01 þ m0d 1d2 ðz0 0 z~Þ1

Given the steady-state of the aggregate econom y, and his initial endow m ent of relative w ealth, ki,0, Eq. (22 0) determ ines
the agent’s steady-state relative stock of capital, ðk~i 0 1Þ, w hich together w ith Eq. (22) then yields its entire tim e path, ki (t)
and, further, w ith Eq. (20) determ ines the agent’s (constant) relative leisure, pi .20
Since all distributional variables are expressed relative to the m ean, w e can m easure their dispersion in canonical form ,
by using the coeffi cient of variation. 21 Given the linearity of Eqs. (22) and (22 0) in ki , w e can im m ediately transform these
 19
This is consistent with various sources of empirical evidence that finds a negative relationship between wealth and relative labor supply; see for
example, Holtz-Eakin et al. (1993), Cheng and French (2000), and Algan et al. (2003).
20
The ranking of agents according to their wealth remains unchanged throughout the transition.
21
Several measures of inequality have been proposed, of which the Gini coefficient and the coefficient of variation are the most prevalent; see e.g.
Atkinson (1970) and Ray (1997). Ray has enunciated four principles that desirable measures should satisfy, and which both these measures indeed meet.
 S. Chatterjee, S.J. Turnovsky / European Economic Review 56 (2012) 1730–1745 1737

equations into corresponding relationships for the coeffi cient of variation for the distribution of capital, w hich serves as
a convenient m easure of w ealth inequality. Therefore,

s k ðt Þ¼ 01 þ m0d 1d2 ðzðt Þ0 ~k , ð23aÞ

z~Þ1

w here s k(t) denotes the coeffi cient of variation for relative w ealth at tim e t, and s~k is the corresponding m easure at the
steady-state. Setting t ¼0 in Eq. (23a), the relationship betw een the initial distribution of w ealth and its steady-state
distribution is given by

~k s k,0 ð23bÞ
¼ 01 þ m0d 1d2 ðz0 0

z~Þ10 1

Thus Eqs. (23a) and (23b) com pletely characterize the evolution of w ealth inequality, given its initial distribution, s k,0,
and the initial stock of the infrastructure to private capital ratio, z0.22

4.2. Labor income inequality

Given that there is no heterogeneity in labor skills in our m odel, labor incom e inequality is purely driven by the
distribution of hours w orked by households. W age incom e for the ith household relative to the m ean w age incom e is
given by
lðt Þ
wi ðt Þ0 1 ¼ lðt Þ0 l i ðt Þ¼ ð10 pi Þ ð24Þ
10 lðt Þ 10 lðt Þ

using Eq. (20), the evolution of inequality in labor incom e is therefore given by

lðt
~ !Þ
Dðz~,l
Þ
s w ðt Þ¼ 10 ~ ~ ~k ð240Þ
10 lðt Þ G z~l l
,

w here s w (t) denotes the coeffi cient of variation of relative w age incom e at tim e t. From Eq. (24 0), w e see that labor incom e
inequality is determ ined by (i) the evolution of the average labor–leisure choice and (ii) the steady-state distribution of
relative w ealth. Given the viability condition from Eq. (19), i.e., G ~~ ~
, l l 4 D , l Þ, and the steady-state distribution of w ealth,
an increase in average labor supply in the econom y w ill increase labor incom e inequality. Since capital-poor agents
allocate m ore tim e to labor than the capital-rich, an increase in average labor supply causes their w age incom es to increase
at a slow er rate relative to that of the capital-rich, due to dim inishing returns.

4.3. Distribution of (after-tax) income

W e defi ne after-tax relative incom e for the ith household as

yi ðt Þ¼ ð10 t k ÞrK i þ ð10 t ÞwKð10 l i ðt ÞÞ ð25Þ

w w ÞwKð10 lÞ

ð10 t k ÞrK þ ð10

t
after-tax incom e inequality can then be expressed as
0 ðt w 0 t k Þsk ðt Þ 1
s y ðt Þ¼ zðt Þþ ð10 zðt ÞÞ sk ðt Þ ð26aÞ
s kðt Þð10 t k Þþ ð10 s kðt ÞÞð10 t w Þ

" #
~
zðt Þ7 sk ðt Þþ 0 sk ðt 8 lðt Þ 10 DÞ
ðz~,l ð26bÞ
10 lðt Þ 01 þ m0d 1d2 ðÞzðt Þ0 z~10 1
~ ~~
Gðz,l Þl

w here s (t) denotes the coeffi cient of variation of relative after-tax incom e at tim e t, and s (t) 7 r(z,l)/y(z,l) is the
y k

equilibrium share of output received by capital. W hile w ealth inequality, s k(t), evolves gradually, the initial jum p in
leisure, l(0), w hich affects short-run incom e inequality, s y(0), through its effect on labor incom e inequality, im plies that a
structural or policy shock causes an initial jum p in incom e inequality, after w hich it evolves continuously. As a result,
short-run incom e inequality, s y(0), m ay over (under)-shoot its long-run equilibrium , s~y .
(footnote continued)
In addition, Atkinson compares the inequality rankings yielded by various measures and the finds that the coefficient of variation and Gini coefficient are
generally pretty close. In the light of this, given its convenience, we adopt the coefficient of variation as our inequality measure.
22
The fact that the long-run distribution depends upon the initial distribution reflects a hysteresis property resulting from the ‘ zero root’ associated
with Eq. (11). This turns out to have important implications for wealth and income inequality that are explored in another context by Atolia et al. (2012).
1738 S. Chatterjee, S.J. Turnovsky / European Economic Review 56 (2012) 1730–1745

4.4. Distribution of welfare

Econom ic w elfare is another key indicator of the im pact of governm ent policies on national w ell-being, and given the
unequal distribution of private w ealth and incom e in the econom y, it is im portant to study its distribution. Recalling the
utility function Eq. (4a), the instantaneous level of w elfare for individual i at tim e t is
1 1
Wi ¼ ½C0i u þ yðz10 j li KÞ0 u80 ðg=uÞ¼ 0u
½Oðz,lÞ þ yzuðj 0 1Þ80 ðg=uÞðl
KÞg ð27Þ
g g
w hile the average level of instantaneous w elfare is given by
1 0u
W ¼ g ½Oðz,lÞ þ yzuðj 0 1Þ80 ðlKÞg ð270Þ
g=u

at each instant of tim e, agent I’s relative w elfare rem ains constant, so that his intertem poral relative w elfare is constant as
w ell. Using Eq. (20) w e can express relative w elfare in the form
" ~ !Þ #g
Ui W i 8 li 9 g Dðz~,l
wi 7 ¼ 1 þ 10 k~0 1Þ
i ð28Þ
U ¼ W ¼ l ~ ~~
Gðz,l Þl

by applying the m onotonic transform ation (w )1/g7 u(v ), w e obtain

i an expression for the relative w elfare of individual i
i

expressed in term s of equivalent units of w ealth. The dispersion of w elfare across agents is then given by its coeffi cient of
variation, s u23:
~!
Dðz~,l
s u ¼ 10 Þ ~k ð29Þ
G ~~
ðz~,l
Þl

5. Fiscal policy, gr ow t h, an d in equality: a n um er ical an alysis

Given the com plexity of the m odel, w e analyze it using num erical sim ulations. The objective is to determ ine the effect
of an increase in infrastructure investm ent by the governm ent on grow th and the various distributional m easures
described above. In doing so, w e com pare the dynam ic adjustm ent of the econom y under four alternative fi nancing
schem es, nam ely w here the long-run increase in governm ent investm ent is fully fi nanced by a (i) lum p-sum tax ( t ),
(ii) capital incom e tax ( t k), (iii) labor incom e tax ( t w ), or (iv) consum ption tax ( t c). W e begin w ith the follow ing
param eterization of a benchm ark econom y:

Preferences g¼ 0 1.5, b ¼0.04, y ¼1.75, u¼0

Production A¼0.6, a ¼0.6, r ¼0
Externalities j ¼ e¼0.6
Fiscal g¼0.05, t ¼0.05

The preference and production functions are non-controversial. Setting u¼0 and r ¼0 yields the conventional case,
w here utility is of the constant elasticity form and production is Cobb–Douglas, so that the respective elasticities of
substitution are both unity. The rate of tim e preference b ¼0.04 is standard, w hile setting g¼ 0 1.5 yields an intertem poral
elasticity of substitution of 0.4, w hich is consistent w ith the bulk of the em pirical estim ates; see Guvenen (2006). The
relative w eight of leisure in utility, y ¼1.75 im plies an equilibrium value of leisure, ~ l ¼ 0:714, w hich is consistent w ith
em pirical evidence from the RBC literature; see e.g. Cooley (1995). Finally, the scale param eter A is set to yield a plausible
equilibrium grow th rate of 2.29%.
The less fam iliar aspects of our param eterization concern the specifi cation of the com posite externalities in production
and utility and are guided by the follow ing considerations. For the Cobb–Douglas specifi cation, the representative fi rm ’s
production function is of the form Y ¼A(K )1 0 a(L )aKae(K )a( 1 0 e). The conventional
i i i G Rom er (1986) m odel corresponds to
e¼1, and for a ¼0.6 im plies that the external effect of the aggregate capital stock (K)is signifi cantly m ore productive than
is the fi rm ’s ow n capital (K ). The stock version of the Barro (1990) m odel corresponds to e¼0, w hich w ith a ¼0.6 yields an
i

im plausibly large output elasticity for public capital. Neither of these param eterizations is realistic. Setting e¼0.6,
how ever, helps in resolving both problem s. First, the fi rm ’s ow n capital stock is now m ore productive than the externality
(0.40 vs. 0.36), w hile the elasticity of governm ent capital is reduced to 0.24, thus placing it w ithin the plausible range
reported by Bom and Ligthart (2010). W ith the externality constrained in this w ay, w e set j ¼ e¼0.6, since w e fi nd no
com pelling reason to assum e that there should be any system atic difference in the construction of the tw o externalities in
the benchm ark econom y. W e do, how ever, perform a sensitivity analysis w ith respect to these param eters in Section 6.

23
Eq. (29) also measures the dispersion of consumption and leisure across agents.
 S. Chatterjee, S.J. Turnovsky / European Economic Review 56 (2012) 1730–1745 1739

Table 1

A. Benchmark steady-state equilibrium. g¼0.05. e¼ j ¼0.6 (composite externality), q¼s¼1

Financing policy z~ l~ y~ c~ (%)

Lump-sum tax-financing, t ¼0.05 0.531 0.714 0.243 2.29

B. Increase in government spending: aggregate and distributional effects. Benchmark specification g¼from0.05 to 0.08

i. Steady-state aggregate effects

Policy change dz~ dl ~ dc~

Lump-sum tax-financed increase in g, dt ¼0.030 0.259 0 0.01 0.206

Capital income tax-financed increase in g, dt ¼0.075 k 0.353 0 0.006 0.101
Labor income tax-financed increase in g, dt ¼0.05 w 0.268 0.002 0.168
Consumption tax-financed increase in g, dt ¼0.037 c 0.265 0 0.001 0.179
ii. Distributional effects (short-run and long-run percentage changes)a

Policy change Wealth inequality, ds k Incom e inequality, ds y

ds k(0) ds~k ds y(0) ds~y

Lump-sum tax-financed increase in g 0 2.736 0 2.602 4.996

Capital income tax-financed increase in g 0 3.527 0 9.174 0 0.149
Labor income tax-financed increase in g 0 2.805 0 0.110 7.933
Consumption tax-financed increase in g 0 2.952 0 3.117 4.955

a
All distributional effects are reported as percentage changes relative to their pre-shock levels: ds ¼ ½ðs ðtÞ0 s~ Þ=s~ 82
j j j,0 j,0 100, j ¼ k,y,u.

The benchm ark governm ent spending ratio, g, is assum ed to be 5%of GDP, w hich is roughly consistent w ith evidence on
the rate of public infrastructure spending for m ost OECD countries. Table 1A sum m arizes the benchm ark equilibrium , w ith
an equilibrium ratio of public–private capital of 0.53 and output-private capital ratio of around 0.24.

5.1. Increase in government spending on infrastructure

W e consider the effect of an unanticipated and perm anent increase in g from its benchm ark rate from 5%to 8%of GDP,
and com pare the dynam ic responses under the four fi nancing schem es noted above. In all cases w e assum e that the
econom y starts from an initial benchm ark equilibrium in w hich governm ent expenditure is fully fi nanced by lum p-sum
taxes, and all distortionary tax rates are zero, i.e. t ¼ t ¼ t ¼0, so that g ¼ t ¼0.05 in Eq. (10). For the distortionary taxes,
c w k 0 0

w e assum e that the corresponding tax rate is set such that it fully fi nances the long-run change in governm ent expenditure
(see Table 1B). Thus, during the transition as the tax base changes, residual lum p-sum tax fi nancing m ust be em ployed to
ensure that the budget rem ains balanced at all tim es.

5.1.1. Aggregate effects

Table 1B(i) show s the effect of an increase in governm ent spending on the steady-state of the aggregate econom y. In all
cases, the direct stim ulus to public investm ent causes the equilibrium ratio of public to private capital, z~, to increase.
Except w hen spending is fi nanced by a tax on labor incom e, leisure falls in the long run, as the higher spending raises the
m arginal product of labor through the com posite externality in the production function. In contrast, w hen g is fi nanced by
a tax on labor incom e, the tim e allocated to leisure increases, as the higher tax rate reduces the after-tax return on labor.
But in all cases the effects are sm all. For all form s of fi nancing, the productive benefi ts of public capital spending and the
consequent private capital accum ulation ensure that the equilibrium grow th rate increases. They dom inate any negative
tax effects, although in the case of capital incom e tax-fi nancing w ith its direct adverse im pact on the return to capital, the
positive grow th effects are sm all. Overall, the differential im pacts on grow th, leisure and the ratio of public to private
capital refl ect the varying degrees of distortions associated w ith the different tax rates.24

5.1.2. Distributional effects

Table 1B(ii) reports the short-run (instantaneous) and long-run effects on w ealth and incom e inequality. All these
effects are calculated as percentage changes in the coeffi cient of variation relative to its pre-shock steady-state level, i.e.,
ds ðt Þ¼ ½ðs ðt Þ0 s~ Þ=s~ 82 100,
j j j ,0 j ,0 j ¼ k,y,u:

24
We do not discuss the transitional adjustment paths for the aggregate economy, as these are well-known from the public investment-growth
literature; see Turnovsky (1997) for an early example. The results are available upon request.