Inequality and Growth-Theory and Policy Implications-Theo S. Eicher, Stephen J. Turnovsky

Inequality and Growth
The CESifo Seminar Series

Hans-Werner Sinn, editor
Inequality and Growth: Theory and Policy Implications, ed. Theo S.

Eicher and Stephen J. Turnovsky (2003)
Public Finance and Public Policy in the New Century, ed. Sijbren Cnos-
sen and Hans-Werner Sinn (2003)
Inequality and Growth
Theory and Policy
Implications
Theo S. Eicher and

Stephen J. Turnovsky,
editors
The MIT Press

Cambridge, Massachusetts
London, England
( 2003 Massachusetts Institute of Technology
All rights reserved. No part of this book may be reproduced in any form by any
electronic or mechanical means (including photocopying, recording, or information
storage and retrieval) without permission in writing from the publisher.
This book was set in Palatino on 3B2 by Asco Typesetters, Hong Kong, and was
printed and bound in the United States of America.
Library of Congress Cataloging-in-Publication Data
Inequality and growth : theory and policy implications / Theo S. Eicher and
Stephen J. Turnovsky, editors.
p. cm. (The CESifo seminar series ; 1)
Conference papers.
Includes bibliographical references and index.
ISBN 0-262-05069-2 (hc. : alk. paper)
1. Income distributionCongresses. 2. Economic developmentCongresses.
I. Eicher, Theo S. II. Turnovsky, Stephen J. III. CESifo (Organization). IV. Series.
HC79.I5I493 2003
339.20 2dc21 2002043162
Contents
Series Foreword vii

Preface ix
I Measuring the Impact of Growth on Inequality 1
1 The Growth Elasticity of Poverty Reduction: Explaining

Heterogeneity across Countries and Time Periods 3
Francois Bourguignon
2 One Third of the Worlds Growth and Inequality 27

Danny Quah
II Population, Education, and Inequality 59
3 Fertility Clubs and Economic Growth 61

Avner Ahituv and Omer Moav
4 Human Capital Formation, Income Inequality, and

Growth 89
Jean-Marie Viaene and Itzhak Zilcha
III European Transition and Inequality 119
5 Social Transfers and Inequality during the Polish

Transition 121
Michael P. Keane and Eswar S. Prasad
vi Contents
6 Growth and Inequality: Evidence from Transitional Economies

in the 1990s 155
Oleksiy Ivaschenko
IV The Political Economy of Inequality 199
7 (Re-)Distribution of Personal Incomes, Education, and

Economic Performance across Countries 201
Gunther Rehme
8 The Impact of Tax Policy on Inequality and Growth: An

Empirical and Theoretical Investigation 227
Theo S. Eicher, Stephen J. Turnovsky, and Maria Carme Riera
Prunera
V Technology and Natural Resources and Inequality 253
9 Inequality and Economic Growth: Do Natural Resources

Matter? 255
Thorvaldur Gylfason and Gyl Zoega
10 Wage Inequality and the Effort Incentive Effects of Technical

Change 293
Campbell Leith, Chol-Won Li, and Cecilia Garca-Penalosa
Index 319
Series Foreword
This book is part of the CESifo Seminar Series in Economic Policy,

which aims to cover topical policy issues in economics from a largely
European perspective. The books in this series are the products of
the papers presented and discussed at seminars hosted by CESifo, an
international research network of renowned economists supported
jointly by the Center for Economic Studies at Ludwig-Maximilians-
Universitat, Munich, and the Ifo Institute for Economic Research. All
publications in this series have been carefully selected and refereed
by members of the CESifo research network.
Hans-Werner Sinn
This page intentionally left blank
Preface
Economists and policymakers alike are inherently interested in the

process of economic growth. As Nobel Laureate Robert E. Lucas
(1988) pointed out in On the Mechanics of Economic Development,
the nature of growth is such that the welfare implications of small
(and possibly simple) policy changes can be staggering. Even minute
increases in the growth rate can compound into dramatic changes
in living standards over just one generation. However, the precise
nature of such growth enhancing policies is often elusive, which
highlights that the execution of such policies presupposes a clear
understanding of the mechanics of economic growth.
Even if growth is achieved, its benets may not be shared equally.
Some may gain more or less than others, and a fraction of the
population may actually be disadvantaged by the enhanced growth
performance of the aggregate economy. The relationship between
inequality and growth is especially topical since the most recent
economic literature reports both positive and negative relationships
between growth and inequality across nations (for a review, see
Aghion, Caroli, and Garca-Penalosa 1999). If policies are designed
to foster growth, what is their impact on inequality? Aside from
possible normative desires to reduce inequality, does (in)equality by
itself facilitate or detract from economic growth, and does it amplify
or diminish policy effectiveness? Once the relationship between
growth and inequality is claried, one can attempt to answer these
questions and inquire about the role for public policy in enhancing
growth and equality.
The questions we raise here have generated great interest in
both academic journals and in the popular press. In this volume we
seek to provide a forum for economists to examine the theoretical,
x Preface
empirical, and policy issues involved. The ultimate goal is to provide

a framework for an informed policy debate based on sound analyti-
cal frameworks and solid empirical evidence.
The chapters that comprise this volume illuminate the range of
economic fundamentals and policies that inuence growth and
equality positively. However, two features stand out as almost com-
mon among all chapters. The rst is the unusual diversity of contexts
within which the issue of inequality and growth is discussed. The
chapters in this volume encompass topics from natural resources,
taxation, fertility, redistribution, technological change, transition,
labor markets, to education in both developed and developing
countries. We believe that this indicates the breadth of the theory
and policy implications that are the subject of inequality and growth.
On the other hand, despite the diversity of topics, one policy pre-
scription is virtually common among just about all papers: It is the
importance of education in reducing inequality and increasing eco-
nomic growth.
Francois Bourguignon (chapter 1) demonstrates that both the level
of development and the degree of inequality inuence the speed by
which economic growth reduces poverty. Growth is shown to gen-
erate more rapid poverty reduction in low-inequality, high-income
countries. These results show a strong bias against the poorest, most
unequal economies, as growth is shown to be least effective in alle-
viating poverty in countries that need it most. On the other hand,
Bourguignon highlights the double dividend of redistribution, which
increases growth as well as the speed with which growth decreases
poverty. Danny Quah (chapter 2) takes Bourguignons argument one
step further to address directly the concerns of those who worry
exclusively about the impact of growth on poverty. He documents
that neither the belief that inequality drives growth, nor the asser-
tion that growth generates inequality, has support in economic
data. Quah reminds us that much of the variation in inequality is
across countries, but that over time, this variation does not change
much. Conversely, growth rates hardly vary across countries, but
vary dramatically over time. Using India and China between 1980
and 1992 as examples, Quah uses simple arithmetic to show that,
even if growth increased inequality, more than 300 million were
helped out of poverty.
Avner Ahituv and Omer Moav (chapter 3) highlight the impor-
tance of education in increasing growth and decreasing fertility. The
Preface xi
authors provide both theory and empirical evidence to support the

hypothesis that reductions in fertility improve growth and increase
education. The effect is amplied by a positive feedback: More edu-
cated parents become progressively better at providing excellent
education. Policy implications are clear: Subsidies to parents should
focus on educational incentives rather than simple child support.
Jean-Marie Viaene and Itzhak Zilcha (chapter 4) show how different
types of education affect human capital accumulation and respond to
technical change. They warn of the dangers of a digital divide, where
technical change in the computer industry could lead to greater het-
erogeneity that would hamper growth.
Instead of relying on aggregate data, Michael P. Keane and Eswar
S. Prasad (chapter 5) examine actual records of Polish budget survey
microdata to construct estimates of inequality that are signicantly
smaller than those previously reported. Large transfer payments
have long been considered detrimental to the transition process, but
the authors show, using the microdata, that Polish shock therapy
worked because of smart redistribution not only to the very poor but
also to the structurally unemployed. Large and specically targeted
transfers to pensioners increased the political support for reforms of
those who were potentially most hurt. At the same time, these spe-
cically targeted transfers increased productivity, since they served
as incentives for retirement of the least productive. Taking the anal-
ysis from Poland to the entire group of transition countries, Oleksiy
Ivanschenko (chapter 6) highlights the differences in European and
the former Soviet Union (FSU) inequality and growth contributors.
Ination is a major issue in the FSU. However, in the more stable
macroeconomic environments, reform indicators are much more sig-
nicant in determining growth and inequality. The precise relation-
ship is shown to be a function of income thresholds.
Gunther Rehme (chapter 7) examines alternative education me-
chanics from Viaene and Zilcha, to focus on the dual function of
education as a public institution and as a means of redistribution.
Essentially he shows that the benets from redistribution via educa-
tion signicantly outweigh those associated with simple income dis-
tribution. Theo Eicher, Stephen J. Turnovsky, and Maria Carme Riera
Prunera (chapter 8) employ a formal growth model that includes the
incentives to invest in education to simulate both short- and long-
term policy experiments, such as the Reagan tax cuts. They conclude
that the recent, second generation of endogenous growth models is
xii Preface
well suited to predict growth and inequality in the short and long
term.
Thorvaldur Gylfason and Gyl Zoega (chapter 9) sharpen the
growing literature popularized by Sachs and Warner (1995) on
natural resources and growth. The contribution of the chapter is to
demonstrate how increased dependence on natural resources tends
to go along with less economic growth and greater inequality in
the distribution of income within countries. The apparently inverse
empirical relationship between growth and inequality may hence
not imply any causal relationship between the two. Instead, natural
resource dependence may be affecting both equality and growth.
Subsidies to education can mitigate the negative effect of natural
resources and their unequal distribution. Finally, Campbell Leith,
Chol-Won Li, and Cecilia Garca-Penalosa (chapter 10) investigate
how the impact of technical change on inequality and growth is
amplied by labor market frictions. They show that when technol-
ogy generates job destruction, relative wages rise or fall depending
on the magnitude of the labor market frictions. Such dynamics are
shown to be capable of explaining U.S./European unemployment
and relative wage differentials. Most interesting is that the authors
highlight the dilemma of policymakers. While job protection may
lower inequality and increase growth, it would hurt those workers
who are now stuck longer in structural unemployment.
The two conferences on Inequality and Growth that preceded this
volume were made possible in large part by the support of Hans-
Werner Sinn and the CESifo network. We thank him for his un-
wavering support. The quality of the conferences and papers was
assured by extensive comments from discussants including Daniele
Checchi, Rafael Domenech, Walter Fisher, Carola Grun, Raji Jayara-
man, Louise Keely, Henryk Kierzkowski, Stephan Klasen, Antonio
Garcia Pascual, Cecilia Garca-Penalosa, Maria Carme Riera Prunera,
Ray Riezman, Mathias Thoenig, Gyl Zoega, and a number of
anonymous referees. CESifo, and specically Roisin Hearn and
Frank Westermann, provided excellent and efcient administrative
support throughout, and they were ably assisted by Silke U bel-
messer and Marko Kothenburger. The Castor Endowment at the
University of Washington also supported this project, a subject in
which Cecil and Jane Castor held a lifelong interest. Financial sup-
port was also provided by the European Union Center at the Uni-
versity of Washington. We are grateful to all contributors for making
Preface xiii
this venture possible and worthwhile. Finally we would like to

extend our deep appreciation to our wives, Regina Lyons and
Michelle Turnovsky, for their loving support during the completion
of this volume.
References
Aghion, Philippe, Eve Caroli, and Cecilia Garca-Penalosa. 1999. Inequality and Eco-
nomic Growth: The Perspective of the New Growth Theories. Journal of Economic Lit-
erature 37(4): 16151660.
Lucas, Robert E. 1988. On the Mechanics of Economic Development. Journal of Mon-
etary Economics 22(1): 342.
Sachs, Jeffrey D., and Andrew M. Warner. 1995. Natural Resource Abundance and
Economic Growth. National Bureau of Economic Research Working Paper no. 5398.
I Measuring the Impact of
Growth on Inequality
1 The Growth Elasticity of
Poverty Reduction:
Explaining Heterogeneity
across Countries and
Time Periods
1.1 Introduction
Part of the ongoing debate on poverty reduction strategies bears on

the issue of the actual contribution of economic growth to poverty
reduction. There is no doubt that faster economic growth is asso-
ciated with faster poverty reduction. But what is the corresponding
elasticity? If it is reasonably high, then poverty reductions strategies
almost exclusively relying on economic growth are probably justi-
ed. If it is low, however, ambitious poverty reduction strategies
might have to combine both economic growth and some kind of
redistribution. Ravallion and Chen (1997) estimated that, on aver-
age on a sample of developing countries, the growth elasticity of
poverty, as measured by the number of individuals below the con-
ventional $1-a-day threshold, was around 3namely, a 1 percent
increase in mean income or consumption expenditures in the popu-
lation reduces the proportion of people living below the poverty line
by 3 percent. As emphasized in Attacking Poverty, World Development
Report 2000/2001, however, there is very much cross-country hetero-
geneity behind this average gurewhich, as a matter of fact, was
found there closer to 2 than 3.1 Several countries knew only limited
changes in poverty despite satisfactory growth performances where-
as poverty fell in some countries where growth had yet been disap-
pointing. Understanding the causes of that heterogeneity is clearly
crucial for the design of poverty reduction strategies.
To get some idea of the actual heterogeneity in the relationship
between changes in poverty and changes in income, and of the
ambiguity of average cross-sectional data, gure 1.1 plots observa-
tions that come from a sample of growth spells taken over various
periods in selected countries. These spells are essentially dened by
4
Figure 1.1
The relationship between poverty reduction and growth in a sample of growth spells
The Growth Elasticity of Poverty Reduction 5
the availability of fully comparable household surveys at the two

ends of the spell, only the most distant ends being considered in
the case of adjacent annual observations. The sample comprises
114 spells covering approximately 50 countriessee the list in the
appendix. The common poverty line that is used is the $1-a-day line
after local currency expenditures data have been translated into dol-
lars and PPP correction.2 The poverty measure being used is the
headcount ratio, that is the proportion of the population below the
poverty line. As expected, the scatter of observations shows a de-
clining relationship between the change in poverty and the change
in the mean income. The tted OLS straight line has a slope equal to
1.6, suggesting an average elasticity somewhat below the value of
2 reported in World Bank (2000). Differences in the mean elasticity in
different samples of countries or growth spells is not the issue, how-
ever. The issue is the difference across countries or growth spells.
From that point of view gure 1.1 is disappointing. Changes in
the mean income of the population explain only 26 percent of the
variance of observed changes in poverty headcounts. If this gure
is taken seriously, would it make sense to base poverty reduction
so much on growth strategies, as is often suggested? Wouldnt it
be better to identify rst the nature of the remaining 74 percent
and then the reason why poverty in various countries tend to react
very differently to the same increase in the mean income of the
population?
Analytically, an identity links the growth of the mean income in a
given population, the change in the distribution of relative incomes,
and the reduction of poverty. Formally, the relationship between
poverty and growth may be obtained from that identity in the case
where there would be no change in the distribution of relative indi-
vidual incomes, or, in other words, if income growth were the same
in all segments of society. Even in that case, however, the growth-
poverty relationship is not simple and the corresponding elasticity is
certainly not constant across countries and across the various ways
of measuring poverty. In effect, the growth elasticity of poverty is a
decreasing function of the development level of a country and of
the degree of inequality of the income distribution, this function
depending itself on the poverty index that is being used. A rather
precise characterization of that relationship is offered in this chapter
under some simplifying assumption about the underlying distribu-
tion of income.
6 Francois Bourguignon
The second source of heterogeneity is of course the change in the

distribution of relative incomes over time. Measuring the actual
contribution of that source to the observed evolution of poverty is of
utmost importance since this should give an indication of the practi-
cal relevance of distributional concerns in comparison with pure
growth concerns in poverty reduction policies. Because the actual
contribution of growth to poverty reduction is not precisely identi-
ed by the practice that consists of assuming a constant elasticity, it
follows that the contribution of the distributional component is also
imprecisely estimated. The methodology proposed in this chapter
permits correcting that imprecision too and has implications for
the interpretation of available evidence on the growth-poverty
relationship.
Many recent papers focused on the statistical relationship between
economic growth and poverty reduction across countries and time
periods. Many of themsee, for instance, de Janvry and Sadoulet
(1995, 2000), Ravallion and Chen (1997), Dollar and Kraay (2000)
are based on linear regressions where the evolution of some pov-
erty measure between two points of time is explained by the growth
of income or GDP per capita and a host of other variables, the
main issue being the importance of GDP and these other variables
in determining poverty reduction. By adopting a linear regres-
sion framework, or by investing too little in functional specication
testing, however, these papers miss the earlier point, that is that
of a complex but yet identity-related relationship between mean
income growth and poverty change. On the contrary, other authors
for instance, Ravallion and Huppi (1991), Datt and Ravallion
(1992), Kakwani (1993)3 fully take into account the poverty/mean-
income/distribution identity in studying the evolution of poverty
and its causes. In particular, they are all quite careful in distinguish-
ing precisely the effects on poverty reduction of growth on the one
hand and distributional changes on the other. At the same time, their
analysis is generally restricted to a specic country or a limited
number of countries or regions: Indonesia, regions of Brazil and
India, the Cote dIvoire, respectively.
This chapter stands midway between these two approaches.
Ideally, international comparisons in the evolution of poverty should
all rely on the methodology based on the poverty/mean-income/
distribution identity. But, because it requires using the full micro-
economic information on the distribution of income or expenditures
in each country or region, it may seem cumbersome. Instead, this

chapter proposes a methodology that is less demanding. It relies on
functional approximations of the identity, and in particular on an
approximation based on the assumption that the distribution of
income or consumption expenditure is lognormal.
A simple application to the sample of growth spells shown in
gure 1.1 shows that these approximations t the data extremely
well and do incomparably better than the linear model that is gen-
erally used. It also suggests that only half of the observed changes in
poverty in the sample may be explained by economic growth, the
remaining half being the result of changes in the distribution of rel-
ative incomes.
The chapter is organized as follows. Section 1.2 introduces and
discusses the analytical identity that links poverty, growth, and dis-
tribution. Closed-form formulae for the growth elasticity of poverty
when the distribution is lognormal are derived. They are used to
analyze in some detail the theoretical relationship that exists among
poverty reduction, the level of development, and the inequality of
the distribution when the distribution of income is remaining con-
stant over time. Section 1.3 tests the empirical validity of various
approximations to the preceding identity, including the lognormal
approximation, and compares the results with the standard linear
specication. A concluding section 1.4 draws several implications of
the main argument in the chapter for the empirical analysis of the
relationship between growth and poverty and for the design of pov-
erty reduction strategies.
1.2 The Arithmetic of Distributional and Poverty Changes
Any poverty index may be seen as a statistics dened on all indi-

viduals in a population whose standard of living lies below some
predetermined limit. In what follows, it is assumed that there is no
ambiguity on the denition of that poverty line; that is, whether it
is dened in terms of income or consumption, the kind of equiva-
lence scale being used to account for heterogeneity in household
composition, and indeed the level of that poverty line. This poverty
line will also be assumed to be constant over timeat least during
the period being analyzed. Since the argument will implicitly refer to
international comparisons, it makes also sense to assume that this
poverty line is dened in absolute terms and the same across
countries, for instance, the familiar $1 or $2 a day after correction for

purchasing power parity.
Given this denition of poverty, let y be a measure of individual
living standardsay, income per adult equivalentand let z be the
poverty line. In a given country, the distribution of income at some
point of time, t, is represented by the cumulative distribution func-
tion Ft Y, which stands for the proportion of individuals in the
population with living standard, or income, y, less than Y. The most
widely used poverty index is simply the proportion of individuals
in the population below the poverty line, z. This index is generally
referred to as the headcount. With the preceding notations, it may
be formally dened as
Ht Ft z: 1
For the sake of simplicity, the analysis will be momentarily restricted
to that single poverty index.
The denition of the headcount poverty index implies the fol-
lowing denition of change in poverty between two points of time,
t and t 0 :
DH Ht 0 Ht Ft 0 z Ft z:
To show the contribution of growth to the change in poverty, it is

convenient to dene the distribution of relative income at time t as the
distribution of incomes after normalizing by the population mean.
This is equivalent with dening the distribution of income in a way
that is independent of the scale of incomes. Let F~t X be that distri-
bution. With this denition, any change in the distribution of income
may then be decomposed into (a) a proportional change in all
incomes that leaves the distribution of relative income, F~t X, un-
changed; and (b) a change in the distribution of relative incomes,
which, by denition, is independent of the mean. For obvious rea-
sons, the rst change will be referred to as the growth effect
whereas the second one will be termed the distributional effect.
This decomposition was discussed in some detail by Datt and
Ravallion (1992) and Kakwani (1993). It is illustrated in gure 1.2.
This gure shows the density of the distribution of incomethat is,
the number of individuals at each level of income represented on a
logarithmic scale on the horizontal axis. In that gure, the function
F appears only indirectly as the area under the density curves.
The move from the initial to the new distribution goes through an
The Growth Elasticity of Poverty Reduction
9
Figure 1.2
Decomposition of change in distribution and poverty into growth and distributional effects
intermediate step, which is the horizontal translation of the initial

density curve to curve (I). Because of the logarithmic scale on the
horizontal axis, this change corresponds to the same proportional
increase of all incomes in the population and thus stands for the
growth effect. Then, moving from curve (I) to the new distribution
curve occurs at constant mean income. This movement thus corre-
sponds to the change in the distribution of relative income, or the
distribution effect. Of course, there is some path dependence in
that decomposition. Instead of moving rst rightward and then up
and down as in gure 1.2, it would have been possible to move rst
up and down and to have the distribution effect based on the mean
income observed in the initial period, and then to move rightward.
Presumably, these two paths are not necessarily equivalent except
for innitesimal changes. This issue is ignored in what follows,
assuming in effect that all changes are sufciently small for path
dependence not to be a problem.4
Figure 1.2 illustrates the natural decomposition of the change in
the whole distribution of income between two points of time. Things
are somewhat simpler if the focus is exclusively on poverty, as mea-
sured by the headcount. In gure 1.2, the poverty headcount is sim-
ply the area under the density curve at the left of the poverty line,
arbitrarily set at $1 a day. The move from the initial curve to the
intermediate curve (I) and then from that curve to the new distribu-
tion curve have natural counterparts in terms of changes in this area.
More formally, this decomposition may be written as

~ z ~ z ~ z ~ z
DH Ht Ht Ft
0 Ft Ft 0 Ft : 2
yt 0 yt yt 0 yt 0
This expression is the direct application in the case of the head-

count poverty index of the general formula proposed by Datt and
Ravallion (1992), which they applied to Brazilian and Indian distri-
bution data. It is indeed a simple identity since it consists of adding
and subtracting the same term F~t z=y 0 in the original denition of
t
the change in poverty. The rst expression in square bracket in (2)
corresponds to the growth effect at constant relative income dis-
tribution, F~t , that is the translation of the density curve along
the horizontal axis in gure 1.2, whereas the second square bracket
formalizes the distribution effectthat is, the change in the relative
income distribution, F~t 0 X F~t X, at the new level of the relative
poverty line, that is the ratio of the absolute poverty line and the
mean income, X z= yt 0 .
Shifting to elasticity concepts, the growth elasticity of poverty may
thus be dened as

~ z ~ z z
Ft Ft F~t
yt 0 yt yt
e lim : 3
t 0 !t yt 0 yt =yt
The distribution effect is more difcult to translate in terms of elas-

ticity because it generally cannot be represented by a scalar.
The terms entering the decomposition identity (2) may be evalu-
ated as long as one observes some continuous approximation of the
distribution functions F at the two points of time t and t 0 . Contin-
uous kernel approximations of the density and cumulative relative
distribution functions may be computed from available microeco-
nomic data. With this kind of tool, evaluating the decomposition
identity for any growth spell for which distribution data are avail-
able at the two ends of the spell should not be difcult. In a
cross-country framework, however, this might require manipulat-
ing a large number of microeconomic data sets and may be found
cumbersome.
Interestingly enough, a very simple approximation of (2) may be
obtained in the case where the distributions may be assumed to be
lognormal, probably the most standard approximation of empirical
distributions in the applied literature. The relative income distribu-
tion writes in that case:

~ logX 1
Ft X P s ;
s 2
where P is the cumulative distribution function of the standard

normal and s is the standard deviation of the logarithm of income.
Substituting this expression in (2) shows that the change in poverty
headcount between time t and t 0 depends on the level of mean
income at these two dates, y=z, expressed as a proportion of the
poverty line, and on the standard deviation, s, of the logarithm of
income at the two dates. Allowing t 0 to be close to t and taking limits
as in (3) then leads to

DH logz=yt 1 D log y 1 logz=yt
l s Ds ; 2 0
Ht s 2 s 2 s2
where l stands for the ratio of the density to the cumulative

functionor hazard rateof the standard normal, D logy is the
growth rate of the economy and Ds is the variation in the standard
deviation of the logarithm of income. Based on that expression and
following (3), the growth elasticity of poverty, e, may be dened as
the relative change in the poverty headcount for 1 percent growth in
mean income, for constant relative inequality, s:

DH 1 logz=yt 1
e l s : 3 0
D logyHt s s 2
In that expression, the growth elasticity of poverty appears ex-

plicitly as an increasing function of the level of development, as
measured by the inverse of the ratio z=yt , and a decreasing function
of the degree of relative income inequality as measured by the stan-
dard deviation of the logarithm of income, s.5
The preceding relationship is represented in gure 1.3 by curves in
the development inequality space along which the growth elasticity
of poverty is constant. The inverse of the development level is mea-
sured along the horizontal axis by the ratio of the poverty line to the
mean income of the population. Inequality is measured along the
vertical axis by the Gini coefcient of the distribution of relative
income rather than the standard deviation of logarithm. This mea-
sure of inequality is more familiar than s but is known to be an
increasing function of it.6
Figure 1.3 is useful to get some direct and quick estimate of the
growth elasticity of poverty. Consider, for instance, the case of a
poor country where the mean income is only twice the poverty line
at the right end of the gure. Reading the gure, it may be seen that
the growth elasticity is around 3 if inequality is lownamely, a Gini
coefcient around 0.3but it is only 2 if the Gini coefcient is
around the more common value of 0.4. If the economy gets richer,
then the elasticity increases. But at the same time it becomes more
sensitive to the level of inequality. For instance, when the mean
income of the population is four times the poverty line, the growth
elasticity of poverty is 5 for low-income inequalitynamely, a Gini
equal to 0.3but 2 again if inequality is highnamely, a Gini coef-
cient equal to 0.5. These various combinations are illustrated by the
example of a few countries in the mid-1980s. The ratio of the poverty
line to the mean income is taken to be the $1-a-day line related to
Figure 1.3
Poverty (headcount)/growth elasticity as a function of mean income and income inequality, under the assumption of constant (lognormal)
distribution
13
GDP per capita, whereas Gini coefcients are taken from the Dein-
inger and Squire (1996) database. Growth elasticities of poverty
reduction consistent with the lognormal assumption vary between 5
for a country like Indonesia to 3 for India and the Cote dIvoire,
which were poorer and/or had a higher level of inequality. The
elasticity is around 2 for Brazil, despite the fact that it is considerably
richer than the other countries. Income inequality makes the differ-
ence. Finally, the elasticity is much below 2 in the case of Senegal
and Zambia, both of which are poor and unequal.
Figure 1.3 and the underlying lognormal approximation to the
growth elasticity of poverty in (5) refers to the headcount as the
poverty index. Similar expressions and curves may be obtained
using alternative indices. For instance, the poverty gap is a measure
of poverty obtained by multiplying the headcount by the average
relative distance at which the poor are from the poverty line. The
advantage of that measure over the headcount is obviously that it
takes into account not only the proportion of people being poor but
also the intensity of poverty. Under the lognormal assumption, it
may be shown that the growth elasticity of the poverty gap is given
by the following formula:7
Plogz=yt =s s=2
eyPG ; 4
z=yt Plogz=yt =s s=2 Plogz=yt =s s=2
whereas the elasticity with respect to the standard deviation of the

logarithm of income is given by
s plogz=yt =s s=2
esPG ;
z=yt Plogz=yt =s s=2 Plogz=yt =s s=2
where P and p are respectively the cumulative distribution and

the density functions of the standard normal variable. As before,
the growth elasticity depends on the level of development as mea-
sured by the ratio z=yt and the inequality of the distribution of
income as given by the standard deviation of the logarithm of in-
come. Although not shown here, the iso elasticity contours in the de-
velopment/inequality space corresponding to function (4) are very
similar to what was shown in gure 1.3 for the headcount index.
Equivalent formulas and similar gures could be derived for other
poverty measures, in particular those belonging to the well-known
Pa familysee Foster, Greer, and Thorbecke (1984).8
1.3 Revisiting the Empirical Evidence on the Growth-Poverty

Relationship
The identity linking growth, poverty reduction, and distributional

changes should be readily apparent from data on growth-poverty
spells. This section shows that this is indeed the case, provided the
adequate specication is used. In particular the lognormal approxi-
mation proves to be extremely precise.
The data being used are the same as those used for gure 1.1. The
sample includes mostly developing countries but a few transition
economies are also present. The only modication made to the orig-
inal data was to eliminate all spells where the percentage change in
the poverty headcount was abnormally large in relative value. Cases
left aside essentially correspond to situations where the poverty
headcount went from zero or an almost negligible gure to some
positive value, or the opposite, in a few years. This was made in
order to comply with the denition of the elasticity concepts given
earlier and the corresponding requirement of small changes.9 The
mean annual growth in mean income over all these spells is 2.7 per-
cent, the mean relative change in the poverty headcountwith a
$1-a-day poverty lineis close to zero and the mean change in the
Gini coefcient is .0022.
Based on this data set, four different models are compared against
each other. The rst corresponds to the naive view alluded to above
that there is a constant elasticity between poverty reduction and
growth. It consists of regressing observed changes in the poverty
headcount on observed changes in mean income. The second model,
which is termed the standard model in table 1.1 includes the
observed change in income inequality, as measured by the Gini
coefcient as an additional explanatory variable. It is thus consistent
with a decomposition of type (2) above, except for the fact that both
the growth elasticity and the Gini elasticity of poverty reduction are
taken to be constant. The third model improves on the previous one
by allowing the growth elasticity to depend on the inverse level of
development, as measured by the poverty-line/mean-income ratio,
and on the initial degree of inequality as measured by the Gini co-
efcient. Nothing is imposed a priori on that relationship, though.
Finding the optimal functional specication is left to econometrics.
To do so, two additional variables are introduced in the regression:
the interaction between growth and the preceding two variables. The
nal model relies on the lognormal approximation discussed earlier.

The explanatory variables are the theoretical elasticity dened in (3 0 )
times the observed growth rate of mean income10 and the change in
the Gini coefcient. If the lognormal approximation discussed earlier
is not too unsatisfactory, then one should nd that the coefcient of
the theoretical elasticity in that regression is not signicantly differ-
ent from unity.
The estimation of these four models is reported in the rst four
columns of table 1.1. Results fully conrm the identity relationship
discussed in this chapter. First, one can see that the naive model
suggests a signicant negative elasticity of poverty with respect to
growth but its explanatory power is low. This strictly corresponds
to the tted line in gure 1.1 with a R 2 equal to 26 percent. Things
improve quite substantially when one shifts to the standard model
by adding distributional changes in the regression equation. Adding
the change in the Gini coefcient in the linear specication practi-
cally doubles the R 2 coefcient, suggesting that the heterogeneity in
distributional changes is as much responsible for variation in pov-
erty reduction across growth spells as the heterogeneity in growth
rates itself.
Interacting growth with the initial poverty-line/mean-income
ratio and the initial Gini coefcient in the improved standard model
1 yields still another signicant improvement in explanatory power.
The two interaction terms are very signicant and go in the right
direction. As expected, both a lesser level of development and a
higher level of inequality reduce the growth elasticity of poverty.
Both effects are signicant and sizable. At the mean point of the
samplenamely, for a growth spell leading to an annual 2.7 percent
rise in mean incomean increase of the initial level of development
by one standard deviation of the poverty-line/mean-income variable
increases poverty reduction by some 3 percentage points annually.
In the same conditions, an increase in the initial Gini coefcient by
one standard deviation diminishes poverty reduction by a little less
than 1 percentage point.
No assumption is made in the preceding regression on the way
income growth, the development level and the initial degree of
inequality interact to determine poverty reduction. In the fourth col-
umn of table 1.1, on the contrary, it is assumed that the joint effect of
these three variables is in accordance with the theoretical elasticity
Table 1.1
Explaining the evolution of poverty across growth spellsa
(dependent variable percentage change in poverty headcount during growth spell)
Identity Identity
Improved check: Improved check:
Naive Standard standard Lognormal standard Lognormal
model model model 1 model 1 model 2 model 2
Explanatory variables (1) (2) (3) (4) (5) (6)
Intercept 0.0826 0.0972 0.0837 0.0752 0.0977 0.0692
0.0434 0.0364 0.0349 0.0325 0.0321 0.0281
Y percentage change in mean income 1.6493 2.0124 6.3518 7.8706
0.2585 0.2223 1.2451 1.1310
DGini Variation in Gini coefcient 4.7178 5.2863 5.0769 21.5606
0.6731 0.6529 0.6118 4.1205
Y * poverty-line/mean-income 3.9678 3.9481
1.1662 1.0286
Y * initial Gini coefcient 7.0039 9.6869
2.4586 2.2104
DGini * poverty-line/mean-income 16.3898
2.8255
DGini * initial Gini coefcient 20.3600
7.4388
Y * theoretical value of growth elasticity under 0.8727 0.9261
lognormal assumption 0.0778 0.0679
DSigma * theoretical value of poverty inequality 0.6824
elasticity under lognormal assumptionb 0.0601
R2 0.2666 0.4916 0.555 0.5857 0.6651 0.6892
a Ordinary least squares estimates, standard errors in italics. The sample includes the 114 growth spells listed in the appendix. All coefcients
17
are signicantly different from zero at the 1 percent probability level except the intercept.
b Switching from the coefcient of Gini to the standard deviation of the logarithm of income is done through the formula in note 6.
derived in the preceding section under the assumption that the

underlying distribution of relative income is lognormal. The result-
ing explanatory variable in the poverty reduction regression thus
is this theoretical elasticity times the observed growth of the mean
income. This test of the identity that links poverty reduction and
growth is very successful. On the one hand, the R 2 coefcient proves
to be substantially higher than when the various explanatory vari-
ables are entered without functional restriction as in the preceding
regression, despite less degrees of freedom being used. On the other
hand, the coefcient of the theoretical value of the elasticity proves
to be only slightly below unity. Overall, it must then be concluded
that the best single explanation of observed poverty reduction in a
sample of growth spells is indeed provided by the identity that logi-
cally links poverty and growth under the lognormal assumption.
In all the preceding regressions, no care is taken of the fact that,
according to the identity discussed above, the role of the change in
inequality on poverty reduction is unlikely to be linear. The last two
columns of table 1.1 correct the preceding results for this. In column
(5) the effect of the distributional change on poverty reduction is
assumed to depend both on the initial level of development and
the initial level of inequality. In column (6), the two explanatory
variables are the growth and inequality elasticities, as dened in (2 0 )
under the lognormal assumption, multiplied respectively by the rel-
ative change in mean income and the change in the Gini coefcient.
The improved standard model 2 in column (5) proves to do much
better than the improved standard model 1 in column (3) where dis-
tributional changes were only captured by the change in the Gini
coefcient. The R 2 coefcient gains more than 10 percentage points,
solely because of the interaction terms between the change in the
Gini coefcient and the two key level variables, the initial inequality
level and the initial development level. Surprisingly enough, still a
better score is achieved in column (6) with the lognormal approxi-
mation with only two explanatory variables that stand for the
growth and inequality elasticity of poverty. It may be seen, more-
over, that the coefcient of the growth-elasticity term is not sig-
nicantly different from unity, something that would seem to be in
favor of the lognormal approximation. Unfortunately, this is not true
of the inequality-elasticity variable, which in effect would lead to
reject that approximation. Note, however, that an additional hy-
pothesis that cannot be tested independently of the lognormality is

the innitesimal approximation behind all types of elasticity calcu-
lation. As may be seen in gure 1.1, some observations are such that
the annual absolute relative change in the mean income of the pop-
ulation is greater then 25 percent. It is not clear that the standard
elasticity calculation would apply in this case. Yet, no attempt has
been made in this chapter to use nondifferential expressions.
That the lognormal approximation plays an effective role in the R 2
of the last column of table 1.1 being still far from unity is conrmed
when the preceding exercise is repeated with the poverty gap, rather
than the poverty headcount. Table 1.2 is the equivalent of table 1.1
with the poverty index being the poverty gap rather than the pov-
erty headcount. The striking feature there is of course that all R 2 are
much lower than what was found with the poverty headcount, and
also that the improvement in the t of poverty reduction observa-
tions due to the approximations discussed in this chapter is less
dramatic.
This last set of results may be somewhat disappointing but they
are easily understandable. Using the headcount leads somehow to
make predictions or simulations that concerns the value of the
cumulative distribution function of income, F~t X, at a single point,
namely, X z=yt . Using the poverty gap, on the contrary, requires
using predictions on the mean value of income for all people below
z=yt , that is about the full range of values of Ft X below the value X.
This is much more demanding. The lognormal approximation may
not perform too badly at a specic point close to the poverty line. It
may do much worse if all incomes below that value are to be taken
into account too.
This experiment is interesting because it suggests that, if one
wants to go beyond the poverty headcount in poverty measurement,
then functional approximations to growth and distribution elas-
ticities of poverty reduction may simply be unsatisfactory. Dealing
with the issue of the determinants of poverty reduction will then
require working with the full distribution of income or living stan-
dards rather than a few summary measures. This will probably
prove to be the only satisfactory solution in the long run and the
sooner poverty specialists will get used to dealing systematically
with distribution data, rather than inequality or poverty summary
measures, at the national level, the better it will be.
Table 1.2
Explaining the evolution of poverty across growth spells (poverty gap)a
20
(dependent variable percentage change in poverty gap during growth spell)

Identity Identity
Improved check: Improved check:
Naive Standard standard Lognormal standard Lognormal
model model model 1 model 1 model 2 model 2
Explanatory variables (1) (2) (3) (4) (5) (6)
Intercept 0.1434 0.1660 0.1683 0.1412 0.1823 0.1272
0.0857 0.0777 0.0796 0.0787 0.0812 0.0759
Y percentage change in mean income 1.2388 1.8003 0.4101 2.3059
0.5101 0.4747 2.8388 2.8658
Variation in Gini coefcient 7.2962 7.1231 7.1781 27.8647
1.4375 1.4885 1.4806 10.4411
Y * poverty-line/mean-income 0.9647 0.9486
2.6588 2.6065
Y * initial Gini coefcient 2.4774 0.8514
5.6055 5.6010
DGini * poverty-line/mean-income 20.0988
7.1598
DGini * initial Gini coefcient 26.8577
18.8496
Y * theoretical value of growth elasticity under 0.4961 0.5750
lognormal assumption 0.1661 0.1627
DSigma * theoretical value of poverty inequality 0.7035
elasticity under lognormal assumptionb 0.1218
R2 0.05 0.229 0.2308 0.1938 0.2849 0.249
a Ordinary least squares estimates, standard errors in italics. The sample includes the 114 growth spells listed in the appendix. All coefcients
are signicantly different from zero at the 1 percent probability level except the intercept.
b Switching from the coefcient of Gini to the standard deviation of the logarithm of income is done through the formula in note 6.
1.4 Some Implications
This being said, the identity that links poverty reduction, mean in-
come growth, and distributional change has several implications for
policymaking and economic analysis in the eld of poverty that are
worth stressing.
On the policy side, it was shown in this chapter that this identity
permits identifying precisely the potential contribution of growth
and distributional change to poverty reduction. However, it also
introduces a point that is often overlooked in the debate of growth
vs. redistribution as poverty reduction strategiesan exception be-
ing Ravallion (1997). To the extent that growth is sustainable in the
long run whereas there is a natural limit to redistribution, it may
reasonably be argued that an effective long-run policy of poverty
reduction should rely primarily on sustained growth. According to
the basic identity analyzed in this chapter, however, income redis-
tribution plays essentially two roles in poverty reduction. A per-
manent redistribution of income reduces poverty instantaneously
through what was identied as the distribution effect. But, in
addition it also contributes to a permanent increase in the elasticity
of poverty reduction with respect to growth and therefore to an
acceleration of poverty reduction for a given rate of economic growth.
This is quite independent of the phenomenon emphasized in the
recent growth-inequality literature according to which growth
would tend to be faster in a less inegalitarian environment.11 If this
were true, there would then be a kind of double dividend asso-
ciated with redistribution policy since it would at the same time
accelerate growth and accelerate the speed at which growth spills
over onto poverty reduction.
From the point of view of economic analysis, the basic argument
in this chapter has clear implications for the understanding of pov-
erty reduction. The common practice of trying to explain the evolu-
tion of some poverty measure over time, or across various countries,
as a function of a host of variables including economic growth has
something tautological. The preceding section has shown that the
actual growth elasticity of poverty reduction in a given country
could be estimated with considerable precision, even under the log-
normal approximation. Running a regression like the standard
model above where growth appears among the regressors would
thus make sense only in the case where one does not observe the
actual determinants of the theoretical value of the growth-poverty

elasticity, that is the level of development and inequality. This would
seem very unlikely, though, and it must be admitted that the way
poverty reduction depends on growth is in effect perfectly known.
Under these conditions, the only thing that remains to be explained
in the basic poverty reduction decomposition formula is the pure
distributional change effect. Somehow, all the variables that may
be added in the regression after the growth effect has been rigor-
ously taken into account should track the change in the distribution
and its effects on poverty. In other words, the only thing that
poverty change regressions should try to do is really to identify the
causes of distributional changes and their effects on poverty indices.
This last remark relates to an earlier point about the nature of
poverty being analyzed. In an international context, it seems natural
that cross-country comparisons of poverty reduction bears on an
absolute concept of povertynamely, $1 or $2 a day. However,
most of the argument in this chapter can be reinterpreted as saying
that changes in absolute poverty may be decomposed into changes
in the mean income of the population and changes in relative
poverty, as measured for instance by the number of people below
some xed proportion of the mean or median income, or simply the
poorest x percent of the population. Viewed in this way, the argu-
ment in the preceding paragraph basically says that, in understand-
ing the evolution of absolute poverty, the main object of analysis
should really be the evolution of relative poverty, as the effect of a
change in the mean income on absolute poverty is practically tauto-
logical. Although they do not formulate it in this way, this is what
Dollar and Kraay (2000) attempt to do by focusing on the mean
income or the income share of the bottom 20 percent of the popula-
tion. When they show that this income share does not seem to move
in any systematic direction with growth across countries, they in
effect validate the use of the identity relationship to compute the
effect of growth on absolute poverty reduction.
1.5 Appendix
The countries and growth spells used in the empirical part of this
chapter appear in the following table. Spells are dened by the initial
and terminal years.
Table 1.A.1
Countries and spells in database
Country Spell Country Spell
Algeria 8895 Honduras 9496
Bangladesh 8485 India 8386
Brazil 8588 India 8990
Chile 8790 India 9697
Chile 9092 Indonesia 8487
Chile 9294 Indonesia 8790
China 9293 Indonesia 9093
China 9596 Jamaica 8990
China 9697 Jamaica 9396
China 9798 Jordan 9297
Colombia 8891 Kazakhstan 9396
Colombia 9596 Kenya 9294
Costa Rica 8690 Kyrgyz Republic 9397
Costa Rica 9093 Lesotho 8693
Costa Rica 9396 Madagascar 8093
Cote DIvoire 8788 Malaysia 8487
Dominican Rep 8996 Malaysia 9295
Ecuador 8894 Mauritania 8893
Ecuador 9495 Mauritania 9395
Egypt 9195 Mexico 8492
El Salvador 8995 Mexico 8995
El Salvador 9596 Morocco 8590
Estonia 9395 Nepal 8595
Ethiopia 8195 Niger 9295
Ghana 8789 Nigeria 8592
Ghana 8992 Nigeria 9297
Guatemala 8789 Pakistan 8790
Honduras 8990 Pakistan 9093
Honduras 9294 Pakistan 9396
Table 1.A.1
(continued)
Country Spell Country Spell
Panama 8991 Thailand 88 (2)92
Panama 9195 Thailand 9296
Panama 9596 Thailand 9698
Panama 9697 Trinidad and Tobago 8892
Paraguay 9095 Tunisia 8590
Peru 9496 Turkey 8794
Philippines 8588 Uganda 8992
Philippines 8891 Ukraine 9596
Philippines 9194 Venezuela 8187
Philippines 9497 Venezuela 8789
Romania 9294 Venezuela 8993
Russia 9396 Venezuela 9395
Russia 9698 Venezuela 9596
Senegal 9194 Yemen? 9298
Sri Lanka 8590 Zambia 9193
Sri Lanka 9095 Zambia 9396
Thailand 8188 (1)
Notes
A preliminary version of this chapter was presented at the conference World Poverty:
A Challenge for the Private Sector, organized by the Amsterdam Institute for Interna-
tional Development, Amsterdam, October 34, 2000. The present version beneted
from comments made by participants to a seminar at CESIfo (Munich, May 2001) and
by participants to LACEA 2001 in Montevideo. I am particularly indebted to Daniele
Checchi, Gyl Zoega, and an anonymous referee. I am also indebted to Martin Rav-
allion for making the data set used in this chapter available to me. Of course, I remain
solely responsible for any remaining error. Views expressed in this chapter are those
of the author and should not be attributed to the World Bank or any afliated organi-
zation.
1. See World Bank (2000, 47).
2. For a full description of these data, see Chen and Ravallion (2000).
3. See also the presentation of that methodology in the short literature survey pro-
vided by Fields (2001).
4. Figure 1.2 provides a very handy general representation of distributional changes

and has been used in a number of circumstances. See, for instance, Quah (chapter 2).
5. These properties derive from the fact that l is known to be an increasing function
in the case of the standard normal distribution. Indeed it may be shown that lx is the
incomplete mean of the standard normal variable over the range a; x.
6. In the case of a lognormal distribution, both magnitudes are related by the follow-
ing relationship (see Aitchinson and Brown 1966): G 2Ps=2 1=2 1.
7. It is also known that eyPG PG H=PG, where PG is the poverty gap.
8. Instead of considering poverty measures, it would also be interesting to consider
aggregate measures of social welfare. From that point of view, the measure W
y 1 G, where y is the mean income and G is the Gini coefcient that was origi-
nally proposed by Sen, lends itself to the same simple decomposition into growth and
distribution effects as the poverty indices considered here.
9. Of course, it would have been possible to keep these observations if the original
decomposition formula (2) with the lognormal approximation, rather than (2 0 ), had
been used in the econometric analysis that follows.
10. Expression (3 0 ) relies on the standard deviation of the logarithm of income
whereas inequality is measured by the Gini coefcient in the data set. However, a one-
to-one relationship exists between these two magnitudes when the underlying distri-
bution is lognormal (see note 5). That relationship was used to derive s from the
observed value of the Gini.
11. On this see the survey by Aghion, Caroli, and Garca-Penalosa. (1999).
References
Aghion, P., E. Caroli, and C. Garca-Penalosa. 1999. Inequality and economic growth:
The perspective of new growth theories. Journal of Economic Literature 37(4): 16151660.
Aitchison, J., and J. Brown. 1966. The Log-Normal Distribution. Cambridge: Cambridge
University Press.
Chen, S., and M. Ravallion. 2000. How did the worlds poorest fare in the 1990s?
Mimeo., The World Bank.
Datt, G., and M. Ravallion. 1992. Growth and redistribution components of changes in
poverty measures: A decomposition with application to Brazil and India in the 1980s.
Journal of Development Economics 38(2): 275295.
de Janvry, A., and E. Sadoulet. 1995. Poverty alleviation, income redistribution and
growth during adjustment. In Coping with Austerity: Poverty and Inequality in Latin
America, ed. N. Lustig. Washington, DC: Brookings Institution.
de Janvry, A., and E. Sadoulet. 2000. Growth, poverty, and inequality in Latin Amer-
ica: A causal analysis, 197094. Review of Income and Wealth 46(3): 267287.
Deininger, K., and L. Squire. 1996. A new data set measuring income inequality. The
World Bank Economic Review 10: 565591.
Dollar, D., and A. Kraay. 2000. Growth is good for the poor. Mimeo., The World Bank.
Fields, G. 2001. Distribution and Development: A New Look at the Developing World.
Cambridge: The MIT Press.
Foster, J., J. Greer, and E. Thorbecke. 1984. A class of decomposable poverty measures,
Econometrica 52: 761766.
Kakwani, N. 1993. Poverty and economic growth with application to Cote dIvoire.
Review of Income and Wealth 39: 121139.
Ravallion, M. 1997. Can high-inequality developing countries escape absolute pov-
erty? Economic Letters 56: 5157.
Ravallion, M., and S. Chen. 1997. What can new survey data tell us about recent
changes in distribution and poverty? The World Bank Economic Review 11: 357382.
Ravallion M., and M. Huppi. 1991. Measuring changes in poverty: A methodological
case study of Indonesia during an adjustment period. The World Bank Economic Review
5: 5782.
World Bank. 2000. Attacking Poverty, World Development Report 2000/2001. Oxford:
Oxford University Press.
2 One Third of the Worlds
Growth and Inequality
Danny Quah
2.1 Introduction
Three concerns underlie all research on income inequality and eco-

nomic growth. First, inequality might be causal for growth, raising
or lowering an economys growth rate. Understanding the mecha-
nism then becomes paramount. How do alternative structures of
political economy and taxation matter for this relation between in-
equality and growth? Does income inequality increase the rate of
capital investment and therefore growth? Do credit and capital mar-
ket imperfections magnify potentially adverse impacts of inequality,
thereby worsening economic performance and growth? For con-
creteness, I refer to this circle of related questions as the mechanism
concern.
Second, even as economic growth occurs, the simultaneous rise
in inequalitysometimes hypothesized, other times asserted
might be so steep that the very poor suffer a decline in their incomes.
This is one of a set of beliefs underlying the anti-capitalism, anti-
globalization, anti-growth movement. Although, not exhaustively
descriptive, anti-globalization is the term I use to refer to this second
concern.
Third is an all-else category of analyses that fall outside the rst
two. This incorporates concerns such as envy, equity, risk, peer
group effects, or the economics of superstars (where the distribution
of outcomes turns out more skewed than that of the important
underlying characteristics). Thus, this category includes the more
traditional motivations in research on income distribution and in-
equality, but that have become less emphasized in recent research
that focus more on the mechanism and anti-globalization concerns.
28 Danny Quah
This chapter is part of a body of research that argues that the

mechanism and anti-globalization concerns are empirically unten-
able. It seeks to sharpen the general points made in Quah (2001b) by
concentrating on the worlds two most populous nation states, China
and Indiaonly two points in a cross-country analysis, but fully
one-third of the worlds population.
The remainder of this chapter is organized as follows. Section 2.2
describes related literature, and section 2.3 develops the class of
probability models underlying the approach in this chapter. Sections
2.4 and 2.5 present the empirical results. Section 2.6 concludes. The
technical appendix, section 2.7, contains details on the estimation
and data.
2.2 Related Literature
A conventional wisdom recently emerged from empirical research

on inequality and growth is how fragile empirical ndings are,
varying with auxiliary conditioning information, functional form
specication, assumed patterns of causality, and so on (e.g., Banerjee
and Duo 2000).
This state of affairs is unlike that at the origins of inequality and
growth as a distinct eld of study. Then, Kuznets (1955) had asked if
personal income inequality increased or declined in the course of
economic growth. He documented both: Looking across countries,
from poorest to richest, he noted that within-country income in-
equality rst rose and then fell.
Since most of the work there entailed dening and collecting data,
it was painstaking and laborious. By contrast, modern researchers
now using readily available observations on growth and inequality
can easily and routinely reexamine Kuznetss inverted U-shaped
curve (e.g., Deininger and Squire 1998). Interest therefore has shifted
to more subtle issues: causality and mechanisms relating inequality
and growthsee, for example, Aghion, Caroli, and Garca-Penalosa
(1999), Benabou (1996), Galor and Zeira (1993), and the literature
surveyed in Bertola (1999).
On these more complex questions, however, the data have given a
less clearcut message. Results have varied, depending on auxiliary
conditioning information and econometric technique. For instance,
Alesina and Rodrik (1994), Perotti (1996), and Persson and Tabellini
(1994) concluded that inequality and growth are negatively related,
One Third of the Worlds Growth and Inequality 29
while Barro (2000), Forbes (2000), and Li and Zou (1998) reported a
positive or varying relation. To some researchers, the situation has
seemed so bad that they have simply concluded the data are not
informative for interesting issues in inequality and growth, and have
attempted to explain why this is so, within a particular model of
inequality and growth (e.g., Banerjee and Duo 2000). In this view,
the data are noisy.
This chapter takes no explicit stance on causality between in-
equality and growth, nor on the functional form relating them.
Instead, it models inequality and growth jointly as part of a vector
stochastic process, and calibrates the impact each has on a range of
welfare indicators and on the individual income distributionsrst
within China and India, and then taking the two countries together.
The chapter addresses simpler questions than those treated in the
ambitious work attempting to trace out causality across growth and
inequality.
This chapter asks, When growth occurs, how do the poor fare?
What difference have the historical dynamics of inequality and
growth made for the incomes of one third of the worlds population?
If inequality were, indeed, to fall when growth is lower, does it fall
enough to overcome the negative impact on the poor of slower eco-
nomic growth overall? Alternatively, if within-country inequality
were to rise, does that occur simultaneously with Chinese and India
per capita incomes converging, so that overall individual income
inequality across these two economies is falling?
Given the data extant, arithmetic alone sufces to retrieve useful
answers to these questions. Here, the data are loud, not noisy: For
a universe comprising China and Indiaone-third of the world
for understanding the secular dynamics of personal incomes against
a setting of cross-country inequalities, those forces of rst-order
importance are macroeconomic ones determining national patterns
of growth and convergence. Rising average incomes dominate
everything else. Within-country inequality dynamics are insigni-
cant for determining inequality across people internationally.
Several earlier papers motivate my approach here. Deininger and
Squire (1998) addressed questions closely related to those Ive just
posed. They used regression analysis and more elaborate data, in
contrast to the minimalist, arithmetic approach of this chapter. They
concluded, though, much the same as I do below: The poor benet
more from increasing aggregate growth by a range of factors than
30 Danny Quah
from reducing inequality through redistribution. Deininger and

Squires view of growth and inequality as the joint outcome of some
underlying, unobserved development process matches that in sec-
tion 2.3.
Dollar and Kraay (2001) studied directly average incomes of the
poorest fth of the population across many different economies.
They noted those incomes rise proportionally with overall average
incomes, for a wide range of factors generating economic growth.
Put differently, it is difcult to nd anything raising average incomes
that doesnt also increase incomes for the very poor. They concluded,
as I will in what follows, that the poor benet from aggregate eco-
nomic growth, whatever is driving the latter. Similarly, Ravallion
and Chen (1997) found in survey data that changes in inequality are
orthogonal to changes in average living standards.
All these papers, in my interpretation, point to a consistent, quan-
titatively important characterization of the relation between growth
and inequality. The characterization is one naturally viewed in terms
of gure 2.2 below and rounded out by the arithmetic calculations in
this chapter and in Quah (2001b).
More recent papers are related as well. Bourguignon (chapter 1)
performed calculations like those in section 2.7.4.1. Sala-i-Martin
(2002) used within-country income shares as vector I (section 2.3
below) and therefore was able to specialize the calculations from
what I give in sections 2.3 and 2.7. Although his techniques and
emphases differ from those in this chapter and in Quah (2001b), our
motivations and conclusions are close and complement each others.
Finally, Heston and Summers (1999), Milanovic (2002), and Sala-i-
Martin (2002) constructed world income distributionsnamely,
across over a hundred countries, not just those for the two I use
herebut that similarly put together individual country statistics.
Their methods, approaches, and data sources differ from mine, but
the underlying ideas are the same.
Critics of this work have pointed out that income inequality sta-
tistics such as Gini coefcients, 90-10 ratios, mean-median income
ratio, log standard deviations, and so on were never intended for
examining the kinds of issues that I treat in what follows nor for
merging with the per capita income and population measures that I
analyze. Thus, for instance, more detailed investigations into the
very poor in any single country, made possible from surveys, eld
research, or other individual-level microeconomic data, could well
display tendencies different from those I derivesee, for example,
Atkinson and Brandolini (2001) or Dreze and Sen (1995). My calcu-

lations then, it is asserted, do no more than reveal the misleading
nature of many income inequality statistics.
Perhaps so. However, it is also exactly these same statistics with
which Benabou (1996) begins his powerful and inuential state-
ment on how inequality matters importantly in economic growth
merging inequality and aggregate statistics, comparing Korea and
the Philippines, in a similar spirit to what I do in what follows.
If measures like those I use mislead, then all such research is
awed (which I dont believe)not just those studies, like the cur-
rent one, that argue inequality is unimportant. The single set of
data that researchers all use has different dimensions to it, and we
cannot selectively ignore some and heed only othersthereby im-
posing biases based on whether certain conclusions seem a priori
sensible.
2.3 Probability Models for Income Distribution Dynamics
Fix a country at a point in time, and let Y denote income, I a vector

of income inequality measures, and F the distribution of Y across
individuals. One entry in I might be the Gini coefcient; another
might be the mean-median income ratio; yet a third might be the
standard deviation of log incomes; further entries might be (within-
country) income shares; and so on. Each element of I is a functional
or a statistic of the distribution F. To emphasize that per capita
income is the arithmetic mean or expectation of F, write it as E. Eco-
nomic growth is E_ =E.
Asking about causality between growth and inequality is asking
about the functions
E_ =E fI or I cE_ =E
(as in, e.g., gure 2.1). If that is the interest, econometric analysis can
trace out f and c.
By contrast, this chapter models E_ =E and I jointly, taking them to
be elements of equal standing in a vector stochastic process Z. Let Z0
denote the vector of other variables in the system, including popu-
lation P, so that
0 1
E_ =E
def B C
fZt : t b 0g; with Z @ I A
Z0
32 Danny Quah
Figure 2.1
Inequality and growth
constitutes the object to investigate. The current study can be viewed

as describing an unrestricted vector autoregression in Z; it makes no
assumptions on causality relations across the different entries of Z.
The law of motion describing the dynamics of the income distribu-
tion F implies a law of motion for the vector Z. Conversely, when Z0
is sufciently extensive, Zs dynamics imply Fs; when Z0 is not
complete, Zs dynamics restrict but do not fully specify Fs.
Figure 2.2 illustrates this. The right side shows the density f corre-
sponding to the distribution F, at two time points t0 and t1 , with
the dashed line indicating f at t0 , the earlier time, and the solid
line indicating that at time point t1 . Associated with f0 is its mean
E0 ; similarly, associated with f1 is its mean E1 . Figure 2.2 has, as
an illustration, E1 > E0 so that economic growth, as measured by
national income statistics, has occurred. If Y is some arbitrary but
xed income level, one can estimate the fraction of the population
that remains with income below Y by calculating YaY fY dY from
knowledge of f, from time t0 to time t1 . If one knows the population
P as well, then one can use this calculation to estimate the number of
people living
at incomes less than Y. In the inequality literature, the
statistic YaY fY dY is sometimes called the poverty headcount index,
and written HCY ; while the size of the population with incomes at
most Y is written PY (e.g., equations (14) and (15) in section 2.7).
The problem is one typically has only incomplete information on Z
and f. But the researcher can use knowledge on Z to infer restrictions
on f, and then estimate statistics of interest like HCY and PY .
Figure 2.2
Income distribution dynamics
To illustrate the idea, suppose F were assumed (or otherwise

inferred) to be Pareto, and known up to the two parameters y1 and
y2 :
Fy 1 y1 y1 y2 ; y1 > 0; y b y1 ; y2 > 1: 1
What restrictions does knowledge of Z imply for F? Equation (1)

gives per capita income E and Gini coefcient IG as
y
def
E y dFy y2 11 y2 y1 ;
y
y
def 1 1
1
IG 2 EF F y y dF y 2y2 11 ;
y 2
so that knowledge of the rst two entries of Z alone gives
y^2 1 IG1 =2;

y^1 1 y^1
2 E:
With more information in Z, the researcher can either estimate y

more precisely, using a method of moments technique as de-
scribed in section 2.7.1 below, or alternatively, relax the Pareto
assumption for F. In either case, HCY and PY can then be esti-
mated straighforwardly.
Putting together the implied Fs for different countries in the
world would allow mapping the worldwide income distribution
(Quah 2001b). To appreciate the value of doing this, consider gure
2.3, which shows what has sometimes been referred to as an emerg-
ing twin peaks in cross-country income distribution dynamics (e.g.,
34 Danny Quah
Figure 2.3
Emerging twin peaks
Quah 2001a). This twin-peaks characterization, as many others in the

macroeconomic growth literature, takes a country as a unit of obser-
vation. Thus, countries as large as China are treated the same way as
those as small as Singapore, and income distributions within coun-
tries are ignoredthe analysis takes everyone in the economy to
have the same (per capita, average) income.
Information on income distributions within a country allow
enriching the picture in gure 2.3 to something like gure 2.4. The
black dots at time t indicate the per capita incomes of two hypo-
thetical countries, with the darker shaded area around each depict-
ing within-country individual income distributions. Thus, even as
national per capita incomes evolve according to an emergent twin-
peaks dynamic law, the distribution of incomes across people, with-
in and across countries, can evolve and overlap in intricate ways.
2.4 China and India
China and Indiaalthough only two countries out of over one hun-
dred in gure 2.3carry within them a third of the worlds popula-
tion. They thus provide substantial insight into the dynamics in
gure 2.4.
Table 2.1 records that between 1980 and 1992 Chinas per capita
income grew from US$972 to US$1,493, an annual growth rate of
Figure 2.4
Individual income distributions
Table 2.1
Aggregate income and population dynamics: China, India, and the United States
Per capita incomes (US$) Population (10 6 )
1980 1992 E_ =E 1980 1992
China 972 1493 3.58% 981 1162

India 882 1282 3.12% 687 884
United States 15295 17945 1.33% 228 255
3.58 percent. Over this period, India grew at a lower annual rate
of 3.12 percent, increasing its per capita income from US$882 to
US$1,282. (Per capita incomes are purchasing power parity adjusted
real GDP per capita in constant dollars at 1985 prices, series rgdpch
from Summers and Heston 1991.) For comparison, table 2.1 also
contains a row for the United States, showing its 1.33 percent an-
nual growth over 19801992, taking per U.S. capita income from
US$15,295 to US$17,945. The last two columns of table 2.1 contain
population gures, again from Summers and Heston (1991). By 1992,
China had grown to over 1.1 billion people, with India approaching
0.9 billion.
It has been remarked many times elsewhere that Chinas fast-
increasing per capita income came together with rises in inequality.
Table 2.2 shows Gini coefcients for the same three countries, China,
India, and the United States, over 19801992. Inequality in China, as
36 Danny Quah
Table 2.2
Inequality in China, India, and the United States, by Gini coefcient (from Deininger
and Squire 1996)
Gini coefcient IG
1980 1992 Min. (year)
China 0.32 0.38 0.26 (1984)
India 0.32 0.32 0.30 (1990)
US 0.35 0.38 0.35 (1982)
Table 2.3
Fraction of population and number of people with incomes less than US$2 per day.
The range of estimates spans the different distributional assumptions described in
section 2.7.4.
Y 2; HCY PY ; 10 6
1980 1992
China 0.370.54 (360530) 0.140.17 (158192)
India 0.480.62 (326426) 0.120.19 (110166)
measured by the Gini coefcient, increased from 0.32 to 0.38, while

that in India remained constant at 0.32. While the last column of
table 2.2 shows the increase in Chinas inequality is not monotone
and Indias inequality was not constant throughoutChina had its
low Gini coefcient of 0.26 over this period in 1984, while Indias
low was 0.30 in 1990it is tempting to conclude from looking cross-
sectionally at China and India that a fast-growing economy also has
its inequality rise rapidly, while a slower-growing economy can keep
inequality in check.
But what do tables 2.1 and 2.2 imply for, say, the number of peo-
ple in China and India living below a specic xed income level?
How rapidly were people exiting low-income states, given aggregate
growth and actual changes in measured inequality? If, counter-
factually, inequality had remained unchanged, how would aggre-
gate growth alone have changed conditions for the poor? Or, again
counterfactually, how much would inequality have had to increase,
for the poor not to have beneted at all from aggregate growth?
Tables 2.3 and 2.4 provide answers to these questions, obtained
using the calculations detailed below in section 2.7. First, from the
actual historical record in per capita income growth E_ =E, popula-
tion, and Gini coefcients, one can, with weak additional assump-
tions on the parametric form of density f, work out how the entire
distribution of individual income shifted between 1980 and 1992.
Table 2.3 shows how the situation for the very poor changed over
this time period. The fraction of the population living on less than
US$2 per day (US$730 annually) varied from 0.37 to 0.54 in China in
1980; this corresponded to between 360 million to 530 million peo-
ple.1 By 1992, the fraction of population in that income range had
fallen to 0.14 to 0.17, implying only between 158 million to 192
million people, given the population size then. In other words,
over 19801992 China reduced the population in this very poor
income range by between 210 million to 338 million people, even as
inequality and total population rose.
The situation for India is less surprising, as measured inequality
there remained constant, and only aggregate economic growth
occurred.2 But since the total Indian population also rose, it might
well have been that the poor did increase in number. Table 2.3 shows
that did not happen. Between 1980 and 1992, the number of Indians
living on less than US$2 a day fell from 326426 million to less than
half that, 110166 million. The fraction of the Indian population in
this income range fell from approximately half to perhaps one-fth,
likely less. India reduced the population living in the very poor
income range by about a quarter of a billion, a number comparable
to the change in China.
If the world comprised only China and India put together, it
would show a number of interesting features. First, the country
that grew faster on aggregate also had inequality rise morethe
upward-sloping schedule in gure 2.1 is that that is relevant. Sec-
ond, however, even despite this positive relation between growth
and inequality, overall the worlds poor beneted dramatically from
economic growth. Over the course of little more than a decade, about
half a billion peopleout of a total population across the two coun-
tries of about 1.6 to 1.9 billionexited the state of extreme poverty.
This decline in sheer numbers of the very poor divided about equally
between China and India.
Table 2.4 takes the argument further. Suppose population were
held constant at 1980 levels in China and India. The left panel in the
table shows that if inequality too were held constant at its 1980
levels, then aggregate growth alone would have removed from being
38 Danny Quah
Table 2.4
From 1980 perspective: Given aggregate growth, reduction in numbers of poor if in-
equality unchanged, and proportional inequality increase per year to maintain poverty
numbers
IG , P constant: P_ Y HCY constant: I_ G =IG

China 33 million/year 8.3%/year
India 17 million/year 8.8%/year
very poor 33 million people a year in China, and 17 million people a

year in India. The right panel shows that to keep constant the num-
ber of people living on incomes less than US$2 a day as aggregate
growth proceeded, the Gini coefcient would have had to rise at
a proportional growth rate of 8.3 percent per year in China, and
8.8 percent per year in India. Such rapid and large increases in
inequality are unprecedented in world history.3 Chinas increase in
inequality would have had to more than double, and be kept at that
rate for a dozen years, to nullify the benecial effects of its high
aggregate growth rate.
I conclude from the discussion here that, given the historical
experience in China and India, aggregate economic growth might
well come about only with increases in inequality. However, given
magnitudes that are historically reasonable, growth is unambigu-
ously benecialespecially for the poor in general, and even for the
poor in particular when inequality rises.
2.5 Extensions
This section expands on the discussion in section 2.4 above. It pro-

vides further quantication and illustration to the conclusions there.
Deininger and Squire (1996), Li, Squire, and Zou (1998), and
Quah (2001b) have presented calculations formalizing how most of
the variation in measured inequality is across countries. Inequality
changes hardly at all in time. It is not that inequalityfor physical
reasons or otherwisecannot vary much; it is that the workings of
economies lead to inequality hardly changing through time. Hon-
durass 1968 income inequality of 62 percent (Gini coefcient) is 2.4
times that of Belgiums 26 percent in 1985. But at the same time, over
the entire postwar era, income inequality in Belgium never rose
above 28 percent while Hondurass never fell below 50 percent.
Figure 2.5
Estimated absolute poverty reduction
This fact has implications for panel data analyses of inequality and
growth (Quah 2001b). Panel data econometric methods that condi-
tion out individual effectsalmost all doend up removing all the
important variation in inequality data. Moreover, since economic
growth has its principal variation in the orthogonal directionin
time rather than across countries (e.g., Easterly et al. 1993)panel
data methods shoehorn inequality and growth data into spuriously
better t with each other. In other words, econometric methods that
condition out individual effects represent here a methodologically
suspect attempt to remove statistical biases. That justication has
often been simply imported from other elds of economics, without
due attention to why applying such techniques to study inequality
and growth might be inappropriate.
From table 2.3 one concludes that China and India together
reduced the number of people living on less than US$2 a day by
between 36 million and 50 million each year. Figure 2.5 graphs this
same reduction from 1980 to 1992 in the number of people with
incomes less than US$2 per day, given the actual historical outcomes
in income inequality and economic growth. China and India alone
shifted 508 million people, more than 12 percent of the worlds pop-
ulation, about one-third of the worlds then-poor, out of poverty.
Figure 2.6 shows, for each economy, the amount of poverty re-
duction per year that would have occurred from aggregate growth
alone, had inequality and population size remained constant at their
40 Danny Quah
Figure 2.6
Growth alone
Figure 2.7
Inequality to nullify growth
1980 values. Obviously, the faster is economic growth, the faster

would PY fall. The gure emphasizes that to reduce poverty world-
wide it is in the very large economies like China and India where
high growth is needed.
Figure 2.7 shows, again for each economy, the proportional
growth rate of inequality that would be required to nullify the ben-
ets of growth, had population remained constant at its 1980 value.
The dark dots toward the bottom of the gure shows the actual
growth rate in inequality that occurred over the sample. (These
numbers would be lower than those in note 3, for the multiple/
single time-period reason described there.) Counterfactual increases

in inequality of the magnitude that would be needed to overcome
the poverty-reducing impact of economic growththe upper part
of the picture, in light dotsare far outside the range of historical
realization.
Quah (2001b) presents counterparts of gures 2.52.7 for the 100
or so countries for which data are available. The previous con-
clusions are reinforced; unsurprisingly, China and India dominate
the picture.
2.6 Conclusions
Much recent research on inequality and growth has taken one of

two possible approaches: The rst is explicit, and that is to see if
inequality causes growth. The second, typically left implicit, is to see
if, even as growth occurs, the poor might be disadvantaged anyway,
because inequality has risen so dramatically. This chapter has shown
that for China and Indiaonly two points in a cross-section but
one third of the worlds populationneither of these possibilities is
empirically tenable.
More traditional motivationsrisk, poverty, equityfor studying
inequality, however, remain. Indeed, they are reinforced as world-
wide inequality continues to evolve, driven by powerful economic
forces. But these motivations have little to do with the more recent
analyses of the relation between inequality and growth.
This chapter has applied a simple arithmetic approach to obtain
its ndings. It has asked, given historical patterns of growth and
inequality, how have income distributions evolved in China and in
India and across both countries? How have the poor fared in the
two countries as per capita incomes and rich-poor differences have
changed? Whether the growth and inequality data are unable to
speak clearly on questions of causality or whether they only imply
weak empirical relations, the data are unequivocal on the questions I
posed. The data are loud, not noisy.
The principal nding of the chapter is twofold: First, only under
inconceivably high increases in inequality would economic growth
not benet the poor. The magnitudes of improvement in living stan-
dards due to aggregate economic growth simply overwhelm any
putative deterioration due to increases in inequality. Second, any
mechanism where inequality causes economic growth, positively or
42 Danny Quah
negatively, is empirically irrelevant for determining outcomes for

individual income distributions. I have obtained these results for
China and India in this chapter. A complete studyrounding out
the remaining two-thirds left over from the title of this chapteris in
Quah (2001b).
This chapter has taken care to assume no single view on the
causal mechanisms relating inequality and growth. It has pointed
out that whatever it is that drives economic growth in the large,
those forcesbe they macroeconomic, technological, political, or in-
stitutionalare dramatically important for improving the lot of the
poor when they lead to economic growth. And similarly so in the
opposite direction when they lead to economic stagnation.
What might seem an appealing possibility to raise here is that
income inequality could, positively or negatively, truly cause aggre-
gate economic growth, so that this chapters principal nding would
then only reinforce the importance of distributional and inequality
concerns over macroeconomic growth. However, even in reduced-
form regressions of growth on inequality, the R 2 t can never be
very high: The directions of principal variation in the two variables
are just too different (Quah 2001b). Therefore, even in the best of
circumstances, even with no ambiguity on the direction of causality,
many other factors beyond inequality inuence economic growth.
And all of them, through their impact on the aggregate income level,
affect the poorindependent of inequalitys effect on economic
growth.
Finally, it is not a telling criticism of the work in this chapter to say
that because it uses Gini coefcients or other standard inequality
measures, it does not get at the true nature of inequality, whatever
that might mean. Because the chapters methods characterize the
entire income distribution, the focus on specic inequality measures
is only for practical convenience, not conceptual necessity. However,
precisely the same measures are used in all other studies of growth
and inequality I knowin particular, in the many regression studies
that claim to show one causal relation or another between these two
quantities. If the data used are inappropriate here, then they are
similarly so there too. Indeed, one might view the calculations here
as simply taking a logical step prior to other work in its drawing out
an interpretation to the indexes used in studies of inequality and
growth.
2.7 Technical Appendix
If data existed on individual incomes accruing to different economic

agents, at each point in time, then the empirical analysis would be
straightforward. One can directly estimate the entire income distri-
bution across agents on the planet, and characterize its dynamics
through time. The problem, however, is that such data are unavail-
able and are unlikely to be produced anytime soon.
I develop here an alternative empirical framework that is general,
exible, and convenient. The approach is designed to be capable of
incorporating a wide range of alternative distributional hypotheses
and a variety of measurements on different characteristics of income
inequality. Thus, the empirical analysis is intended to apply readily
as more and better data on income inequality characteristics become
available.
I seek to uncover characteristics of the global distribution of
income across individuals. One knows characteristics of income dis-
tributions within countries, over time for a number of countries. A
traditional approach then to analyzing inequalities across progres-
sively larger subsets of individual incomesproceeding from yet
ner subgroupsis to ask if the inequality index aggregates (e.g.,
Milanovic 2002). The approach I take here differs. It begins with
noting that if one had the actual distribution Fj; t for economy j at
time t, where the population size is Pj; t , then the worldwide income
distribution FW; t , in a world of economies j 1; 2; . . . ; N, is
X
N
FW; t y P1
W; t Fj; t y Pj; t ; y 2 0; y 2
j1
with the world population
X
N
PW; t Pj; t :
j1
Differentiating (2) with respect to y gives the implied density for the
worldwide distribution of income as the weighted average of indi-
vidual country income distribution densities:
X
N
f W; t y fj; t y Pj; t =PW; t ; y 2 0; y: 3
j1
44 Danny Quah
Knowing the distribution FW means one can calculate directly all the
inequality indexes one wisheswhether or not particular indexes
aggregate becomes irrelevant.
2.7.1 Estimating Individual Income Distributions
Given the quantities on the right of equation (3) the worldwide

income distribution is straightforward to calculate. However, the
individual distributions Fj; t are, generally, unknown. Instead, typi-
cally, the researcher has data on a number of diverse functionals of
themfor example, Gini coefcients, quintile shares, averages, and
so on. This section describes obtaining an estimate for Fj from data
on such functionals.
Since the remainder of this section concentrates on what happens
with a single economy, the j subscript is taken as understood and
deleted to ease notation.
Fix an economy j. Suppose in each period t, one observes realiza-
tions on Pt ; Xt , where P is the population size and Xt A R d is a d-
dimensional vector of functionals of the underlying unobservable
income distribution Ft and population Pt . For example, when the rst
entry of Xt is the average or per capita income, then
y y
X1; t y dFt y y dFt y:
y 0
Let R; R denote the pair comprised of the real line R together

with the collection R of its Borel sets. Let BR; R denote the Banach
space of bounded nitely additive set functions on the measurable
space R; R endowed with total variation norm
X
Ej in BR; R: jjj sup jjAk j;
k
where the supremum in this denition is taken over all

fAk : j 1; 2; . . . ; ng
nite measurable partitions of R.

Distributions on R can be identied with probability measures
on R; R. Those are, in turn, just countably additive elements in
BR; R assigning value 1 to the entire space R. Let B denote the
Borel sigma-algebra generated by the open subsets (relative to total
variation norm topology) of BR; R. Then B; B is another mea-

surable space.
Write the vector of potentially observable functionals as a col-
lection
Tl : B R; B R ! R; R; l 1; 2; . . . ; d
(where B R denotes the sigma-algebra generated by the Cartesian

product of B and R). Thus, for distribution Ft associated with prob-
ability measure jt A B; B,
Xl; t Tl jt ; Pt ; l 1; 2; . . . ; d: 4
Without loss or ambiguity, I also write Tl Ft ; Pt to denote the right
hand side of (4). Write T to denote the vector of observed func-
tionals, that is,
TFt ; Pt T1 Ft ; Pt ; T2 Ft ; Pt ; . . . ; Td Ft ; Pt 0 :
Assume, nally, that the distribution Ft is known up to a p-
dimensional vector yt A R p ,
def
Ft F j yt Fyt : 5
(In equation (5) the symbol F is used to mean a number of different
mathematical objects, but this will be without ambiguity, as the
context will always be revealing.)
Equation (5) restricts in two distinct ways. First, the functional
form Ft is assumed known. Second, time variation in the sequence of
distributions Ft is assumed mediated entirely through the nite-
dimensional parameter vector yt .
If for some yt , distribution Fyt is the true model, then
Tl Fyt ; Pt Xl; t ; l 1; 2; . . . ; d:
At xed t, dene the estimator y^t for yt as

def
y^t arg minTFy ; Pt Xt 0 WTFy ; Pt Xt ;
y A Rp
W d d positive denite: 6
Each different weighting matrix Wincluding, notably, the identity
matrixproduces a different estimator. Under standard regularity
conditions (as in GMM or related analogue estimation; e.g., Hansen
1982; Manski 1988), each W-associated estimator is consistent when
46 Danny Quah
Xt is itself replaced with a consistent estimator for the underlying

population quantity. Moreover, dening the minimand
QXt y TFy ; Pt Xt 0 WTFy ; Pt Xt ; 7
and denoting yt; 0 as the probability limit of (6), standard reasoning

using
!1
d 2
Q dQ
y^t yt; 0 ;
dy dy 0 yt; 0 dy yt; 0
allows a limit distribution theory for these estimators, provided the

quantities Xt have a characterizable distribution around their under-
lying population counterparts.
Using yt from the estimating equation (6) in (5) gives an estimator
for Ft in each economy j. Plugging the result for each j in turn into
(2)(3) gives an estimator for the worldwide distribution of income.
Tracking yj; t as they evolve through time then gives worldwide indi-
vidual income distribution dynamics.
Section 2.7.4 provides some explicit analytically worked out
examples of this procedure.
2.7.2 Alternative Functionals Tl
This section provides examples of some candidate functionals

Tl . When observations on them are availableas assumed in the
notation for section 2.7.1they are readily used in estimating and
characterizing the distributions Fj; t . Conversely, if they are not ob-
servable but an estimate of Fj; t is available, then estimates for Tl
can, instead, be induced.
For mean or per capita income, take
y
def
EF; P y dF y: 8
y
The Gini coefcient is standard in analysis of income inequality.

Associate with it the functional
y
def 1 1 1
IG F; P 2 EF F y y dF y 9
y 2
(see, e.g., Cowell 2000).

A different set of functionals standard in inequality analyses is the
set of cumulative quintile shares. To dene these, set for integer i from
1 to 4,
def
Y0:2i F; P supf y j Fy a 0:2ig; 10
yAR
Y0:2i F; P !
def
S0:2i F y dFy EF; P1 : 11
y
The rst of these, equation (10), denes the 20 i-th percentile

income level; the left-hand side is also known as the i-th quintile. The
pair (10)(11) generalizes to arbitrary percentile shares, but in prac-
tice the more general versions are rarely used (see, however, equa-
tions (12), (13), and (17)).
Concepts (9)(11) are those traditionally used in studies on in-
equality. Reliable observations on them are now widely available
across time and economies (Deininger and Squire 1996).
Recently, Milanovic (2002) has used household data to construct
within-decile average incomes across many different countries. These t
within the framework as follows. Dene
def
Y0:1i F; P supf y j Fy a 0:1ig; i 0; 1; . . . ; 9; 12
yAR
and let
Y0:1i
def
E0:1i F; P y dF y; i 1; . . . ; 9;
Y0:1i1
y 13
def
E1 F; P y dFy:
Y0:9
Similar to (10) above, equation (12) denes the 10 i-th percentile

income level, with the left-hand side also known as the i-th decile.
The analysis in Milanovic (2002) can thus be merged with that which
follows if one uses the decile averages E0:1i from (13) (or even the
deciles themselves Y0:1i in (12)) as candidate Tl s.
Yet other ways to extract or summarize information from F; P are
relevant when interest lies in poverty specically (e.g., Ravallion
1997; Ravallion and Chen 1997; World Bank 1990). Fix a low but
otherwise arbitrary level of income Y, and let
Y
def
HCY F; P FY dF y: 14
y
Equation (14) gives a poverty headcount index, namely, the fraction of

48 Danny Quah
population below a given income level Y. Record also the absolute

size of the population with those incomes:
def
PY F; P P FY: 15
Finally, dene
Y
def y y dF y
PGIY F; P : 16
Y
This is a poverty gap index, namely, a (normalized) average income

distance from a given income level Y.
When researchers are interested in whether a gap is emerging
between groups of high-income and low-income individuals, a con-
cept more useful than just inequality is polarization (e.g., Esteban
and Ray 1994; Quah 1993, 1997; Wolfson 1994). To obtain a func-
tional that captures such an effect, follow the notation of (10) and let
Y0:5 denote the median

def 1
Y0:5 F; P sup y j F y a ; 17
yAR 2
and then, using (8), (9), and (17), dene a polarization index
" Y0:5 #
def y y dF y 2
PZ F; P 1 IG E Y0:5 : 18
dFy Y0:5
y
The rst term in square brackets is the Gini-adjusted per capita

income; the second is the average level of incomes below the median
(this is a special case of a conditional expectation that appears again
in what follows). The greater this separation, the higher will be the
value taken by the polarization index in (18).
All the functionals so far consideredapart from PY in (15)vary
only with the distribution F, and not the size of the population P.
The next functional takes both into account; it describes a dynamic
property of the evolving distributions. From the headcount index
(14), one might be interested in the rate of ow of people past the
xed income level Y. This is
def d
FlY Fyt ; Pt Fy Y Pt
dt t

d dPt
Pt Fyt Y Fyt Y : 19
dt dt
Equation (19) shows interaction among a range of factors, including

in particular per capita income growth E_ =E and static, point-in-time
inequality IG . I use this simultaneous relationship in sections 2.7.4.1
and 2.7.4.2. Using different techniques, it is exactly this interaction
that Ravallion (1997) studies for developing countries, using house-
hold survey data with direct observations on FlY .
The previous examples should certainly not be viewed to be
exhaustive. I have given explicit Tl calculations only for those
functionals readily found in the empirical literature and for which
observations are available. As progressively more rened income
distribution data are constructed, the reasoning here is easily ex-
tended to take those into account.
2.7.3 Distribution F as Organizing Principle
As the discussion makes clear, my approach in this chapter is to

use the distribution dynamics in FW; t as the core concept around
which I organize all subsequent discussion. Equation (2) is the key
compositional relation from individual economies to the world. All
induced statisticsGini coefcients, poverty headcounts, poverty
gap indexes, polarization indexes, and so onderive from it. In this
exercise, it is not key whether those statistics retain compositional
integrity, or have an axiomatic justication, or satisfy other reason-
able criteria. They are not special in this analysis. I use them in
what follows because they are easily interpretable and are stan-
dard in discussions on income distributions, thus allowing to reduce
the dimensionality of (the information in) estimated distribution dy-
namics. As formulated here, when independently available, these
statistics can be used to augment the estimation (6); when not, they
can be straightforwardly derived from an estimate of Fj; t . Everything
centers on the distributions.
Admittedly, backing out estimates of individual-economy distri-
butions Fj; t as in equation (6)might be viewed as a contrived
problem. If a researcher had the original individual-level incomes
data, then Fj; t (and thus FW; t ) could be estimated directly by stan-
dard methods (e.g., Milanovic 2002; Silverman 1981). One should
never need to construct any of (9)(19), and go through (6), to
characterize the distribution Fj; t . It is because such individual-level
data are not readily availableinstead statistical agencies have cal-
culated and made available only different, aggregative statistics of
the underlying datathat one is led to estimation by (6).
50 Danny Quah
By the same token, one might wish to take care not to view y
as deep structural parameters in any sense of the term. Instead,
a useful perspective is to treat the ys as simply convenient ways
hyperparametersto keep control on the high-dimensional cal-
culations that would be otherwise involved in tracing through
distribution dynamics. The analysis in this chapter is obviously not
one that sets out to test a multivariate regression or simultane-
ous equations model. It studies historical tendencies, notto a
large degreethe effects of articial growth paths and inequality
dynamics.
Standard econometric analysis of (6)(7) allows consistency and
limit distribution results for the hyperparameters y. Measurement
errors in the data Xt , in sample, do not logically pose any difculties.
However, whether Xt can be guaranteed to converge to underlying
population quantities, and in a manner where the limiting distribu-
tion can be characterized falls outside the domain of analysis in this
chapter.
Finally, to state the obvious, this approach is one that makes
sense when the individual distributions Fj; t are comparable. If they
are not, then the whole enterprise of trying to study worldwide in-
equality is awed from the beginning, regardless of the approach
taken.
2.7.4 Induced Statistics and Parametric Examples
I now turn to some explicit parametric examples to provide intuition

for the remainder of the analysis. In describing the distribution dy-
namics, it is useful to establish some additional notation.
Suppose that in a given economy per capita income E increases at
a positive constant proportional growth rate:
E_ =E x > 0: 20
I wish to compare dynamically evolving income distributions
against a xed (feasible and low, but otherwise arbitrary) threshold
income level Y. One statistic I am concerned with in particular is the
rate of ow of people past Y, namely, equation (19). I am interested
in the value of (19) when inequality, as measured by the Gini coef-
cient IG say, is held constant. Alternatively, I am interested in nd-
ing how fast IG has to change to set (19) to zero.
Write Fy to denote a parametrized income distribution function,

and let fy be its associated density function
y
Fy y fy y~ d~
y; y A R:
y
Any given distribution also implies the conditional expectation

function

y dFy y
Ey Y j Y in set A A :
A dFy y
This is the expectation of a random variable Y, distributed Fy , con-

ditional on Y falling in set A of possible values.
I abuse notation by using subscripts such as Ny, Ly, or Py to
the functions F, f, and E, to denote specic functional formsin this
case the normal, the lognormal, and the Pareto type 1 distributions,
respectively. In the general case (with no explicit functional form
restriction), the subscript will be simply y.
To begin discussing explicitly parametrized distributions, record
that the normal distribution characterized by mean y1 and variance
y2 has density

1
fNy y 2py2 1=2 exp y y1 2 ; y2 > 0:
2y2
The standard normal sets y1 0 and y2 1 so that then

y
1=2 1 2
FN0; 1 y 2p exp y~ dy~:
y 2
2.7.4.1 Lognormal
The lognormal distribution is widely used in traditional studies of
personal income distributions. Its density is

1=2 1 1 2
fLy y 2py2 y exp log y y1 ; y2 > 0; y > 0:
2y2
For this distribution the T functionals in (8)(11) of section 2.7.2 are

EFLy exp y1 12y2 ;
1=2
p
IG FLy 2 FN0; 1 y2 = 2 1;
S0:2i FLy FLy1 y2 ; y2 Y0:2i FLy ;

52 Danny Quah
with
1=2
Y0:2i FLy expfF1
N0; 1 0:2i y2 y1 g:
An alternative expression for the cumulative quintile share is
!
log Y0:2i y1 y2
S0:2i FLy FN0; 1 1=2
y2
1=2
FN0; 1 F1
N0; 1 0:2i y2 :
If estimation (6) used only E and IG , and ignored information on

other elements of T (or if those observations were unavailable),
then an exact analytical formula for the estimator can be given as
follows:
2
y^2 F1
N0; 1 IG 1=2 2;
y^1 log E y^2 =2:

These can be used, in any case, as starting values in an iterative
solution to (6). Heston and Summers (1999) used these as the esti-
mates for their study.
Explicit formulas for some of the dynamics are then available:
E_ =E y_1 12y_2 ;
fN0; 1 y2 =2 1=2
I_G =IG y2 =2 1=2 y_2 =y2 ;
2FN0; 1 y2 =2 1=2 1
and
Y
d d
FLy Y fLy dy:
dt 0 dt
(The Pareto case that follows permits explicit calculation for all the
dynamics of interest, in particular, for all the numerical results in
section 2.4. Other distributional hypotheses will, as with the log-
normal, require at least some of the results calculated numerically as
closed-form expressions are intractable.)
When IG is held xed, y_2 is zero. Then
y_1 E_ =E x;
so that for any xed y,

d 1=2 1 1 2
fLy y 2py2 y exp log y y1
dt 2y2
y1 _
2 log y y1 y1

1=2 log y y1 _
y2 fLy y p y1 :
y2
But then,
Y
d 1=2 log y y1
FLy Y y2 p fLy y x
dt 0 y2

1=2 log Y y1
y2 EN0; 1 Z j Z a p
y2

log Y y1
FN0; 1 p x:
y2
With xed inequality at a constant IG , this expression says that the
ow of population past a given threshold level Y is proportional to
the aggregate growth rate x. The constant of proportionality, more-
over, is easily calculated from knowledge of y.
The value of I_G =IG that sets the ow dFLy Y=dt to zero can be
obtained only by numerical simulation.
2.7.4.2 Pareto (Type 1)

A different widely used parametrization for personal income dis-
tributions is the Pareto (type 1) distribution:
FPy y 1 y1 y1 y2 ; y1 > 0; y b y1 ; y2 > 1;

with density

0 if y a 0,
fPy y
y2 y1 y1 y2 y1 otherwise.
The implied T functionals in (8)(11) of section 2.7.2 then are

EFPy y2 11 y2 y1 ;
IG FPy 2y2 11 ;

Y0:2i FPy FPy1 ; y2 1 S0:2i ;
with
54 Danny Quah
S0:2i FPy 1 0:2i1=y2 y1 :
As with the previous lognormal (similarly having two parameters),

an exact formula for the estimator (6) is available when only E and
IG are observed:
y^2 1 IG1 =2;
y^1 1 y^21 E:
In this case the dynamics in y and E; IG can be easily seen to be
related by
y_1 y_2
E_ =E y2 11 ;
y1 y2
_
2y2 y2
I_G =IG :
2y2 1 y2
Moreover, direct calculation shows

" #
d d y1 y 2
FPy Y
dt dt Y
" _ #
y_1 y1 y2
1 FPy Yy2 log :
y1 Y y2
When inequality in the form of IG is held xed, one has
y_1 E_
x
y1 E
and
d
FPy Y 1 FPy Yy2 x:
dt
Alternatively, to x FPy Y instead, require
y_1 =y1 logy1 =Yy_2 =y2

or
1
y_2 =y2 logy1 =Y y2 11 x:
To achieve this, one needs

2y2 1
I_G =IG logy1 =Y y2 11 x: 21
2y2 1
Equation (21) shows, at a given aggregate growth rate x, the rate of

change in inequality required to hold xed the proportion of the
population below income Y. The increase in IG is proportional to x.
When Y is sufciently low, namely, when

y2
FPy Y < 1 exp
y2 1
(which happens to be the case of interest), the constant of propor-

tionality is necessarily positive.
For the purposes of this chapter, the lognormal and Pareto cases
are interesting only because they permit explicit (closed-form) anal-
yses of the distribution dynamics of interest. They provide intu-
ition for how the general case will work. In the latter, typically only
numerical solutions are available.
2.7.5 Data
This chapter merges data from Deininger and Squire (1996),

Summers and Heston (1991), and UNU (2000). The updated and ex-
panded inequality data in Deininger and Squire (2002) are not, as of
this writing, yet distributed for general use.
Notes
I thank the MacArthur Foundation for nancial support. I have received many helpful
suggestions from colleagues, including Abhijit Banerjee, Tim Besley, Richard Blundell,
Andrew Chesher, Frank Cowell, Bill Easterly, Theo Eicher, Raquel Fernandez, Chico
Ferreira, James Feyrer, Oded Galor, Cecilia Garca-Penalosa, Louise Keely, and Branko
Milanovic. Also useful were Clarissa Yeaps research assistance in the early stages of
this work and Claudia Biancottis Ec473 LSE seminar presentation, in the spring of
2001. All calculations and graphs were produced using LATEX and the authors econo-
metrics shell tsrf.
1. Each entry in table 2.3 is a range rather than just a single number since alternative
distributional assumptions can be used in the calculation; see section 2.7.4.
2. See again, however, the discussion at the end of section 2.2 and the more detailed
picture available from studies such as Dreze and Sen (1995).
3. The only possible exceptions are the transition economies and Russia after the col-
lapse of the Soviet Union: see, for example, Ivaschenko (chapter 6) and Shorrocks and
Kolenikov (2001). The seven instances that Li, Squire, and Zou (1998) identied with
statistically and quantitatively signicant time trends in Gini coefcients only saw
proportional growth rates of 1.02 percent (Australia), 1.04 percent (Chile), 3.18 percent
(China), 1.71 percent (France), 1.18 percent (Italy), 1.61 percent (New Zealand), and
56 Danny Quah
1.46 percent (Poland) (this authors calculations, from Table 4 in Li, Squire, and Zou
(1998)) taken linearly over a single time periodmultiple time periods would imply
yet smaller growth rates.
References
Aghion, Philippe, Eve Caroli, and Cecilia Garca-Penalosa. 1999. Inequality and eco-
nomic growth: The perspective of the new growth theories. Journal of Economic Liter-
ature 37(4): 16151660, December.
Alesina, Alberto, and Dani Rodrik. 1994. Distributive politics and economic growth.
Quarterly Journal of Economics 109(2): 465490, May.
Atkinson, Anthony B., and Andrea Brandolini. 2001. Promise and pitfalls in the use
of secondary datasets: Income inequality in OECD countries as a case study. Journal
of Economic Literature 39(3): 771799, September.
Banerjee, Abhijit V., and Esther Duo. 2000. Inequality and growth: What can the
data say? Working Paper, Economics Department, MIT, May.
Barro, Robert J. 2000. Inequality and growth in a panel of countries. Journal of Eco-
nomic Growth 5(1): 532, March.
Benabou, Roland. 1996. Inequality and growth. In Macroeconomics Annual, vol. 11,
ed. Ben Bernanke and Julio Rotemberg, 1174. Cambridge: The MIT Press.
Bertola, Giuseppe. 1999. Macroeconomics of distribution and growth. In Handbook
of Income Distribution, vol. 1, ed. Anthony B. Atkinson and Francois Bourguignon,
477540. Amsterdam: North-Holland/Elsevier Science.
Cowell, Frank. 2000. Measurement of inequality. In Handbook of Income Distribution,
vol. 1, ed. Anthony B. Atkinson and Francois Bourguignon, 87166. Amsterdam:
North-Holland/Elsevier Science.
Deininger, Klaus, and Lyn Squire. 1996. A new data set measuring income inequal-
ity. World Bank Economic Review 10(3): 565591, September.
Deininger, Klaus, and Lyn Squire. 1998. New ways of looking at old issues: Inequal-
ity and growth. Journal of Development Economics 57: 259287.
Deininger, Klaus, and Lyn Squire. 2002. Revisiting inequality: New data, new
results. Working Paper, The World Bank, Washington, DC, March.
Dollar, David, and Art Kraay. 2001. Growth is good for the poor. Working paper,
DRG, The World Bank, Washington, DC, March.
Dreze, Jean, and Amartya Sen. 1995. India: Economic Development and Social Opportu-
nity. Oxford: Oxford University Press.
Easterly, William, Michael Kremer, Lant Pritchett, and Lawrence H. Summers. 1993.
Good policy or good luck? Country growth performance and temporary shocks.
Journal of Monetary Economics 32(3): 459483.
Esteban, Joan-Mara, and Debraj Ray. 1994. On the measurement of polarization.

Econometrica 62(4): 819851, July.
Forbes, Kristin J. 2000. A reassessment of the relationship between inequality and

growth. American Economic Review 90(5): 869887, December.
Galor, Oded, and Joseph Zeira. 1993. Income distribution and macroeconomics.
Review of Economic Studies 60(1): 3552, January.
Hansen, Lars Peter. 1982. Large sample properties of generalized method of moments
estimators. Econometrica 50(4): 10291054, July.
Heston, Alan, and Robert Summers. 1999. The world distribution of income: A
synthesis of intercountry and intracountry data to measure worldwide material well-
being. Working paper, University of Pennsylvania, Philadelphia, August.
Kuznets, Simon. 1955. Economic growth and income inequality. American Economic
Review 45(1): 128, March.
Li, Hongyi, and Hengfu Zou. 1998. Income inequality is not harmful for growth:
Theory and evidence. Review of Development Economics 2(3): 318334, October.
Li, Hongyi, Lyn Squire, and Hengfu Zou. 1998. Explaining international and inter-
temporal variations in income inequality. Economic Journal 108, 118, January.
Manski, Charles F. 1988. Analog Estimation Methods in Econometrics. London: Chapman
and Hall.
Milanovic, Branko. 2002. True world income distribution, 1988 and 1993: First
calculation based on household surveys alone. Economic Journal 112(476): 5192,
January.
Perotti, Roberto. 1996. Growth, income distribution, and democracy: What the data
say. Journal of Economic Growth 1(2): 149187, June.
Persson, Torsten, and Guido Tabellini. 1994. Is inequality harmful for growth?
American Economic Review 84(3): 600621, June.
Quah, Danny. 1993. Empirical cross-section dynamics in economic growth. European

Economic Review 37(2/3): 426434, April.
Quah, Danny. 1997. Empirics for growth and distribution: Polarization, stratication,
and convergence clubs. Journal of Economic Growth 2(1): 2759, March.
Quah, Danny. 2001a. Cross-country growth comparison: Theory to empirics. In
Advances in Macroeconomic Theory, vol. 133 of Proceedings of the Twelfth World Congress
of the International Economic Association, Buenos Aires, ed. Jacques Dreze, 332351. Lon-
don: Palgrave.
Quah, Danny. 2001b. Some simple arithmetic on how income inequality and eco-
nomic growth matter. Working paper, Economics Department, LSE, London, June.
Given as the rst Journal of Applied Econometrics Lecture at the Econometric Society
Meetings, Auckland, New Zealand, July.
Ravallion, Martin. 1997. Can high-inequality developing countries escape absolute
poverty? Economics Letters 56(1): 5157.
Ravallion, Martin, and Shaohua Chen. 1997. What can new survey data tell us
about recent changes in distribution and poverty? World Bank Economic Review 11(2):
357382, June.
58 Danny Quah
Sala-i-Martin, Xavier. 2002. The disturbing rise of world inequality. Working

paper, Columbia University, New York, March.
Shorrocks, Anthony F., and Stanislav Kolenikov. 2001. Poverty trends in Russia dur-
ing the transition. Working paper, United Nations University/WIDER, Helsinki,
May.
Silverman, Bernard W. 1981. Using kernel density estimates to investigate multi-
modality. Journal of the Royal Statistical Society, Series B 43(1): 9799.
Summers, Robert, and Alan Heston. 1991. The Penn World Table (Mark 5): An
expanded set of international comparisons, 19501988. Quarterly Journal of Economics
106(2): 327368, May.
UNU. 2000. UNU/WIDER-UNDP World Income Inequality Database. Version 1.0.
United Nations University/WIDER, Helsinki.
Wolfson, Michael (1994) Diverging inequalities. American Economic Review 84(2):
353358, May.
World Bank. 1990. World Development Report: Poverty. New York: Oxford University
Press.
II Population, Education,
and Inequality
3 Fertility Clubs and
Economic Growth
Avner Ahituv and

Omer Moav
3.1 Introduction
This chapter offers an empirical and theoretical investigation of the

relationships among fertility, education, and economic growth. The
theoretical model offers an explanation to the positive cross-country
relationship between education and growth and the negative cross-
country relationship between fertility and growth. The model also
offers some insights explaining dramatic demographic transitions
observed in some countries in the last few decades and bears policy
implications for the promotion of economic growth. We examine the
models predictions empirically using a rich panel data set. We nd
signicant negative relationships between fertility and economic
growth and between fertility and education. In addition, consistent
with the model, we nd that improvements in education have
increased the probability that a country would experience a demo-
graphic transition of fertility reduction that may pave the way to
improved economic performance.
A comparison of living standards among countries shows wide
diversity in both income and fertility levels. For example, the mean
value of GDP per capita, for our sample of 114 countries, in 1985 was
$4,204, with a larger standard deviation of $4,381. The richest coun-
try was the United States with $16,559 and the poorest was Ethiopia
with only $283. The average number of children per woman was
4.43, with standard deviation of 2.03. Rwanda was highest, with 7.74
children; West Germany and Italy were lowest, with 1.41 children
each. Thus, one woman in Rwanda produces the same number of
children as ve women in Italy.
More important, fertility and education have a strong negative
correlation across countries, and education has a positive strong
62 Avner Ahituv and Omer Moav
correlation with income per capita. As shown by Ahituv (2001) in

low-income countrieswith GDP of less than $1,000 per capita
the average fertility rate was 6.5 children per woman in 1965, as
opposed to only 3.7 in the high-income group with income above
$2,500. During the same period only 6.0 percent of the relevant
school age children were enrolled in secondary school in low-income
countries, compared to 49.2 percent in the high-income group of
countries. Twenty years later, the picture was very similar. In 1985
the average fertility rate in the low-income countries was still above
six, and in the high-income group the average had declined to 2.4.
These facts, along with the aforementioned dramatic demographic
transitions experienced by some countries, motivate the model,
which is based on Moav (2001), and generates a high-income equi-
librium, characterized by high investment in education and low fer-
tility, along with a poverty trap equilibrium characterized by low
investment in education and high fertility. At the same time, the theo-
retical model offers insights into the understanding of the economic
forces leading to the demographic transitions that are observed in
the data and suggests policy measures that will trigger and enhance
these processes. In particular, the theory highlights the policy mea-
sures that would encourage a shift in household resource allocation
from child quantity (fertility) to child quality (education), starting a
process that may eventually lead to a catch-up of the poor countries
with the rich.
Households, in the model, optimally allocate resources between
the quality and quantity of children, capital transfers, and consump-
tion. These decisions made by households in the economy determine
subsequent aggregate supply of production factors. Due to the trade-
off between child quality and quantity, economies characterized by
high fertility rates sustain low levels of human capital and therefore
low income per capita. The high fertility rates in the poor countries
further dilute the accumulation of physical capital per capita, ampli-
fying the impact of child quality choice on economic development.
As a result of the assumption that human capital has a positive
effect on the productivity in teaching, it is shown in the model that
high-wage (educated) individuals have a comparative advantage
in raising quality children, whereas the poor have a comparative
advantage in child quantity. Thus, households in poor economies
choose high fertility rates with low investment in their offsprings
education and low levels of capital transfers; and therefore, poverty
Fertility Clubs and Economic Growth 63
persists. In contrast, families in the rich countries choose low fertility

rates with high investment in education and high levels of capital
transfers, and therefore, high income also persists.
The microfoundations of the model follow the principle observa-
tion of Becker and Lewis (1973) that the cost of an additional child
increases with the desired level of child quality. In contrast to Becker
and Lewis who assume high-income elasticity for child quality, here
the key assumption is that individuals productivity in educating
children increases with their own human capital. For instance, indi-
viduals effectiveness in helping their children with homework, in
contrast to feeding a child, increases with their own level of educa-
tion. This assumption is consistent with empirical evidence. Lam and
Duryea (1999) argue that mothers with more schooling, despite the
increase in their market wages, do not increase their labor supply
because of their higher productivity in producing well-educated chil-
dren. In addition, Psacharopoulos, Valenzuela, and Arends (1996)
nd that teachers in Latin American countries are rewarded for their
education similarly to the average reward in the economy. Card and
Krueger (1992) argue that schooling rates of return are higher for
individuals from states with better-educated teachers and Angrist
and Lavy (2001) nd that teachers training led to an improvement
in students test scores.
Economic growth models of fertility designed to explain the pos-
sibility of multiple equilibria with persistence of poverty go back
to Nelson (1956). He shows that in an environment in which fer-
tility and saving rates increase with income, an underdevelop-
ment trap with low savings is plausible. In this trap, even if capital
is accumulated, population rises at an equal rate. More recently,
Becker, Murphy, and Tamura (1990) have developed a model in
which individuals face the trade-off between the quality and quan-
tity of their offspring.1 Their model generates multiple steady states
consistent with the cross-country relationship between fertility, edu-
cation, and growth. The source of multiplicity of equilibria in their
model is the assumption that the return to education is lower in poor
economies. In particular, they assume that the return to human cap-
ital increases with its aggregate level in society. Their approach
suffers from both theoretical and empirical limitations. The poverty
trap equilibrium is a result of a market failure in a framework in
which education has a positive externality, where existing evidence
contradicts the underlying assumption of increasing returns on
education with income.2 Furthermore, while fertility decisions in

the analysis of Becker, Murphy, and Tamura (1990) may amplify the
negative impact of low education investment on per capita income,
they are not the source of multiple equilibria. In this chapter, in
contrast, a simple dynamical system generates multiple steady states
that emerge from the comparative advantage of educated individ-
uals in the production of educated children. The model is based
neither on restrictive assumptions concerning preferences, or the
return on human capital, nor on any nonconvexity in the production
process.3
The model offers explanations for cross-country output differences
and for the phenomenon of club convergence.4 Consistent with the
evidence, as shown for instance by Quah (1997), countries in the
club of the rich converge to a high-income per capita steady state,
whereas countries in the club of the poor converge to a low income
level. The poor countries, as consistent with the observation of
Cohen (1996), fail to catch up with the rich because of insufcient
progress in education, which, according to the model, is due to high
fertility rates. Cross-section observations from 1965 suggest that
countries in the club of the poor are also the countries in the high-
fertility club, and the club of the rich consists of the countries that
have low fertility rates. Our empirical investigation of the fertility
trends, however, shows a different pattern than Quahs (1997) nd-
ings of increased polarization over time. On the one hand, fertility
rates have further declined in the rich countries, but on the other,
many of the high-fertility countries have dramatically reduced fer-
tility rates, joining the club of the low-fertility countries. Based on
our theory, we predict that some of these countries will eventually
also converge to the club of the rich, which is now only a subset of
the club of the low-fertility countries.
The empirical section of this chapter examines the models predic-
tions and implications. Our data contains rich social, demographic,
political, and income indicators from 1965 to 1989, and fertility
rates up to 2000, across 114 countries. Using this data, we rst
show an inverse relationship between fertility and GDP. Then we
present evidence supporting the existence of a poverty trap and
the phenomenon of club convergence. Finally, the analysis turns
to the documentation of the various demographic transitions that
have recently occurred in many of the poor countries of the world.
We nd that education, and in particular womens education, ini-
tiates fertility reduction and consequently brings about faster eco-

nomic growth. These ndings offer an explanation for the signicant
negative effect that fertility has on long-term economic growth, and
the nonsignicant effect on short-term growth, as shown by Barro
(1991).
The chapter proceeds as follows: Section 3.2 develops the basic
model of fertility, capital accumulation, and child educational
choices and discusses the results. Section 3.3 presents the empirical
results. It rst shows that fertility differs by economic cluster. Then
it describes results of simple regression analysis that examine the
interplay among the previous three variables. Section 3.4 concludes.
3.2 The Model
We develop a closed economy model of homogenous individuals

following Moav (2001). The homogeneity assumption ts well the
focus of this chapter: differences across countries and trends of
countries over time, where it should be noted that empirical evi-
dence show that fertility rates are indeed more heterogeneous across
countries than within a country. Consider an overlapping generation
economy in which activity extends over innite discrete time. In
every period the economy produces a single homogenous good,
using two factors of production: physical and human capital. The
supply of human capital is determined by households decisions in
the preceding period regarding the number of their children and
the level of education investment in each child; the supply of physi-
cal capital is determined by individuals choice of bequest.
3.2.1 Production
Production occurs within a period according to a neoclassical,

constant-returns-to-scale, production technology. The output pro-
duced at time t, Yt , is given by
Yt FKt ; Ht 1 Ht f kt Ht Akta ; kt 1 Kt =Ht ; 1
where Ht and Kt are human (measured in efciency units of labor)

and physical capital employed in production in period t.
Producers operate in a perfectly competitive environment and
therefore production factors are paid according to their marginal
products,
wt 1 aAk a 1 wkt ;
2
rt aAk a1 1 rkt ;
where wt is the wage rate per efciency unit of labor in time t, and rt
is the capital rate of return. For sake of simplicity, physical capital is
assumed to fully depreciate at the end of each period.
3.2.2 Individuals
In each period a generation of individuals, who each has a single

parent, is born. Individuals live two periods: In childhood they
acquire human capital; in adulthood they are endowed with one unit
of time, which they allocate between child rearing and participation
in the labor force.
The preferences of members of generation t (born in t 1) are
dened over consumption as well as over the quality and quantity of
their children, where quality is measured by their offsprings full
income (potential income). Preferences are represented by the utility
function
ut 1 b log ct blog nt logwt1 ht1 rt1 st1 ; 3
where b A 0; 1, ct is the consumption in the household of members

of generation t, nt is the number of children in this household, ht1 is
the level of human capital of each child measured in efciency units,
and st1 is physical capital transferred to each child.
3.2.3 The Formation of Human Capital
In the rst period of their lives individuals devote their entire time
for the acquisition of human capital (measured in efciency units of
labor). The acquired level of human capital increases if their time
investment is supplemented by investment in education. However,
even in the absence of investment in education, individuals acquire
one efciency unit of laborbasic skills. The level of investment in
education of each individual in period t, et1 , is measured in ef-
ciency units of labor. The resulting number of efciency units of
labor in period t 1, ht1 , is a strictly increasing, strictly concave
function of investment in education in period t, et1 ,
ht1 het1 ; 4
where h0 1, lim e!0 h 0 et g, lim e!y het > 1=tg,5 and

lim e!y h 0 et 0. The assumption that the slope of the production
function of human capital is nite at the origin, along with the ability
of individuals to supply some minimal level of labor regardless of
the investment in human capital (beyond time), assure that under
some conditions raising quality children is not optimal.6 Since et1 is
measured in efciency units of labor, the real cost of the investment
in education is wt et1 .7
3.2.4 Budget Constraint
Let t be the minimum time cost required for raising a child; addi-
tional time allocated to children positively affects their quality. That
is, t is the fraction of the individuals unit of time endowment that is
required in order to raise a child, regardless of quality. Following
Becker and Lewis (1973), Rosenzweig and Wolpin (1980), and Galor
and Weil (2000), it is assumed for the sake of simplicity that the
quantity cost per child does not vary with family size. The eco-
nomic force behind the fertility gap between rich and poor is the
lower quantity cost faced by the poor. Therefore, incorporating a
range of decreasing quantity cost would increase the marginal cost
difference faced by poor and rich, amplifying the fertility gap, and
thus strengthening the results. Increasing returns to investment in
quality would work in the opposite direction, but, up to a limit,
should not have a qualitative effect on the models results.8 As will
become apparent, fertility rates are bounded from above by b=t,
therefore it is assumed that t < b. It is further assumed that t is suf-
ciently small, so that individuals with a low level of human capital
choose the corner solution of zero investment in child education,
t < 1=g: A1
Consider an adult agent of generation t who is endowed with ht

efciency units of labor and st units of capital. Full income, wt ht rt st ,
is divided between expenditure on child rearing (including bequests)
and consumption, ct . The (opportunity) cost of raising each child,
regardless of quality, is equal to wt ht t, and the cost of quality of
each child is equal to wt et1 . The cost of raising nt children, with
an education level of et1 and a bequest st1 is given therefore,
by nt wt ht t wt et1 st1 , and the individual faces the budget
constraint
nt wt ht t wt et1 st1 ct a wt ht rt st ; 5
where wt ht rt st is full income of each individual in period t, and

ht het . As captured by the budget constraint, given in equation
(5), the cost of child quantity, wt ht t, in contrast to the cost of child
quality, wt et1 , increases with the level of human capital of the t
genereation agent, ht . This is a result of the assumption that individ-
uals productivity as educators, in contrast to their productivity in
child quantity, increases with their own human capital.
3.2.5 Optimization
Members of generation t choose the number, nt , and quality, et1 , of

their children, the quantity of physical capital they transfer to each
child, st1 , and the household consumption, ct , so as to maximize
their utility. It follows from the optimization that consumption is
ct 1 bwt ht rt st ; 6
that is, a fraction 1 b of full income is devoted to consumption and
hence a fraction b of full income is devoted to child rearing in terms
of quality and quantity.
The optimization with respect to capital transfers, st1 , is given
by
8
het1 wt1 <> 0 st1 0;
rt1 0 st1 A 0; y; 7
tht et1 wt :
< 0 st1 ! y:
where st1 ! y implies that nt ! 0. In equilibrium, however, since
all individuals are identical, it follows from (7) that
het1 wt1
rt1 : 8
tht et1 wt
Otherwise, if the left-hand side is larger (smaller), there is no physi-

cal (human) capital in period t 1, and the left-hand side is smaller
(larger) in contradiction. Given (8), it follows from the optimization
with respect to et1 that9

het1 b 0 if et1 0;
h 0 et1 9
thet et1 0 if et1 > 0:
It follows, therefore, that the equilibrium level of investment in edu-

cation in period t, et1 , is uniquely determined by the investment in
education in the previous period, et .
Lemma 1 Under (A1) there exists a single valued function fet

such that

0 if et a ê;
et1 fet
> 0 if et > ê;
where f 0 et > 0 for et > ê and ê > 0 is unique and given by
1=g thê.
Proof Dene Get1 1 het1 =yh 0 et1 . It follows from (9) that
Get1 b thet et1 : 10
As follows from (A1), qGet1 =qet 0, qthet et1 =qet > 0, and
G0 1=g > th0 t, and as follows from the properties of (4)
G0 1=g < lim e!y thet , and h 0 et > 0. Therefore, as follows from
the intermediate value theorem, there exists a unique et ê > 0,
given by 1=g thê; such that G0 thê and G0 >< thet
for all et <> ê implying that fet 0 for all et a ê and fet > 0
otherwise.
As follows from implicit differentiation of the rst-order condition
of the maximization problem as given by (9), noting that second-
order condition hold for a maximum, fet is single valued and
f 0 et > 0 for et > ê. r
It follows from lemma 1, and equations (5) and (6) that the number
of children of each member of generation t, nt is given by

b=t if et a ê;
nt net 11
bhet =thet fet if et > ê;
where b=t b bhet =thet fet . That is, fertility rates in low edu-
cation economies (et a ê), where there is no investment in the quality
of children, are higher than in those countries with higher education
levels who invest in childrens education.
3.2.6 The Evolution of the Economy
Figure 3.1 depicts the evolution of education in the economy, fet ;

(A1) assures the existence of a low education steady state, f0 0.
Figure 3.1
The evolution of education
However, in order for fet to generate a high education steady state,

there must exist a range of et in which fet is sufciently sensitive to
changes in et . That is, a range must exist in which small changes in
the parents education bring about large changes in the offspring
education and as a result fet > et for some et .
Proposition 1 There exists a human capital production function,
that satises (A1) and the properties of het , in equation (4), such
that, fet , is characterized by multiple steady states.
Proof The proof follows from an example. Consider the following

human capital production function:

1 get1 if et1 < e;
ht1 het1 12
1 ge if et1 b e;
where e > 1 tg=tg 2 . Hence, it follows from the optimization
problem of a member i of generation t, that is endowed with het
efciency units of human capital that
8
< 0 if et < ê;
et1 fet A 0; e if et ê; 13
:
e ^
if et > e;
where as follows from (12) and (A1) ê 1 tg=tg 2 > 0 and e > ê.
The dynamic system fet is characterized by two stable steady
states, e and 0, and an unstable steady state ê, which is the threshold
level of education. The lemma thus follows from continuity consid-
erations.10 r
It follows from (2), (8), (9), and lemma 1 that the dynamical system
governing the evolution of the economy is uniquely determined by
the sequence fkt ; et gyt0 , such that

et1 fet ;
14
kt1 cet ; kt aAkta thet fet =hfet ;
where k0 and e0 are given. Note that kt is the physical human capital
ratio, and that physical capital per worker is equal to het kt st > kt .
Output per worker, yt , as follows from (1), is therefore uniquely
determined by the dynamical system
yt het Akta Aht1a sta :
3.2.6.1 The kk Locus

Let kk be the locus of all pairs kt ; et such that physical capital per
efciency unit of labor, kt , is in a steady state: kk 1 fkt ; et : kt1 kt g.
As follows from (14) there exists a function

aAthet fet 1=1a
k kk et ; 15
hfet
such that if kt k kk et , then kt1 cet ; kt kt that is, the kk locus

consists of all the pairs k kk et ; et .
Lemma 2 dk kk et =det > 0. That is, as depicted in gure 3.2, the kk
locus is strictly increasing in the plan et ; kt .
Proof For et a ê, as follows from lemma 1, fet 0, and therefore,
hfet 1, and since h 0 et > 0, it follows that dk kk et =det > 0 for
et a ê.
For et > ê, it follows from (9) and lemma 1 that hfet =thet
fet h 0 fet . Furthermore, as established in lemma 1 f 0 et > 0
for et > ê, and since h 00 et < 0 it follows that hfet =thet fet is
strictly decreasing with et , and therefore dk kk et =det > 0 for et > ê. r
3.2.6.2 The ee Locus

Let ee be the locus of all pairs kt ; et such that the level of investment
in human capital per capita, et , is in a steady state: ee 1 fkt ; et :
Figure 3.2
The dynamical system
et1 et g. However, as follows from (14), the evolution of et is inde-

pendent of the evolution of the physical human capital ratio, and
hence ee 1 fet : fet et g. As follows from the properties of fet and
as depicted in gure 3.1, et fet for et 0, et e T , and et e,
therefore the ee locus consists of three vertical lines in gure 3.2:
e 0, e e T , and e e.
The evolution of et as follows from (14) and depicted in gure 3.1
is given by et fet . Therefore, as depicted in gure 3.2, et1 > et for
all et A e T ; e, whereas et1 < et for all et < e T and all et > e. The
dynamics of kt follow from (14). As depicted in gure 3.2, kt1 > kt
for all kt < k kk et , and vice versa. Hence, kt is increasing below the kk
locus and decreasing above it.
3.2.7 Steady States and Implications of the Model
The model generates two locally stable steady states. If the initial
level of education is above the threshold level, e0 > e T , the economy
converges monotonically to the high capital labor ratio, high educa-
tion steady state, characterized by a low fertility rate. If however
e0 < e T , the economy converges to the low capital labor ratio, low
education steady statethe poverty trapthat is characterized by a
high fertility rate.
The mechanism generating multiple steady states is based on the

effect of parental education on child quantity cost. The lower the
parents education, that is, the cheaper the parents time, the cheaper
the children and the parents choice shifts to higher fertility rates and
lower investment in offsprings human capital. Note that a decline in
parental education, and hence in their income, leaves less resources
for childrens education even if their number is unchanged. The
increased fertility further reduces investment in education that is, in
addition, divided between more children. Therefore, consistent with
proposition 1, differences in parental education can be amplied
when it comes to differences in offspring education. The effect of
quality choice on output per capita is amplied by its consequence
on fertility and the diluting effect fertility has on capital accumula-
tion. In the poverty trap, therefore, the high fertility rate yields a low
capital labor ratio.
It is interesting to note that in the context of a closed economy,
the low output steady state is not a result of capital market imper-
fections nor any other market failure, and resource allocation is
dynamically efcientthe marginal return to education is not higher
than the marginal return to physical capital. Hence, even if individ-
uals could borrow to nance their own education they choose not
to do so. A shift of the economy from the low- to the high-output
steady state can be achieved only if one generation gives up child
quantity in favor of child quality and suffers from a utility cost. Of
course, if countries differ from each other, in particular, if some
economies are in the high output steady state, than the low output
steady state is an outcome of imperfection in international capital
markets (taken to the extreme of closed economies in the model).
The effect of changes in the quantity cost parameter, t, on the
dynamical system and its steady states follows from equation (9) and
lemma 1. An increase in t increases the relative cost of quantity,
inducing a shift to child quality. Hence, it reduces the level of the
threshold below which individuals choose not to purchase any edu-
cation for their offspring, ê, and increases the level of human capital,
fet , above ê. This implies that the dynamical system depicted in
gure 3.1 shifts upward, the threshold level, e T , declines, and the
high-income steady state, e, increases. Therefore, the ee locus, de-
picted in gure 3.2, shifts accordingly, that is, the vertical line at e
shifts to the right, while the vertical threshold line at e T shifts to the
left. Furthermore, since fet increases with t for e > ê, and since it is
constant with respect to et for e a ê, it follows from lemma 1, the
concavity of het , (9) and (15) that the kk locus, depicted in gure
3.2, shifts upward as a result of an increase in t. Hence, in the
extended model, the impact of changes in the cost of child quantity
is amplied by the diluting effect on physical capital (the change
in the kk locus). Furthermore, since the threshold level of education,
e T , declines with t, a sufcient increase in the quantity cost can fa-
cilitate a demographic transition and release the economy from the
poverty trap.
The effect of changes in the cost of education is not so straight-
forward. The analysis of public schooling on fertility and education
follows.11 Suppose the government supplies schooling free of charge
g
at a level et , which is nanced by foreign aid or taxation.12 It follows
g
from (9) that if fet < et parental investment in education would
g
be zero, or otherwise it would equal fet et . Public schooling has,
therefore, a positive effect on the offspring level of education (for
g
fet < et ), but it also reduces parental expenditure on education
if fet > 0 and therefore gives rise to a reallocation of resources
g
for increased fertility. In the long run, however, if et > e T , public
schooling will shift the economy to a path of increased education
and income and reduced fertility.
3.3 Evidence on Fertility and Economics Growth
The empirical part of the chapter uses aggregate international data

from the United Nations Statistical Yearbook. We start with descrip-
tive statistics that portray the joint time trends of fertility, education,
and income of the last three decadesa period associated with rapid
demographic transition in many countries around the world. Then,
we use regression analysis to test whether the cross-country rela-
tionships between these variables are consistent with the predictions
of the model.
We use a panel of data containing rich social, demographic, politi-
cal, and income indicators from 1965 to 1989, in addition to fertility
rates up to 2000, across 114 countries. The database for this study
covers approximately fty indicators. For each country, the data set
contains annual time-series data from 1960 to 1989. Most of the
analyses presented in this chapter use only six data points (1965,
1970, 1975, 1980, 1985, and 1989), because the early years are used as
lag variables, and ve-year spans are less likely to be serially corre-
lated. The variables in this database are divided into four categories:
(1) national account variables related to expenditure and income;
(2) social and demographic indicators, including school and fertility;
(3) political indicators, indexing political instability, regime and
market distortions; and (4) geographic regions in which the country
is located. Most of the variables are in time-variant format, but
some of them, mainly the political and geographic regions, are time-
invariant.
In 1965, the average number of children per womanthe fertility
ratewas above 6 in more than half of the countries sampled, while
in only six countries they were below 2.5. Even in 2000, in some East
African countries fertility rates are still above seven. On the other
hand in many countries (such as Spain, Italy, and Korea) they are
well below 1.5. Figure 3.3 displays the distributions of fertility rates
across countries and over time, showing a sharp decrease in fertility
rates, and interesting changes in the shape of the distribution. In
1965, there were, roughly speaking, two separate groups of countries
with respect to fertility levels. The distribution peaks around seven,
capturing the mass of high-fertility countries, and it also peaks
below three children per woman capturing the mass of the low fer-
tility countries. Surprisingly, only a few countries had fertility rates
between four and six children per woman. This twin-peak structure
did not change during the next twenty-ve years, in spite of the fact
that many countries experienced a rapid demographic transition,
and thus the mass of countries in the high fertility range has
declined. These facts support the basic result of the model that there
are strong economic forces driving toward polarization in fertility
rates.
Since 1965, fertility declined in all but twelve African counties in
our 114-country sample. Interestingly, fertility in the group of the
rich countries, which was low in 1965, further declined. At the same
time, many poor countries went through a major demographic tran-
sition, signicantly reducing their fertility rates and improving their
education. In 2000, fertility rates are below 3 in more than sixty
countries, and below 1.5 in twelve, which is signicantly below the
replacement rate. On the other spectrum, only twenty-ve countries
have fertility rates above 5, and only four of them still have fertility
rates of above 7 children per woman. These observations raise inter-
Figure 3.3
Fertility rates across countries
Table 3.1
Means of fertility by country group
Low-income Middle-income High-income
economies economies economies
1965 6.51 6.26 3.69
1980 6.41 5.07 2.64
1989 5.89 4.42 2.32
2001 4.90 3.42 1.89
Number of countries 34 39 38
in each group
Note: The three groups are based on GDP per capita in 1965. Group 1: 01000; Group
2: 10012500; Group 3: 2501.
esting questions. What are the economic forces or policy measures

initiating the demographic transition? What will be the long-run
consequences of these policies? Could this transition take place
without outside intervention, from the United Nations and the de-
veloped countries?
We add income as our next stage of the analysis, to study the
evolution of fertility rates by income group. Table 3.1 shows that
fertility and GDP are closely related by displaying gures for four
representative years. Low-income countries had 6.5 children per
woman in 1965, as opposed to only 3.7 in the high-income group.
Until 1980, fertility rates in the low-income countries did not change,
while those in the two other groups declined signicantly. During
the last decade, however, fertility rates have been dropping in the
low-income countries also, converging at 4.9 children per woman in
2000approximately the rate in the middle-income group twenty
years earlier. Fertility rates in the middle-income group declined
during that period to 3.4, similar to those of the high-income coun-
tries in the early 1970s. The evidence reveals that the transition from
high to lower fertility occurred in different periods across these three
groups. In the high-income group the transition occurred before
World War II; in the middle-income group the sharper decline
started in the late 1960s, while in the low-income groups the transi-
tion started during the last ten years.
Table 3.2 shows results complementary to those in table 3.1, by
presenting indicators for economic activities by groups of countries
based on their fertility rates in 1965. The rst line shows that eco-
nomic growth declines with fertility rates. The trend, however, is not
Table 3.2
Summery of selected indicators by fertility rates
Fertility rates
13 34 45 56 67 79
1. Annual GDP growth 2.70% 2.60% 3.20% 2.09% 0.90% 1.25%

(6585 avg.)
2. Changes in fertility 0.83 1.07 1.49 0.86 0.92 0.81
rates (6585)
3. Secondary school 60% 53% 30% 14% 12% 12%
enrollment (65)
4. Secondary school 89% 86% 62% 32% 28% 34%
enrollment (85)
5. Log of physical 9.27 8.82 7.65 6.43 6.48 6.61
capital (65)
6. Log of physical 10.33 9.91 9.03 7.74 7.49 7.75
capital (85)
Number of countries 19 9 10 14 43 17
in each group
linear. While the rates of economic growth in the low-fertility coun-

tries are very similar to those in the middle-fertility group, the
countries with high-fertility experienced signicantly lower rates of
economic growth. The second line shows that middle-fertility coun-
tries experienced a sharper reduction in fertility. This is consistent
with the models prediction of two steady states. Lines 3 and 4 show
that enrollment in secondary school improved dramatically across
all countries, leading, according to our theory, to the observed fer-
tility declines of the 1990s. Finally when we compare physical capital
across groups, we see a strong negative correlation with fertility
rates, suggesting that capital accumulation, as consistent with our
model, is diluted by fertility.
Table 3.3 adds a conditional correlation analysis to our descrip-
tive examination. The table presents results from regressions in
which the dependent variables are log-fertility, log-GDP, and school
enrollment (a proxy for education) as well as 1965 to 1985 growth
rates of these indicators (columns 46). The rst three regressions, in
which the explained variables are the levels of log-fertility, log-GDP,
and school enrollment, are panel regressions with six observa-
tions per country. The main focus here is to display the interrelation
between these three variables, controlling for other characteristics of
each country. The growth regressionsthe growth rates of these
three variablesare cross-section regressions, where the values of

the independent variables are from the initial year (1965). The main
focus here is to examine the effects of the initial condition on the
countrys growth rates.
The results presented in columns 13 support one of our main
hypotheses by showing clearly that fertility is inversely correlated
with physical capital, GDP and education. In addition, countries
with a wide gender education gap have higher fertility, and coun-
tries with long life expectancy have lower fertility rates. Controlling
for a countrys characteristics, fertility rates in Europe are still lower,
while in Latin America and the Muslim countries they are signi-
cantly higher. The political conditions of the countries (civil liberties,
number of revolutions, and social regime) do not have a signicant
effect on fertility rates. In addition to the negative effect of fertility on
income, we also nd a positive effect of education, and a negative
effect of Africa and civil liberties. The regression in school enroll-
ment reveals positive correlation with physical capital, again similar
to the prediction of the model.
Shifting our analysis to the determinants of the growth rates of
these three indicators, we nd that initial GDP does not have a sig-
nicant effect on fertility changes.13 On the other hand, education
has a strong negative effect on fertility changes, suggesting that the
timing and speed of the demographic transition is affected by the
initial level of education more than the initial level of GDP. This
suggests that investment in education is an important element in
initiating the fertility transition. Focusing on the other coefcients,
fertility rates in African, Muslim, and socialist countries decline
more slowly than in the rest of the world.
According to our theory, countries that are in a poverty trap char-
acterized by high fertility will not exhibit high growth rates. Coun-
tries that manage, however, to escape the poverty trap will benet
from high growth rates. Accordingly, the effect of initial fertility
rates on growth depends on the subsequent behavior of the economy
in terms of changes in fertility rates. Our data shows that initial fer-
tility has a negative effect, but not a very signicant one, on eco-
nomic growth. The coefcient of the changes in fertility, when we
add this variable to the right hand side of the regression (not shown
in table 3.3) is indeed signicant and negative. In addition, countries
that have not reduced their fertility rates are concentrated in Africa,
and, indeed, the coefcient in African countries in explaining growth
Table 3.3
80
Coefcient estimates from cross-country regressions

(6)
(3) (4) (5) Growth of
(1) (2) Secondary Fertility GDP secondary
Log Log school growth growth school enrollment
Dependent variable fertility GDP enrollment (19651985) (19651985) (19651985)
Log fertility 0.198 0.208 0.355 0.232 0.644
(3.34)** (10.21)** (3.35)** (1.27) (0.66)
Log GDP 0.033 0.226 0.044
(0.83) (3.25)** (0.12)
Log capital 0.023 0.453 0.053
(2.27)* (28.25)** (9.23)**
Secondary school enrollment 0.550 0.291 0.414 0.271 3.838
(7.87)** (2.60)** (2.16)* (0.82) (2.12)*
Years of school attainment 0.021 0.009
(3.30)** (1.00)
Gender gap in secondary school 0.438 5.465
(5.26)** (2.24)*
Life expectancy 0.011 0.003 0.008 0.037
(6.29)** (1.24) (8.89)** (1.09)
EUROPE 0.231 0.019 0.031 0.052 0.045 0.357
(7.98)** (0.4) (1.77) (0.67) (0.34) (0.51)
AFRICA 0.019 0.092 0.047 0.218 0.459 0.553
(0.73) (2.26)* (3.08)** (3.49)** (4.26)** (0.92)
Latin-America 0.103 0.063 0.073 0.084 0.248 0.436
(2.69)** (1.06) (3.27)** (0.89) (0.53) (0.51)
Avner Ahituv and Omer Moav
CATHOLIC 0.013 0.036 0.013 0.04 0.008 0.031
(0.39) (0.72) (0.69) (0.51) (0.06) (0.04)
MUSLIM 0.115 0.065 0.032 0.104 0.023 0.718
(5.00)** (1.78) (2.22)* (1.83) (0.23) (1.38)
Civil liberties (index) 0.012 0.05 0.009 0.029 0.007 0.017
(1.52) (4.17)** (2.08)* (1.49) (0.20) (0.11)
Number of revolution and coups 0.017 0.041 0.043 0.078 0.401 0.377
(0.49) (0.02) (2.07)* (0.90) (2.70)** (0.51)
Socialist economy 0.046 0.041 0.029 0.117 0.079 0.562
(1.94) (1.09) (2.06)* (2.03)* (0.79) (1.11)
Constant 2.423 4.226 0.138 0.403 2.693 5.82
(23.59)** (19.48)** (1.71) (1.05) (4.07)** (1.59)
R2 0.86 0.92 0.86 0.58 0.38 0.39
Number of observations 675 675 675 112 112 112
Fertility Clubs and Economic Growth
Note: Absolute value of t-statistics in parentheses.

* signicant at 5 percent level; ** signicant at 1 percent level
81
is negative and signicant. Surprisingly, most of the coefcients in

school enrollment growth are insignicant. The only two signicant
coefcients are the negative sign on initial values of school enroll-
ment and wide gender education gap. We do not have good expla-
nations for these last results.14
The analysis in this section provides evidence supporting the
building blocks of the model, and its main results. It shows an
inverse relationship between fertility and education and a positive
relationship between education and economic performance, sup-
porting our presumption that there exists a trade-off between child
quality and quantity, and that education has a positive effect on
economic growth.15 In addition, the fact that twenty-ve countries in
our sample are still characterized by fertility rates of above 5 chil-
dren per women and by very low levels of education and income
suggests that poverty is indeed persistent and highly related to fer-
tility rates. Moreover, the rapid demographic transition experienced
by many countries, combined with the fact that only a few countries
are characterized by intermediate levels of fertility, supports the
theoretical result of two locally stable equilibria.16 These observa-
tions suggest that the next decade can bring signicant economic
growth in many countries that have recently experienced a rapid
demographic transition.
3.4 Concluding Remarks
This chapter offers an empirical and theoretical investigation of the

relationships among fertility, education, and economic growth. It
argues that the trade-off between child quality and quantity, along
with the comparative advantage of the poor in child quantity, in
contrast to the comparative advantage of the rich in child quality,
clusters countries into two fertility clubs. The club of high fertility, in
contrast to the club of low fertility, is characterized by low invest-
ment in education, low capital ratios, and low income. Moreover, we
show that improvements in education have increased the probability
that a country will experience a sharp reduction in the fertility rates,
joining the club of low-fertility countries. We suggest that these
demographic transitions, experienced in the last few decades in
many of the worlds poor countries, will eventually pave the way to
signicant improvements in economic performance.
The theory presented in this chapter has interesting policy impli-

cations. An increase in the cost of quantitythe cost of a child
regardless of the childs qualityinduces a reallocation of resources
to child quality. It therefore reduces the threshold above which
economies reach the high-income steady state, allowing an easier
escape from poverty. It would, in addition, further increase the
investment in education, and hence increase income in the high-
income steady state. Hence, an increase in the quantity cost posi-
tively affects economic growth and could, furthermore, release an
economy from the trap of poverty, setting the stage for a demo-
graphic transition and economic growth. Thus, policies that reduce
the quantity cost, such as tax discounts on large families, child
allowances, and subsidized daycare and meals, have, according to
the theory, a negative effect on income in the long run. Therefore,
though not in harmony (at least in the short run) with a humanitar-
ian approach, policy implications are straightforwardcanceling
or even reversing the aforementioned policies. Furthermore, since
public schooling can release the economy from the poverty trap,
growth-encouraging policies include the reallocation of government
or foreign aid resources from quantity-cost reduction measures to
the nance of schooling. The nding that a temporary improvement
in education opportunities could have a permanent effect on the
economy was established in previous literature (e.g., Galor and Zeira
1993). The contribution of this chapter, in this respect, is the linking
of education to fertility, amplifying the economic consequences of
the education policy.
Notes
We wish to thank Eric Gould, Joram Mayshar, M. Carme Riera Prunera, Mathias
Thoenig, an anonymous referee and participants in CESifo conference on Growth and
Inequality: Issues and Policy Implications, Munich 2001 and Elmau 2002, for helpful
discussions.
1. See also Hazan and Berdugo (2002), who study the relationship between child
labor, fertility, and growth, and Kremer and Chen (2002), who study the effect of
income inequality on economic growth in an endogenous fertility framework. In
addition, Veloso (1999) analyzes the effect of different compositions of wealth and
human capital on education. Finally, some recent literature focuses on the long-run
phenomena of fertility and the take-off from economic stagnation to sustained eco-
nomic growth, including the work of Kremer (1993), Tamura (1996), Galor and Weil
(1996, 2000), Dahan and Tsiddon (1998), Morand (1999), Hansen and Prescott (2002),
Lagerloff (2000), Jones (2001), and Galor and Moav (2002).
2. See, for instance, Psacharopoulos (1994), Angrist (1995), and Acemoglu and Angrist
(2000).
3. In contrast to this chapters thesis, existing literature explaining poverty traps is
based on nonconvexities in the technology. In particular, Banerjee and Newman
(1993), Galor and Zeira (1993), Benabou (1996), Durlauf (1996a), Piketty (1997), Maoz
and Moav (1999), Ghatak and Jiang (2002), and Mookherjee and Ray (2000) show that
credit constraints combined with investment thresholds generate persistence of pov-
erty. In the model developed by Piketty (1997), the effort level, rather than capital
investment, is indivisible. Mookherjee and Ray (2000) show that while inequality per-
sists irrespective of the divisibility of investment, the multiplicity of steady states
requires indivisibilities in the return to education. An exception is presented by Moav
(2002), where increasing saving rates with income, replace the role of nonconvexities
in the technology in generating multiple steady states.
4. See Azariadis (1996) and Galor (1996) for surveys of the theoretical and empirical
literature and the summery by Durlauf (1996b).
5. Where t, formally dened in what follows, is the time cost for raising a child.
6. The Inada conditions are typically designed to simplify the exposition by avoiding
a corner solution, but they are surely not realistic assumptions.
i
7. Note that the cost of education is wt et1 , whether this is viewed as a direct cost of
hiring a teacher or an opportunity cost of teaching ones own children.
8. It is implicitly assumed that the time cost of raising a child can not be reduced by
child care employment.
9. Note that individuals optimization assures efcient investment. The marginal
return to physical capital and human capital are equal in an interior solution and the
return to physical is larger in a case of a corner solution with no investment in human
capital.
10. While the model is not designed to perform calibrations, a numerical illustration
reveals that reasonable parameters, in particular, the lifetime dollar return to each
dollar invested in education, as given by g (see Psacharopoulos 1994) can generate
multiple equilibria. For g 2, y 1, t 0:15, b 1=2 and e 1:5. earnings at the high-
income equilibrium are four times higher than earnings at the low-income equilib-
rium, and fertility rates are 1.9 children per household (of two parents) in the high-
income equilibrium and 6.6 at the low. Note that multiplicity will hold for any set of
parameters restricted by (A1) and e > ê.
11. The analysis abstracts from the potential effect of public schooling on the quantity
cost.
12. An indirect consumption tax, for instance, will not have an impact on the quality
and quantity choice of individuals.
13. Most countries have negative fertility growth, and the minus sign means faster
decline in fertility.
14. As previously reported in the literature, for example, Barro (1991), initial GDP
has a negative effect on GDP growth, supporting the conditional convergence hypo-
thesis.
15. This is consistent with the ndings of Rosenzweig and Wolpin (1980) and
Hanushek (1992), who provide evidence that there exists a trade-off between quantity
and quality of children.
16. Consistent with our theoretical argument and with our empirical ndings, Altonji
and Dunn (1996) nd that parents education positively affects childrens education,
supporting the existence of persistence of low education. In addition, Behrman et al.
(1999) nd that increases in the schooling of women enhance the human capital of the
next generation, and, consistent with the underlying mechanism developed here, they
argue that a component of the signicant and positive relationship between maternal
literacy and child schooling reects the productivity effect of home teaching.
References
Acemoglu, D., and J. D. Angrist. 2000. How Large Are the External Returns to Edu-
cation? Evidence from Compulsory Schooling Laws. NBER Macroannual 15: 959.
Ahituv, A. 2001. Be Fruitful or Multiply: On the Interplay between Fertility and Eco-
nomic Development. Journal of Population Economics 14: 5171.
Altonji, J. G., and T. A. Dunn. 1996. The Effect of Family Characteristics on the Return
to Education. Review of Economics and Statistics 78: 692704.
Angrist, J. D. 1995. The Economic Returns to Schooling in the West Bank and Gaza
Strip. American Economic Review 85: 10651087.
Angrist, J. D., and V. Lavy. 2001. Does Teacher Training Affect Pupil Learning? Evi-
dence from Matched Comparisons in Jerusalem Public Schools. Journal of Labor Eco-
nomics 19: 343369.
Azariadis, C. 1996. The Economics of Poverty Traps, Part One: Complete Markets.
Journal of Economic Growth 1: 449486.
Banerjee, A., and A. Newman. 1993. Occupational Choice and the Process of Devel-
opment. Journal of Political Economy 101: 274298.
Barro, R. J. 1991. Economic Growth in a Cross Section of Countries. Quarterly Journal
of Economics 106: 407443.
Becker, G. S., and H. G. Lewis. 1973. On the Interaction between Quantity and Qual-
ity of Children. Journal of Political Economy 82: S279S288.
Becker, G. S., K. M. Murphy, and R. F. Tamura. 1990. Human Capital, Fertility, and
Economic Growth. Journal of Political Economy 98: S12S37.
Behrman, J. R., A. D. Foster, M. R. Rosenzweig, and P. Vashishtha. 1999. Womens
Schooling, Home Teaching, and Economic Growth. Journal of Political Economy 107:
682714.
Benabou, R. 1996. Equity and Efciency in Human Capital Investment: The Local
Connection. Review of Economic Studies 63: 267264.
Card, D., and A. B. Krueger. 1992. Does School Quality MatterReturns to Educa-
tion and the Characteristics of Public-Schools in the United States. Journal of Political
Economy 100: 140.
Cohen, D. 1996. Tests of the Convergence Hypothesis: Some Further Results.

Dahan, M., and D. Tsiddon. 1998. Demographic Transition, Income Distribution, and
Economic Growth. Journal of Economic Growth 3: 2952.
Durlauf, S. N. 1996a. A Theory of Persistent Income Inequality. Journal of Economic
Growth 1: 7593.
Durlauf, S. N. 1996b. ControversyOn the Convergence and Divergence of Growth
RatesAn Introduction. Economic Journal 106: 10161018.
Galor, O. 1996. Convergence? Inferences from Theoretical Models. Economic Journal
106: 10561069.
Galor, O., and O. Moav. 2002. Natural Selection and the Origin of Economic
Growth. Quarterly Journal of Economics 117: 11331191.
Galor, O., and D. Weil. 1996. The Gender Gap, Fertility and Growth. American Eco-
nomic Review 86: 931953.
Galor, O., and D. Weil. 2000. Population, Technology, and Growth: From Malthusian
Stagnation to Demographic Transition and Beyond. American Economic Review 90:
806828.
Galor, O., and J. Zeira. 1993. Income Distribution and Macroeconomics. Review of
Economic Studies 60: 3552.
Ghatak, M., and N. H. Jiang. 2002. A Simple Model of Inequality, Occupational
Choice and Development. Journal of Development Economics 69: 205226.
Hansen, G., and E. Prescott. 2002. Malthus to Solow. American Economic Review 92:
12051217.
Hanushek, E. A. 1992. The Trade-Off between Child Quantity and Quality. Journal of
Political Economy 100: 84117.
Hazan, M., and B. Berdugo. 2002. Child Labor, Fertility, and Economic Growth.
Economic Journal 112: 810828.
Jones, C. I. 2001. Was the Industrial Revolution Inevitable? Economic Growth over
the Very Long Run. Advances in Macroeconomics 1, Article 1. Available online.
Kremer, M. 1993. Population Growth and Technological Change: One Million B.C. to
1990. Quarterly Journal of Economics 108: 681716.
Kremer, M., and D. Chen. 2002. Income Distribution Dynamics with Endogenous
Fertility. Journal of Economic Growth 7: 227258.
Lagerloff, N. P. 2000. From Malthus to Modern Growth: The Three Regimes Re-
visited. Mimeo., Sydney.
Lam, D., and S. Duryea. 1999. Effects of Schooling on Fertility, Labor Supply and
Investments in Children, with Evidence from Brazil. Journal of Human Resources 34:
160192.
Maoz, Y. D., and O. Moav. 1999. Intergenerational Mobility and the Process of
Development. Economic Journal 109: 677697.
Moav, O. 2001. Cheap Children and the Persistence of Poverty. CEPR Discussion
Paper No. 3059.
Moav, O. 2002. Income Distribution and Macroeconomics: The Persistence of In-
equality in a Convex Technology Framework. Economics Letters 75: 187192.
Morand, O. F. 1999. Endogenous Fertility, Income Distribution, and Growth. Journal
of Economic Growth 4: 331349.
Mookherjee, D., and D. Ray. 2000. Persistent Inequality. Mimeo., Boston University.
Nelson, R. R. 1956. A Theory of Low-Level Equilibrium Trap in Underdeveloped
Economies. American Economic Review 46: 894908.
Piketty, T. 1997. The Dynamics of the Wealth Distribution and Interest Rate with
Credit-Rationing. Review of Economic Studies 64: 173189.
Psacharopoulos, G. 1994. Returns to Investment in EducationA Global Update.
World Development 22: 13251343.
Psacharopoulos, G., J. Valenzuela, and M. Arends. 1996. Teacher Salaries in Latin
America: A Review. Economics of Education Review 15: 401406.
Quah, D. T. 1997. Empirics for Growth and Distribution: Stratication, Polarization,
and Convergence Clubs. Journal of Economic Growth 2: 2759.
Rosenzweig, M. R., and K. I. Wolpin. 1980. Testing the Quantity-Quality Fertility
Model: The Use of Twins as a Natural Experiment. Econometrica 48: 227240.
Tamura, R. 1996. From Decay to Growth: A Demographic Transition to Economic

Growth. Journal of Economic Dynamics and Control 20: 12371261.
Veloso, F. A. 1999. Wealth Composition, Endogenous Fertility and the Dynamics of
Income Inequality. Mimeo., University of Chicago.
4 Human Capital
Formation, Income
Inequality, and Growth
Jean-Marie Viaene and

Itzhak Zilcha
4.1 Introduction
It has been established in many studies by economists (and sociolo-

gists) that education plays a signicant role in shaping the income
distribution and the growth process. We observe in recent decades
increasing awareness of governments in the education process and,
consequently, in enhancing investments to promote human capital
skills. In recent years, as information technology advances and com-
putors are being integrated into the learning technology, we are
witnessing some important technological progress in the process of
human capital formation. In this chapter we investigate the effects
of various kinds of technological improvements on growth and the
intragenerational distribution of income.
Education/training lies in the heart of our model, and it is com-
posed of two parts: the parental role that takes place at home,
mainly during the period of youth, and the out of home school-
ing, or the public part that, in most cases, is provided by the gov-
ernment and inuenced by the environment. Home education is
provided by the close family, and it is carried out through parental
tutoring, social interaction, learning devices available at home (such
as computors), and so forth. In this case the human capital of parents
and the time they dedicate to teaching/tutoring play an important
role. The government in our economy has two main tasks: rst, in
organizing the public provision of education and determining the
level of public schooling and, second, in nancing the public pro-
vision of education via taxes on wage income. We do not attempt in
this chapter, except in our numerical simulations, to study the pro-
cess that determines the level of public schooling, but rather take
it as given in each period. Clearly, given the initial distribution of
90 Jean-Marie Viaene and Itzhak Zilcha
human capital (and of income), some democratic process will lead

to certain decisions, based on the principle that education is pro-
vided equally to the younger generation, while the taxes paid by
each individual to nance public education depend on his level of
income.
We consider an overlapping generations economy that produces a
single good using two types of production factors: physical capital
and human capital. It starts at date 0 with some given initial distri-
bution of human capital and physical capital stock. Due to invest-
ments in human capital of the younger generation, the economy
exhibits endogenous growth. Each individual lives for three periods:
the youth period in which no economic decisions are made but
education is acquired, the working period where this individual
earns wage income, and the retirement period in which only con-
sumption takes place. Intergenerational transfers in our economy
take place only in the form of investment, made by parents, in edu-
cating their offspring and in the provision of public education.
When looking at the effects of technological changes in human
capital formation, we nd that in some cases a more equal intra-
generational income distribution coincides with higher output, while
in other cases certain technological improvements enhance growth
but make income distribution less equal. Basically, an important
result of this work is to point out that the way in which technological
progress affects the process of human capital accumulation matters.
If improvements occur mainly in home education, we nd that
growth increases while inequality in income distributions increases.
On the other hand, when the technological improvement affects
mostly public education, then we witness higher growth but less
inequality in income distribution.
The remainder of the chapter is organized as follows. The next
section reviews some related literature. Section 4.3 presents a process
of human capital formation that is part of an OLG model with
altruistic heterogenous agents and characterizes the equilibrium of a
closed economy. Numerical simulations illustrate the properties of
the model. Section 4.4 studies and simulates the effects of changes in
educational technology and externalities on growth and intragen-
erational income distributions. Section 4.5 presents numerical simu-
lations of our dynamic general equilibrium model when majority
voting determines the level of public schooling. Section 4.6 spells out
Human Capital Formation 91
the policy implications of the chapter, and the appendix contains

proofs to facilitate the reading.
4.2 Related Literature
Endogenous growth models have attracted tremendous attention in

economics in the last two decades. As was demonstrated in various
ways in the literature, they provide an extremely efcient analytical
tool in studying issues related to growth, convergence, and distri-
bution of income in equilibrium (see, e.g., Loury 1981; Becker and
Tomes 1986; Lucas 1988; Azariadis and Drazen 1990; Tamura 1991;
Glomm and Ravikumar 1992; Eckstein and Zilcha 1994; Fischer and
Serra 1996; Eicher 1996; Fernandez and Rogerson 1998; van Mar-
rewijk 1999; Galor and Moav 2000; Viaene and Zilcha 2002a). A cen-
tral feature in all these studies is the way in which the evolution
process of human capital is modeled. This process is complex since
the accumulation of human capital or skills depends not only on
parents, the environment, teachers, schools, and investment in
education, but also on technology and culture. However, the pro-
duction function for human capital used in economic models con-
centrate, for tractability reasons, on very few parameters (see, e.g.,
Jovanovic and Nyarko 1995). Like that part of the literature, pro-
duction in our framework is constrained by education and work
experience. Our model in the stationary state is an AK-type endoge-
nous growth model where all variables grow at the same rate as
effective labor. The advantage of our OLG framework is that, in
contrast to the existing literature, it allows for a comparison, period
by period, of nonstationary competitive equilibria.
Statistical ofces of international organizations compile extensive
lists of indicators that describe and compare educational achieve-
ments across countries (see, e.g., OECD 1997). While these features
vary from country to country and thus there may not be a single
theory that characterizes all the observed developments, two main
common elements have inspired our framework of analysis. First,
the production function for human capital exhibits the property that
individuals from below-average families have a greater return to
human capital investment derived from public schooling than those
from above-average human capital families. Also, the effort, and
therefore cost, of acquiring human capital for the younger generation
is smaller for societies endowed with relatively higher levels of

human capital (see, e.g., Tamura 1991; Fischer and Serra 1996). Sec-
ond, parental tutoring plays an important role. For example, Glaeser
(1994) divides the educations positive effects on economic growth
into parts and concludes that children in families with educated
parents seem to obtain a better education than do those children
without that supportive context. Also, Burnhill, Garner, and Mc-
Pherson (1990) nd that parental education inuences entry to
higher education in Scotland over and above the inuence of paren-
tal social class. A reason that is put forward is that parental edu-
cation elicits more parental involvement at home. An important
difference between our process of human capital accumulation and
most cases discussed in the literature is the representation of private
and public inputs via time in the production of human capital. Our
approach suggests that the time spent learning, coupled with the hu-
man capital of the instructors, and not the expenditures on educa-
tion, should be the relevant variables in this process. This distinction
is important since in a dynamic framework the cost of nancing a
particular level of human capital uctuates with relative factor
rewards.
There is some analogy between the objectives of our chapter and
those analyzed in Eicher (1996). The latter looks at the endogenous
absorption of new technologies into production on endogenous
growth and the wage of skilled relative to nonskilled labor. While
technological change is exogenous in our model, we have a contin-
uum of skills that provides insight into how technological change
inuences the equilibrium income distribution, partly through in-
centives to acquire human capital. Unlike Eicher (1996), individ-
uals do not invest in their own human capital. With compulsory
schooling in mind, it seems that the acts of training are not fully
decided by the young generations.
Income distribution is another key economic issue, and its impor-
tance is forcing economists and policymakers to improve their
understanding of its underlying determinants. Evidence of a rise in
income inequality has been observed in a large number of OECD
countries. Some believe that social norms are crucial determinants of
earnings inequality (e.g., Atkinson 1999; Corneo and Jeanne 2001). In
contrast, there is a widely held belief that this rise is driven by events
like progress in information technology and integration of world
trade and nancial markets. The role of human capital accumulation
on income distribution was thoroughly studied by many researchers

in various contexts (see, e.g., Loury 1981; Becker and Tomes 1986;
Galor and Zeira 1993; Fernandez and Rogerson 1998; Viaene and
Zilcha 2002a). Others have shown great interest in the impact of
income inequality on economic systems. For example, it was shown
by Glomm and Ravikumar (1992) that majority voting results in a
public educational system as long as the income distribution is neg-
atively skewed. Cardak (1999) strenghens this result by considering
a voting mechanism where the median preference for education
expenditure, rather than median income household, is the decisive
voter. There is also the popular claim that income inequality is
harmful to economic growth. Some empirical ndings indicate in-
deed that the conjecture of a negative effect holds (see, e.g., Pers-
son and Tabellini 1994). More recent evidence differs, however,
depending on the sample period, on the sample of countries and on
whether time-series or cross-section estimation techniques are used
(see, e.g., Forbes 2000), a fact that is also obtained in our theoretical
work.
4.3 The Model
4.3.1 Human Capital Formation
Consider an overlapping generations economy with a continuum of

consumers in each generation, each living for three periods. During
the rst period, each child gets education but takes no economic
decision. Individuals are economically active during a single work-
ing period, which is followed by the retirement period. At the
beginning of the working period, each parent gives birth to one
offspring. An agent is characterized by his or her family name
o A 0; 1, population is normalized to unity. Denote by W the set of
families in each generation: W is time independent since we assume
no population growth. Denote by m the Lebesgue measure on W.
Agents are endowed with two units of time in their second period:
one is inelastically supplied to the labor market, while the other is
allocated between leisure and time invested in generating human
capital of the offspring. The motivation for parental tutoring is the
utility parents derive from the future lifetime income of their child.
Besides self-educating their own child, parents also pay (by taxes)
for formal education, to enhance the human capital of their child.
Consider generation tnamely, all individuals o born at the outset

of date t, denoted Gt and denote by ht1 o the level of human
capital of family os child. We assume that the production function
for human capital is composed of two components: informal educa-
tion provided by the parents, and public education provided by
teachers and the social environment. Informal education depends
on the time allocated by the parents to this purpose, denoted by
et o, and on the quality of tutoring represented by the parents
human capital level ht o. The time allocated to schooling by the
public education system is denoted by egt , and we assume that the
human capital of the teachers determines the quality of this con-
tribution to the formation of human capital. We assume that, for
some constants b1 > 1, b2 > 1, u > 0 and h > 0, a familys human
capital evolves as follows:
ht1 o b1 et ohtu o b2 egt hth ; 1

where the average human capital of teachers is the average human
capital of generation t, denoted ht . This can be justied if we assume
that the individuals engaged in education in each generation, called
teachers, are chosen randomly from the population of that gener-
ation. The parameters u and h measure the intensity of the external-
ities derived from parents and societys human capital, respectively.
The constants b 1 and b2 represent how efciently parental and public
education produce human capital: b 1 is affected by facilities and the
environment at home, while b2 is affected by facilities, the schooling
system, the neighborhood, social interactions, organization, and so
forth. A similar human capital formation process to this one has been
used in Eckstein and Zilcha (1994).
The assumption that teachers have the average level of human
capital has a number of implications for our analysis. On the one
hand, it allows a feedback to occur between the rest of the model
and teacher quality, an element of complication. On the other hand,
it leads to a simplication in that the tax rate on labor is equal to
time allocated to schooling by the public education. To see that,
consider the lifetime income of individual o, denoted by yt o. Since
the human capital of a worker is observable and constitutes the only
source of income, it depends on the effective labor supply
yt o wt 1 tt ht o; 2
where wt is the wage rate in period t and tt is the tax rate on labor
income.1 Under the public education regime taxes on incomes
nance the costs of educating the young generation. Making use of
(1) and (2), balanced government budget means

wt egt ht dmo tt wt ht o dmo;
W W
or equivalently,
egt tt ; 3
that is, the tax rate on labor is equal to the proportion of the econ-
omys effective labor used for public education.2
4.3.2 Equilibrium
Production in this economy is carried out by competitive rms that

produce a single commodity, using effective labor and physical cap-
ital. This commodity serves for consumption and also as an input
in production. There is a full depreciation of the physical capital.
The per capita human capital in date t, ht (not including the human
capital devoted to formal education), is an input in the production
process. In particular, we take the aggregate production function to
be
qt Fkt ; 1 egt ht ; 4
where kt is the capital stock and 1 egt ht 1 tt ht is the effective

human capital used in the production process. F ; is assumed
to exhibit constant returns to scale, it is strictly increasing, concave,
and continuously differentiable and satises Fk 0; 1 tt ht y,
Fh kt ; 0 y, F0; 1 tt ht Fkt ; 0 0.
Each agent o at time t maximizes the following lifetime utility:
max ut o c1t o a1 c2t o a2 yt1 o a3 1 et o a4 ; 5
et ; st
subject to
c1t o yt o st o b 0; 6
c2t o 1 rt1 st o: 7
Given the optimum for et ; st in (5)(7), the following equilibrium

conditions hold:
wt Fh kt ; 1 egt ht ; 8
1 rt Fk kt ; 1 egt ht ; 9

kt1 st o dmo; 10
W
where income yt o is dened by (2) and human capital ht1 o

is given by (1). The a i s are known parameters and a i > 0 for i
1; 2; 3; 4; c1t o and c2t o denote, respectively, consumption in rst
and second period of the individuals life; st o represents savings;
leisure is given by 1 et o; 1 rt is the interest factor at date t.
The offsprings income, given by yt1 o, enters parents preferences
directly and represents the motivation for parents tutoring and for-
mal education expenditure. Equation (6) is individual os budget
constraint. Equations (8) and (9) are the clearing conditions on factor
markets. Condition (10) is a market-clearing condition for physical
capital, equating the aggregate capital stock at date t 1 to the
aggregate savings at date t.
After substituting the constraints, the rst-order conditions that
lead to the necessary and sufcient conditions for optimum are
c1t a1
; 11
c2t a2 1 rt1
a4 b a3 1 tt1 wt1 htu o

1 ; if et o > 0; 12
1 et o yt1 o
b if et o 0: 13
The last equation allocates the unit of nonworking time between
leisure and the time spent on education by the parents. The latter,
et o, increases with the parents human capital htu and the wage,
net of taxes, at the future date. Equation (12) establishes a negative
relationship between types of education, that is, public education
substitutes for parental tutoring as tt1 increases. Hence, for each
individual there exists a particular value of the tax rate such that
et o 0. This is obtained when the marginal utility of leisure is
larger than the net future wage received from a marginal increase in
the human capital of the younger generation as a result of parental
tutoring. From (6), (7), and (11), we also obtain

a1
c1t o yt o; 14
a1 a2

a2
st o yt o: 15
a1 a2
It is useful to derive the evolution of human capital from the rst

order conditions. Making use of (12), the human capital of a dynasty
given by (1) can be rewritten as follows:

a3
ht1 o b1 htu o b 2 tt hth : 16
a3 a4
Dene the growth factor of aggregate labor as

ht1 h o dmo
gt 1 1 W t1 : 17
ht W ht o dmo
Substitution of (16) in (17) gives us an alternative expression for gt :

u
a3 h o dmo
gt b 1 W t b2 tt hth1 : 18
a3 a4 W ht o dmo
It is clear from (18) that the growth factor of effective labor is the
sum of two terms, one representing the contribution of parental
tutoring, the other the contribution of public education. While the
latter is inuenced by the tax rate the former depends upon the dis-
tribution of human capital at each date.
4.3.3 Numerical Simulations
The aim of this section is to introduce a dynamic computable general

equilibrium model with heterogenous agents and to characterize
the properties of the equilibria of the model discussed so far. In par-
ticular, we are interested in establishing the relationship between
changes in some parameters, and the growth and distribution of
income that can be sustained in equilibrium. To facilitate the inter-
pretation of our theoretical results, the rst set of numerical simu-
lations assumes that the sequence of tt is exogenously given. Later in
section 4.5, we allow for the tax rate to be endogenously determined
through majority voting.
In our numerical examples, we replace (4) by the Cobb-Douglas

production function qt Akty 1 tt 1y ht1y , that is, wt A1 y
kt =1 tt ht y and 1 rt Ay1 tt ht =kt 1y . In the baseline case,
we assume that the economy is in a steady state. To characterize the
latter, consider equations (2), (10), (15), and the Cobb-Douglas pro-
duction function to obtain
kt1 1 ya2
1 rt : 19
kt ya1 a2
Making use of (17) and of the expression for the rental rate:
y
kt1 Aa2 1 y 1y 1 kt
1 tt gt ; 20
ht1 a1 a2 ht
which describes the dynamic path of the capital-labor ratio of the

economy. In the long run, kt1 =ht1 kt =ht is a constant k=h if tt t
and gt g. The time independence of g can be obtained by incorpo-
rating externalities that yield constant returns to scale to parents
and societys human capital in (1), namely, assuming u h 1. In
that case, we obtain the long-run capital-labor ratio from (20):

k a2 1 yA 1=1y
1 t : 21
h ga1 a2
From the previous equations, we obtain the expression for long-

run output and income growth:

qt1 W yt1 o dmo a2 1 yA h1 t 1y
:
qt W yt o dmo a1 a2 k
Substituting (21) in this last expression gives
qt1
g:
qt
Long-run economic growth coincides with the growth factor of

effective labor g, regardless of initial conditions. Our model in the
stationary state is therefore an AK-type endogenous growth model
where all variables grow at the rate g 1.
Besides u h 1, we assume that the other baseline parameters
are k1 70:019, t 0:2, a1 a2 a4 1, a3 2, A 4, y 0:3, and
b1 b2 1:6. We consider a discrete number of heterogenous fami-
lies, namely 11, with a human capital at t 1 taking the values
1; 2; . . . ; 8; 11; 14; 16. The initial endowments in physical and human

capital were chosen with three criteria in mind. First, the values of
the endogenous variables that follow from these initial conditions
and parameter values are long-run values at all dates. Second, the
initial heterogeneity in human capital calibrates an exact Gini coef-
cient close to the European average, namely 0.309 in period 0. Third,
the distribution of human capital is negatively skewed, a fact that is
observed in many countries. The median lies therefore to the left of
the mean. The following formula for the Gini coefcient is used:
1 X n X n
gt j yi yj j; 22
2n 2 yt i1 j1
where n represents the number of families, yt is average income, yi

and yj are individual incomes.
Given the set of baseline parameters of the model, the equilibrium
path of all variables belonging to a particular family is obtained
in two steps. First, the human capital of any individual at date t
is given by (16). Aggregating the levels of human capital across
individuals and equating the aggregate capital stock at date t to the
aggregate savings at date t 1 (see (10)), we obtain aggregate pro-
duction qt , the equilibrium wt and 1 rt . Upon this information,
each individual derives his or her income yt o from (2) and sum-
mary statistics like the Gini coefcient can be computed. Second,
given the time path of wages, marginal returns to physical capital,
and income of each dynasty, each individual can compute et o,
c1t o, c2t o, and ut o.
Table 4.1 presents the solution for our baseline case and the equi-
librium corresponding to a 10 percent increase in each parameter of
the utility function. Changes in other parameters are emphasized in
the next section. In the numerical simulations, given the chosen
parameters, we solve the model for two hundred periods. As pat-
terns emerge within twenty periods we discard the last 180 periods
and compute the relevant statistics averaging over the rst ten peri-
ods and over the second ten periods.
A feature of the baseline in column 1 is the decreasing inequality
among dynasties. Though families start in period 0 with very differ-
ent endowments, they tend to be similar after twenty periods. This
shows the strength of public education relative to parental tutoring
in the accumulation of human capital. This result is obtained even
Table 4.1
Baseline and parameters of the utility function
(1)
a1 a2 1
a3 2 (2) (3) (4) (5)
a4 1 a1 1:1 a2 1:1 a3 2:2 a4 1:1
Relative factor 0.471 0.504 0.442 0.492 0.451

returns 1 rt =wt 0.471 0.505 0.441 0.493 0.450
Gini coefcient gt 0.155 0.155 0.155 0.155 0.155
0.025 0.025 0.025 0.025 0.025
Growth rate (%) 28 27.7 28.3 31.8 24
(aggr. output) 28 28 28 32 23.9
Parental education et
Poorest agent 0.550 0.550 0.550 0.578 0.521
0.596 0.596 0.596 0.621 0.570
Richest agent 0.622 0.622 0.622 0.646 0.598
0.605 0.605 0.605 0.630 0.580
Note: Column 1 is the baseline scenario assuming t 0:2, a1 a2 a4 1, a3 2,
A 4, y 0:3, b1 b 2 1:6, u h 1. Each row reports the average over the rst ten
periods and the average of the second ten periods.
though families have different degrees of parental tutoring as indi-

cated by the last two rows. Changes in the parameters of the utility
function do not affect the income distribution results. Though they
modify individual income levels, these are modied in the same
proportion as all families share the same utility function. Results
in columns 2 and 3 can be best explained by referring to (14) and
(15). While an increase in a1 is conducive to less savings and more
current consumption, an increase in a2 leads to the reverse. While an
increase in a1 leads to somewhat lesser growth in the short run and
higher rental rates when compared to the baseline solution, the
opposite occurs in column 3. Columns 4 and 5 contrast stronger
altruistic preferences with stronger preferences for leisure, respec-
tively. It is important to note the marked differences in growth rates
and parental education. More altruism leads to higher levels of hu-
man capital via more parental efforts in education and ultimately to
higher long-run growth rates. The opposite occurs with a higher a4 .
4.4 Income Distribution and Growth
The focus of this section is to consider the inequality in the intra-

generational income distribution, in equilibrium, and relate it to the
various parameters of our dynamic model. At the same time, we

wish to explore the relationship between inequality and growth. Our
explanation is based on the extent of efciencies and externalities in
the process of human capital accumulation.
We use the relations that we derived in the previous section to
obtain an expression for income at date t 1, yt1 o. To that end,
isolate yt1 o in (13) and make use of (1), (2), and (3) to obtain

a3
yt1 o 1 tt1 wt1 b1 htu o b2 egt hth : 23
a3 a4
Equation (23) determines income at the future date in terms of the

net wage at date t 1, the parents and societys level of human
capital at date t, the current education input tt egt , and the exter-
nalities in education. Note that in this framework there is no direct
dependence of incomes across generations.
We use a denition to compare distribution functions. Let X and
W be two random variables with values in a bounded interval in
y; y and let mx and mw denote their respective means. Dene
X^ X=mx and W ^ W=mw . Denote by Fx and Fw the cumulative dis-
tribution functions of X ^ and W^ , respectively. Let a; b be the smallest
interval containing the supports of X ^ and W^.
Denition Fx is more equal than Fw if, for all t A a; b,

t
a Fx s Fw s ds a 0.
Thus, Fx is more equal than Fw if Fx dominates in the second-degree

stochastic dominance Fw . This denition, due to Atkinson (1970), is
equivalent to the requirement that the Lorenz curve corresponding
to X is everywhere above that of W. We say that X is more equal than
^ and W
W if the cumulative distribution function (c.d.f.) of X ^ satisfy:
Fx is more equal than Fw . Henceforth the relation X is more equal
than W is denoted X g W. We say that X is equivalent to W; and
denote this relation by X A W, if X g W and W g X.
4.4.1 Initial Conditions
Consider two similar economies that differ only in the initial distri-
butions of human capital: One economy has higher levels of human
capital but the same inequality of human capital distribution. Can
we compare the equilibrium intragenerational income distributions
of these two economies over time? The next proposition provides an

answer.
Proposition 1 Consider two economies that differ only in their
initial human capital distributions, h0 o and h0 o. Assume that
h0 o > h0 o for all o, but h0 o A h0 o, namely, these two dis-
tributions have the same level of inequality. Then, the equilibrium
from h0 o will have less unequal intragenerational income distri-
butions at all dates t, t 1; 2; 3; . . .
Proof See the appendix.

This result indicates that a country that starts with higher levels of
human capital, not necessarily more equal, has a better chance to
maintain more equality in its future income distributions. A larger
endowment of human capital increases the contribution of public
education to human capital relative to that of parental tutoring. This
evens the family attributes and reduces income inequality because,
as public education is common to all, below-average families have
a greater return to public schooling than above-average families.
In general, policy and parameter changes of this model will reduce
(increase) income inequality when the weight of public education in
human capital formation increases (decreases).
4.4.2 Public Education
Throughout this section, we assume that public provision of educa-

tion is determined by the government, say, by elections or other
social decision mechanism, and it is equal to egt in date t and
nanced by taxing labor income at a xed rate tt > 0. From now on
we assume that v a 1 and that h a 1. Let us consider the variation
over time of the inequality in the distribution of income.
Proposition 2 If the same tax rate applies to all periods, the in-
equality in intragenerational income distribution at date t 1 is
smaller than the inequality at date t.

Column 1 of table 4.1 indicated already that under the assumption
that the tax rate is the same for all levels of income, inequality
declines over time. The inequality in income distribution at date t 1
is indeed smaller than the inequality in income distribution at date t

and in the limit families tend to the same level of human capital and
income. Let us show now that a higher provision of public education
reduces inequality in the distribution of income in each generation.
Proposition 3 In the previous economy, let h0 o be the initial
human capital distribution. Increasing the public provision of edu-
cation results in less inequality in the intragenerational income dis-
tribution in each date.

This result may not be surprising since public education is pro-
vided equally to all young individuals (of the same generation),
while it is nanced by a at tax rate on wage income. However, its
importance lies in the fact that it is proved in equilibrium and that it
holds in all future periods. It is also clear from (18) that, when
u h 1, more public education contributes to a higher long-run
growth rate of effective labor.
These results are quantied in the two columns of table 4.2 where
tt takes two values, 0.20 and 0.22, respectively. Besides increasing
the long-run growth rate of output and decreasing the inequality in
the income distribution, table 4.2 conrms the substitution among
education types. Public education crowds out parental tutoring
though the elasticity computed at steady-state values is about 0.1
and thus quite small.
Table 4.2
Baseline and public education
(1) (2)
t 0:2 t 0:22
Relative factor returns 0.471 0.493
1 rt =wt 0.471 0.495
Gini coefcient gt 0.155 0.145
0.025 0.020
Growth rate (%) 28 29.7
(aggr. output) 28 30.1
Poorest agent 0.550 0.543
0.596 0.590
Richest agent 0.622 0.616
0.605 0.598
4.4.3 Efciencies and Externalities
Consider some technological change that affects the production

of human capital. We say that the provision of public education
becomes more efcient if, in the human capital process (1), b2 =b1
becomes larger without lowering neither b1 nor b2 .3 We say that the
private provision of education becomes more efcient if, in the process
(1), b 1 =b 2 becomes larger while neither b1 nor b2 declines. Likewise, a
technological improvement in the production of human capital is
said to be neutral if the ratio b2 =b1 remains unchanged while both
parameters increase. Let us consider now the effects of each type of
technological improvement in the education process on intragenera-
tional income inequality.
Proposition 4 Consider the previous economy. A technological
improvement in the production of human capital, given by equation
(1), results in
a. If public provision of education becomes more efcient, the
intragenerational distribution of income becomes less unequal in all
periods.
b. If private provision of education becomes more efcient, income
inequality becomes larger in all periods.
c. If the technological improvement is neutral, the inequality in
income distribution remains unchanged at period 1 but declines for
all periods afterward.

Let us consider now another type of a change in the home-
component of the production of human capital and its economic
implications in equilibrium. Observe the process represented by
(1). Let us vary the parameters v and h, which relate to the role
played by human capital of the parents or the environment. Since
we assume that v a 1 and h a 1, let us consider the effect that
lower values will have on the inequality in income distributions in
equilibrium.
Proposition 5 Consider the process of production of human capital

given by (1). Then
a. Comparing two economies which differ only in this parameter v,

we nd that the economy with the lower v will have less inequality
in the intragenerational income distribution in all periods.
b. Comparing two economies which differ only in the parameter h,
we nd that the economy with the lower value of h will have more
inequality in the income distribution in all periods.

Let us consider now the effect that technological improvement in
the production of human capital will have on output in equilibrium.
Consider (1) and remember that we call the rst term on the RHS,
b 1 et ohtu o, the home component, and the second term, b 2 egt hth , the
public component. Now we prove:
Proposition 6 Consider the human capital production process
given by (1) and the following types of technological improve-
ments:
a. Increasing b1 , or increasing v or both, will increase output in all

dates.
b. Increasing b 2 , or increasing h or both, will result in higher output
in all periods.
If we consider the computor information revolution as a techno-
logical improvement in enhancing knowledge, then we ask whether
the home component benets more than the public component in the
formation process of human capital. We believe that computors and
the Internet have enhanced home education considerably, while
schools benet only in a limited manner. The following corollary
may provide some explanation to the recent widespread phenomena
(mostly during the 1990s) that in the OECD countries, economic
growth is accompanied by increasing inequality in the distribution
of income.
Corollary 7 (a) In the following two cases of technological im-

provement in the home component, we obtain higher economic
growth coupled with more inequality in the distribution of income:
(i) an increase in b1 , (ii) an increase in v.
(b) In the following two cases of technological improvement in the
public component, we obtain higher economic growth coupled with
Table 4.3
Baseline and other specication externalities
Externalities (1) (2) (3) (4) (5)
u 1 0.9 1 1.1 1
h 1 1 0.9 1 1.1
Relative factor returns 0.471 0.366 0.445 0.816 0.513
1 rt =wt 0.471 0.345 0.430 3.29 0.581
0.025 0.003 0.050 0.327 0.008
Growth rate (%) 28 7.1 23.0 86.9 35.9
(aggr. output) 28 2.9 20.0 large 48.1
Poorest agent 0.550 0.550 0.570 0.548 0.518
0.596 0.578 0.621 0.633 0.527
Richest agent 0.622 0.598 0.636 0.649 0.597
0.605 0.580 0.632 0.661 0.533
less inequality in the distribution of income: (i) an increase in b2 , (ii)

an increase in h.
In terms of results, it is remarkable that both cases of technological

improvement yield similar predictions on growth but opposite on
income distribution. In this regard, table 4.3 adds that inequality as
measured by Gini coefcients is more sensitive to externalities aris-
ing from the home component than from those arising from the
public part of human capital formation. Decreasing returns in par-
ents human capital (column 2) reduce inequality substantially, all
individuals becoming equal in the long run. In contrast, substantial
income inequality is observed with increasing returns (column 4).4
Columns 2 and 4 establish a positive correlation between growth
and income inequality. In column 2, decreased inequality is obtained
at the expense of growth, whether measured in terms of income or
human capital (not shown), and vice versa in column 4. In contrast,
when looking at changes in h, the correlation between growth and
income inequality is negative as indicated in columns 3 and 5.
Table 4.4 looks at a technological improvement in human capital
formation represented here by rises in the bs. Columns 2 to 4 show
that a greater efciency in education is always conducive to growth
while hardly affecting income distributions. A comparison of col-
umns 2 and 3 shows the stronger impact that parental education has
on output growth.
Table 4.4
Baseline and other specication efciency
Efciency (1) (2) (3) (4)
b1 1.6 1.76 1.6 1.76
b2 1.6 1.6 1.76 1.76
Relative factor returns 0.471 0.526 0.482 0.537
1 rt =wt 0.471 0.528 0.483 0.540
Gini coefcient gt 0.155 0.166 0.145 0.155
0.025 0.031 0.020 0.025
Growth rate (%) 28 38.2 30.0 40.2
(aggr. output) 28 38.7 30.1 40.8
Poorest agent 0.550 0.556 0.543 0.550
0.596 0.601 0.590 0.596
Richest agent 0.622 0.627 0.616 0.622
0.605 0.612 0.598 0.605
4.5 Majority Voting
Though there is a growing awareness of governments in education,

enhancing human capital skills require nancial resources to cover
the investment. Though the majority of constituents recognize the
importance of learning, they are not prepared to contribute nan-
cially via income taxes in the same way. To establish the preferences
of each individual with respect to tt o let us compute the reduced-
form utility of each agent. Substituting the rst-order conditions in
(5), lifetime utility of agent o can be rewritten as
ut o Wt 1 tt o a1 a2 1 tt1 o a3 b1 ht o v b2 tt ohth a3 a4
24
where Wt groups all parameters and variables like factor rewards
that are given to atomistic individuals. Knowing that each agent
cannot enforce any tax rate at the future date, namely, tt1 o is
given to him, the maximization of (24) with respect to tt o gives
a3 a4 a1 a2 b1 ht on
tt o : 25
a1 a2 a3 a4 a1 a2 a3 a4 b2 hth
Each agent chooses the optimal tt o such that the cost of cur-
rent spending on education (in terms of foregone current and future
consumption) is equal to the reward of a marginal increase in the
human capital of their children. It is clear that the heterogeneity in
Table 4.5
Externalities and median voter
Externalities (1) (2) (3) (4) (5)
u 1 0.9 1 1.1 1
h 1 1 0.9 1 1.1
Tax rate tt 0.222 0.299 0.109 0.101 0.347
0.203 0.323 0.031 0.019 0.460
Relative factor returns 0.496 0.447 0.384 0.684 0.751
1 rt =wt 0.476 0.440 0.315 2.31 1.67
0.021 0.001 0.127 0.586 0.001
Growth rate (%) 30.1 12.4 9.3 80.6 52.8
(aggr. output) 28.6 8.5 10.3 large large
Poorest agent 0.539 0.496 0.602 0.585 0.422
0.595 0.511 0.657 0.641 0.189
Richest agent 0.616 0.556 0.651 0.657 0.519
0.603 0.511 0.662 0.666 0.191
tt o derives from the heterogeneity in human capital. When h a 1

and v a 1 above-average agents are willing to pay a lower tax rate
than below-average agents. In terms of numerical results, the rst
step in our simulations computes a vector of tt o based on (25).
Given this vector of individual preferences for education expen-
diture, we assume that the level of public schooling is obtained at
each date through majority voting. Numerically, majority voting
boils down to identifying the median voters preference for public
schooling.
Tables 4.5 and 4.6 repeat the exercises performed in tables 4.3 and
4.4, now with endogenous public education. What difference does it
make? Baseline is different because the distribution of human capital
being negatively skewed, the median voters human capital lies to
the left of the mean and therefore he or she wishes a higher tax
rate. Tables 4.5 and 4.6 reproduce the substitution in equilibrium
between public education and parental education observed before:
Any increase in tt decreases the time spent on parental education et
and hence, raises leisure. This substitution among types of provision
of education has a number of implications, one of them being that
Gini coefcients vary more. It is important to note that simulation
results conrm the robustness of corollary 7 whose results apply also
to the case of endogenous public education. A positive correlation
Table 4.6
Efciency and median voter
Efciency (1) (2) (3) (4)
b1 1.6 1.76 1.6 1.76
b2 1.6 1.6 1.76 1.76
Tax rate (tt 0.222 0.188 0.254 0.222
0.203 0.166 0.238 0.203
Relative factor returns 0.496 0.516 0.545 0.565
1 rt =wt 0.476 0.490 0.532 0.546
Gini coefcient gt 0.143 0.166 0.124 0.143
0.021 0.037 0.011 0.021
Growth rate (%) 30.1 37.8 35.0 42.6
(aggr. output) 28.6 35.4 34.7 41.4
Poorest agent 0.539 0.560 0.517 0.539
0.595 0.611 0.577 0.595
Richest agent 0.616 0.630 0.600 0.616
0.603 0.622 0.583 0.603
between income inequality and income growth is obtained when

externalities or efciencies arising from parents human capital vary.
In contrast, this correlation is negative when externalities or efcien-
cies in the public contribution to human capital are considered.
4.6 Policy Implications
The chapter has studied the determinants of income distribution and

growth in an overlapping generations economy with heterogenous
households. Heterogenity results simply from the initial nondegen-
erate distribution of human capital across individuals. Both parental
tutoring and public education contribute to human capital accumu-
lation. Theoretical and numerical results from this model are consis-
tent with some key features of the data.
A rst effect analyzed in this chapter is the role of initial condi-
tions. We nd that a country that starts with a higher level of human
capital, not necessarily less equal, has a better chance to maintain
less inequality in the future. Hence, communities that create a cul-
ture of literacy and knowledge are more likely to experience lower
income inequality. In this regard, International Adult Literacy Sur-
veys show that between one-quarter and three-quarters of adults in
a comparison of OECD countries fail to attain a literacy level con-
sidered by experts as a suitable minimum to cope with the demands

of contemporanous societies. Our rst proposition suggests that
these cross-country differences in literacy skills may be important
determinants of the observed cross-country differences in income
inequality.
There is an important role for public education. Under the as-
sumption that the tax rate that nances education is positive and
xed over time, inequality declines over time. Increasing this tax rate
in an attempt to enhance the provision of public education results in
less income inequality. These theoretical results do not depend on
the levels of public educations, fegt g. However, most governments
are concerned with the optimality of the resources invested in edu-
cation (see, e.g., Viaene and Zilcha 2002b). In economic theory, the
choice of some optimal level of public education requires some
social welfare function that, due to the heteregeneity of households,
may not represent a political equilibrium. Instead, the majority vot-
ing criterion is widely used and one can determine this level by
using the median voters optimal choice. Our numerical simulations
make use of this concept to endogenize public education. It follows
that the median voters preferred tax rate to nance public schooling
is increasing in (1) the negative skewness of the distribution of hu-
man capital, (2) the externality derived from societys human capi-
tal, (3) the efciency of public schooling, and (4) the parameters of
altruism and leisure in the utility function. The prefered tax rate is
decreasing in (1) the externality derived from familys human capital,
(2) the efciency of parental tutoring, and (3) the parameters for
current and future consumption in the utility function.
The claim that income inequality is harmful to economic growth
has been tested in a number of empirical studies. While the rst evi-
dence indicated that the conjecture of a negative sign indeed holds,
more recent empirical studies nd the effect to be ambiguous, a fact
that is also obtained in our work. If technological improvements in
the human capital accumulation process occur mainly in home edu-
cation, both growth and income inequality increase. In contrast, if
the technological improvement affects mostly public education then
higher growth and less income inequality are obtained. In terms
of results, it is remarkable that both types of technological changes
yield dissimilar predictions on income inequality. This creates chal-
lenges for policymakers because independent policies affecting pa-
rental education only cannot serve two masters at the same time
while those affecting public education can. For example, policies

aimed at the use of computers and telecommunication in homes (via
tax deductions) magnify differences in family attributes while the
same policies in schools diminish them.
The information revolution of the last decade is said to lead to a
digital divide. This phenomenon means a split of a countrys popu-
lation between those without and those with Internet access, which
are mainly the better-educated groups. A divide may also arise
globally as the Internet density is higher in industrial countries than
elsewhere. In both cases, this may enhance the existing intragenera-
tional socioeconomic disparities within the worlds population. In
our framework, a divide is obtained when human capital formation
is characterized by increasing returns to scale in familys human
capital. In this case, income inequality increases (see column 4 of
tables 4.3 and 4.5): Some families experience an ever increasing
human capital and some face a development trap. That is, the latter
have too low an endowment of human capital and, as a result, ex-
perience a continuous decrease in their human capital over time. The
existence of a development trap gives a role for public education to
offset this gap. Numerically, we can show that, even in the presence
of externalities in familys human capital, there exists a level of
public education sufcient to lead to a drop in income inequality.
Many industrialized countries are experiencing an increase in
income inequality. Using Gini coefcients, there is evidence of a
sharp rise for the United States, the United Kingdom, and New Zea-
land. Increased inequality is not uniform across the OECD as some
countries like France and Canada experience decreasing inequality.
Our framework offers a simple explanation that focuses mainly on
education technology and education policy: It depends on the size
of the public component of education relative to the home contribu-
tion. For example, increased inequality is explained by the decline
through time of the relative contribution of public education where-
by family attributes gain a larger role. As indicated in tables 4.24.4,
this could be the result of numerous factors or any combination of
them. Likely explanations are the lower quality of public education
due to lower scholastic standards, lower quality of teachers, lower
efciency or disorganization of the schooling system, and techno-
logical improvement in parental education. As for the digital divide,
another possible explanation is the presence of increasing returns to
scale in familys human capital that, if combined with increases in
public education, can duplicate the erratic pattern of inequality

shown in the data of some countries.
Our model makes some specic and simplifying assumptions, and
it is therefore subject to the robustness issue. First, the production
function for human capital is additive in the contributions of paren-
tal tutoring and public education. Though this specication is quite
standard, there is hardly no empirical test of the assumed functional
form. It follows that both types of education are substitutes and that
our theoretical results on inequality depend on the weight of public
education (common to all), relative to parental tutoring (which is
family-specic). Second, it is important to note that introducing
intergenerational monetary transfers in our model will modify the
results. In such a case, we lose the one to one relationship between
the distribution of income and the distribution of human capital.
Technological progress in the aggregate production function may
have different effects on the intragenerational income distributions
(see Karni and Zilcha 1994). Likewise, the framework can be gener-
alized by introducing an additional redistributive measure by the
government, such as social security. This may also vary some of our
conclusions.
4.7 Appendix
Proof of Proposition 1 Consider two equilibria in which human cap-

ital accumulation is described by (1). Variables under the second
equilibrium are marked by *. Let us rewrite equation (23) for both
equilibria:

b
yt1 o Ct htu o 2 egt hth ;
b1

u b2 h
yt1 o Ct ht o egt ht
b1
where Ct and Ct are positive constants. Since h0 and h0 are equally
distributed, the same holds for h0v o and h0 o v , since v a 1.
Moreover, since h0 < h0 we obtain that h1 o is more equal than
h1 o (see Lemma 2 in Karni and Zilcha 1994). It is easy to verify
from (18) that h1 o are lower than h1 o for all o. In particular, we
obtain that h1 o v is more equal than h1 o v (see Shaked and
Shanthikumar 1994, Theorem 3.A.5). Also we have h1 h < h1 h . This

implies, using (16), that h2 o is more equal than h2 o. It is easy
to see that this process can be continued to generalize this to all
periods. r
Proof of Proposition 2 Let us show rst that in each generation indi-
viduals with higher level of human capital choose at the optimum
higher level of time to be allocated for private education of their off-
spring. To see this, let us derive from the rst-order conditions,
using some manipulation, the following equation:

b 1 a4 a4 b 2
1 1 et o egt hth hu
t o; 26
a3 a3
which demonstrates that higher ht o implies higher level of et o.

Let us show that such a property generates less equality in the
distribution of yt1 o compared to that of yt o. It is useful how-
ever, to apply (16) for this issue. In fact it represents the period t 1
income yt1 o as a function of the date t income yt o via the
human capital evolution. Dene the function Q : R ! R such that
Qht o ht1 o using (16). This monotone-increasing function
satises: Qx > 0 for any x > 0 and Qx=x is decreasing in x.
Therefore (see Shaked and Shanthikumar 1994), the human capital
distribution ht1 o is more equal than the distribution in date t,
ht o. This implies that yt1 o is more equal than yt o. r
Proof of Proposition 3 Let us consider equation (16) for t 0. Since

h0 o is given, h0v o and h0 are xed. By raising eg0 the distribution
of the human capital for generation 1, h1 o becomes more equal.
This follows from Lemma 2 in Karni and Zilcha (1994). Moreover,
we claim from (16) that the average human capital in generation 1
increases as well. Increasing eg0 will result in higher h1 o for all o
and higher level of h1 . Moreover, it also implies that h1v o will have
a more equal distribution (see Shaked and Shanthikumar 1994, Theo-
rem 3.A.5).
Now, let us consider t 1. Increasing eg1 will imply the following
facts: h1v o becomes more equal and b 2 eg1 h1h is larger than its value
before we increase the levels of public education. Using (16) and the
same lemma as before, we obtain that h2 o becomes more equal.
This process can be continued for t 3; 4; . . . ; which establishes our
claim. r
Proof of Proposition 4 Let the initial distribution of human capital

h0 o be given. Compare the following two equilibria from the same
initial conditions: one with the human capital formation process
given by (1), and another with the same process but b2 is replaced by
a larger coefcient b 2 > b2 . Clearly, we keep b1 unchanged. Let us
rewrite equation (23) as follows:

b
yt1 o Ct htu o 2 egt hth ;
b1

u b 2 h
yt1 o Ct ht o egt h t ;
b1
where Ct and Ct are some positive constants. Since h0 o is xed at

date t 0 we nd (using once again Lemma 2 from Karni and Zilcha
1994) that b2 =b1 > b2 =b1 imply that y1 o is more equal than y1 o.
We also derive that h1 o are lower than h1 o for all o and, hence,
h1 < h1 . By (16), using the same argument as in the last proof, hv 1 o
is more equal than h1v o and b2 =b1 eg1 hh 1 > b 2 =b e h
1 g1 1
h
, hence
h2 o is more equal than h2 o. This same argument can be con-
tinued for all dates t 3; 4; 5; . . . ; which completes the proof of part
(a) of this proposition. The proof of part (b) follows from the same
types of arguments using the fact that if b1 < b1 then b2 =b1 > b2 =b1
and, hence, h1 o is more equal than h1 o and h1 > h1 . This process
leads, using similar arguments as before, to yt o more equal than
yt o for all periods t. Consider now the claim in part (c). From
(16) we see that inequality in the distribution of h1 o remains
unchanged even though all levels of h1 o increase due to this tech-
nological improvement. In particular, h1 increases. Now, since in-
equality of h1v o did not vary but the second term in the RHS of (16)
has increased due to the higher value of h1 , we obtain more equal
distribution of h2 o. Now, this argument can be used again at dates
3; 4; . . . ; which completes the proof. r
Proof of Proposition 5 Assume, without loss of generality, that

h0 o b 1 for all o. Since the two economies have the same initial
distribution of human capital h0 o the process that determines h1 o
differs only in the parameter v. Denote by v < v a 1 the parameters,

then it is clear that h0 o v is more equal than h0 o v since it is
attained by a strictly concave transformation (see Shaked and Shan-
thikumar 1994, Theorem 3.A.5). Likewise, the human capital distri-
bution h1 o is more equal than the distribution h1 o. This implies

that y1 o is more equal than y1 o. Now we can apply the same

argument to date 1: The distribution of h1 o v is more equal than
that of h1 o v ; hence, using (16) and Shaked and Shanthikumar

(1994), we derive that the distribution of h2 o v is more equal than
that of [h2 o v . This process can be continued for all t.
When we lower the value of h, keeping all other parameters con-
stant, we basically lower the second term in (16), h0 h , while h0 o v
remains unchanged. By Lemma 2 in Karni and Zilcha (1994), we
obtain that the distribution of h1 o becomes less equal. This can be
continued for t 2 as well since it is easy to verify that h1 h
decreases while h1 o v becomes less equal. This process can be
extended to t 2; 3; . . . ; which complete the proof. r
Proof of Proposition 6 Let us just sketch the proof of this claim. Any
technological improvement, either in the public component or the
home component, will imply higher human capital stock as of period
1 and on. Since, the initial capital stock is given this will increase the
output in date 1 and, hence, the aggregate savings in this period.
Thus the output in date 2 will be higher and hence the capital stock
to be used as well. This process continues in all coming periods. r
Notes
We are grateful to the participants of the CESifo conferences on Growth and Inequal-
ity: Issues and Policy Implications for contributing remarks. We also wish to express
our thanks to an anonymous referee and to T. Eicher, H. Kierzkowski, H. de Kruijk,
R. Riezman, H.-W. Sinn, O. Swank, and S. Turnovsky for numerous suggestions.
Research assistance by D. Ottens is gratefully acknowledged.
1. The heterogeneity of consumers stems from the heterogeneity of income. As wt and

tt are common to all agents, (2) clearly indicates that heterogeneity of incomes derives
from the distribution of human capital across individuals.
2. In contrast, under a decentralized system, both tt o and egt o are decision vari-
ables of agents and the individuals budget constraint on private education is
tt owt ht o wt egt oht ;
where the level of teachers instruction is chosen freely from the market but their
average human capital is the same as the economys. Aggregate resources invested in
education then become

1
egt o dmo tt oht o dmo;
W ht W
which depend upon the distribution of human capital in each date. This is not the case
under public education.
3. A growing empirical literature has given much attention to efciency of public

education on pupils current and later achievements. One issue that has been high-
lighted is the causal effect of class size on human capital. For example, Lindahl (2001)
nds that smaller classes in Sweden generate higher educational attainments.
4. Externalities that yield increasing returns to scale to parents human capital, that is
u > 1, have been observed in China (Knight and Shi 1996) and are therefore not a mere
theoretical curiosum.
References
Atkinson, A. B. 1970. On the measurement of inequality. Journal of Economic Theory 2:

244263.
Atkinson, A. B. 1999. Is rising income inequality inevitable? A critique of the Trans-
atlantic Consensus. UNU/WIDER Publication WAL3.
Azariadis, C., and A. Drazen. 1990. Threshold externalities in economic development.

Quarterly Journal of Economics 105: 501526.
Becker, G. S., and N. Tomes. 1986. Human capital and the rise and fall of families.
Journal of Labor Economics 4(3): S1S39.
Burnhill, P., C. Garner, and A. McPherson. 1990. Parental education, social class and
entry to higher education 197686. Journal of the Royal Statistical Association, Series A
153(2): 233248.
Cardak, B. A. 1999. Heteregeneous preferences, education expenditures and income
distribution. The Economic Record 75(228): 6376.
Corneo, G., and O. Jeanne. 2001. Status, the distribution of wealth, and growth. The
Scandinavian Journal of Economics 103(2): 283293.
Eckstein, Z., and I. Zilcha. 1994. The effects of compulsory schooling on growth,
income distribution and welfare. Journal of Public Economics 54: 339359.
Eicher, T. S. 1996. Interaction between endogenous human capital and technological
change. Review of Economic Studies 63: 127144.
Fernandez, R., and R. Rogerson. 1998. Public education and income distribution:
A quantitative dynamic evaluation of education-nance reform. American Economic
Review 88(4): 813833.
Fischer, R. D., and P. J. Serra. 1996. Income convergence within and between countries.
International Economic Review 37(3): 531551.
Forbes, K. J. 2000. A reassessment of the relationship between inequality and growth.
American Economic Review 90(4): 865887.
Galor, O., and O. Moav. 2000. Ability biased technological transition, wage inequality
and growth. Quarterly Journal of Economics 115: 469497.
Galor, O., and J. Zeira. 1993. Income distribution and macroeconomics. Review of Eco-
nomic Studies 60: 3552.
Glaeser, E. L. 1994. Why does schooling generate economic growth? Economics Letters
44(3): 333337.
Glomm, G., and B. Ravikumar. 1992. Public versus private investment in human
capital: Endogenous growth and income inequality. Journal of Political Economy 100:
818834.
Jovanovic, B., and Y. Nyarko. 1995. The transfer of human capital. Journal of Economic
Dynamics and Control 19: 10331064.
Karni, E., and I. Zilcha. 1994. Technological progress and income inequality: A model
with human capital and bequests. In The Changing Distribution of Income in an Open
U.S. Economy, ed. J. H. Bergstrand et al., 279297. Amsterdam: North-Holland.
Knight, J., and L. Shi. 1996. Educational attainment and the rural-urban divide in
China. Oxford Bulletin of Economics and Statistics 58(1): 83117.
Lindahl, M. 2001. Home versus school learning: A new approach to estimating the
effects of class size on achievement. Discussion Paper No. 261. IZA (Bonn).
Loury, G. 1981. Intergenerational transfers and the distribution of earnings. Econo-
metrica 49(4): 843867.
Lucas, R. 1988. On the mechanics of economic development. Journal of Monetary Eco-
nomics 22: 342.
OECD. 1997. Education at a Glance: OECD Indicators. Paris: OECD.
Persson, T., and G. Tabellini. 1994. Is inequality harmful for growth? American Eco-
nomic Review 84(3): 600621.
Shaked, M., and J. G. Shanthikumar. 1994. Stochastic Orders and Their Applications.
Boston: Academic Press.
Tamura, R. 1991. Income convergence in an endogenous growth model. Journal of
Political Economy 99: 522540.
van Marrewijk, C. 1999. Capital accumulation, learning and endogenous growth.
Oxford Economic Papers 51: 453475.
Viaene, J.-M., and I. Zilcha. 2002a. Capital markets integration, growth and income
distribution. European Economic Review 46(2): 301327.
Viaene, J.-M., and I. Zilcha. 2002b. Public education under capital mobility. Journal of
Economic Dynamics and Control 26: 20052036.
III European Transition and
Inequality
5 Social Transfers and
Inequality during the
Polish Transition
Michael P. Keane and

Eswar S. Prasad
5.1. Introduction
Beginning in the late 1980s and early 1990s, all the countries of the
former Soviet Union and Eastern Europe have, to a greater or lesser
extent, undertaken a process of transition away from central plan-
ning toward more market-oriented economic systems. However,
different countries have chosen very different approaches to the
process of transition. The outcomes, in terms of macroeconomic and
social indicators, have been quite diverse as well. In general, how-
ever, most countries of the former Eastern bloc have experienced
poor growth performance and large increases in income inequality
during the transition process. The most obvious success story in the
process of transition to date has been Poland, which has outstripped
other transition economies in terms of growth performance, while at
the same time experiencing only modest increases in inequality.
Hence, a detailed analysis of the Polish experience, with a view
toward asking what the country did right, is of considerable interest.
There has been a lively debate on the efcacy of alternative
market-oriented reform strategies, both in terms of the sequencing
and magnitude of reforms (see, e.g., Aghion and Blanchard 1994;
Dewatripont and Roland 1996). Poland pursued a strategy of rapid
and decisive reform that has often been referred to as shock ther-
apy. This process began in August 1989January 1990, a period that
has become known as the big bang. Price controls on food were
lifted in August 1989 and those on most other products were lifted in
January 1990. Numerous other macroeconomic and microeconomic
reforms, including restraints on credit for state-owned enterprises,
the hardening of their budget constraints, and the opening up of
122 Michael P. Keane and Eswar S. Prasad
the economy to import competition, were also instituted during this

period.
One area in which the Polish reforms lagged behind those of many
other transition economies was the privatization of state-owned
enterprises (SOEs). Nevertheless, as argued by Pinto, Belka, and
Krajewski (1993), even the SOEs that lagged behind in terms of
changes in ownership and governance did undertake signicant
adjustment in response to hard budget constraints and import com-
petition. Thus, the transformation of Poland from a command econ-
omy to a market-oriented economy was quite rapid by any measure.
Conrming this, Poland has consistently ranked among the top
reformers in terms of the various indicators used by the European
Bank for Reconstruction and Development (EBRD) in its annual
Transition Report.1
Poland, like other former Eastern bloc countries, experienced sub-
stantial declines in output and employment in the early phase of
transition.2 Between 1989 and 1992, the cumulative decline in output
was about 20 percent and the ofcial unemployment rate rose from
near zero to about 14 percent (table 5.1). Since then, however, the
growth performance of Poland has been among the best of the tran-
sition economies. Real GDP in 1999 was 22 percent higher than in
1989. In contrast, only a few other transition economies (including
Albania, the Czech Republic, Hungary, the Slovak Republic, and
Slovenia) managed to keep output to within a few percent, above or
below, their pretransition levels. Most other transition economies
have still to recover to anywhere close to their pre-transition peaks.
How did Poland succeed in rapidly instituting durable market-
oriented reforms despite falling income and rising unemployment?
A widely held view is that the process of transition has led to sub-
stantial increases in income inequality in almost all the transition
economies, thereby complicating the process of reform (see, e.g.,
Aghion and Commander 1999). How, then, was it possible to main-
tain broad-based support for market-oriented reforms in Poland?
In this chapter, we use microdata from the Polish Household
Budget Surveys (HBS) to show that, in contrast to most other transi-
tion economies, the increase in income inequality in Poland during
the transition was actually quite modest. In fact, our preferred esti-
mate of the Gini coefcient for the overall individual income distri-
bution actually declined from 0.256 in 1988 to 0.230 in 1992. It then
Table 5.1
Selected macroeconomic indicators for Poland
(annual percentage changes, unless indicated otherwise)
1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Real GDP 4.2 2.1 4.0 0.3 11.4 7.0 2.6 3.8 5.2 7.0 6.1 6.9 4.8 4.1 5.0
Consumer price index 16.5 26.4 60.2 251.1 585.8 70.3 43.0 35.3 32.2 27.8 19.9 14.9 11.8 7.3 9.9
(annual average)
Employment (end-year) 0.3 0.0 1.0 0.8 6.2 3.9 3.1 1.7 1.1 0.3 3.5 1.3 1.4 1.5 6.7
Unemployment rate (%) 0.1 6.1 11.8 14.3 16.4 16.0 14.9 13.2 8.6 10.5 13.0 15.1
(end-year)
Levels of real GDP in 1999 (1989 100) for selected transition economies
Poland 122 Czech Republic 95
Slovenia 109 Albania 95
Slovak Republic 100 Uzbekistan 94
Hungary 99 Belarus 80
Note: Dashes indicate data are not available.

Sources: IMF (1994) and EBRD Transition Report (various years).
Social Transfers and Inequality during the Polish Transition
123
began a gradual increase, reaching levels comparable to the pre-

transition period in 19941996 and then rising to 0.276 by 1997.
To put this overall increase of 0.02 in context, it is actually
rather modest, being only two-thirds the magnitude of the increase
observed in the United States in the 1980s (see, e.g., Atkinson, Rain-
water, and Smeeding 1995). It is far smaller than the average in-
crease of 0.09 reported by Milanovic (1998) for eighteen Eastern bloc
countries. Indeed, our cross-county analysis (see section 5.5 below)
suggests that Poland had the smallest increase in inequality of any
transition country. Also, note that this still leaves Poland with a Gini
value closer to those of the Scandinavian countries (around 0.25)
than that of the United States (0.41). (See World Bank 2000.)
We nd that social transfers played a critical role in mitigating the
increase in income inequality during the Polish transition. In contrast
to overall income inequality, inequality in labor earnings increased
steadily and substantially during the transition period of 19891997.
For instance, we estimate that the Gini measure of inequality for
individuals in worker-headed households, based only on the labor
earnings of those households, increased steadily from 0.252 in 1988,
the last full year prior to the transition, to 0.298 in 1997. Thus, the
increase in the Gini coefcient for labor earnings (0.046) was more
than twice that of the Gini for overall income (0.020).
By attenuating the rise in overall income inequality that might
have been generated by the increase in earnings inequality, social
transfers may have contributed to maintaining the social and politi-
cal cohesion that was essential for the reform process. In this regard,
it is interesting that social transfers also played a key role in the
evolution of between-group income dynamics. A marked increase in
the generosity of public sector pensions in 1991 led to a substantial
exit of older workers from the labor force onto the pension rolls in
19911992 and improved the relative income position of pensioner-
headed households. At the same time, other social transfers were
increased from 3 percent in 1989 to about 5 percent of GDP by 1992.
As Dewatripont and Roland (1996) point out, such large pensions
and social transfers can be rationalized as necessary to achieve initial
political support for the big bang reform strategy. Generous pen-
sions, in particular, may be efcacious (in the short run) for reform
since they serve a dual purpose: rst, paying off the older workers
whose earnings capacity seems to have been most adversely affected
by the reforms, and second, enabling enterprises to shed less pro-
Social Transfers and Inequality during the Polish Transition 125
ductive workers through early retirement. EBRD statistics indicate

that Poland experienced among the most rapid increases in labor
productivity of the transition countries.
Analysis of the targeting of transfers in Poland indicates that a
substantial proportion was directed not toward households at the
bottom of the income distribution but rather toward the middle class
and, as noted earlier, toward older workers. Thus, it appears that the
transfer system could have been better designed if the goal was
purely income support. However, the focus on the middle class and
older workers may have been critical for ensuring social stability and
setting the stage for rapid reforms.
Recent developments in growth theory (see, e.g., Benabou 1996)
suggest mechanisms whereby greater income inequality may be det-
rimental for growth. Thus, aside from the political economy consid-
erations noted earlier, maintenance of a high degree of income
equality may have also enhanced growth in Poland by broadening
the base of individuals with the resources to engage in small-scale
entrepreneurial activity. Indeed, according to the EBRD reports, the
transition in Poland has been characterized (more than in any other
transition country) by an explosion in small-scale entrepreneurial
activity. For instance, Poland had almost 2 million private entrepre-
neurs and 125,000 private commercial enterprises by 1996.
Beginning in 1993, the Polish government reined in the growth
of social transfers. At that point, overall inequality began to rise
gradually. This suggests that the drop in income inequality from
19891992 and the rise from 1993 onward did not follow as a direct
consequence of the transition process, but may have been driven
largely by scal policy choices. Thus, it appears that an increase
in inequality is not a necessary consequence of successful transi-
tion, but, rather, that inequality dynamics during transition may be
strongly inuenced by policy choices.
The outline of the chapter is as follows: Section 5.2 describes the
data and methods used in our analysis. Section 5.3 presents our
results on changes in inequality during the Polish transition, and
the role of social transfers in the evolution of inequality. Section
5.4 provides a cross-country perspective on the relationship among
transfers, inequality, and growth in the transition countries. Section
5.5 relates the analysis in this paper to the broader literature on
the relationship between inequality and growth, and section 5.6
concludes.
5.2 Data and Methods
The empirical analysis in this chapter is based on the Household

Budget Surveys (HBS), a representative sample of Polish households
conducted by the Polish Central Statistical Ofce (CSO).3 The CSO
has been collecting detailed microdata on household income and
consumption for more than two decades. Households were surveyed
for a full quarter (until 1992) or for a full month (from 1993 onward)
in order to monitor their income and spending patterns. Supple-
mentary information, including household demographics, is collected
from the same households once every year. The typical sample size
is about 25,000 households per year.
The HBS contains detailed information on sources and amounts of
income for both households and individuals within each household.
Total income is broken down into four main categories: labor income
(including wages, salaries, and nonwage compensation); pensions;
social benets and other transfers; and other income. Social benets
include income from unemployment benets that were introduced in
late 1989. A key point is that the data include measures of the value
of in-kind payments from employers to workers, which have been
an important part of workers compensation in Poland and other
transition economies. For farm households, farm income and expen-
ditures, as well as consumption of the farms produce, are also re-
ported. There were no taxes on personal income until 1992. After
that year, we use net incomes in the analysis. The HBS also contains
detailed information on demographic characteristics of all household
members.
We use the aggregate consumer price index (CPI) as the price
deator. Since there were large price changes in the early years of
transition, we match the price data to the survey period for each
observation by using quarterly CPI data for 19851992 and monthly
data for 19931997.
The change from quarterly reporting to monthly reporting in 1993
has serious consequences for cross-sectional inequality measure-
ment. Since income and consumption tend to be more volatile at the
monthly compared to the quarterly frequency, this change could
result in an exaggeration of any increase in inequality. In Keane and
Prasad (2002a), we develop a technique for adjusting the 19931997
income and consumption data for the increased variability that may
be attributable solely to the shift from quarterly to monthly report-
ing. The basic idea of our approach is to assume that income consists
of a permanent or predictable component (determined by education,
age, and other observable characteristics of households and their
members) plus a mean zero idiosyncratic component. We rst re-
gress household income on a large set of controls; these regressions
were run separately for each quarter from 19851992 and for each
month from 19931997. We then assume that the variance of the
idiosyncratic component would not have jumped abruptly between
the fourth quarter of 1992 and the rst month of 1993, since, to our
knowledge, no dramatic policy changes or exogenous shocks oc-
curred at that point in time. Rather, we assume that the variance of
the idiosyncratic component varies smoothly over time (measured
in months) according to a polynomial time trend. We estimate this
polynomial trend, along with a dummy for the post-1992 period that
captures the discrete jump in variance that occurred with the change
to monthly income reporting.
The second-stage estimation, where the standard deviations of
the income residuals from the rst-stage regressions are used as the
dependent variable, is done separately for households with differ-
ent primary income. The results, shown in table 5.2, indicate that
the adjustment factor ranges from 0.369 for farmers to 0.044 for
Table 5.2
Regressions using income residuals
Workers/
Workers Farmers farmers Pensioners
Time 0.001 0.006 0.006 0.000
(0.002) (0.006) (0.006) (0.001)
Time 2 /10 3 0.000 0.188 0.171 0.002
(0.000) (0.092) (0.092) (0.023)
Time 3 /10 6 0.001 0.919 0.859 0.009
(0.010) (0.365) (0.364) (0.095)
Mean real 0.207 0.126 0.116 0.367
income (0.014) (0.013) (0.021) (0.027)
Dummy for 0.165 0.369 0.201 0.044
9397 (0.028) (0.102) (0.098) (0.026)
Adjusted R 2 0.88 0.65 0.51 0.85
Number of 92 92 92 92
observations
Note: The dependent variable is the log standard deviation of the residuals from
equation (1). Standard errors are reported in parentheses.
pensioner-headed households.4 Once we have obtained these ad-

justment factors, we then scale down the idiosyncratic component
(the residuals from the rst-stage regressions) of the post-1992 in-
come data for each household to eliminate this jump in variance.
Failure to account for the change in survey frequency may have
caused prior studies to greatly overstate the increase in inequality in
Poland. For instance, based on statistics computed by the CSO,
OECD (1997, 86) reports that the Gini for Poland rose by 0.02 points
between 1992 and 1993 alone, which is as large as the increase we
nd for the entire transition period (see section 5.3.1).5
Prior studies of changes in inequality in Poland have often relied
on the aggregate data on quantiles of the income distribution pub-
lished by the CSO in the annual publication Budzety Gospodarstw
Domowych, which we henceforth refer to as the Surveys. Unfortu-
nately, the aggregate income statistics reported by the CSO, as well
as those reported by other former communist countries, differ in a
number of important ways from economically meaningful measures
of income. The ofcial statistics appear to reect total revenues or
inows since they include loans, dissaving, and cash holdings at
the beginning of the survey period. For farmers, income includes
gross farm revenues, rather than net revenues. This is an important
issue as approximately one-fth of Polish households are either
farm households or mixed worker-farmer households. Access to the
detailed microdata enables us to make important adjustments in
order to obtain a more meaningful measure of income (by excluding
non-income revenue items and by calculating net farm income).6
Both our procedure for adjusting for the spurious increase in
inequality stemming from the switch to the monthly reporting inter-
val, and our corrections for the denitions of income and consump-
tion, rely on access to the HBS microdata.7 In particular, the variance
correction requires access to the data for an extended period of time.
Our study is unique in that it is based on the HBS microdata for a
long sample period, extending from ve years prior to the big
bang to eight years after. To our knowledge, no prior study of
inequality in Poland has adjusted for the change in survey design in
1993, and most have not adjusted for the denitional problems noted
earlier.8
Table 5.3 reports sample means for some of the variables used
extensively in our analysis of inequality.9 Two interesting features
are that the average share of income from transfers and the share of
pensioner-headed households increase markedly after the transition.

We discuss this in greater detail in what follows. The demographic
characteristics of households and household heads remain quite
stable during and after the transition. The means of the education
dummies indicate a small increase in average levels of educational
attainment of household heads in the 1990s (a similar increase occurs
in the general population as well).
5.3 Changes in Inequality during the Transition
In this section, we examine various aspects of inequality in Poland

over the period 19851997. For the years 19931997, we use the
income and consumption measures that are adjusted (using the pro-
cedure described in the previous section) for the increase in idiosyn-
cratic variance that occurred with the shift to a monthly reporting
period.
The measures of inequality we examine are based on the distribu-
tion of individual income, unless explicitly noted otherwise. A key
problem in inequality measurement is how to account for household
composition and household economies of scale when measuring
household well being, or when assigning individual income or
consumption levels to household members. In Keane and Prasad
(2001a), we constructed food share (FS)based equivalence scales for
Poland using the Engel (1895) method, which assumes that two
households with different demographic composition are equally well
off at income levels that enable them to have equal food shares (ratio
of expenditure on food to total expenditure on nondurables).10 In
Keane and Prasad (2002a), we document that our results on the evo-
lution of inequality are not sensitive to the choice of equivalence
scale.11 Hence, in what follows, we present results using only the
food sharebased equivalence scale.12
5.3.1 Measures of Overall Inequality
We rst examine the evolution of summary measures of overall

inequality.13 Table 5.4 reports Gini coefcients based on total in-
comes adjusted by the FS equivalence scale. According to the Gini,
inequality increased from 0.256 in 1988 to 0.263 in 1989, but then
declined in 19901992. In fact, by 1992, the Gini had declined to
0.230, which is below the pretransition level.
Table 5.3
130
Household budget surveys: Sample means for selected years

1988 1989 1990 1991 1992 1993 1995 1997
Real household income (shares)
Labor income 0.52 0.53 0.51 0.49 0.49 0.50 0.52 0.56
Transfers 0.23 0.22 0.26 0.32 0.34 0.33 0.33 0.32
Farm income 0.18 0.19 0.16 0.12 0.12 0.11 0.11 0.08
Other income 0.06 0.05 0.06 0.06 0.05 0.07 0.05 0.05
Real household consumption (shares)
Durables 0.13 0.14 0.11 0.10 0.08 0.08 0.08 0.10
Nondurables 0.87 0.86 0.89 0.90 0.92 0.92 0.92 0.90
Food 0.45 0.46 0.53 0.47 0.44 0.43 0.41 0.38
Household characteristics
Urban 0.51 0.51 0.51 0.52 0.64 0.66 0.65 0.67
Number of persons in household 3.27 3.27 3.24 3.16 3.14 3.15 3.18 3.12
Primary income source of household
Workers 0.55 0.55 0.53 0.50 0.49 0.44 0.42 0.42
Farmers 0.10 0.10 0.10 0.09 0.09 0.08 0.08 0.06
Mixed, worker-farmers 0.07 0.07 0.07 0.07 0.06 0.06 0.06 0.06
Pensioners, others 0.28 0.28 0.30 0.34 0.36 0.38 0.39 0.40
Self-employed 0.05 0.06 0.06
Michael P. Keane and Eswar S. Prasad
Household head characteristics
Male, 1830 0.11 0.10 0.10 0.10 0.10 0.10 0.10 0.11
Male, 3160 0.58 0.59 0.57 0.57 0.57 0.59 0.59 0.58
Male, >60 0.13 0.14 0.14 0.14 0.14 0.13 0.13 0.13
Female, 1830 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01
Female, 3160 0.09 0.09 0.09 0.09 0.09 0.09 0.09 0.09
Female, >60 0.08 0.08 0.08 0.09 0.09 0.08 0.08 0.08
Age 47.54 47.78 47.90 48.30 48.45 47.96 48.03 48.09
College degree 0.07 0.06 0.06 0.07 0.08 0.09 0.09 0.09
Some college 0.00 0.00 0.00 0.00 0.00 0.01 0.01 0.01
High school 0.20 0.19 0.20 0.21 0.23 0.24 0.24 0.26
Some high school 0.01 0.01 0.02 0.01 0.01
Basic vocational training 0.31 0.33 0.33 0.33 0.33 0.34 0.36 0.35
Primary school 0.34 0.34 0.32 0.32 0.30 0.28 0.26 0.25
Primary school not completed 0.07 0.06 0.05 0.05 0.04 0.04 0.03 0.02
Number of observations ( households)
1985 21,560 1989 29,366 1992 10,642 1995 31,874
1986 25,475 1990 29,148 1993 31,966 1996 31,782
1987 29,510 1991 28,632 1994 31,942 1997 31,659
1988 29,287
Note: The components of income and consumption are shown as (mean) shares of total income and consumption, respectively. Dashes indicate
data are not available.
131
Starting in 1993, however, inequality began to rise and, by 1997, it

was at a level signicantly higher than the peak attained in 1989. It is
important to note, however, that the increase in inequality even by
1997 is hardly dramatic. The increase of 0.020 in the Gini coefcient
from 1988 (the year before the transition) to 1997 is smaller than the
increase of 0.03 reported for the United States in the 1980s by Atkin-
son, Rainwater, and Smeeding (1995), or the increase from 0.326 to
0.361 reported for the United Kingdom from 1986 to 1991 in World
Bank (1999, 2000).
Next, we examine inequality based on income net of transfers
(table 5.4, second row).14 Interestingly, this reveals a very different
picture. The Gini coefcient for income excluding transfers increased
by 0.066 from 1988 to 1997, more than three times the increase in the
Gini for overall income.
Table 5.4 also reports Gini coefcients for labor income for worker
headed households only. These show that inequality in labor earn-
ings increased substantially during the transition (i.e., by 0.046
points between 1988 and 1997), with much of the increase taking
place in the early phase. Thus, we see that inequality in labor earn-
ings grew substantially more than inequality in the overall income
distribution.
Together, these results suggest that social transfers played a cru-
cial role in inequality dynamics during the Polish transition. Speci-
cally, they appear to have substantially mitigated the increase in
income inequality that might otherwise have occurred, given the
substantial increase in inequality in labor earnings. We examine the
magnitude and targeting of social transfers in more detail later.
Gini coefcients for consumption inequality, based on either total
or nondurables consumption, show a pattern similar to those for
income inequality. Thus, to conserve space, we focus only on income
inequality in much of the analysis that follows.
Next, we examine whether our main results are sensitive to the
choice of inequality measure. The Gini coefcient is known to be
particularly sensitive to changes around the median of the distribu-
tion. The coefcient of variation (and its monotonic transforms, one
of which we use here) is more sensitive to changes at the high end
of a distribution, while the mean logarithmic deviation is more sen-
sitive to changes near the low end. We report these alternative
inequality measures in the bottom rows of table 5.4, in order to
determine if they tell a consistent story. In fact, they do. When we
Table 5.4
Poland: Measures of overall inequality
1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997
Gini coefcients
Total income 0.252 0.254 0.246 0.256 0.263 0.250 0.235 0.230 0.248 0.262 0.255 0.265 0.276
Income excluding transfers 0.373 0.375 0.368 0.385 0.384 0.389 0.404 0.416 0.416 0.437 0.432 0.448 0.451
Nondurables consumption 0.196 0.200 0.205 0.211 0.219 0.209 0.208 0.205 0.222 0.228 0.222 0.227 0.235
Total consumption 0.230 0.234 0.239 0.244 0.258 0.241 0.233 0.227 0.247 0.254 0.247 0.262 0.271
Gini coefcients for worker-headed households
Labor Income 0.237 0.243 0.240 0.252 0.262 0.268 0.278 0.289 0.285 0.292 0.288 0.295 0.298
Half the square of the coefcient of variation
Total income 0.085 0.090 0.085 0.091 0.105 0.086 0.079 0.077 0.097 0.103 0.096 0.105 0.112
Mean log deviation
Total income 0.075 0.079 0.077 0.078 0.087 0.075 0.071 0.069 0.079 0.086 0.081 0.086 0.093
Note: The inequality measures shown here are for the individual distributions of income and consumption. Household income and consump-
tion are adjusted using the food sharebased equivalence scale and allocated equally to individuals in the household. Income and consumption
data for 19931997 are adjusted for the change in survey frequency.
133
use total income, both these measures of inequality also show an

upward spike in 1989, followed by a decline in 19901992 to be-
low the pretransition level, and a subsequent steady increase dur-
ing 19931997 to a level modestly above that in the pretransition
period.15
When we look at income net of transfers, both the coefcient of
variation (CV) and mean logarithmic deviation show far greater
increases in inequality over the transition period than for total
income. This pattern is particularly interesting in the case of the CV
measure, which is most sensitive to changes at the high end of the
distribution. This result stems from the fact that transfers in Poland
are focused not only at the low end of the income distribution but
extend well into the high end. We give more details on the targeting
of transfers below.
To summarize, we nd no evidence to support the view, based on
aggregate CSO statistics, of a sharp increase in total income inequal-
ity following the transition in Poland. For instance, OECD (1997),
which uses the CSO aggregates, reports a Gini increase of 0.05 for
Poland over the 19891996 period, while we nd essentially no
increase over the same period. Our results also differ markedly in
terms of the timing of changes in inequality. The OECD-CSO gures
imply that inequality grew tremendously from 1989 to 1993, and that
it then stayed rather at through 1996. Our results indicate that
inequality actually fell from 19891992. But we nd that inequality
rose noticeably after 1993 and, especially, in 1996 and 1997. Thus, we
nd that most of the increase in inequality occurred several years
after the big bang, and long after the OECD-CSO gures imply the
increase had already ceased.
This difference in timing also has important implications for the
interpretation of what occurred during the transition. The OECD-
CSO gures for Poland, as well as the comparable gures for other
transition economies (e.g., Milanovic 1999), are often interpreted as
evidence that substantial increases in inequality are an inevitable
concomitant of the process of transition to a market economy. Our
results, however, indicate that the change in overall income inequal-
ity during the rst six years of the transition in Poland was quite
modest. Thus, these results suggest that changes in inequality during
transition are not inevitable but, rather, may result from particular
policy choices.
5.3.2 Winners and Losers in the Transition
We have found no evidence of an increase in overall income

inequality in Poland in the immediate aftermath of the big bang,
regardless of which of several inequality measures we consider.
However, this does not mean that there were not winners and losers
in the transition. We now turn to a discussion of how different
groups fared in terms of relative income.
In Keane and Prasad (2002a), we report how median total income
evolved for four types of households differentiated by main income
source of the household head: workers, farmers, mixed worker-
farmers, and pensioners. Pensioner-headed households had lower
median total income than other groups during the 19851989 period,
but their relative position improved dramatically after the big bang
so as to bring their income up to levels that are comparable the other
groups. The main impetus behind the improved relative position of
pensioners was a substantial increase in pension availability and
generosity that took place in 1991. As a result, we nd that pensions
contributed importantly to a reduction in inequality.16
In addition to the improved relative position of pensioners, other
notable features of the transition were an improvement in the rela-
tive position of more educated workers17 and a deterioration in the
relative position of more experienced (i.e., older) workers. In Keane
and Prasad (2002b), we perform a detailed regression analysis of
individual labor earnings data from the HBS. The results indicate
that the earnings premium for a college degree relative to a primary
school degree increased from 47 percent in 1987 to 102 percent in
1996. But the premium for labor market experience fell sharply in
1990, before recovering to pretransition levels after 1992.
The deterioration in experience premia in the immediate aftermath
of the big bang is consistent with the notion of rapid obsolescence of
rm- or industry-specic skills during a period of rapid technologi-
cal change and industrial restructuring (see Svejnar 1996). This,
combined with the increased generosity of pensions, explains the
surge in the number of pensioner-headed households in 19911992
that is obvious in table 5.3. Indeed, self-selection into retirement
probably accounts for the recovery in experience premia for older
workers that occurred after 1992, since a large number of older
workers, particularly in the 5565 age bracket, retired in 19911992.
Specically, the number of newly granted pensions increased from

about 0.6 million per year in 19881989 to almost 1.4 million in 1991
(OECD 1998, 65).
Since older workers had the most to lose from the privatization
or closure of existing state-owned rms, giving them the option of
moving on to the pension rolls may have been a key factor in re-
moving a potential political obstacle to enterprise restructuring and
privatization. Furthermore, these early retirements may have con-
tributed to the exceptional rise in labor productivity that occurred
in Polish industrial enterprises after 1992 (EBRD 1997).
5.3.3 The Targeting of Transfers
Having discussed the role of transfers in mitigating the rise in over-

all income inequality during the transition, we now turn to a more
detailed analysis of the targeting of transfers.
One simple way to analyze the equalizing role of transfers is to
run nonparametric regressions of transfers on income net of trans-
fers. We ran these regressions separately for each year. Figure 5.1
shows the results for 1991. In this gure, households are sorted
Figure 5.1
Households sorted by income net of transfers
along the X-axis in terms of their level of pretransfer income. The

horizontal line shows median real household income net of trans-
fers in 1991; the vertical line shows the observation at the median
of the 1991 sample. The key point of this gure is that transfers
are not heavily concentrated among households with low pre-
transfer income. Rather, there are substantial transfers even to
households at and even above the median of the pretransfer income
distribution.
This pattern is present not just in 1991, but in every year from 1985
to 1997. However, the targeting of transfers changed in important
ways during the transition. As a way of summarizing these changes,
gure 5.2 shows the fractions of total income accounted for by
transfer income, at different percentile points of the pretransfer
income distribution, for every year from 1985 to 1997. The key
observation from this gure is that ratio of transfer income to total
income increased signicantly in the early years of transition, at
virtually all points of the pretransfer income distribution. At the
median, the average share of transfers in total income rose from
approximately 20 percent in the pretransition period, to about 35
percent by 1992, before stabilizing at around 28 percent thereafter.
Clearly, these results suggest that transfers could have been better
targeted if the goal was purely income support. However, since
individuals in the middle class tend to have a signicantly higher
propensity to vote than individuals at lower income levels, the
extension of transfers higher up into the income distribution that
occurred in 19901992 may have been a mechanism for buying the
social stability that characterized the transition period.
Also noteworthy is the importance of pensions as a transfer
mechanism, and the extent to which transfers were targeted at older
workers. We already noted in section 5.3.2 that pension expenditures
and the size of the pension rolls increased enormously in the early
years of the transition. As shown in table 5.5, total public pension
expenditure as a percent of GDP rose from 8 percent in 19891990 to
almost 15 percent by 1992. The HBS data show a similar pattern,
with the share of total income accounted for by pensions rising from
16 percent in 1989 to 25 percent in 1992. The pension replacement
rate (average pension as a ratio of average wage) rose from about 52
percent in 19881989 to 65 percent in 1991 and remained above 60
percent through 1997 (OECD 1998).
Figure 5.2
(a) Percentiles of pretransfer distribution: 10, 25; (b) Percentiles of pretransfer distri-
bution: 50, 75, 90
The HBS data indicate that, among households headed by a male

in the 5565 age range, the share of pension income in total income
rose from 26 percent in 1989 to 50 percent in 1992, remaining at
around that level through 1997. For households headed by women in
the 5262 age range, this share rose from 34 percent in 1989 to 49
percent in 1992, before declining slightly to 45 percent by 1997.18
These gures show the importance of early retirement during the
transition.
Table 5.5
Social transfers
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997
General government expenditures
(in percent of GDP)
Cash transfers to individuals 9.4 11.2 10.6 17.3 19.9 20.4 20.2 19.7 18.7 19.4
Pensions 7.1 8.2 8.1 12.2 14.8 15.0 14.9 14.5 14.3 14.4
Unemployment benets 0.0 0.0 0.2 1.2 1.7 1.2 1.2 1.2 1.1 1.0
Other benets 2.3 3.0 2.3 3.9 3.4 4.2 4.1 4.0 3.3 4.0
Mean cash transfers ( HBS data)
Total transfers (avg. ratio to 41154 41792 36254 44948 44694 43486 44171 44860 46786 48197
total income) (23.4) (21.8) (26.3) (32.2) (33.6) (31.6) (32.8) (32.7) (32.4) (31.3)
Pensions (avg. ratio to total 29857 30497 27307 33520 33346 33172 34672 36240 38008 40715
income) (17.0) (15.9) (19.8) (24.0) (25.1) (24.1) (25.8) (26.4) (26.3) (26.4)
Other cash benets (incl. UI) 11280 11279 8927 11404 11323 10315 9498 8620 8777 7482
(avg. ratio to total income) (6.4) (5.9) (6.5) (8.2) (8.5) (7.5) (7.1) (6.3) (6.1) (4.9)
General government balance (in 0.0 7.4 3.1 6.5 6.7 2.9 3.0 3.1 3.4 3.1
percent of GDP)
Real GDP (annual % change) 4.0 0.3 11.6 7.0 2.6 3.8 5.2 7.0 6.1 6.9
Note: The data on real GDP and government expenditures are taken from various IMF sources. The gures in the middle panel (mean transfers
in HBS data) are expressed in terms of 1992Q4 prices.
139
5.3.4 Summary
Our results suggest that social transfers, especially pensions, played

an important role in the evolution of income inequality during the
Polish transition. The broad patterns are summarized in gure 5.3.
The top panel plots Gini coefcients for both total income and
labor income (for worker-headed households). We see that inequal-
ity in labor income grew fairly steadily during the entire 19891997
period, but with most of the increase occurring during the early
transition (19891992). Despite this, overall income inequality de-
clined substantially in the early transition phase, and only began
to rise in 1993. The bottom panel of gure 5.3 shows social transfers
as a percent of GDP. The dramatic increase in transfers during the
early transition period was sufcient to actually outweigh the effect
of increased earnings inequality, leading to a decline in overall in-
come inequality up through 1992. However, the growth of transfers
was reined in beginning in 1993. And, from 1993 onward, inequal-
ity does grow substantially. By 1995, inequality had returned to the
pretransition level, and it rose modestly above that level in 1996
1997.
Clearly, the growth in social transfers was crucial in dampening
the rise in overall inequality during the early years of the Polish
transition. Social transfers as a percent of GDP averaged 17.7 percent
during 19901997, the highest level in any transition country. The
mean level of transfers across the eighteen transition economies for
which we could nd data was substantially lower at 10.8 percent
(see section 5.4). This at least partially explains the fact that Poland
had the smallest increase in inequality during the transition.
In fact, Gomuka (1998) refers to a Polish model of transition,
which is distinguished by an exceptionally large volume of social
transfers, especially . . . pensions that helped to reduce the social
cost of reform, but is inhibiting Polands ability to sustain rapid
growth. This themethat the level of transfers in Poland will hin-
der future growthhas been sounded by many authors, including
the International Monetary Fund (1994), World Bank (1995), and
OECD (1997). But such dire predictions have yet to be borne out. In
19982000, Poland continued to experience strong growth, making it
by far the best-performing transition economy in terms of cumula-
tive growth during the transition. As we describe in sections 5.4
Figure 5.3
Gini coefcients
and 5.5, cross-country evidence, along with recent developments in

growth theory, suggest the intriguing possibility that the high level
of social transfers in Poland may in fact have been conducive to
growth.
5.4 A Cross-Country Perspective on Transfers, Inequality and

Growth
In our view, the evidence we provide on transfers and inequality in

Poland is also relevant to the broader literature on inequality, redis-
tribution, and growth. Here, we examine the relationship between
social transfers, inequality, and growth across the Eastern bloc
countries.
The top panel of gure 5.4 plots the relationship between transfers
as a percent of GDP and inequality changes during the transition, as
measured by changes in the Gini coefcient. The gure contains
results for all fourteen countries for which we were able to obtain
data on transfers as well as pre- and post-transition Gini coef-
cients.19 As expected, countries with higher levels of transfers typi-
cally had smaller increases in inequality. The countries with the
highest levels of transfersPoland, Hungary, and Sloveniaalso
had among the smallest inequality changes during transition.20
The bottom panel of gure 5.4 plots the relationship between GDP
growth and government transfers as a percent of GDP, for all eigh-
teen countries for which we were able to obtain data on transfers
and GDP. The X-axis shows cumulative GDP growth in the rst
eight years of transition for each country. The relationship is
strongly positive, with a simple correlation of 0.67. Note that nding
a positive correlation between transfers and growth is particularly
surprising given the blatant denominator bias driving the correlation
in the opposite direction (i.e., higher output growth increases the
denominator of the transfer to GDP ratio).
These ndings raise the intriguing possibility that policies that
foster income equality may also be conducive to successful tran-
sition. In fact, the relationships among transfers, inequality, and
growth that we nd here for transition economies have also been
reported by authors such as Perotti (1996) for a different but much
larger sample of industrial and developing countries. This suggests,
more generally, that policies that foster income equality may be
conducive to growth.
Figure 5.4
Transfers as share of GDP
Of course, the results we have reported only indicate a correlation

rather than a causal relationship. It is possible that some omitted
factor may actually account for good performance on both the
inequality and growth dimensions. However, the patterns we have
noted are at least not inconsistent with recent developments in
growth theory that imply that redistribution to enhance equality
may actually enhance rather than dampen growth.21 In the next sec-
tion, we discuss these theories and relate them to the Polish case.
5.5 A Perspective from New Growth Theory
A traditional view of the relationship between inequality and

growth is that a certain degree of income inequality is conducive to
economic growth. The basic idea is that, in a country at a low level of
development and with imperfect capital markets, wealth must be
concentrated in the hands of a few so they can invest in physical
and/or human capital. Kuznets (1955) presented evidence that
appears consistent with this view. He found a hump-shaped rela-
tionship between inequality and per capita GNP, which he inter-
preted as evidence that inequality increases in the early stages of
development and falls thereafter. It can be further argued that
inequality may foster growth even in more advanced economies,
provided liquidity constraints remain important, because only weal-
thy individuals can bear the sunk costs of starting industrial activ-
ities. Consistent with this view, Evans and Jovanovic (1989) provide
some evidence that capital market constraints affect the decision to
become an entrepreneur even in the United States, a country with
highly developed capital markets.
But the relationship between growth and inequality has been
reexamined by various authors in recent years (see the survey by
Aghion, Caroli, and Garca-Penalosa 1999). Some of this work has
challenged the view that higher inequality is associated with higher
rates of growth. Persson and Tabellini (1994), among others, present
evidence indicating a negative correlation between inequality and
growth.
Recent work in growth theory has rationalized a negative relation
between inequality and growth by invoking either borrowing con-
straints or political economy considerations. For instance, Galor and
Zeira (1993) turn on its head the argument that wealth concentration
encourages growth when there are liquidity constraints. They pre-
sent a model with borrowing constraints in which individual pro-

ductivity is a concave function of human capital and show that re-
distribution of wealth from the rich to the poor enhances growth
because the poor have a higher marginal productivity of investment.
Related results have been obtained by Banerjee and Newman (1993),
Aghion and Bolton (1997), and Benabou (1996).
Turning to the political economy models, Alesina and Rodrik
(1994) show that income redistribution can enhance growth by
reducing political support for taxation of capital. And Perotti (1996)
nds empirical support for the view that redistribution can enhance
growth by fostering sociopolitical stability. In a similar vein, Roland
(1993) and Dewatripont and Roland (1996) argue that transfers can
reduce the desire of potential economic losers to block reforms.
How are these models relevant to the Polish case? Clearly, these
models do not imply that social transfers are themselves directly
benecial for growth, but rather that transfers affect growth through
one of several mechanisms. In the Polish case, the plausible political
economy mechanisms seem fairly obvious. Our analysis of winners
and losers in the Polish transition suggests that older workers were
quite adversely affected, due to declines in experience premia. How-
ever, these workers were cushioned by a substantial increase in the
availability and generosity of pensions. This led many workers in
their 1950s and early 1960s to take early retirement, at generous re-
placement rates. Furthermore, the mass retirements by older workers
in the early phase of the transition had the added benet that they
facilitated enterprise restructuring. This helped Poland achieve bet-
ter increases in labor productivity than most other transition econ-
omies, while at the same time securing the support of pensioners for
market-oriented reforms.
The mechanism whereby social transfers mitigate borrowing con-
straints, thus making it possible for a broader segment of the popu-
lation to invest in human capital, does not seem relevant to the
Polish case. Too little time has passed since the start of the transition
for a substantial increase in the stock of human capital (via new
investment) to have taken place.
On the other hand, generous transfers may have helped reduce
liquidity constraints on small-scale entrepreneurs and the self-
employed. Both of these groups have played a key role in Polands
vibrant economic performance since 1990. Indeed, according to
OECD (1998, 107), Polands recent growth performance rests on
a strong entrepreneurial basis, with many dynamic small and

medium-sized enterprises (SMEs) and creations of new rms . . .
SMEs make up the bulk of Polands 2.2 million registered non-
agricultural enterprises . . . almost 90 percent [are] micro-enterprises
(employing 1 to 5 persons). Indeed, Poland has seen greater growth
in small-scale entrepreneurial activity than has any other transition
country.
But evaluating the importance of mechanisms whereby transfers
mitigate borrowing constraints on entrepreneurs is made difcult by
the potential for intrafamily and intrahousehold transfers. Speci-
cally, the person who receives a transfer may not be the person
whose constraint is relaxed and who can therefore start or expand
entrepreneurial activity. For example, a pension transfer to a retired
older worker may in fact enable his or her children, who might
otherwise have had to expend resources caring for the worker, to
instead acquire capital. Thus, a simple test of whether transfer re-
cipients were likely to become small-scale entrepreneurs would not
shed light on the issue. Further research on mechanisms underlying
the exceptional explosion of small-scale entrepreneurial activity in
Poland is clearly needed.
5.6 Conclusions
Poland is perhaps the greatest success story of transition. It insti-

tuted drastic market-oriented reforms early in the transition process
and has received relatively high marks for the extent of its subse-
quent market-oriented reforms (e.g., in the EBRDs annual Transition
Report). Like other transition countries, Poland experienced a severe
contraction of GDP in the rst few years of transition (19901992).
However, Poland subsequently experienced rapid growth, starting
in 1993. Its GDP stood at 22 percent above the pretransition level
in 1999, which is a substantially better performance than any other
country in the former Eastern bloc.
Another positive aspect of the Polish transition is that it has
resulted in very little increase in overall income inequality. This is in
sharp contrast to most other transition countries, which have typi-
cally seen drastic increases in inequality. Our results indicate that
Poland has experienced among the smallest (if not the smallest)
increase in inequality of any transition country, and that measures
of inequality for Poland remain only modestly above those for the
Scandinavian countries.
Our results further indicate that inequality in labor earnings did
increase substantially during the Polish transition. However, a strik-
ing aspect of government policies during the early years of transition
was a sharp increase in social transfers, from about 10 percent of
GDP to roughly 20 percent. This is the highest level of transfers (as a
share of GDP) of any transition country. We have argued that this
increase in transfers substantially mitigated the increase in overall
income inequality that might otherwise have resulted from increased
earnings inequality.
From these patterns we draw one clear policy conclusion: The
Polish experience makes clear that a substantial increase in income
inequality is not a necessary concomitant of successful transition.
Clearly, it is possible to use high levels of social transfers to maintain
income equality, without hindering the transition process or choking
off economic growth.
A more speculative hypothesis that is not inconsistent with our
ndings is that the generous social transfer policy pursued by
Poland may have actually promoted the transition process and pro-
moted growth. One mechanism through which social transfers
could promote successful transition is by cushioning (or buying
off) groups that would have been most adversely affected by
reforms.
For instance, older workers had the most to lose from the privati-
zation or closure of state-owned rms and would have been most
adversely affected by enterprise restructuring. But, early in the
transition process, Poland substantially increased the availability
and generosity of pensions.22 This induced large numbers of older
workers to take early retirement. This may have facilitated transition
both by removing potential opposition to reforms by a powerful
interest group and by helping to reduce employment at enterprises
to more efcient levels and promoting other aspects of enterprise
restructuring.23
More generally, we found that transfers in Poland have to a great
extent been targeted at the middle class rather than just the poor. A
number of authors have argued, on political economy grounds, that
the development of a solid, property-owning middle class is essen-
tial to the consolidation of capitalism (Kornai 2000). For further
details on these type of arguments, we refer the reader to Havryly-

shyn and Odling-Smee (2000), Easterly (2002), and Keane and Prasad
(2001b).
Recent developments in growth theory stress another mecha-
nism through which social transfers could promote growth. In
environments characterized by liquidity constraints, a more equal
income distribution increases the fraction of the population with the
resources to engage in small-scale entrepreneurial activity. Thus,
social transfers can lead to more equal income distribution, which in
turn leads to more small-scale entrepreneurial activity, which in
turn promotes growth. In Poland, which had almost 2 million pri-
vate entrepreneurs by 1996, small-scale entrepreneurial activity has
clearly been a key engine of growthmore so than in any other
transition country. Further work is needed to determine the extent
to which this exceptional explosion in entrepreneurial activity in
Poland may be attributable to government transfer policies.
Of course, high social transfer payments do not come without sig-
nicant costs. These could take the form of distortions induced by
disincentives for employment (the result of a generous social safety
net), distortions caused by taxation required to nance these trans-
fers, and, more generally, the effects of government budget decits
on overall macroeconomic performance.
In the Polish case, the increase in social transfers during the early
phase of the transition was accompanied by a substantial increase in
the general government budget decit (see the bottom panel of table
5.5). De Crombrugghe (1997), for instance, traces the destabiliza-
tion of the Polish budget in 19911992 directly to the rise in transfer
expenditures. By 1993 the growth in transfer expenditures (as a per-
cent of GDP) had been halted, although pensions and other social
benets remained at a higher level than in the pretransition years.
Our view is that, at least in the early stages of transition, social
transfers may have been crucial in setting the stage for rapid
reforms, and that this benet probably outweighed the direct
short-term costs of a rising budget decit. The Polish experience
suggests that the prerequisites for this approach to work are a well-
functioning set of transfer mechanisms and the commitment to
implementing institutional and macroeconomic reforms at a rapid
pace. Of course, a danger for the future is that generous social
transfers, especially pensions, which in early transition were useful
for cushioning older workers and facilitating enterprise restructur-
ing, will remain as entitlements long after their original purpose is

no longer operative and become a long-term scal burden.
Notes
We thank the staff at the Polish Central Statistical Ofce, especially Wiesaw agod-
zinski and Jan Kordos, for assistance with the data. We also thank Krzystof Przyby-
lowski and Barbara Kaminska for excellent translations of the survey instruments and
Branko Milanovic for generously sharing his cross-country data with us. We received
helpful comments on earlier drafts from Theo Eicher, Omer Moav, Stephan Klasen, an
anonymous referee, and the participants of the 2001 and 2002 CESIfo Growth and In-
equality conferences. Financial support was provided by the National Council for
Eurasian and East European Research. The views expressed in this chapter do not
necessarily represent those of the IMF.
1. These indicators measure the progress made by transition countries in many areas
of market-oriented reforms including enterprise privatization and reform; price and
trade liberalization; and establishment of the rule of law, property rights, and well-
functioning nancial markets.
2. The reasons for this transition recession, experienced by all transition economies,
are controversial. Some authors stress the contraction of credit to state enterprises (see
Calvo and Coricelli 1992), while others stress the aggregate demand contraction, due
in part to the opening to import competition and the sharp contraction in exports to
other former communist bloc countries (see Berg and Blanchard 1994).
3. Although the survey sample is designed to be representative of the underlying
population, nonresponse rates tend to differ across demographic groups. This neces-
sitates the use of sampling weights, although these weights had little effect on any of
our main results.
4. Higher-order polynomial terms were insignicant in these regressions, which t the
time path of the idiosyncratic component of variance quite well.
5. One other important change that was made in the 1993 survey was an attempt
to obtain a more representative sample of the self-employed. This groups size is
believed to have increased markedly since the transition began, resulting in its under-
representation in the HBS data during the period 19901992. However, as shown in
Keane and Prasad (2002a), underrepresentation of the self-employed is likely to have
led to only a marginal understatement of the extent of inequality in the early years of
the transition.
6. It is possible to make some (but not all) of the necessary adjustments to income
using information in the aggregate data on categories of income.
7. The aggregate consumption gures published by the Polish CSO, as well as by

other former communist countries, often correspond to a measure of total outows,
including saving and repayment of loans. For farm households, consumption includes
farm investment and purchases of supplies. An indication of the strange nature of the
aggregate consumption data is provided by Milanovic (1998, 41), who reports that in
1993 the Gini for consumption is 0.31, which substantially exceeds the Gini of 0.28 for
income. He also reports (33) that in 1993 the ratio of consumption to income is 1.30, an
unreasonably high gure. Our access to the detailed micro data enables us to make
necessary adjustments to the categories that are included in consumption. We then

nd the more plausible results that consumption Ginis are smaller than income Ginis
and that the aggregate consumption to income ratio falls in the 0.89 to 0.96 range
during the 19851997 period.
8. At the time we began our study, the Polish CSO had never before released the HBS
microdata. A long negotiation process by the rst author during 19921993 led to its
release. Subsequently, the microdata for the rst half of 1993 was released to the
World Bank and this data is used in World Bank (1995) and Milanovic (1998). More
recently, the data for 19931996 have been obtained by researchers at the World Bank.
As noted above, a subsample of the HBS is also now available in the through the
Luxembourg Income Survey (LIS) for 1987, 1990 and 1992. Thus, no prior researchers
have had access to the microdata for the entirety of the extended period that we
examine.
9. The sample size falls in 1992 since half of the total sample in that year was used to
test the new monthly survey. These monthly data from 1992 were considered unreli-
able and not made available to us.
10. For details on the estimation of the food share based equivalence scales, and for a
comparison with other commonly used equivalence scales, see Keane and Prasad
(2001a). We are aware of the potential problems associated with the use of food share
based equivalence scales. However, we were concerned about estimating a complete
demand system under conditions when rationing of certain commodities was proba-
bly an issue in some years of our sample, but where we do not observe the rationing
regimes.
11. Besides our own FS scale, we also used the OECD scale, the McClements scale
(which is commonly used in Britain), and the simple per capita scale. Appendix Table
B1 in Keane and Prasad (2001a) shows values of different equivalence scales for a
representative set of household types.
12. We recomputed many of the results in this chapter using different equivalence
scales. Although the levels of inequality were slightly affected by the choice of equiv-
alence scale, patterns of the evolution of inequality over time were robust to this
choice.
13. In all cases, we examine the distribution of individual income (or consumption),
assigning to each individual the per equivalent income for the household in which the
person resides.
14. Since transfers tend to be stable over time, the adjustment factors (used to adjust
for the change in survey frequency in 19931996) for income net of transfers were
nearly identical to those we computed for income including transfers.
15. In Keane and Prasad (2001a, 2002a), we show that similar results hold if one
examines alternative inequality measures such as quantile ratios, quantile shares, or
kernel density estimates of the income distribution.
16. A similar result is reported by Garner and Terrell (1998), who nd that pensions
substantially reduced inequality during the early transition years in the Czech and
Slovak republics.
17. In Keane and Prasad (2002a), we report that changes in within-group inequality
were very different across different groups. Within-group inequality actually fell
among farmer and mixed worker-farmer households during the transition, which, in
addition to the roles of pensions and other social transfers, also helps to account for
the rather small increase in overall income inequality. The Gini for household income
for individuals in worker-headed households rose from 0.189 in 1988 to 0.248 in 1997.
This increase of 0.059 is almost three times as great as the 0.020 increase in the Gini
for the overall income distribution, and is consistent with the increase in labor income
inequality noted earlier.
18. The typical retirement age in Poland is 65 for men and 62 for women. Among
households with heads in the 4555 age range and in lower age ranges, there was a
small drop from 1989 to 1992 in the share of income from labor income, but this was
mostly offset by an increase in other social benets rather than pensions. Among
households with heads aged 65 and older, pensions constitute 8590 percent of total
income, with labor income accounting for barely 2 percent.
19. The level of transfers is expressed as a percent of GDP and is the average, for each
country, from the rst year of its transition through 1997. The Gini coefcients are for
per capita income and the change is the difference between the value of the coefcient
four to ve years into the transition and the value of this coefcient before transition.
For data and sources, see Keane and Prasad (2002a).
20. Note that higher transfers do not necessarily imply more redistribution. An
extreme example is provided by Commander and Lee (1998), who note that transfers
in Russia have actually become regressive in the transition.
21. In Keane and Prasad (2002a), we provide a more detailed regression analysis of
the relationship between inequality changes and growth in the transition economies.
We nd a negative relationship even after controlling for initial conditions (including
the pretransition level of inequality) and measures of the extent of market-oriented
reforms. Grun and Klasen (2001) report similar ndings.
22. The Polish situation is considerably different from that in numerous other transi-
tion economies, where the real value of pensions and other transfers fell precipitously
during transition (both in absolute terms and relative to average wages).
23. Fidrmuc (2000) presents an interesting empirical analysis of voting patterns in
transition countries. He notes that various politically powerful groupsincluding
unions and retireeswere more likely to vote for left-wing parties. Notwithstanding
their political leanings, these were in fact the regimes that had enough political capital
to institute signicant reforms. The Polish experience can be seen as one where suc-
cessive (relatively short-lived) governments during early transition used generous
transfers to acquire such political capital and appease groups that had the most to
lose, in the short run, from market-oriented reforms.
References
Aghion, Philippe, and Olivier Blanchard. 1994. On the Speed of Transition in Central
Europe. NBER Macroeconomics Annual 9: 283320.
Aghion, Philippe, and Simon Commander. 1999. On the Dynamics of Inequality in
the Transition. Economics of Transition 7(2): 275298.
Aghion, Philippe, and Patrick Bolton. 1997. A Trickle-Down Theory of Growth and
Development with Debt Overhang. Review of Economic Studies 64(2): 151162.
nomic Growth: The Perspective of the New Growth Theories. Journal of Economic
Perspectives 37(4): 16151660.
Alesina, Alberto, and Dani Rodrik. 1994. Distributive Politics and Economic
Growth. Quarterly Journal of Economics 109(2): 465490.
Atkinson, Anthony B., Lee Rainwater, and Timothy M. Smeeding. 1995. Income Dis-
tribution in OECD Countries: Evidence from the Luxembourg Income Study. OECD
Social Policy Studies No. 18. Paris: OECD.
Banerjee, Abhijit, and Andrew F. Newman. 1993. Occupational Choice and the Pro-
cess of Development. Journal of Political Economy 101(2): 274298.
Benabou, Roland. 1996. Inequality and Growth. NBER Macroeconomics Annual 11:
1174.
Berg, Andrew, and Olivier Blanchard. 1994. Stabilization and Transition: Poland
19901991. In The Transition in Eastern Europe, Vol. 1: Country Studies, ed. Olivier
Blanchard, Kenneth Froot, and Jeffrey Sachs. Chicago: University of Chicago Press.
Calvo, Guillermo A., and Fabrizio Coricelli. 1992. Stabilizing a Previously Centrally
Planned Economy: Poland 1990. Economic Policy 14: 176226.
Commander, Simon, and Une Lee. 1998. How Does Public Policy Affect the Income
Distribution? Evidence from Russia, 19921996. Manuscript, EBRD, London.
de Crombugghe, Alain. 1997. Wage and Pension Pressure on the Polish Budget.
World Bank Policy Research Working Paper No. 1793.
Dewatripont, Mathias, and Gerard Roland. 1996. Transition as a Process of Large-

Scale Institutional Change. Economics of Transition 4(1): 130.
Engel, E. 1895. Die Lebenskosten belgischer Arbeiter-Familien fruher und jetzt. Bul-
letin de lInstititut International de Statistique 9: 1129.
Easterly, William. 2002. The Middle Class Consensus and Economic Development.
Working paper, World Bank.
European Bank for Reconstruction and Development (EBRD). Various years. Transition
Report.
Evans, Davis S., and Boyan Jovanovic. 1989. An Estimated Model of Entrepreneurial
Choice under Liquidity Constraints. Journal of Political Economy 97(4): 808827.
Fidrmuc, Jan. 2000. Political Support for Reforms: Economics of Voting in Transition
Countries. European Economic Review 44: 14911513.
Galor, Oded, and Joseph Zeira. 1993. Income Distribution and Macroeconomics.
Review of Economic Studies 60(1): 3552.
Garner, Thesia, and Katherine Terrell. 1998. A Gini Decomposition of Inequality in
the Czech and Slovak Republics during the Transition. Economics of Transition 6(1):
2346.
Gomuka, Stanisaw. 1998. The Polish Model of Transformation and Growth. Eco-
nomics of Transition 6(1): 163171.
Grun, Carola, and Stephan Klasen. 2001. Growth, Income Distribution and Well-
Being in Transition Countries. Economics of Transition 9(2): 359394.
Havrylyshyn, Oleh, and John Odling-Smee. 2000. Political Economy of Stalled
Reforms. Finance and Development 37(3): 710.
International Monetary Fund (IMF). 1994. Poland: The Path to a Market Economy.
Occasional Paper No. 113.
Keane, Michael P., and Eswar S. Prasad. 2001a. Consumption and Income Inequality
during the Transition to a Market Economy: Poland, 19851992. IMF Staff Papers 47:
121154.
Keane, Michael P., and Eswar S. Prasad. 2001b. Poland: Inequality, Transfers and
Growth in Transition. Finance and Development 38(1): 5053.
Keane, Michael P., and Eswar S. Prasad. 2002a. Inequality, Transfers and Growth:
New Evidence from the Economic Transition in Poland. Review of Economics and Sta-
tistics 84(2): 324341.
Keane, Michael P., and Eswar S. Prasad. 2002b. Changes in the Structure of Earnings
during the Polish Transition. IMF Working Paper No. 02/135.
Kornai, Janos. 2000. Making the Transition to Private Ownership. Finance and
Development 37(3): 1213.
Kuznets, Simon. 1955. Economic Growth and Income Inequality. American Economic
Review 45(1): 128.
Milanovic, Branko. 1998. Income, Inequality and Poverty during the Transition from
Planned to Market Economy. World Bank Regional and Sectoral Studies, Washing-
ton, DC.
Milanovic, Branko. 1999. Explaining the Increase in Inequality during Transition.
Economics of Transition 7(2): 299341.
OECD. 1997, 1998. OECD Economic Surveys: Poland. Paris: OECD.
Perotti, Roberto. 1996. Growth, Income Distribution and Democracy: What the Data
Say. Journal of Economic Growth 1: 149187.
Persson, Torsten, and Guido Tabellini. 1994. Is Inequality Harmful for Growth?
American Economic Review 84(3): 600621.
Pinto, Brian, Marek Belka, and Stefan Krajewski. 1993. Transforming State Enter-
prises in Poland: Evidence on Adjustment by Manufacturing Firms. Brookings Papers
on Economic Activity, 213270.
Roland, Gerard. 1993. The Political Economy of Restructuring and Privatization in
Eastern Europe. European Economic Review 37(23): 533540.
Svejnar, Jan. 1996. Enterprises and Workers in the Transition: Econometric Evidence.
American Economic Review Papers and Proceedings 86(2): 123127.
World Bank. 1995. Understanding Poverty in Poland. Country Study, World Bank,
Washington, DC.
World Bank. 19972000. World Development Indicators. World Bank, Washington,
DC.
6 Growth and Inequality:
Evidence from Transitional
Economies in the 1990s
Oleksiy Ivaschenko
6.1 Introduction
The relationship between income inequality and economic growth

has received much attention in the economic literature. The impact
of economic development on income inequality, however, remains
ambiguous. Even if found to be signicant in univariate regressions
of income inequality on per capita GDP, the parameter estimate on
aggregate income loses its strength and can even reverse sign when
other explanatory factors or country-specic dummies are intro-
duced (Deininger and Squire 1998).
However, a common trait of the previous studies linking income
inequality and economic growth is that they concentrated primarily
on what happens to income distribution during the process of devel-
opment, that is of rising per capita income. In contrast, the countries
of Eastern Europe (EE) and the former Soviet Union (FSU) witnessed
a sharp contraction in output during the initial stage of the transi-
tion.1 This decline has been accompanied by a marked increase in
income inequality, though, not at a uniform rate across the region. In
many transitional economies, inequality has reached levels compa-
rable to that observed in highly unequal countries of Asia and Latin
America.
These developments in transitional countries pose many intrigu-
ing questions. What is the role of economic decline (and recovery) in
changing income distribution? What specic factors lie behind a
noticeable increase in income inequality over the transition? How
well do the same factors explain the changes in inequality across
different counties?
I attempt to answer these questions in this chapter using a unique
panel of inequality estimates constructed for twenty-four transitional
156 Oleksiy Ivaschenko
countries of EE and the FSU and embracing the period from 1989 to
1998. The fact that the combined population of these countries
exceeds 400 million people makes the understanding of the factors
driving the changes in income distribution go far beyond a purely
research interest. Although it is often argued that policymakers
should be more concerned about absolute poverty than income
inequality, there are several reasons why one may (or should) care
about the latter as well. At a given rate of economic growth, more
unequal distribution of income would be associated with a lower
rate of poverty reduction, assuming, of course, that the poor partici-
pate fully in sharing the gains from growth. Moreover, as suggested
in many studies (e.g., Alesina and Rodrik 1994; Birdsall, Ross, and
Sabot 1995; Deininger and Squire 1998; Persson and Tabellini 1994;
Sylwester 2000; Easterly 2001), an unequal income distribution might
itself be detrimental to long-run economic growth for a variety of
reasons.2 The most common arguments for this are that an unequal
distribution of income creates pressure for redistributional policies,
and hence distorts incentives for working and investing; that it leads
to abuse of power by the elite and to sociopolitical instability and,
thus, harms the investment environment; and nally that, in the
presence of imperfect capital markets, it reduces opportunities for
accumulating human capital (such as education and health) and
physical assets. From a social welfare point of view, it has also been
argued that both utilitarian and nonutilitarian views of welfare sug-
gest that income inequality reduces aggregate well-being.3 These
considerations leave no doubt that inequality indeed matters, and
in this chapter I investigate which factors underlie the trends in
inequality observed in transitional economies.
There is a growing amount of research which attempts to explain
the rise in income inequality during the transition. Many existing
studies try to gure out the possible factors behind the changes in
the distribution of income using either theoretical models of transi-
tion (Aghion and Commander 1999; Ferreira 1999; Milanovic 1999)
or a Gini decomposition analysis (by income component or recipient)
applied to a single country or a set of countries (Garner and Terrell
1998; Milanovic 1999; Yemtsov 2001). Yet a third approach employs
cross-country regressions to examine why income inequality is dif-
ferent across countries at a given point in time (World Bank 2000).
This chapter represents the rst attempt to identify factors under-
lying the changes in income inequality over time within countries
Growth and Inequality 157
rather than to explain differences in inequality levels across coun-

tries. Until now a lack of compatible time-series data with sufcient
geographical coverage ruled out the possibility of doing this, and I
undertake the task using the assembled panel of inequality estimates
comparable over time and across countries. I use panel data estima-
tion methods to control for unobservable country-specic effects that
result in a missing-variable bias in cross-sectional studies.
The remainder of the chapter is structured as follows. In section
6.2 I present some evidence on the evolution of income inequality
and economic growth during the transition. Section 6.3 discusses
potential determinants of rising inequality in transitional countries
with a reference to existing literature. Section 6.4 describes the data
used in the empirical analysis. Section 6.5 is devoted to model speci-
cation and description of the estimation technique. Section 6.6 de-
scribes regression results. In section 6.7 I examine the robustness
of results. Section 6.8 offers a conclusion and presents some policy
implications of the ndings.
6.2 Growth and Inequality during the Transition
Figure 6.1 shows the relationship between the changes in income

inequality and changes in real GDP during 19891998. It clearly
suggests that better growth performers experienced much smaller
increases in income inequality.
There is substantial variation in the regional performance, with the
transitional economies of EE performing much better, both in terms
of economic growth and distributional outcomes, than the FSU
countries. However, there are signicant differences within these
two groups of countries as well.
Although quite illustrative, inequality and growth dynamics pre-
sented in gure 6.1 may be misleading as they do not fully reect
what happened at different stages within this period. For instance,
given the evidence presented in gure 6.1, one may mistakenly con-
clude that Poland (POL) and Slovenia (SVN) were growing con-
sistently through the 1990s while other countries were declining, and
that inequality was uniformly trending upward during the period.
Therefore, in table 6.1 I present the evidence on the evolution of
inequality and growth separately for economic decline and economic
recovery episodes. This analysis gives us a better idea of the rela-
tionship between income distribution and economic growth.
Figure 6.1
The dynamics of income inequality and GDP growth in transitional economies,
19891998
Source: Authors calculations using a constructed panel of inequality estimates and the
Real GDP index from the Trans/MONEE2000 database, UNICEF, Florence.
Several major observations emerge out of the data in table 6.1.

First, no single country escaped economic decline and an increase in
income inequality (except Poland) at the start of the transition (see
columns 5 and 7, table 6.1).
Second, after the sharp economic decline in the initial period, most
of the countries started to recover at some later stage. In general,
Eastern Europe and the Baltic states began growing in 19921994,
while the non-Baltic FSU countries started to grow later or continued
to decline as of 1998 (see column 3, table 6.1).
Third, it appears that the economic recovery in the FSU countries
was generally associated with declining income inequality. Con-
versely, recovery in EE countries was accompanied by rising income
inequality, although at very modest rates (see columns 6 and 8, table
6.1). This is a very interesting observation since it indicates that the
mechanisms behind the inequality trends in EE and the FSU are not
necessarily the same.
In what follows, I discuss potential determinants of the changes
in income distribution in transitional countries. This discussion
serves as a basis for the choice of variables used later in the empirical
analysis. Most of the factors that I consider are those commonly
found in the literature on the determinants of cross-country inequal-
ity, while others are specic to the transitional region circumstances
that I expect to be inuential in explaining the pattern of income
inequality.
6.3 Potential Determinants of Rising Inequality in Transitional

Countries
A vast amount of literature exists on the determinants of income

inequality that considers both the individual (e.g., increasing returns
to skills) and macro-level factors (e.g., ination, political democracy)
affecting income distribution. In this chapter I focus on the latter,
although the former might be equally important.
The main factors that I anticipate to affect income inequality in
transitional countries are: the level of economic development (mea-
sured by per capita GDP), macroeconomic conditions (ination, un-
employment), government involvement in the economy (government
consumption, social transfers), structural changes (economic liberal-
ization, privatization, deindustrialization), and forces outside eco-
nomic domain (political freedom, civil conicts).
Many attempts to identify a link between income inequality and
the level of economic development have been undertaken since
the seminal work of Kuznets (1955), who argued for an inverted
Ushaped relationship between income inequality and economic
development. Although several studies (e.g., Paukert 1973; Ahluwa-
lia 1976) have found a support for such a relationship, most of the
recent research does not nd economic development to affect income
distribution (e.g., Anand and Kanbur 1993; Deininger and Squire
1998; Ravallion 1995).
However, the striking economic decline in EE and the FSU coun-
tries in the initial years of the transition, and the subsequent eco-
nomic recovery, are expected to have had signicant implications for
income distribution. That is because economic decline and recovery
were associated with dramatic and heterogeneous shocks to real
incomes, the changes in the real value of social transfers, and other
developments in social and economic conditions. Figure 6.1 pro-
vides strong support for anticipating a negative relationship between
Table 6.1
160
The dynamics of income inequality and GDP growth in transitional economies of EE and the FSU
Avg. annual change Avg. annual change in
Real GDP index in the real GDP index the Gini index
Population (1989 100) (percentage points) (percentage points)
Region/ (mid-1997, at the bottom decline growth decline growth
country millions) of decline (year) in 1998 period period period period
1 2 3 4 5 6 7 8
I. FSU
a. Baltic states
Estonia 1.5 60.76 (94) 75.70 7.85 3.74 3.31 0.65
Latvia 2.5 51.04 (95) 59.30 8.16 2.75 1.00 1.20
Lithuania 3.7 64.83 (94) 79.53 7.03 3.67 2.51 0.26
b. Western CIS
Belarus 10.2 62.69 (95) 77.75 6.22 5.02 0.31 0.44
Moldova 4.0 32.00 8.11 1.95
Russia 147.0 55.89 4.90 2.58
Ukraine 50.4 36.61 7.04 0.69
c. Caucasus
Armenia 3.8 31.63 (93) 41.68 15.29 2.08 8.42 0.40
Azerbaijan 7.8 41.86 (95) 49.40 9.69 2.51 2.05 0.33
Georgia 5.4 24.60 (94) 31.70 8.64 2.57
d. Central Asia
Kazakhstan 15.3 61.20 5.49 0.84
Kyrgyz Rep. 4.6 50.39 (95) 60.30 5.09 1.97
Tajikistan 6.0 39.19 (96) 41.90 6.44 1.69
Turkmenistan 4.6 41.99 (97) 43.75 6.25 1.04
Uzbekistan 23.6 83.36 (95) 89.50 3.18 0.48
Oleksiy Ivaschenko
II. Central EE
Czech Rep. 10.3 84.58 (92) 94.90 5.14 2.53 0.33 1.46
Hungary 10.2 81.89 (93) 95.20 4.53 2.66 0.32 0.52
Poland 38.7 82.21 (91) 117.15 5.22 5.47 0.34 1.33
Slovak Rep. 5.4 74.97 (93) 99.60 6.26 4.09 0.41 0.74
III. South EE
Bulgaria 8.3 63.69 (97) 65.90 4.54 1.27
Romania 22.6 74.99 (92) 82.08 8.34 3.30 0.63 1.37
IV. FY
Croatia 4.7 59.54 (93) 77.70 10.11 3.65 0.35 1.36
Macedonia 2.0 67.99 (95) 71.50 5.34 0.79 0.60 0.41
Slovenia 2.0 82.04 (92) 103.90 5.99 3.66 0.37 0.40
Note: Dashes indicate data are not available.
Note: Dashes in columns means that by the end of 1998 a country under consideration continued to decline. South EE also includes Albania;
the former Yugoslavia (FY) also includes Yugoslavia, FR, and Bosnia-Herzegovina. These countries are not included in the table due to the lack
of data.
Sources: Authors calculations using a constructed panel of inequality estimates and the real GDP index and population data from the Trans-
MONEE2000 database, UNICEF, Florence.
161
income inequality and economic development for transitional coun-

tries. Nevertheless, as the evidence from table 6.1 indicates, this
relationship is hardly universal across countries.
Ination may have a strong redistributional impact through its
effect on individuals whose nominal incomes are not adjusted pro-
portionally to increases in prices; mostly state-sector employees,
pensioners, and beneciaries of various social benets. That would
be an argument for a positive relationship between income inequal-
ity and ination. However, ination may also have an equalizing
impact on income distribution through a progressive tax system by
pushing wage earners into higher tax brackets, thus implying less
inequality in disposable income. These two effects may well coun-
terbalance each other. In a study of the determinants of inequality
for OECD countries (Gustafsson and Johansson 1999), ination was
not found to be signicant in explaining inequality. That may not
be the case for transitional economies, however, as most of them
experienced a sharp rise in ination at the start of the transition.4
Moreover, the progressivity of the ination tax is unlikely to be a
mechanism at work in most of the transition countries due to a high
occurrence of tax evasion. Hence, I expect ination to be positively
associated with income inequality in the transition region.5
Destruction of the old economic system and signicant structural
changes during the transition caused a substantial rise in unem-
ployment across the region. In many countries the unemployment
rate grew from virtually zero to 1015 percent even when measured
by the number of ofcially registered unemployed.6 Unemployment
is likely to largely affect those in the lower percentile of income dis-
tribution. Milanovic (1998) indicates that unemployment in transi-
tion countries increased the most among women, young people, and
those with lower education. A negative impact of unemployment on
income distribution has been conrmed in a number of studies of
industrialized countries (Gustafsson and Palmer 1997; Weil 1984),
and I anticipate unemployment to have an inequality-increasing
effect in transitional economies as well.
In times of economic hardship and increasing unemployment,
government-nanced projects (e.g., construction) may provide a
source of employment and income (with low-skilled labor probably
beneting the most), which serve as a buffer to widening income
inequality.7 The size of the public sector is found to reduce inequal-
ity in cross-country studies by Stack (1978) and Boyd (1988). Gov-
ernment involvement in the economy, measured as a share of gov-

ernment consumption in GDP, decreased between 1989 and 1998 in
eight out of fteen states of the FSU (including the Baltic states),
increased in ve states, and was practically unchanged in the rest. In
the EE region, government consumption has declined in only two
countries, while in the remainder it has either grown or has been
relatively stable. In this chapter I look at the effect of government
consumption on income distribution in transitional economies.8
Centrally planned economies were dominated to a various extent
by state enterprises with administratively set wages. The over-
whelming predominance of the state sector in EE and the FSU
economies is widely regarded as a main reason for low income
inequality in the region before the transition.9 The process of transi-
tion brought about a massive expansion of the private sector and the
share of the private-sector employment.10 This process is likely to
increase income inequality due to the wage differential between the
state and private sectors. Moreover, the distribution of earnings
within the private sector is usually more unequal than in the state
sector. That privatization can lead to rising income inequality is
argued in theoretical models of transition by Milanovic (1999) and
Ferreira (1999). However, due to the poor data on the scope of pri-
vatization in transitional countries, the impact of rising private sec-
tor on income distribution has not been empirically tested, until this
chapter.
Economic liberalization also led to profound changes in the sec-
toral composition of the economy. There is a clear trend for the
industrial sector to shrink, while the evidence for the agricultural
sector is mixedin some countries its relative importance has
declined, while in others it has increased.11 The share of industry in
total output in the region declined on average by 25 percent from
1989 to 1998, and in several countries the drop was even more pro-
found. For instance, in the ten years after 1989 the share of the
industrial sector declined from 52 percent to 32 percent in Poland,
from 58 percent to 33 percent in Slovak Republic, from 59 percent to
25 percent in Bulgaria, and from 50 percent to 35 percent in Russia. It
is very likely that the declining industrial-sector employment may
have an inequality-increasing impact due to an outow of labor to
sectors with higher wage differentials, for instance, services.12 A
negative relationship between industrial-sector employment and in-
come inequality is conrmed in studies of industrialized countries
by Gustafsson and Johansson (1999) and Levy and Murnane (1992),

and in this chapter I investigate the effect of de-industrialization on
income distribution in the transitional region.
The process of economic transition in EE and the FSU was gener-
ally accompanied by the expansion of political democracy. Although
a common argument in the literature is that the higher degree of
political democracy should be accompanied by a more equal distri-
bution of income (e.g., Gradstein, Milanovic, and Ying 2001; Rodrik
1999), the existing evidence does not show any robust relation-
ship between democracy and inequality in a cross-country regres-
sion analysis. Here I investigate whether political democracy affects
income inequality in transitional countries.
A number of countries in EE (republics of the former Yugoslavia)
and the FSU (Armenia, Azerbaijan, Georgia, Tajikistan, Moldova,
and Russia) experienced persistent internal conicts over the last
decade. Since civil conicts are likely to have strong distributional
consequences, I analyze their impact on income inequality in the
transitional region. The data used in the empirical analysis and their
sources are described in detail in the next section.
6.4 The Data
I construct a panel of inequality estimates using time-series data

on income inequality across transitional countries. The majority
of observations are drawn from the UNU/WIDER-UNDP World
Income Inequality Database (WIID) (Version 1.0, September 2000),
which to date represents the latest and most extensive data on
inequality for both developed and developing countries.13 In addi-
tion, I augment these data with a few observations from Milanovic
(1998) (mainly for 1989) and the latest household surveys conducted
by the World Bank (2000) (mainly for 19981999).
To minimize problems with data comparability across countries
and over time I require inequality data that I select for the panel to
be based on the same living standard indicator, have the same sam-
ple and enumeration unit, be drawn from nationally representative
surveys, and, whenever possible, come from one source.14 Income
inequality is measured by the Gini coefcient with individuals rep-
resenting the unit of analysis. The coefcients are calculated based
on household per capita income. The compiled panel of inequality
estimates represents perhaps the most consistent and extensive cov-
erage available for transitional countries to date. It consists of 149

observations covering twenty-ve countries in transition from 1989
to 1999.15 However, due to either missing observations on other
variables, or the deletion of observations based on the inuence
diagnostics tests (as discussed later), only 129 out of 149 originally
assembled Gini coefcients are used in the estimation. A detailed
description of the data on income inequality used in the empirical
analysis and their sources is presented in table 6.A1.
The constructed panel of inequality estimates is far from be-
ing perfect, however, as not all of the above comparability require-
ments could always be met, and the resulting inequality measures
are still subject to potential measurement error problems. The use
of panel data and panel data estimation methods (to be discussed),
however, help diminish some problems with data consistency.
The country-specic intercepts in the xed-effects model setting
can absorb, among all other unobservable characteristics, the differ-
ences in inequality denition across countries (Deininger and Squire
1998).16 Nevertheless, the use of panel data cannot remedy all data
limitations, and thus the empirical results must be treated with some
degree of caution.
I now turn to the denitions and sources of data on explanatory
variables. The level of economic development is measured by PPP-
adjusted per capita GDP in constant 1992 USD. The data on PPP-
adjusted per capita GDP in current USD come from the World Bank
World Development Indicators (WDI) 2000 database, and they are
then deated to 1992 prices using the U.S. GDP deator.
Ination is measured as the annual percentage change in the con-
sumer price index (CPI) (end-year). As CPI-based ination is not
available for all countries in our sample for 1989 and 1990, the GDP
deator ination is taken for those years instead. Finally, as neither
of the mentioned above indexes could be obtained for Croatia (1989),
Macedonia FYR (1989, 1990), and Slovenia (1989) ination there is
measured by the food price index (a subindex of the CPI). All ina-
tion data are drawn from the World Bank WDI 2000 database.
Unemployment represents a share of the labor force that is with-
out work but available for and seeking employment. However, as
I have mentioned, unemployment data for transitional countries
may substantially understate the actual scope of unemployment.
Nonetheless, as no better alternative is available, ofcial estimates
are used in most cases. Unemployment data are taken from the
European Bank for Reconstruction and Development (EBRD) Transi-

tion Report 2000, which provides further reference on the origin of
the data for each country.
General government consumption, expressed as a fraction of GDP,
refers to all current spending for purchases of goods and services
(including wages and salaries). It also includes most expenditures on
national defense and security, but excludes government military
expenditures that are part of government capital formation. As such,
government consumption represents a good measure of the gov-
ernments involvement in the economy. The data on government
consumption come from the World Bank WDI 2000 database.
Industrial employment represents a share of industry in total
employment. I was able to obtain only 105 observations covering
twenty-four countries (EBRD 2000). As other explanatory variables
contain more observations, the use of these employment data in the
model estimation would substantially reduce a number of observa-
tions on other variables. Since the sample is relatively small, I con-
sider that inappropriate. Therefore, I use a share of industry value
added in GDP as a proxy for industrial-sector employment.17 This
provides us with a substantially larger number of observations. The
data come from the World Bank WDI 2000 database.
Private-sector employment equals the number of people employed
in the private sector as a percentage of total employment. Data
availability, however, represents a severe constraint here, as practi-
cally no data prior to 19931994 exist. I have managed to collect
fty-one observations using IMF country reports, a number not suf-
cient for our purposes. Thus, in the regression analysis I use a
share of the private sector in GDP as a proxy for the private-sector
employment. Since I have found high correlation between the size of
the private sector and the private-sector employment in our sample
(for those observations that are available), and in view of the lack of
an alternative, such a proxy is considered to be justiable. These
data and their more extensive descriptions are available in EBRD
(2000).
Economic liberalization (which is largely reected in structural
changes) is measured with the Cumulative Liberalization Index (De
Melo, Denizer, and Gelb 1996), which reects the progress with
economic reforms on several fronts: internal (price) liberalization,
external (foreign trade) liberalization, and the extent of privatization
and banking-sector reform.18
The progress in the introduction of political rights and civil

liberties during the transition is measured using the Index of Politi-
cal Freedom (IPF) (Freedom House 2001), which represents an arith-
metic average of the political rights and civil liberties indexes. The
political rights index reects the extent to which people in a country
can participate in the political process. The civil liberties index mea-
sures the freedoms to develop views, institutions, and personal
autonomy apart from the state.
The effect of civil conicts on income inequality is measured using
a dummy variable set equal to one for each year since an internal
conict has taken place in a given country.19 In our sample the
countries affected by civil conicts are Croatia, Macedonia FYR,
Armenia, Azerbaijan, Georgia, Moldova, Tajikistan, and Russia.20
The historical information on civil conicts in the region is obtained
from the Reuters Foundation.21
Table 6.A2 provides the descriptive statistics of the data used in
the regression analysis. Table 6.A3 shows the matrix of the Pearson
correlation coefcients.
6.5 Model Specication and Estimation
The primary interest of this study is to explain the changes in income

inequality in transitional economies, and I thus estimate income
inequality as a function of various potential explanatory variables
presented here. The base model specication is
GINIit a i b 0 GDPPCit b1 GDPPC Sit b2 INFLit

b3 UNEMPit b 4 CONSGit b 6 INDVAit
b7 PRIVSit eit ; i 1; . . . ; N; t 1; . . . ; T; 1
where i represents country index, t denotes time period, GINI is the

Gini coefcient of income inequality, a i is a country-specic inter-
cept, GDPPC is PPP-adjusted GDP per capita (1992 constant USD),
GDPPC_S is its squared value, INFL is annual ination as measured
by the year-to-year change in the consumer price index, UNEMP is a
share of unemployed in total labor force, CONSG is general govern-
ment consumption as a percentage of GDP, INDVA is industry value
added as a percentage of GDP, PRIVS is the private-sector share in
GDP, and eit is an error term. The assumption on eit is that
eit @ IID0; se2 . All variables (including the Gini coefcient) enter
the regressions in the natural log form. The natural log of (1 INFL/
100) is used for INFL variable in the estimations to deal with nega-
tive and very high values of INFL. The natural log of (1 UNEMP)
is used for UNEMP variable in the estimations since the unemploy-
ment rate equals zero for many countries at the start of the transi-
tion. A squared value of GDPPC (with GDPPC expressed in the
natural log form) is included into the regression to account for the
potential quadratic relationship between income inequality and per
capita GDP.
In view of the large body of literature exploring the effect of
income distribution on economic growth, one may be quick to point
out the possible problem with the given model specication arising
from the potential existence of a reverse causality between inequality
and growth. I argue that the transition economies of EE and the
FSU represent a unique case when the possibility of causality from
income distribution to economic growth can be ruled out, at least
for the period under investigation, since the reasons for the eco-
nomic collapse and subsequent recovery in the region had clearly
nothing to do with the distribution of income. This is conrmed by
the Granger causality tests (Granger 1969). I have tested whether
income inequality Granger-causes economic growth using from one
to ve lags, which provides 87 to 22 observations, respectively. In
none of the cases did the test statistic indicate that I could reject the
null hypothesis that the Gini coefcient does not Granger-cause per
capita GDP.22
I estimate equation (1) using the assembled panel for twenty-four
transitional countries covering a period from 1989 to 1998. The use of
panel data produces several well-known advantages. The most
important is that it allows one to control for unobservable time-
invariant country-specic effects that result in a missing-variable
bias, an often encountered problem when cross-section data are
used. This problem is recognized in Bourguignon and Morrison
(1998), Bruno, Ravallion, and Squire (1995), Deininger and Squire
(1998), Forbes (2000), Ravallion (1995), and other studies.
To control for unobservable country-specic characteristics I in-
troduce country-specic intercepts in the xed-effects model setting.
The addition of xed effects to the model also helps alleviate poten-
tial heteroscedasticity problems stemming from possible differences
across countries (Greene 1997).
There might be another reason for preferring the xed-effects

model to the random-effects model. A crucial assumption for the
random-effects model is that country-specic terms (a i ) are uncor-
related with the other explanatory variables. Its violation makes
random-effects estimates biased and inconsistent (Greene 1997). The
use of the xed-effects model avoids this problem as individual
effects are allowed to be correlated with other regressors. To test
whether the country-specic effects are correlated with the exoge-
nous variables, I conduct a Hausman test.23 While the test statistic
suggests that I cannot reject the null hypothesis for the base model,
the results of a Hausman test are found to be sensitive to the model
specication and sample selection. Therefore, I prefer to use the
xed-effects model.
More important, the xed-effects model is chosen since the main
goal of this study is to investigate what factors have caused sub-
stantial changes in income inequality over time within countries
rather than to explain variation in inequality across countries.24
Thus, the use of the xed-effects estimator, which is also called
the within estimator, is very appropriate since it allows one to
focus on how changes in within-country characteristics are related to
changes in within-country inequality. The xed-effects model is also
more suitable when the focus is on a specic set of countries and the
inference is restricted to these countries (Baltagi 1995, 10). Moreover,
the country-specic effects have been found to be statistically sig-
nicant.25 This implies that if the model is estimated without
taking them into account (i.e., if a single overall intercept is in-
cluded instead of country dummies) the estimated coefcients will
be biased.
The xed-effects estimation technique, however, is not perfect.
First, random-effects estimates are more efcient than xed-effects
ones given that all necessary assumptions are satised. Second, the
xed-effects model is very costly in terms of the lost degrees of free-
dom, which may represent a particular problem for the relatively
small sample. To overcome this problem, I estimate the model in
deviations from the country means.26 This within-countries estima-
tor is identical to the least squares dummy variable (LSDV) estimator
obtained if a dummy variable is included for each country (as in the
original formulation of equation (1)), but the resulting R 2 is lower
(Greene 1997, 619).
Third, it has been argued (Barro 1997, 37; Temple 1999) that the
xed-effects technique eliminates the cross-sectional information
and, hence, lowers precision of the estimates. This problem is, of
course, especially acute if most of the variation in the data is due to
cross-country differences. While in nontransition countries most
(approximately 90%) of the variation in inequality is due to variation
across countries (Deininger and Squire 1996; Li, Squire, and Zou
1998; Quah 2001), in transition countries a substantial source of
the variation in inequality over the last decade is attributable to the
profound changes in inequality over time (see table 6.A2). Hence, the
use of the xed-effects estimation for transitional countries seems to
be appropriate. In addition, if one bears in mind that inequality
comparisons across countries are likely to be much less reliable than
inequality comparisons for a single country over time, despite all
efforts to assemble the inequality estimates that are as consistent as
possible both across space and time, the reliance on mostly variation
over time in inequality is even desirable.
Given all considerations outlined here, I prefer to use the xed-
effects model since its advantages seem to outweigh its weaknesses
given the data and research purposes. Nevertheless, to check the
robustness of the results to the estimation technique the random-
effects model is also estimated. The empirical results are described
later in the chapter. The panel that I estimate is unbalanced as a
number of time-series observations differ across countries. However,
assuming that observations are missing randomly, consistency of the
xed-effects estimator is not affected. The xed-effects model is esti-
mated with OLS, which, given the assumed properties of residuals,
is the best linear unbiased estimator (Hsiao 1986).
6.6 The Regression Results
I rst estimate a full version of equation (1) with both the log of
per capita GDP (GDPPC) and its squared value (GDPPC_S) included
among the explanatory variables. This allows capturing a potential
threshold effect in the relationship between income inequality and
per capita GDP.
It is worth noting that up to now most of the attempts to test for a
U-shaped relationship between income inequality and economic
development have used cross-sectional regressions. This approach
may be conceptually incorrect when studying the intertemporal rela-
tionship between income inequality and per capita income. If one

wants to see whether inequality changes with economic develop-
ment, longitudinal data are needed (Deininger and Squire 1998).
Here the use of the panel data provides an obvious advantage over
purely time-series or cross-sectional data.
However, to test for any kind of a quadratic relationship between
income inequality and per capita GDP is not the main purpose of
this study. Here I undertake an attempt to identify specic factors
behind the changes in inequality in the transitional region. While
economic growth represents a good aggregate measure of the econ-
omys health since it reects the outcome of multiple complex pro-
cesses taking place at all levels of the economy, it alone does not
seem to be a satisfactory explanation of the inequality pattern. That
is why in equation (1) I also introduce other potential explanatory
forces into play. In addition, I estimate an alternative specication of
equation (1) where I include only GDPPC (but not GDPPC_S) among
the regressors to test for the linear relationship between income
inequality and per capita GDP. Finally, since some of the explana-
tory variables in the model appear to be signicantly correlated, I try
several alternative specications to investigate the robustness of the
parameter estimates.27 The regression results from estimating differ-
ent modications of equation (1) are reported in table 6.2.28
The regression results indicate a statistically strong relationship
between income inequality and per capita GDP. There is stronger
support for a quadratic rather than a linear relationship.29 The pa-
rameter estimates on GDPPC and GDPPC_S indicate that the rela-
tionship between income inequality and economic growth depends
on where a country stands in terms of its per capita GDP. More spe-
cically, they suggest that for a country below (above) some thresh-
old level of development the process of economic growth would be
associated with falling (rising) income inequality. It is worth noting
that many transition countries (mostly those in the FSU) had either
already been below the estimated threshold level of per capita GDP
(see table 6.2) at the start of the transition, or slipped below this level
as a result of an economic decline, and thus are expected to have
negative relationship between growth and inequality. The linear
specication indicates that a 10 percent decline in per capita income
would increase the Gini coefcient at the mean by 0.48 percentage
points (see column 3, table 6.2). That income inequality might in-
crease during recessions was conrmed in a number of studies of the
Table 6.2
Fixed-effects estimates from the regression of the Gini coefcient on selected explana-
tory variables: All countries
Reduced Reduced
Full model Full model model 1 model 2
Explanatory (quadratic (linear (quadratic (quadratic
variable relationship) relationship) relationship) relationship)
1 2 3 4 5
Intercept
GDPPC 4.190*** 0.157** 3.650*** 4.912***
(1.084) (0.063) (1.131) (1.208)
GDPPC_S 0.239*** 0.203*** 0.272***
(0.064) (0.067) (0.072)
INFL 0.033*** 0.034*** 0.024*
(0.013) (0.013) (0.013)
UNEMP 0.001 0.023 0.035* 0.086***
(0.021) (0.021) (0.020) (0.019)
CONSG 0.014 0.023 0.061 0.090*
(0.047) (0.050) (0.048) (0.052)
INDVA 0.288*** 0.350*** 0.375***
(0.072) (0.074) (0.073)
PRIVS 0.105*** 0.091***
(0.027) (0.028)
Number of 24 24 24 24
countries
Number of 129 129 129 129
observations
R 2 adj. 0.65 0.61 0.61 0.52
F-value 32.66 31.97 33.33 12.14
Estimated 6405 8015 8337
threshold level
(PPP-adjusted
GDP per capita,
1992 USD)
Note: All variables are in the natural log form. Standard errors are presented in
parentheses. The model is estimated in deviations from the group means. The Yule-
Walker (iterated) method was used to correct for serial correlation and hetero-
skedasticity.
* indicates signicance at 10 percent level; ** indicates signicance at 5 percent level;
*** indicates signicance at 1 percent level (two-tailed tests).
United States (Meier 1973; Metcalf 1969; Thurow 1970). So, given
that economic decline in many transition countries reached an un-
precedented scale, the adverse changes in the distribution of income
are not surprising.
Ination is found to increase income inequality, although the esti-
mated coefcient in one of the specications turns out to be only
marginally signicant.30 The magnitude of the effect, however, does
not appear to be large. A 10 percent increase in ination would raise
inequality at the mean by at most 0.08 Gini points (see table 6.2). I
have also tested (using the LSDV specication) for a threshold effect
of ination as one would expect that it is only the ination above
a particular level that affects income distribution. Indeed, when I
include among the explanatory variables in equation (1) dummies
for ination levels instead of INFL, a clear threshold effect is appar-
ent. I nd (see column 2, table 6.3) that hyperination (annual CPI
exceeds 500 percent) is associated with a 9.5 percent higher income
inequality compared to the situation of the relative macroeconomic
stability (annual ination below 20 percent).
Unemployment rate does not seem to have an impact on inequal-
ity in the base model (see column 2, table 6.2). It is very likely, how-
ever, that the parameter estimate on unemployment is contaminated
due to a high correlation of the unemployment rate with other indi-
cators of structural changes, namely de-industrialization and priva-
tization (see table 6.A3). Indeed, when I eliminate PRIVS from the
estimation, the coefcient on UNEMP becomes positive and margin-
ally signicant (see column 4, table 6.2). Finally, when INDVA and
INFL are also omitted from the regression, the parameter estimate on
UNEMP becomes statistically signicant at a 1 percent level (see
column 5, table 6.2). The magnitude of the coefcient suggests that a
1 percentage point increase in the unemployment rate at the mean
would raise the Gini coefcient at the mean by 0.33 points.
The lack of robustness of the effect of unemployment on income
inequality may also be due to the following factors. First, a likely
inequality-increasing effect of growing unemployment can be coun-
terbalanced by an increasing ow of unemployment benets. Sec-
ond, it is also possible that the increase in between-groups inequality
stemming from larger unemployment could be offset by a more
equal distribution of income among transfer recipients (Milanovic
1999). Third, it has been argued that in developing countries the
unemployment statistics can in fact reect the extent of the informal
Table 6.3
Estimates from the regression of the Gini coefcient on selected explanatory variables:
All Countries
Model with Model with
ination Model Model war dummy
Explanatory dummy with CLI with IPF (pooled
variable (LSDV) (LSDV) (LSDV) regression)
1 2 3 4 5
Intercept 4.370***
(0.365)
GDPPC 3.834*** 2.993** 4.005***
(1.225) (1.217) (1.230)
GDPPC_S 0.216*** 0.162** 0.225***
(0.073) (0.073) (0.073)
Ination > 500 0.091** 0.106*** 0.089** 0.163***
dummya (0.038) (0.037) (0.038) (0.036)
UNEMP 0.007 0.014 0.003 0.002
(0.025) (0.025) (0.025) (0.022)
CONSG 0.035 0.044 0.045 0.049
(0.054) (0.051) (0.054) (0.046)
INDVA 0.242*** 0.199** 0.205** 0.360***
(0.081) (0.079) (0.086) (0.074)
PRIVS 0.098*** 0.020 0.097*** 0.117***
(0.032) (0.039) (0.032) (0.028)
CLI 0.034***
(0.011)
IPF 0.016
(0.013)
War dummy 0.125**
(0.050)
Number of 24 24 24 24
countries
Number of 129 129 129 129
observations
R 2 adj. 0.999 0.999 0.999 0.61
F-value 4290.35 4573.36 4194.15 15.73
Estimated 7255 10432 7293
threshold level
(PPP-adjusted
GDP per capita,
1992 USD)
parentheses. Yule-Walker (iterated) method was used to correct for serial correlation
and heteroskedasticity.
a excluded category is ination < 20 percent annual. The coefcients on other catego-
ries (2150, 51100, 101500) are not reported since they are not signicant.
sector employment and self-employment, which are generally asso-

ciated with higher income inequality (Ferreira and Litcheld 1998).
Finally, the quality of the unemployment data can also be a sim-
ple and quite probable explanation of why the effect of unemploy-
ment is not very robust. These data, as was mentioned before, are
mostly based on ofcial unemployment records, which may severely
underestimate not only the actual scope of unemployment, but also
the changes in the rate of unemployment over time.31 Thus, their use
in the estimation may induce a substantial downward bias in the
parameter estimate on unemployment.
With regard to government consumption, I do not nd it to inu-
ence income distribution. Although the sign of the parameter esti-
mate is as anticipated, the coefcient on CONSG is only marginally
signicant in one of the specications (see table 6.2). The size of
government consumption (as a share of GDP) may have poor pre-
dictive power since the total effect of this factor on income distribu-
tion clearly depends on the composition of government expenditures
and progressivity of taxes used to nance them. Also, the variation
in the share of GDP devoted to public consumption is substantially
higher across countries than over time (see table 6.A2). The estima-
tion method that relies on the intertemporal variation does not cap-
ture the potential effect of government consumption on the levels of
income inequality across countries.
The parameter estimate on INDVA supports our hypothesis that
deindustrialization increases inequality.32 The coefcient from the
full model (see column 2, table 6.2) suggests that a 10 percent decline
in the share of the industrial sector would lead to a 0.88 percentage
point increase in the Gini coefcient at the mean. The parameter
estimate implies that in Ukraine, for example, where the share of
industrial sector in total output dropped from 48 percent in 1989 to
34 percent in 1998, the Gini coefcient increased by 2.17 percentage
points over the period due to this factor alone, thus explaining a
third of the total increase in income inequality.
There is a statistically strong positive relationship between income
inequality and the size of the private sector. The estimated coefcient
suggests that a 10 percentage point increase in the share of the pri-
vate sector in the economy would result in a 0.83 point increase in
the Gini coefcient. The magnitude of the effect does not seem to be
very large. However, if one bears in mind that transitional countries
have witnessed a substantial growth of the private sector, it is clear
that the rising private-sector employment plays a crucial role in

explaining the increase in income inequality.
I next look at the effect of economic liberalization on income
inequality using the Cumulative Liberalization Index (CLI) as an
explanatory variable in the regressions. The regression results (see
column 3, table 6.3) indicate that the process of liberalization is
associated with rising income inequality.33 The CLI is by construc-
tion highly correlated with the structural indicators used in my
analysis (see table 6.A3). Nevertheless, the parameter estimate on
CLI is signicant at a 1 percent level even when all other variables
are included. The coefcient on PRIVS, however, becomes insigni-
cant in this case. The parameter estimate on CLI indicates that a 10
percent increase in the CLI at the mean would be associated with a
0.27 point increase in the Gini coefcient at the mean. The effect of
economic liberalization on income distribution is highly robust to the
model specication.
I next investigate the impact on the distribution of income of
the factors outside economic domain. Transitional countries have
made different progresses in the introduction of political rights
and civil liberties during the transition, which makes it interesting to
see whether the progress on the political freedom front has any
implications for the distribution of income. To do this, I use the IPF
described earlier. It is worth noting that the IPF is highly correlated
with some other variables (see table 6.A3). For instance, it is posi-
tively associated with the progress in economic reforms and nega-
tively with the occurrence of civil conicts.34 It is noteworthy that
the regression results do not indicate that the extent of political
rights and civil liberties has an independent impact on income
inequality (see column 4, table 6.3). That does not preclude, though,
the possibility that the degree of political democracy may affect
income inequality indirectly (Gradstein, Milanovic, and Ying 2001).
A number of countries in the transitional region experienced the
periods of civil conicts and wars over the last decade. Therefore, I
also look at the impact of civil wars on income inequality by esti-
mating the pooled regression with a dummy for civil conicts con-
structed as discussed earlier.35 Since this dummy is highly correlated
with the level of per capita GDP (see table 6.A3), I omit GDPPC and
GDPPC_S from the estimation. The regression results (see column 5,
table 6.3) indicate that civil conicts are associated with a 13.3 per-
cent rise in income inequality.36
6.7 Sensitivity Analysis
In this section, I investigate the robustness of my ndings. I rst

check whether the results are sensitive to the denition of the
dependent variable. Although most of the Gini coefcients in the
data set are based on disposable income, there are several data
points based on other welfare concepts (see table 6.A1). As a rst
check of the robustness of the results described in section 6.6, I per-
form the estimation using disposable income Gini coefcients only.
Since several Gini indexes in our sample come from other sources
than the main data series used (the WIDER database) (see table
6.A1), I also examine the sensitivity of the ndings to the omission of
these observations. Finally, I estimate the model by including only
those Gini coefcients that are based on the same welfare denition
within countries.37 The resulting parameter estimates are reported in
table 6.4.38
Similarly to the results reported in table 6.2 (quadratic specica-
tion), all estimations shown in table 6.4 strongly support a U-shaped
relationship between income inequality and per capita GDP. The
estimated coefcients on GDPPC and GDPPC_S are larger than their
counterparts in table 6.2.39 Although the parameter estimates on
INDVA and PRIVS differ somewhat from sample to sample (see
columns 24, table 6.4), which is quite natural given the variation in
the representation of countries and time periods across the samples,
they are generally in line with those reported in table 6.2. I note that
the coefcient on ination is signicant in one case only (see column
2, table 6.4). In the light of our nding that there is a clear threshold
effect in the relationship between ination and inequality this result
is not surprising, since the samples differ in the number of high-
ination observations.
I am aware of the literature that advises to make a regression-
based additive adjustment of the Gini coefcients based on dif-
ferent concepts for the purpose of cross-country comparisons. This
approach, however, hinges on a very strong assumption, namely
that the differences in Gini coefcients based on different concepts
are the same across countries and over time. For transition countries,
this assumption clearly does not hold. For instance, the compari-
son of the consumption-based Gini coefcient with the disposable
income-based one obtained from the same survey data indicate that
the former is 2 percentage points higher than the latter in Poland, is
Table 6.4
tory variables: Robustness to the denition of the dependent variable and the choice of
the data series
Full model Full model Full model (Gini

(disposable (the WIDER coefcients based
income Gini database Gini on the same
Explanatory coefcients coefcients denition within
variable only) only) countries)
1 2 3 4
Intercept
GDPPC 6.623*** 6.910*** 5.727***
(1.395) (1.243) (1.293)
GDPPC_S 0.388*** 0.399*** 0.334***
(0.083) (0.073) (0.077)
INFL 0.032** 0.007 0.013
(0.016) (0.013) (0.015)
UNEMP 0.023 0.025 0.019
(0.025) (0.021) (0.024)
CONSG 0.043 0.079 0.062
(0.067) (0.051) (0.060)
INDVA 0.179* 0.287*** 0.205**
(0.107) (0.069) (0.101)
PRIVS 0.133*** 0.082*** 0.113***
(0.032) (0.029) (0.029)
Number of countries 18 21 23
Number of observations 100 110 113
R 2 adj. 0.58 0.66 0.55
F-value 18.44 29.48 18.35
Estimated threshold level 5084 5758 5284
(PPP-adjusted GDP per
capita, 1992 USD)
skedasticity.
of the same magnitude in Russia, and is 5 percentage points lower

in Georgia (World Bank 2000). Thus, for these countries such an
adjustment can hardly be an improvement. In this situation the only
solution is to use observations on inequality that are as fully con-
sistent as possible (Atkinson and Brandolini 2001).
That is what I attempt to do in this chapter. It is important to note
also that since I use the xed effects estimation any adjustments to
the Gini coefcients based on the same concept within countries
would be canceled out anyway. Moreover, when I estimate the
pooled regression by including dummies for Gini coefcients based
on different concepts, I do not nd those dummies to be signicant.40
I also investigate the robustness of the results to the use of the
random effects rather than xed effects estimation technique. The
regression results (see table 6.5) indicate that the xed effects and
random effects estimates are generally very similar.
I next verify to what extent the results could be driven by obser-
vations for a particular time period. The parameter estimates are
found to be fairly robust to the removal of any single period from the
estimation.41 I note, however, that the coefcient on ination shows
to be insignicant when the model is estimated without data for
1993. That is not surprising since for most countries in the transi-
tional region 1993 was the year of hyperination. Thus, the elimi-
nation of this year from the estimation is likely to substantially
underestimate the impact of ination on the distribution of income.
As the countries in the sample differ widely in their levels of
development and growth experiences during the transition (despite
being collectively referred to as transitional economies), it is neces-
sary to investigate the robustness of the results to the regional
coverage. I rst test the robustness of the results with respect to
countries by removing one country at a time. Although the values of
the coefcients (not reported here) uctuate slightly, their magni-
tudes and signicance levels are largely in line with those reported.
In view of the countries differences in institutional characteristics
and macroeconomic performance during the transition, I then esti-
mate the model separately for the FSU and EE regions. I do not
argue that such a division of countries into subsamples is perfect as
countries within EE and the FSU regions are not homogenous, but
it seems to be a natural choice in many respects. First, economic
decline in EE was on average less profound and persistent than in
the FSU. Second, income inequality in EE has increased much less
Table 6.5
Random-effects estimates from the regression of the Gini coefcient on selected
explanatory variables: All countries
Reduced Reduced
Full model Full model model 1 model 2
Explanatory (quadratic (linear (quadratic (quadratic
1 2 3 4 5
Intercept 19.749*** 6.638*** 17.926*** 20.524***
(4.288) (0.464) (4.218) (4.682)
GDPPC 3.427*** 0.234*** 2.811*** 3.673***
(1.030) (0.046) (1.011) (1.115)
GDPPC_S 0.190*** 0.151*** 0.197***
(0.061) (0.060) (0.066)
INFL 0.027** 0.025* 0.021
(0.013) (0.014) (0.014)
UNEMP 0.003 0.017 0.027* 0.073***
(0.020) (0.020) (0.016) (0.016)
CONSG 0.036 0.057 0.080* 0.094*
(0.047) (0.048) (0.046) (0.052)
INDVA 0.312*** 0.352*** 0.389***
(0.073) (0.075) (0.071)
PRIVS 0.080*** 0.068**
(0.027) (0.028)
Number of 24 24 24 24
countries
Number of 129 129 129 129
observations
R 2 adj. 0.64 0.62 0.63 0.53
Estimated 8374 13422 11260
threshold level
(PPP-adjusted
GDP per capita,
1992 USD)
parentheses.
than in the FSU. Third, in contrast to the FSU, most EE countries

already had at least some rudimentary elements of the market econ-
omy (e.g., a private sector) before the transition, and were much
more effective with reform implementation during the transition.
Finally, social safety nets in EE during the transition are widely rec-
ognized to have been much stronger than in the FSU.
The results of separate estimations for the FSU and EE are pre-
sented in table 6.6. A number of interesting observations can be
made based on them. First, when the estimation is performed sepa-
rately for the FSU and EE countries, a Ushaped relationship be-
tween income inequality and the level of economic development
becomes less evident for the FSU and collapses completely for EE.
The estimation of the respecied model (excluding GDPPC_S to test
for the linear relationship) indicates that the inequality-development
relationship is in fact linear within these regions. Moreover, the para-
meter estimate on GDPPC is negative for the FSU region, but posi-
tive for EE countries (see columns 45, table 6.6). These results are
consistent with the found for the whole sample threshold effect in the
relationship between income inequality and economic development.
In fact, while the FSU countries mostly fell below the estimated
threshold level of per capita GDP, EE countries were positioned
mostly above the threshold. The estimated coefcients on GDPPC
suggest that a $1,000 increase in per capita GDP (constant 1992 pri-
ces) at the region-specic means would be associated with a 2.12
Gini point decrease in income inequality at the mean in the FSU and
a 0.89 Gini point increase in income inequality at the mean in EE.
A question of great interest is what makes economic growth push
income inequality in different directions in the FSU and EE? It could
be that explanation lies in the institutional environment in which the
growth takes place. For instance, the levels of rent seeking and cor-
ruption in the FSU have been much higher than in EE. That may be
one explanation of why income inequality in the former region
has risen despite a dramatic economic decline.42 It is also important
to note that while the FSU countries were experiencing economic
decline over most of the transition decade, the EE countries were
growing.43 The impact of economic recessions and recoveries on
income distribution is not necessarily symmetric.
Ination is found to have a signicant impact on income inequal-
ity in the FSU countries, but not in EE. This result clearly comes from
much higher ination in the former region. In contrast to the FSU
Table 6.6
tory variables: EE versus the FSU
Full model
FSU Full model Full model Full model
Explanatory (quadratic EE (quadratic FSU (linear EE (linear
1 2 3 4 5
Intercept
GDPPC 2.766** 3.424 0.314*** 0.252***
(1.373) (2.679) (0.095) (0.089)
GDPPC_S 0.146* 0.212
(0.082) (0.154)
INFL 0.039** 0.031 0.039** 0.026
(0.019) (0.020) (0.019) (0.020)
UNEMP 0.014 0.047* 0.028 0.031
(0.042) (0.024) (0.042) (0.021)
CONSG 0.006 0.076 0.021 0.046
(0.072) (0.061) (0.073) (0.055)
INDVA 0.301** 0.233** 0.339** 0.203**
(0.127) (0.094) (0.129) (0.091)
PRIVS 0.060 0.100*** 0.044 0.118***
(0.051) (0.038) (0.041) (0.036)
Number of 15 9 15 9
countries
Number of 65 64 65 64
observations
R 2 adj. 0.72 0.67 0.70 0.65
F-value 20.83 20.11 22.88 21.99
Estimated
threshold level
(PPP-adjusted
GDP per capita,
1992 USD)
skedasticity.
countries, many of which experienced hyperination (with annual

ination measured in hundreds percent), most EE countries wit-
nessed ination rates relatively modest by transitional standards.
Deindustrialization appears to be associated with rising income
inequality in both EE and the FSU regions. This nding is robust
to the alternative model specications. The estimated coefcients
from the linear specications (see columns 45, table 6.6) suggest
that a 10 percent decline in the share of industrial sector in the
economy would be associated with a 1.17 and 0.54 percentage point
increase in the Gini index at the region-specic means in the FSU
and EE, respectively.
The parameter estimates on PRIVS suggest that the growing
private sector had an inequality-increasing effect exclusively in EE
countries. This result is robust to the model specication. For in-
stance, when INDVA is omitted from the regression, the coefcient
on PRIVS becomes even more signicant for EE, but remains insig-
nicant for the FSU. The differential impact of privatization in two
regions is striking given that the private sector expanded markedly
in all economies. It is clearly a look at what makes privatization
processes in EE and the FSU different that may provide the explana-
tion. For instance, one consequence of the growing private sector in
EE was a signicant increase in returns to education and a rise in
wage disparities (World Bank 2000). Conversely, privatization in the
FSU countries did not substantially raise educational premiums,
probably because of the excess supply of highly skilled labor.
Unemployment is not found to affect income distribution. The
parameter estimate is only marginally signicant for EE in one spec-
ication (see column 3, table 6.6). I have also tried several other
specications of the model and the results (not shown here) gener-
ally suggest that the exclusion of at least one variable reecting the
structural change in the economy, such as PRIVS or INDVA, from
the linear model makes the coefcient on UNEMP signicant (at a
10% level) for EE, but not for the FSU. The share of government
consumption in GDP does not appear to explain the distributional
outcomes neither in EE nor in the FSU.
With regard to the impact of CLI and IPF on income distribution
in EE and the FSU, the regression results suggest that economic lib-
eralization was associated with rising income inequality in both
regions (see columns 23, table 6.7). The extent of political freedom,
however, does not affect income inequality in either region.44
Table 6.7
Estimates from the regression of the Gini coefcient on selected explanatory variables:
EE versus the FSU
Model Model
with war with war
Model Model dummy dummy
with CLI with CLI (pooled (pooled
Explanatory (LSDV) (LSDV) regression) regression)
variable FSU EE FSU EE
1 2 3 4 5
Intercept 4.440*** 4.510***
(0.488) (0.626)
GDPPC 0.386*** 0.243***
(0.105) (0.096)
Ination > 500 0.122** 0.006 0.208*** 0.026
dummya (0.059) (0.051) (0.047) (0.059)
UNEMP 0.050 0.001 0.005 0.047
(0.041) (0.025) (0.036) (0.031)
CONSG 0.082 0.046 0.024 0.209***
(0.084) (0.055) (0.078) (0.058)
INDVA 0.222* 0.201* 0.376*** 0.286**
(0.133) (0.103) (0.102) (0.136)
PRIVS 0.132*** 0.068*
(0.050) (0.040)
CLI 0.039*** 0.041***
(0.019) (0.008)
IPF
War dummy 0.112* 0.193***
(0.067) (0.061)
Number of 15 9 15 9
countries
Number of 65 64 65 64
observations
R 2 adj. 0.999 0.999 0.69 0.64
F-value 2705.14 8450.25 15.71 10.57
Estimated
threshold level
(PPP-adjusted
GDP per capita,
1992 USD)
parentheses. Yule-Walker (iterated) method was used to correct for serial correlation
and heteroskedasticity.
a excluded category is ination < 20 percent annual. The coefcients on other catego-
ries (2150, 51100, 101500) are not reported since they are not signicant.
I also look at the impact of civil conicts on income distribution

separately for EE and the FSU regions using the pooled regressions.
The regression results (see columns 45, table 6.7) indicate that the
periods of civil wars were associated with rising income inequality
in both regions.
Finally, it is worth noting that the results obtained from the sepa-
rate estimations for EE and the FSU must be treated with caution
due to the relatively small regional samples.
6.8 Conclusions
The main goal of this chapter is to identify the factors that caused
dramatic changes in income inequality in the transitional countries
of EE and the FSU throughout the 1990s. The empirical analysis is
performed using a unique panel of inequality estimates that cover
twenty-four transitional countries over the period 19891998. The
econometric approach employs panel data estimation methods.
I nd support for a normal U-shaped relationship between income
inequality and per capita GDP for the transitional region as a whole.
It suggests that for a country below (above) some threshold level
of development economic growth is associated with falling (rising)
income inequality. Specically, the relationship between income in-
equality and economic growth is shown to be negative for coun-
tries of EE, but positive for those of the FSU. The results suggest that
economic recovery-promoting policies may certainly have an equal-
izing effect on income distribution in some transition countries.
However, at least in the short run, there can be a trade-off between
economic growth and income inequality in other countries.
Although undoubtedly important, the relationship between in-
come inequality and economic growth does not represent the main
focus of this study. I have searched for specic economic factors and
noneconomic forces that determine the changes in income distribu-
tion during the transition.
The empirical results indicate that economic liberalization and
structural adjustments are associated with rising income inequality.
More specically, I nd that a 10 percent increase in the CLI at the
mean is associated with a 0.27 and 0.34 percentage point increase in
the Gini coefcient at the region-specic means in the FSU and EE
countries, respectively. De-industrialization has a strong impact on
income distribution in both regions. A 10 percent decline in the share
of industrial sector in the economy is related to a 1.17 and 0.54 per-

centage point increase in the Gini coefcient at the mean in the FSU
and EE countries, respectively. Although the economies of both EE
and the FSU regions have been substantially privatized during the
transition, the evidence suggests that a rapidly growing private sec-
tor has contributed to rising income inequality in EE countries only.
A 10 percent growth in the share of private sector in the economy is
associated with a 0.31 percentage point increase in the Gini coef-
cient at the mean in these countries.
It is important to note that some increase in income inequality due
to structural reforms associated with the transition from centrally
planned to market economy is largely inevitable and should not
be considered in the negative light. These reforms may have siz-
able longer-term rewards by strengthening incentives, creating new
jobs, and fostering economic growth. Ultimately, it is better to be
unequally rich than equally poor. Nevertheless, the policies aimed
at facilitating the transition of workers from the public to the pri-
vate sector, and from the manufacturing sector to services, may be
of paramount importance for the distributional outcomes of the
reforms. Although there is some evidence to suggest that unem-
ployment may be positively associated with income inequality, the
effect is not robust to model specication.
The degree of government involvement in the economy through
government consumption generally does not seem to have an impact
on income distribution. I nd that hyperination makes the distri-
bution of income more unequal. This nding may certainly contrib-
ute to the explanation of why income inequality in the FSU countries
(most of which experienced hyperination at the start of the transi-
tion) increased much more than in EE countries, where ination
levels have been relatively moderate. The important policy implica-
tion of this nding is that macroeconomic stabilization not only
fosters economic recovery, but is also benecial in terms of dis-
tributional outcomes.
Finally, I have also investigated the role of some forces outside
economic domain in determining income inequality in transition
economies. The empirical evidence indicates that civil conicts are
associated with rising income inequality. On average, they lead to a
13.3 percent higher income inequality. The extent of political rights
and civil liberties, measured by the IPF, is not found to affect income
distribution. Nevertheless, this index is strongly correlated with the
indicators of structural changes in the economy, suggesting that

political rights and civil liberties are likely to affect income distribu-
tion indirectly.
To conclude, the avenue of research undertaken in this chapter
appears promising, for it reveals forces inuencing income dis-
tribution in transitional countries. However, I certainly have not
exhausted all factors explaining the dynamics of income inequality
in the transitional region, and thus further research here may be
benecial.
6.9 Appendix
188
Table 6.A1
Description of the Inequality Data Used in the Empirical Analysis
No. Gini Gini Max. Area/ Sample/
Region/ of index index value Welfare population reference Source
country Start End obs. (start) (end) (year) measure coverage unit (year)
1 2 3 4 5 6 7 8 9 10 12
I. FSU
a. Baltic states
Estonia 1989 1998 9 23.00 36.97 36.97 (98) DI, GI (90) All/All HH/HH pc WIID, BM (89)
Latvia 1989 1998 5 22.50 32.10 32.60 (97) DI, GI (89) All/All HH/HH pc WIID, BM (89)
Lithuania 1989 1999 6 22.50 34.00 35.04 (94) DI, GI (89) All/All HH/HH pc WIID, BM (89),
WB (99)
b. Western CIS
Belarus 1989 1999 5 22.80 26.00 26.00 (99) DI, GI (89) All/All HH/HH pc WIID, BM (89),
WB (99)
Moldova 1993 1997 2 36.50 42.00 42.00 (97) DI, GI (93) All/All HH/HH pc BM (93), WB (97)
Russia 1989 1996 5 23.80 37.83 37.83 (96) GI All/All HH/HH pc WIID, BM (89)
Ukraine 1989 1999 6 25.80 32.00 32.00 (99) DI, GI (97) All/All HH/HH pc WIID, WB (99)
c. Caucasus
Armenia 1990 1998 6 26.90 59.00 62.14 (95) DI, GI (90) All/All HH/HH pc WIID, WB (98)
Azerbaijan 1995 1999 2 44.00 43.00 44.00 (95) CS All/All HH/HH pc WB
Georgia 1989 1997 4 31.30 51.86 58.71 (96) DI, GI (88, 90) All/All HH/HH pc WIID
Oleksiy Ivaschenko
d. Central Asia
Kazakhstan 1990 1996 4 29.70 35.00 35.00 (96) DI, GI (90, 93) All/All HH/HH pc WIID, WB (96)
Kyrgyz Republic 1990 1997 3 30.80 47.00 55.30 (93) DI, GI (90) All/All HH/HH pc WIID, BM (93),
WB (97)
Tajikistan 1989 1990 2 31.80 33.40 33.40 (90) GI All/All HH/HH pc WIID
Turkmenistan 1989 1993 3 31.60 35.80 35.80 (93) DI, GI (89, 90) All/All HH/HH pc WIID
Uzbekistan 1990 1994 3 31.50 33.00 33.30 (93) DI, GI (90, 94) All/All HH/HH pc WIID
II. Central EE
Czech Republic 1989 1997 9 19.36 27.64 28.14 (96) DI All/All HH/HH pc WIID
Hungary 1989 1997 7 21.41 24.58 24.58 (97) DI All/All HH/HH pc WIID
Poland 1989 1998 9 25.05 32.00 34.20 (97) DI All/All HH/HH pc WIID, WB (98)
Slovak Rep. 1989 1997 9 18.06 23.36 24.83 (96) DI All/All HH/HH pc WIID
III. South EE
Bulgaria 1989 1997 8 24.47 34.59 34.78 (96) DI, GI (89, 91) All/All HH/HH pc, WIID
HH (89, 98)
Romania 1989 1997 9 23.24 30.27 31.18 (95) DI All/All HH/HH pc WIID
IV. FY
Croatia 1989 1998 4 25.10 33.30 33.30 (98) DI All/All HH/HH pc WIID
Macedonia 1990 1997 5 34.90 36.65 36.94 (96) DI All/All HH/HH pc WIID
Slovenia 1991 1998 4 22.71 25.00 25.05 (93) DI All/All HH/HH pc WIID, WB (98)
189
Sources: UNU/WIDER-UNDP World Income Inequality Database, Version 1.0, September 12, 2000, provides further reference on the source
190
and estimation methodology for each data point drawn from this database. The data by Branko Milanovic are from appendix 4, The Original
Income Distribution Statistics, in his book Income, Inequality and Poverty during the Transition from Planned to Market Economy (1998). The data
by the World Bank are taken from appendix D, Poverty and Inequality Tables, in the book Making Transition Work for Everyone: Poverty and
Inequality in Europe and Central Asia (World Bank 2000).
Note: Disposable income (DI) is equal to gross income (GI) minus payroll and direct personal income taxes (PIT). Gross income consists of earn-
ings from labor, cash social transfers, self-employment income, other income (gifts, income from property) and in-kind consumption (e.g.,
agricultural products grown on a households plot of land). It is argued (Milanovic 1998) that the difference between gross and disposable
incomes is negligible for transition countries (especially for the pretransition period) as gross income already excludes payroll taxes withdrawn
at the source, and PIT is minimal (less than one percent of gross income). That allows one to use the Gini coefcients based on gross incomes as
the benchmarks for the levels of income inequality observed before the transition (mostly for the FSU countries, for which the pretransition
disposable income Gini indexes are not available). It is very important to have these pretransition observations in the sample since the evidence
suggests that most of the variation in income inequality over time has taken place over the initial period of transition and economic collapse.
As the rst transition period surveys were often conducted a few years into the transition process, by taking the estimates of inequality derived
solely from these surveys one would signicantly underestimate the changes in inequality over time (which is what I want to explain).
The Gini coefcients for Romania and Macedonia are based on disposable monetary income, which does not include in-kind consumption.
These Gini coefcients are likely to overestimate the levels of inequality, but not the changes in inequality. Note that two data points (Azerbai-
jan, 1995, 1999) are Gini coefcients based on consumption (CS). These observations are used due to the lack of alternatives. They are not found
to inuence the overall results. The Gini index for 1988 is used in the absence of 1989 data.
The data coming from the Family Budget Surveys (FBS) (mostly 1989 data in our sample) are not completely representative and may
underestimate inequality as FBS excluded pensioner-headed households and households headed by the unemployed. However, the estimates
of inequality obtained from transition-year surveys can also be downward-biased due to decreased response rates among the rich, inadequate
coverage of informal sector incomes, etc. (for a detailed discussion of these and other data issues see Milanovic 1998). It is not clear, though,
how all these biases would, on the net, affect the changes in inequality. In any case, there is not much that one can do about these sorts of
problems except trying to use observations that are as fully consistent as possible (Atkinson and Brandolini 2001). That was a guiding prin-
ciple in the compilation of the data.
Oleksiy Ivaschenko
Table 6.A2
Descriptive Statistics and Variance Decomposition
Std. Dev.
Variable Mean Min Max Overall Between Within
GINI 30.5 17.8 62.1 9.21 7.87 5.14

GDPPC 6322.4 1649.9 13764.9 2789.40 2680.60 1191.45
INFL 247.7 7.6 9750.0 945.22 669.02 793.09
UNEMP 7.9 0.0 38.8 7.39 6.84 3.89
CONSG 18.0 6.0 27.4 4.96 4.27 2.90
INDVA 38.2 15.3 67.4 8.71 5.30 7.02
PRIVS 38.4 5.0 75.0 21.44 13.64 17.55
CLI 2.6 0.0 7.7 2.11 1.21 1.81
IPF 3.4 1.5 7.0 1.60 1.44 0.89
Source: Authors calculations.
Note: The number of observations for all variables is 129. The overall and within (over
time) standard deviations are calculated over all 129 observations. The between
(across countries) standard deviation is calculated over the means for twenty-four
countries.
GDPPC GDP per capita, PPP-adjusted USD in 1992 prices;
INFL Ination rate, measured by the year-on-year change in CPI (percent);
UNEMP Share of unemployed in total labor force (percent);
CONSG Government consumption as share of GDP (percent);
INDVA Share of industry value added in GDP (percent);
PRIVS Private sector share in GDP (percent);
CLI Cumulative Liberalization Index (score);
IPF Index of Political Freedom (score).
Table 6.A3
192
The matrix of the Pearson correlation coefcients

GINI GDPPC INFL UNEMP CONSG INDVA PRIVS WAR_D CLI IPF
GINI 1.00
GDPPC 0.75 1.00
(0.00)
INFL 0.15 0.25 1.00
(0.10) (0.01)
UNEMP 0.26 0.12 0.05 1.00
(0.00) (0.18) (0.55)
CONSG 0.24 0.14 0.04 0.01 1.00
(0.01) (0.11) (0.68) (0.99)
INDVA 0.62 0.41 0.14 0.42 0.05 1.00
(0.00) (0.00) (0.10) (0.00) (0.58)
PRIVS 0.30 0.11 0.22 0.52 0.12 0.51 1.00
(0.00) (0.22) (0.01) (0.00) (0.17) (0.00)
WAR_D 0.56 0.43 0.35 0.09 0.27 0.29 0.02 1.00
(0.00) (0.00) (0.00) (0.30) (0.00) (0.00) (0.86)
CLI 0.25 0.04 0.27 0.68 0.04 0.51 0.81 0.05 1.00
(0.00) (0.68) (0.00) (0.00) (0.63) (0.00) (0.00) (0.56)
IPF 0.11 0.37 0.27 0.49 0.11 0.21 0.57 0.31 0.59 1.00
(0.20) (0.00) (0.00) (0.00) (0.22) (0.02) (0.00) (0.00) (0.00)
Note: All variables except WAR_D, CLI and IPF are in the natural log form. ln(1 INFL/100) and ln(1 UNEMP) are used for respectively
INFL and UNEMP. WAR_D is a dummy variable equal to 1 for each year since an internal conict has taken place in a given country. The
values in parentheses indicate Prob. > jrj under H0 : Rho 0; (0.00) means that the p-value is less than 0.005. Dashes indicate data are not
available.
Oleksiy Ivaschenko
Notes
I am grateful to Arne Bigsten, Francois Bourguignon, Bjorn Gustafsson, Stephan Kla-

sen, Michael Lokshin, Per Lundborg, Jukka Pirtilla, Ali Tasiran, an anonymous referee
and seminar participants at the 2001 Panel Data Workshop at Gothenburg Univer-
sity, 2001 Jamboree Conference at Mannheim University, 2001 CESifo Conference in
Munich, 2001 WIDER Conference in Helsinki, and 2002 CESifo Conference in Elmau
for useful comments on an earlier draft of this chapter. I am entirely responsible for all
remaining errors.
1. In many countries of the FSU the outputs declined to 3040 percent of their pre-
transition levels.
2. A number of recent studies (e.g., Forbes 2000) suggest, however, that in the short
run the relationship between inequality and growth could be positive.
3. Grun and Klasen (2001) applied a set of inequality-adjusted indicators of well-being
to measure aggregate welfare in transition countries. They found that an adjustment
for inequality signicantly inuences the ranking of transition countries in terms of
their absolute levels of well-being and the achievements in well-being over time.
4. In our sample, the mean annual ination is 248 percent, and the maximum is 9,750
percent (Turkmenistan 1993).
5. The regressivity of the ination tax and imperfect indexation have been found,
using micro-level data, to increase income inequality in Brazil during the years of high
ination (Ferreira and Litcheld 1998).
6. The ofcial unemployment statistics are likely to signicantly understate the real
level of unemployment. Indeed, unofcial estimates indicate signicantly higher levels
of unemployment. Moreover, many people are registered as employed even when
they are not receiving payment for their labor.
7. Government consumption, however, can affect the distribution of income not only
on the expenditure side, but also through the tax collection.
8. Another mechanism for government to inuence income distribution is through

social security transfers. These transfers are found to reduce inequality in a cross-
country study of both advanced and developing countries by Milanovic (1994), but
appear insignicant in a study of OECD countries by Gustafsson and Johansson
(1999). I do not analyze the impact of social transfers in this chapter due to the lack of
data.
9. The level of income inequality in EE and the FSU before the transition was widely
considered to be lower than in the rest of the world (Atkinson and Micklewright 1992).
10. In many transitional countries the share of private sector in total economy has
increased from barely existing to from 50 to 60 percent.
11. Diminishing relative importance of the industrial and/or agricultural sector was
offset by the growing relative importance of services sector, which in many countries
has increased 1.52 times.
12. Note that if the labor force moves from the remaining state-owned industrial sec-
tor to the private sector, the potential effect of this on income distribution would be
captured by the privatization variable.
13. About half of the WIID database is formed by K. Deininger and L. Squires 1996
database. However, for the transitional region most of the data in the database come
from the UNICEF/IRC TransMonee2000 Database, Florence.
14. In the original WIID database most of the countries are represented with multiple
time series of inequality estimates that often are not compatible.
15. As there are only ve observations for 1999, in the estimation they are used as
1998 values.
16. This is true, however, only if these differences are systematic. I am thankful to a
referee for pointing this out to me.
17. As a check of how good this proxy is, I estimate the model with the industrial
employment variable as well. The results (discussed in what follows) indicate that
industry value added is a very good proxy indeed.
18. The data are updated up to 1998.
19. The assumption is that the effect of civil conict (if any) on income inequality is
likely to persist for a certain period.
20. In the case of Russia, I refer to the military conict in Chechnya.
21. See hhttp://www.alertnet.orgi.

22. The results of these tests are not shown but are readily available upon request
from the author.
23. H0 is that there is no correlation. The test statistic is distributed as Xk2 , where k
denotes the dimension of the slope vector b (Baltagi 1995, 68).
24. The random effects estimator combines the within and between estimators, thus
giving some weight to the cross-country variation.
25. F-test of the null hypothesis that country-specic effects are not signicant yields
an F-value of 6.14, which is higher than a critical value of 2.07. This means that I can
reject the hypothesis that there are no country-specic effects of omitted variables. The
same test was conducted for time effects. In this case the F9;88 statistic was 2.10, indi-
cating that it is not possible to reject the hypothesis of no time-specic effects. Hence,
the use of the one-way model is appropriate.
26. Specifying the original formulation of equation (1) as yit a i b 0 Xit eit , the for-
mulation in terms of deviations from the country means becomes:
yit yi b 0 Xit Xi eit ei ; where yi Syit =t; Xi SXit =t; ei Seit =t:
27. I note here that the test of multicollinearity for the linear model indicates that the
highest condition index equals 4.17, which is below a cutoff point of 10 suggested in
the literature. Hence, I do not have a problem of severe collinearity in the estimation of
the model.
28. To detect outliers and inuential cases, I have conducted inuence diagnostics
such as the studentized residuals, the hat matrix, the COVRATIO statistic, DFFITS,
and DFBETAS (Belsley, Kuh, and Welsch 1980; Bollen and Jackman 1985). I then
deleted those observations that were detected inuential by at least three tests. These
observations turned out to be Moldova (1990), Russia (1998), Tajikistan (1998), and
Uzbekistan (1989). However, when the model is estimated with these outliers included
the results are very similar to those reported in table 6.2.
29. I have also estimated the model with GDPPC and GDPPC_S being the only
explanatory variables. The parameter estimates (not shown here) on both terms are
statistically signicant at 1 percent and 5 percent levels respectively (negative for
GDPPC and positive for GDPPC_S). The F-test statistics for their joint signicance is
45.71 (signicant at a 1% level). Adjusted R 2 for the model is equal to 0.46.
30. The recent study by the World Bank (2000) has found that ination volatility is also
associated with distributional costs in the transitional region.
31. Unemployment data can also be a poor indicator of labor market conditions since
in many transition countries adjustments in the labor market manifested themselves in
underemployment and nonpayment of wages rather than in the shedding of labor.
32. I note that when the model is estimated using the available observations on the
industrial sector employment as a share of total employment (100 observations, 23
countries), instead of those on INDVA, the parameter estimate is nearly identical to
the one on INDVA reported in table 6.2. These estimation results are not reported here
for brevity.
33. This nding is in contrast to that reported in the World Bank (2000), where they
use the same reform index as I do, but a smaller sample of countries (20) and a differ-
ent estimation technique (a pooled regression rather than xed effects).
34. Note that the higher value of the IPF means fewer political rights and civil
liberties.
35. The xed effects model cannot be estimated in this setting. Thus I perform a
pooled regression. A pooled regression refers to an OLS regression with a single
overall intercept.
36. The civil war dummy remains signicant at a 5 percent level even when GDPPC
and GDPPC_S enter the regression. The estimated effect, however, is slightly lower in
this case than the one reported.
37. Note that the estimation of the model in deviations from the country means
requires that the Gini coefcients are as comparable as possible within countries over
time but not necessarily across countries. For this reason, the use of consumption-
based Gini coefcients for Azerbaijan (see table 6.A1) does not cause a problem since
like is compared with like. The same applies to the disposable monetary income Gini
coefcients for Romania and Macedonia.
38. Unfortunately, I cannot test the robustness of our ndings to the use of alternative
measures of inequality, as only the Gini coefcients are available.
39. One may not conclude here on the quality of different data series, though, as the
changes in the parameter estimates could be driven by exclusion of the observations
for particular countries and/or time periods.
40. The comparison category is Gini coefcient based on the household per capita
disposable income.
41. These results are not shown here but are available from the author upon request.
42. I do not have sufcient longitudinal data on the levels of corruption in transitional
countries to run the xed effects regression. However, a simple regression of the
average Gini coefcient during the transition period on the corruption perception
index for sixteen transition countries of EE and the FSU indicates that higher corrup-
tion is associated with larger income inequality (the parameter estimate is signicant
at a 7% level). The regression results are not shown here but are available from the
author upon request.
43. This is reected in the composition of our sample, where the majority of observa-
tions for the FSU and EE countries cover the periods of economic decline and eco-
nomic recovery, respectively.
44. The results are not reported here but are available from the author upon request.
I have also estimated several alternative specications with IPF, but the parameter
estimate on IPF is found to be insignicant.
References
Aghion, P., and S. Commander. 1999. On the Dynamics of Inequality in the Transi-
tion. Economics of Transition 7(2): 275298.
Ahluwalia, M. S. 1976. Income Distribution and Development: Some Stylized Facts.
American Economic Review 66: 125135.
Alesina, A., and D. Rodrik. 1994. Distributive Politics and Economic Growth. Quar-
terly Journal of Economics 109(2): 465490.
Anand, S., and S. M. R. Kanbur. 1993. Inequality and Development: A Critique.

Journal of Development Economics 41: 1943.
Atkinson, A., and A. Brandolini. 2001. Promise and Pitfalls in the Use of Secondary
Data Sets: Income Inequality in OECD Countries as a Case Study. Journal of Economic
Literature 39(2): 771799.
Atkinson, A., and J. Micklewright. 1992. Economic Transformation in Eastern Europe and
the Distribution of Income. Cambridge: Cambridge University Press.
Baltagi, B. H. 1995. Econometric Analysis of Panel Data. West Sussex, UK: John Wiley
and Sons Ltd.
Barro, R. J. 1997. Determinants of Economic Growth: A Cross-Country Empirical Study.
Belsley, D., E. Kuh, and R. E. Welsch. 1980. Regression Diagnostics: Identifying Inuential
Data and Sources of Collinearity. New York: Wiley.
Birdsall, N., D. Ross, and R. Sabot. 1995. Inequality and Growth Reconsidered: Les-
sons from East Asia. The World Bank Economics Review 9(3): 477508.
Bollen, K. A., and R. W. Jackman. 1985. Regression Diagnostics: An Expository
Treatment of Outliers and Inuential Cases. Sociological Methods and Research 13:
510542.
Bourguignon, F., and C. Morrison. 1998. Inequality and Development: The Role of
Dualism. Journal of Development Economics 57: 233257.
Boyd, R. L. 1988. Government Involvement in the Economy and the Distribution of
Income: A Cross-National Study. Population Research and Policy Review 7: 223238.
Bruno, M., M. Ravallion, and L. Squire. 1995. Equity and Growth in Developing
Countries: Old and New Perspectives on the Policy Issues. Paper prepared for the
IMF Conference on Income Distribution and Sustainable Growth, Washington, DC.
Deininger, K., and L. Squire. 1996. A New Data Set Measuring Income Inequality.
World Bank Economic Review 10(3): 565591.
Deininger, K., and L. Squire. 1998. New Ways of Looking at Old Issues: Inequality
and Growth. Journal of Development Economics 57: 259287.
De Melo, M., C. Denizer, and A. Gelb. 1996. From Plan to Market: Patterns of Tran-
sition. The World Bank Policy Research Paper No. 1564. Washington, DC.
Easterly, W. 2001. The Middle Class Consensus and Economic Development. The
World Bank Development Research Group Working Paper No. 2346. Washington, DC.
European Bank for Reconstruction and Development (EBRD). Various years. Transition
Report.
Ferreira, F. H. G. 1999. Economic Transition and the Distributions of Income and
Wealth. Economics of Transition 7(2): 377410.
Ferreira, F. H. G., and J. Litcheld. 1998. The Roles of Structural Factors and Macro-
economic Instability in Explaining Brazilian Inequality in the 1980s. LSE Discussion
Paper No. DARP 41.
Forbes, K. J. 2000. A Reassessment of the Relationship between Inequality and
Growth. American Economic Review 90(4): 869887.
Freedom House. Index of Political Freedom. Available online at hwww.freedomhouse.
orgi.
Garner, T. I., and K. Terrell. 1998. A Gini Decomposition Analysis of Inequality in the
Czech and Slovak Republics during the Transition. Economics of Transition 6(1): 2346.
Gradstein, M., B. Milanovic, and Y. Ying. 2001. Democracy and Income Inequality:
An Empirical Analysis. The World Bank Development Research Group Working
Paper No. 2561. Washington, DC.
Granger, C. W. J. 1969. Investigating Causal Relations by Econometric Models and

Cross-Spectral Methods. Econometrica 37(3): 425438.
Greene, W. H. 1997. Econometric Analysis, 3d ed. New York: Prentice-Hall International

Inc.
Grun, C., and S. Klasen. 2001. Growth, Income and Well-Being in Transition Coun-
tries. Economics of Transition 9(2): 359394.
Gustafsson, B., and M. Johansson. 1999. In Search of Smoking Guns: What Makes
Income Inequality Vary Over Time in Different Countries? American Sociological
Review 64: 585605.
Gustafsson, B., and E. Palmer. 1997. Changes in Swedish Inequality: A Study of
Equivalent Income 19751991. In The Changing Distribution of Economic Well-Being
International Perspectives, ed. P. Gottschalk, B. Gustafsson, and E. Palmer, 293325.
Cambridge, UK: Cambridge University Press.
Hsiao, C. 1986. Analysis of Panel Data. Cambridge, UK: Cambridge University Press.
Kuznets, S. 1955. Economic Growth and Income Inequality. American Economic

Review: 128.
Levy, F., and R. J. Murnane. 1992. U.S. Earnings Levels and Earnings Inequality: A
Review of Recent Trends and Proposed Explanations. Journal of Economic Literature
30: 13331381.
Li, H., L. Squire, and H. Zou. 1998. Explaining International and Intertemporal Vari-
ations in Income Inequality. Economic Journal 108: 118.
Meier, T. 1973. The Distributional Impact of the 1970 Recession. Review of Economics
and Statistics 55: 214224.
Metcalf, C. 1969. The Size Distribution of Personal Income during the Business
Cycle. American Economic Review 59: 657668.
Milanovic, B. 1994. Determinants of Cross-Country Income Inequality: An Aug-

mented Kuznets Hypothesis. The World Bank Development Research Group Work-
ing Paper No. 1246. Washington, DC.
Milanovic, B. 1998. Income, Inequality and Poverty during the Transition from Planned to
Market Economy. Washington, DC: The World Bank.
Milanovic, B. 1999. Explaining the Increase in Inequality during Transition. Eco-
nomics of Transition 7(2): 299343.
Paukert, F. 1973. Income Distribution at Different Levels of Development: A Survey
of Evidence. International Labour Review 108: 97125.
Persson, T., and G. Tabellini. 1994. Is Inequality Harmful for Growth? American
Economic Review 84(3): 600621.
Quah, D. 2001. Some Simple Arithmetic on How Income Inequality and Economic
Growth Matter. CESifo Working Paper.
Ravallion, M. 1995. Growth and Poverty: Evidence for Developing Countries in the
1980s. Economic Letters 48: 411417.
Rodrik, D. 1999. Democracies Pay Higher Wages. Quarterly Journal of Economics 114:
707738.
Stack, S. 1978. The Effect of Direct Government Involvement in the Economy on

the Degree of Inequality: A Cross-National Study. American Sociological Review 43:
880888.
Sylwester, K. 2000. Income Inequality, Education Expenditures, and Growth. Journal
of Development Economics 63: 379398.
Temple, J. 1999. The New Growth Evidence. Journal of Economic Literature 37:
112156.
Thurow, L. 1970. Analyzing the American Income Distribution. American Economic
Review 60: 261269.
Weil, G. 1984. Cyclical and Secular Inuences on the Size Distribution of Personal
Incomes in the UK: Some Econometric Tests. Applied Economics 16: 749755.
World Bank. 2000. Making Transition Work for Everyone: Poverty and Inequality in Europe
and Central Asia. Washington, DC.
Yemtsov, R. 2001. Inequality and Income Distribution in Georgia. Discussion Paper
No. 252. World Bank, Washington DC, and IZA, Bonn.
IV The Political Economy of
Inequality
7 (Re-)Distribution of
Personal Incomes,
Education, and Economic
Performance across
Countries
Gunther Rehme
7.1 Introduction
According to the Kuznets (1955), hypothesis redistribution that makes

the income distribution more unequal should be benecial in the
earlier stages of development. The opposite would hold at later
stages of development.
Following Perotti (1996), at least four theoretical approaches can
be identied that link inequality, redistribution, and growth. All of
them predict that growth increases as equality increases. Indeed a
number of studies such as Alesina and Rodrik (1994), Persson and
Tabellini (1994), or Perotti himself nd that growth is negatively
associated with income inequality across countries. This has estab-
lished what may be called the Conventional Consensus View (CCV).
However, based on inequality data compiled by Deininger and
Squire (1996) that consensus has been challenged by, for example, Li
and Zou (1998), Forbes (2000), Barro (2000), and others who nd a
nonrobust or even positive association, especially for rich countries.
These results may thus be called the New Challenge View (NCV).1
Whatever the association between growth and inequality might
be, it would entail important consequences for the effect of redistri-
bution on growth. One should bear in mind that income inequality
and (income) redistribution are two distinct things. But they are
related as follows: The economic system produces an income distri-
bution and then the state intervenes to redistribute income by levy-
ing taxes and granting subsidies to satisfy some welfare target. After
the state intervention, another income distribution emerges that may
look quite different from the one before the intervention. The net
effect of the intervention is usually called redistribution. Thus, a
202 Gunther Rehme
comparison between the distribution of personal incomes before and

after taxes provides one with a picture of the level of redistribution.
The evidence about the link between redistribution and growth
across countries is mixed.2 For instance, Perotti (1993), Alesina and
Rodrik (1994), and Persson and Tabellini (1994) show that redistri-
bution causes lower growth. However, empirical studies such as
Easterly and Rebelo (1993), or Sala-i-Martin (1996) nd that there is a
positive relation across countries. These results can be reconciled
with theory by models along the lines of, for example, Galor and
Zeira (1993), Saint-Paul and Verdier (1996), Chiu (1998), or Aghion,
Caroli, and Garca-Penalosa (1999).
This chapter focuses on schooling and argues that public educa-
tion, its nance and the way it is undertaken (schooling technology)
are important determinants of income inequality and growth.
In the model, human capital simultaneously determines growth
and income inequality. In this framework the chapter identies
two redistribution mechanisms. On the one hand, redistribution
occurs by means of direct scal redistribution from the well-off to the
not-so-well-off. On the other hand, there is redistribution through
taxes used for expenditure on public education, which redistributes
income by changing the relative wages. It is shown that growth
and pre-tax and post-tax income inequalitymeasured by the Gini
coefcientare rst increasing and then decreasing in human capi-
tal. For a given level of human capital a less efcient education
technology implies lower growth, but also lower after-tax income
inequality and higher measured income redistribution. The intuition
for this is straightforward: To maintain a given level of public edu-
cation a less efcient education sector requires higher, redistributive
taxes. In contrast, a more skill-intensive technology does not affect
the education sector directly, but it implies higher growth and more
inequality and lower measured redistribution for a given level of
human capital.
Most of the recent NCV proponents have based their results on
unadjusted inequality measures using the secondary data set of
Deininger and Squire (1996). As shown by Atkinson and Brandolini
(2001), there are pitfalls when using these data. In particular, they
argue that there is no real alternative to seeking data sets where the
observations are as fully consistent as possible; at the same time, the
choice of denition on which to standardize may affect the con-
clusions drawn (796).
(Re-)Distribution of Personal Incomes 203
Therefore, this chapter uses reliable and consistently dened in-

come data from the Luxembourg Income Study (LIS) for a sample
of relatively rich countries. With these data the models implica-
tions are then set against the empirical evidence. The data suggest
the following:
The association between the education as well as the distributional
variables and growth is not very strong. More secondary as well as
tertiary education or more spending on overall education appear to be
associated with higher growth. Pre-tax and post-tax income inequal-
ity are negatively related to growth, even when controlling for fertility.3
Thus, the situation after redistribution is not conducive to growth,
suggesting that more redistribution might raise long-run growth.
These ndings are interpreted in light of research based on un-
adjusted measures of inequality, which mixes Gini coefcients for
income before and after taxes. It is argued that the coefcients on
unadjusted Gini coefcients are most likely to be biased upwards.
Controlling for standard variables such as initial income or fertil-
ity, the data reveal that the government expenditure on (all levels of)
education is negatively associated with pre-tax and post-tax income
inequality and positively related to redistribution.
Thus, the data suggest that the typical (rich) country would have
higher growth and less inequality if it spent more on education
given its education technology. If the latter becomes worse, higher
inequality and lower growth might ensue, once policy reactions have
responded to that change.
The chapter is organized as follows: Section 7.2 presents the theo-
retical model and derives testable predictions. Section 7.3 confronts
the model with empirical evidence. Section 7.4 provides concluding
remarks.
7.2 The Model
Consider an economy that is populated by N (large) members of

two representative dynasties of innitely lived individuals. The two
dynasties are made up of high-skilled people, Lh , and low-skilled peo-
ple, Ll , where Lh ; Ll denote the total numbers of the respective agents
in each dynasty. The difference between the agents is lumpy, that
is, either an individual has received education certied in the form of
a degree and is then considered high-skilled or it has no degree and
remains in the low-skilled labor pool.
204 Gunther Rehme
By assumption the population is stationary with Lh 1 xN and

Ll 1 1 xN, where x denotes the percentage of high-skilled people
in the population. Each individual supplies one unit of either high-
or low-skilled labor inelastically over time. Furthermore, the high-
skilled agents own an equal share of the total capital stock, which is
held in the form of shares of many identical rms operating in a
world of perfect competition. Thus, high-skilled agents receive wage
and capital income and make investment decisions, whereas low-
skilled agents do not, as they do not own capital by assumption.
Aggregating over rms overall output is produced according to
Yt Bt Kt1a H a ; H a Lh Ll a bLha ; 0 < a < 1; 7:1

where Kt denotes the aggregate capital stock including disembodied
technological knowledge, H measures effective labor in production,
and Bt is a productivity index. The production function is a reduced
form of the following relationship: By assumption effective labor de-
pends on tasks requiring basic skills and tasks requiring high skills.
These tasks are imperfect substitutes in production. On the other hand,
low- and high-skilled people are taken to be perfect substitutes in per-
forming basic tasks. Thus, high-skilled people always perform the
tasks of low-skilled people in the model, but low-skilled people can
never execute tasks that require a degree. See Rehme (1999) for more
details.
The parameter b measures skill-biased productivity differences,
that is, it captures how productively tasks, which require high skills,
contribute to the generation of output in relation to tasks requiring
low skills.4 Notice that each type of labor alone is not taken to be an
essential input in production.
The government runs a balanced budget and uses its tax revenues
to nance public education and to grant direct transfers to the low-
skilled workers.5 Thus, the chapter contemplates two redistributive
mechanisms. On the one hand resources are redistributed directly
from the currently working, relatively rich high-skilled people to the
currently working, relatively poor low-skilled individuals. On the
other hand there is intertemporal redistribution from the currently
working high-skilled to the future high-skilled individuals whose
parents in turn may be low- or high-skilled.
For this the government taxes the accumulated factor of produc-
tion, that is, it taxes the high-skilled agents capital income at a con-
stant rate Q 1 t f. The capital stock (wealth) of the representative
high-skilled agent is kht Kt =Lh so that Gt Qrt kht Lh Qrt Kt , where rt

denotes the return on capital. This implies that Gt =rt Kt Q for all t.
Thus, real resources Qrt Kt t frt Kt are taken from the private
sector where the amount trt Kt is used to nance public education,
which generates high-skilled agents.6 In turn the amount frt Kt is
granted as transfers to the low-skilled and captures that the govern-
ment corrects for post-education income differentials.7
In general, public education depends on government resources
and other factors such as high-skilled labor itself. That is captured by
the following reduced form of the education technology8

t
x where 0 < < 1; c b 1 xt > 0; and xtt < 0: 7:2
c
Thus, if the government channels more resources into education,

it will generate more high-skilled people. However, doing this
becomes more difcult at the margin, as more resources provided to
the education sector lead to a decreasing marginal product of those
resources due to congestion effects.
The parameter c measures the efciency by which resources are
used in education. One may think about 1=c is as a survival rate in a
particular education program. If the latter is higher it would make
funds more efcient when generating graduates and the study times
were of equal length.
In turn, measures the productivity of the whole education sector.
A lower implies that the education sector is more productive and
that a marginal increase in taxes would increase education output
relatively more. This productivity may depend on the duration of
studies, the quality of education, student-teacher ratios, or how cap-
ital (computers) and students are combined for given resources in
efciency units t=c.
7.2.1 The Private Sector
There are as many identical, price-taking rms as individuals and

the rms face perfect competition and maximize prots. By assump-
tion they are subject to knowledge spillovers, which take the form
Bt AKt =N h Akth with h b a. Thus, the average stock of capital,
kt Kt =N, which includes disembodied technological knowledge, is
the source of a positive externality.9 Then simplify by setting h a,
which allows one to concentrate on steady state behavior. For a
206 Gunther Rehme
justication see Romer (1986). As the rms cannot inuence the

externality, it does not enter their decision directly so that
r 1 aAkta Kta H a ;
wh aAkta Kt1a Lh Ll a1 bLha1 ; 7:3

wl aAkta Kt1a Lh Ll a1 :
All agents act price-takingly and have logarithmic utility. (In the
rest of the chapter, time subscripts are dropped for convenience.)
The low-skilled do not invest and consume their entire wage and
transfer income. Thus, their intertemporal utility is given by
y 1n
cl 1 rt K
e dt where cl wl fr : 7:4
0 1 n Ll
In contrast, the high-skilled own all the assets that are collateralized
one-to-one by capital. A representative high-skilled agent takes the
paths of r, wh , wl , Q as given and solves
y 1n
ch 1 rt
max e dt 7:5a
ch 0 1n
s:t: k_h wh 1 Qrkh ch 7:5b

kh 0 given; kh y free: 7:5c
This problem is standard and involves the following growth rate of

consumption:
c_h 1 Qr r
g1 ; 7:6
ch n
which depends on the after-tax return on capital. As the high-skilled

agents own the initial capital stock equally and as they have identi-
cal utility functions, their investment decisions are the same. But
then the wealth distribution will not change over time, and only
high-skilled agents continue to own equal shares of the total capital
stock over time.
7.2.2 Market Equilibrium
For the rest of the chapter normalize the population by setting N 1

so that the factor rewards in (7.3) are given by
r 1 aA1 bx a ; wh aAkt 1 bx a1 ; and wl aAkt :
The return on capital is constant over time and wages grow with the
capital stock. Note that wl t does not directly depend on x. It only
does so indirectly through kt and so gx when t 0 0. See, for exam-
ple, Johnson (1984).
As wh wl 1 bx a1 , high-skilled labor receives a premium over
what their low-skilled counterpart gets, regardless of whether high-
skilled labor is taken as scarce or not. This reects that the high-
skilled may always (perfectly) substitute for low-skilled labor so that
both types of labor receive the same wage wl for routine tasks and
that performing high-skilled tasks is remunerated by the additional
amount wl bx a1 .
The wage premium depends on the percentage of high-skilled
labor in the population, grows over time at the rate g, and is
decreasing in x for a given capital stock.10 On the other hand the
(relative) wage premium wh =wl increases when production is getting
more (high-)skill biased (higher b).11
From the production function gy gk so that for given x per capita
output and the capital-labor ratio grow at the same rate. With con-
stant N and x total output also grows at the same rate as the aggre-
gate capital stock. From (7.6) the consumption of the representative
high-skilled agent grows at g.
Each high-skilled worker owns kh0 K0 =Lh units of the initial
capital stock. Equation (7.5b) implies k_h wh 1 Qrkh ch so that
gkh wh ch =kh 1 Qr where 1 Qr is constant. In steady
state, gkh gk is constant by denition. But wi =k, i h; l is constant as
well, because
wh aAkt 1 bx a1 wl
aA1 bx a1 and aA;
kt kt kt
which implies gk g. But then consumption of the low-skilled also

grows at the rate g. Thus, the economy is characterized by balanced
growth in steady state with gY gK gy gk gch gcl . From (7.6),
(7.7), and t cx 1= , one obtains
1 cx 1= f1 aA1 bx a r
g ; 7:7
n
which is rst increasing and then decreasing, that is, concave in x.12
208 Gunther Rehme
Thus, in the model it is possible that an economy has high-skilled

workers, but does not necessarily do better than another economy
with less high-skilled people.
For given x A 0; 1 the effect of a change in the productivity of the
education sector is given by dg=d lnxcx 1= r= 2 n < 0. One also
veries dg=df < 0, dg=dc < 0 and dg=db > 0 for given x.
Proposition 7.1 The long-run growth rate g is rst increasing and
then decreasing in x. For given x, a less productive education tech-
nology (higher ) or a less efcient use of public resources in educa-
tion (higher c) imply lower growth, whereas a more skill-biased
technology (higher b) implies higher growth. An increase in direct,
purely redistributive transfers to the low-skilled (higher f) lowers
long-run growth.
7.2.3 Income Inequality
The chapter concentrates on simple inequality measures for cur-

rent income such as the Lorenz curve and the Gini coefcient,
because the latter has been used extensively in the recent growth
literature. Note that average current income before ( g) and after
(n) taxes depends on time and is given by m g 1 wl 1 x wh x rkt
and m n 1 wl 1 x frkt wh x 1 f trkt . However, the gross
and net income shares of the low-skilled are constant and given
by
g wlt 1 x a1 x
sl 1 g ; 7:8a
mt 1 bx a
wlt 1 x frkt a1 x f1 a1 bx a
sln 1 ; 7:8b
mtn 1 bx a 1 1 acx 1=
where use has been made of t cx 1= . The corresponding Lorenz
curve (LC), which relates cumulative population shares to cu-
mulative income shares, is presented in gure 7.1. The LC has
a kink at the point A at which 1 x percent of the population
receive sl percent of total income. The Gini coefcient is then calcu-
lated as

1 xsl 1 sl x
G12 xsl 1 sl x;
2 2
Figure 7.1
Ordinary Lorenz curve
where the expression in square brackets represents the area under

the LC.
g
It is easy to see that sln > sl so that the low-skilled get a larger
share of income after taxes than before taxes. If point A corresponds
to the situation before taxation then the income distribution after
taxes would have a kink at a point strictly above A and would imply
a less unequal income distribution. See Atkinson (1970). Further-
more, the Gini coefcients for gross G g and for net G n incomes,
namely,
g
G g 1 sl x and G n 1 sln x; 7:9
g
would report that as well, because sln > sl implies G g > G n . Thus,
taxation for education as well as direct transfers have a long-run
redistributive impact reected in the difference between the respec-
tive Gini coefcients. For that reason the chapter uses (see, e.g.,
Lambert 1993, chap. 2)
Denition 7.1 (Redistribution): Income redistribution is measured
by P 1 G g G n and captures the long-run redistributive impact of
taxation used for education and of the direct transfers granted to the
low-skilled workers.
210 Gunther Rehme
Table 7.1
Numerical simulation
x Gg Gn P g t
0.10 0.386 0.342 0.044 0.0195 0.027
0.15 0.392 0.345 0.047 0.0201 0.054
0.20 0.390 0.339 0.051 0.0202 0.089
0.25 0.382 0.327 0.056 0.0200 0.131
0.30 0.370 0.310 0.059 0.0194 0.179
Note: Parameter values: a 0:7, b 1:13, c 1:43, 0:58, f 0:13, n 3:55, r 0:01.
The difference in Gini coefcients is given by
g 1 af1 bx a a1 xcx 1=

P sln sl : 7:10
1 bx a 1 1 acx 1=
Growth, inequality and redistribution are complicated functions of
x. In order to get an impression of its qualitative features, I have
calibrated the model using the chapters data for thirteen OECD
countries. Focusing on the percentage of the population with tertiary
education, which ranges from 10 to 26 percent with a sample mean
of 15 percent, table 7.1 presents simulations based on reasonable
parameter values.
Thus, income inequalityas measured by the Gini coefcientas
well as growth rst increases, and then decreases with a rising
number of people with tertiary education.13 These simulated effects
are small. In particular, they are smaller for the growth rate than for
the distributional variables.
Clearly, measured redistribution is higher if the government di-
rectly transfers more resources to the low-skilled (higher f). How-
ever, in the relevant range measured redistribution is also increasing
in x and hence in t. But there is no clear (functional) relation be-
tween inequality and redistribution. For instance, when plotting
Px against G g x, it would be possible that two values of P are
associated with the same G g . Furthermore, higher x implies higher
redistribution P but also rst higher and then lower growth.
Proposition 7.2 For a given production and education technology
and many parameter constellations b; c; , the Gini coefcients
G g ; G n for pre-tax or post-tax income inequality are inverted U
shaped in x.
Thus, for sufciently high x an increase in it would lower pre-

tax and post-tax income inequality. Furthermore, such an increase
would often also widen the gap between them and so redistribution
as dened here.
The results have to be interpreted with caution as they are sen-
sitive to changes in the institutional and production features. For
changes in the latter it is not difcult to verify
Proposition 7.3 For a given level of human capital x,
1. a more skill-biased production (higher b) entails higher pre-tax

and (even) higher post-tax inequality and so lower redistribution;
2. a less efcient use of public resources in education (higher c) or
a less productive education sector (higher ) imply no change in
pre-tax, but a reduction in post-tax inequality and hence more
redistribution.
The rst result follows because a higher b has a direct and positive
bearing on the wages of the high-skilled and pre-tax capital income,
but has no direct effect on tax revenues. As a consequence, there is
lower redistribution. The intuition for the second result is the fol-
lowing: If it is relatively more difcult to generate more high-skilled
people, higher taxes are called for. Thus, if two economies have the
same x the one with a less productive education sector must use
more resources to have that x, thereby redistributing relatively more
income.14
7.3 Empirical Evidence
7.3.1 Data and Methodology
Human capital is measured by the percentage of the population from

25 to 64 years of age that has attained at least upper secondary edu-
cation (SECP) or at least university-level (tertiary) education (TERP).
Data for these variables are provided by the OECD for 1996 and
thirty-four countries.15 They collapse the time series dimension into
a single number by attaching weights to the human capital compo-
sition of different generations at a particular point in time and are
taken to represent a long-run process which is approximated by their
time-averages over the sample period.16
212 Gunther Rehme
The nature of the human capital data also serves as a justication

for the methodology employed. Although authors such as Caselli,
Esquivel, and Lefort (1996) argue that growth should be investigated
by means of dynamic panel data methods, these methods may have
their own problems as, for example, argued by Barro (1997, 37) or
Temple (1999, 132). Therefore, the chapter uses time-averaged data
and concentrates on simple statistics, the properties of which may
also be relevant for more sophisticated methods.
A valuable source for data on income distributions for many
countries is the Luxembourg Income Study (LIS), which satises
many minimum quality requirements and provides a consistent data
source. Thus, income inequality is measured by Gini coefcients
based on LIS data.17 Then averages of the Gini coefcients for each
country are taken for the period 19701990 and are meant to reect
long-run within-country inequality. For a similar procedure, see
Deininger and Squire (1998, 268).
The income and recipient concept employed here is gross or net
income per household where the latter has been adjusted by the
square root of household members. These concepts are strictly
adhered to. On the importance of income and recipient concepts in
the measurement of inequality, see, for instance, Cowell (1995) or
Atkinson (1983).
Finally, long-run growth rates were calculated using the Penn
World Table (Mark 5.6) from Summers and Heston (1991). All the
other data are taken from Barro and Lee (1994). Combining these
data sources with LIS data yields a sample of thirteen relatively
rich countries for the period 19701990 for which reliable (good)
inequality data are available.
7.3.2 Findings
The Gini coefcient for individual households gross incomes is

denoted by LIS.G and that for individual households net incomes by
LIS.N. Furthermore, RE 1 LIS.G LIS.N denotes redistribution. In
the sample the Gini coefcients are characterized as follows.
Over the sample period income inequality has risen in some
countries, but not in all. Redistribution has increased in almost all of
them. For instance, in the United States the Gini coefcient went
up from 35.05 in 1974 to 41.81 in 1997. Thus, there was a marked
increase in pre-tax income inequality. For the same period the Gini
coefcient for net income goes up from 31.46 in 1974 to 37.24 in 1997.
On the other hand, redistribution (RE) goes up from 3.59 in 1974 to
4.57 in 1997. Thus, policy in the United States has corrected slightly
increasingly for some of the increase in pre-tax inequality. A similar
picture holds for the United Kingdom so that higher inequality in
pre-tax incomes often seems to be associated with more redistribu-
tion within countries over time.
Sweden has low pre-tax inequality that fell over the period (1967:
32.05; 1995: 26.2). It reduced redistribution from 6 in 1967 to 4.2 in
1995. Thus, Sweden and the United States have very different pre-tax
income inequality, but redistribute approximately the same. On
period averages the United States redistributes more (RE: 4.4) than,
for example, Germany (RE: 3.7), France (RE: 2.49) or Canada (RE:
3.4). All the latter countries have lower pre-tax inequality than the
United States.
Clearly, growth of GDP per capita should be controlled for by
many factors. This is done here by means of simple growth regres-
sions and by focusing on parsimonious models. Following a common
procedure, a benchmark model with often used robust regressors
is used to add education and distributional variables to see what
the latter contribute to the explanation of long-run growth across
countries.
The benchmark model used here for i countries is gi a
b 1 LY70i b 2 LAFERTi b3 CVLIBi i , where LY70 denotes the (nat-
ural) logarithm of GDP per capita in 1970, LAFERT represents the
logarithm of the average fertility rate for the period 19601984,
CVLIB is Gastils index of civil liberties (from 1 to 7; 1 most
freedom) for the period 19721989, and i is a disturbance term.
According to the estimated coefcients, CVLIB does not really add to
the explanation of growth and is dropped in the subsequent anal-
ysis, because it is not the main variable of interest.18
In table 7.2 model (1) is the reduced benchmark model. In all
models the estimated coefcient on LY70 is always negative and that
on fertility (LAFERT) is negative in ten out of the sixteen models.
The estimates then suggest the following.
The association between the human capital as well as the dis-
tributional variables and growth is not very strong. Most of the
coefcients on these variables are statistically insignicant. However,
even small effects may have important and economically signicant
consequences in the long run.19
214 Gunther Rehme
Table 7.2
Growth regressions
(1) (2) (3) (4) (5) (6) (7)
Const. 23.225 23.375 16.656 16.374 23.008 24.755 22.129
(6.344) (5.209) (5.959) (6.198) (5.708) (6.867) (8.886)
[0.004] [0.001] [0.023] [0.030] [0.004] [0.006] [0.038]
LY70 2.262 2.587 1.825 1.844 2.563 2.449 2.121
(0.638) (0.541) (0.641) (0.649) (0.580) (0.707) (0.985)
[0.005] [0.001] [0.021] [0.022] [0.002] [0.007] [0.064]
LAFERT 0.534 0.407 2.413 2.253 0.434 0.656 0.025
(0.905) (0.840) (1.340) (1.306) (0.894) (0.945) (1.681)
[0.568] [0.639] [0.109] [0.123] [0.640] [0.506] [0.988]
SECP 0.029 0.036 0.036 0.030
(0.012) (0.011) (0.012) (0.013)
[0.039] [0.014] [0.015] [0.049]
TERP 0.019 0.018
(0.027) (0.028)
[0.501] [0.535]
GEDU
LIS.G 0.077 0.030

(0.043) (0.059)
[0.108] [0.630]
LIS.N 0.067
(0.039)
[0.121]
RE 0.025
(0.102)
[0.814]
R2 0.642 0.783 0.846 0.842 0.784 0.661 0.671
Obs. 13 13 13 13 13 13 13
Note: The dependent variable is the average growth rate of real GDP per capita over
the period 19701990. The estimation method is OLS. Standard errors are shown in
parentheses and t-probabilities are reported in square brackets.
1. The coefcients on education are positive in all models. How-

ever, secondary education (SECP) appears to contribute more to
linear explanations of long-run growth than tertiary education
(TERP) or education expenditure (GEDU). As an indication notice
the relatively high R 2 s in models (2)(5). For instance, the latter
models suggest that an increase of one standard deviation (11 per-
centage points) in the percentage of people aged 2565 who have
at least upper secondary education (SECP) raises long-run growth by
0.3 to 0.4 percentage points. Models (6)(9) suggest that growth
Table 7.2
(continued)
(8) (9) (10) (11) (12) (13) (14) (15) (16)
22.954 25.886 22.403 21.092 22.102 22.972 20.497 20.911 23.334
(9.439) (7.857) (6.585) (8.665) (9.062) (7.091) (8.241) (8.469) (6.931)
[0.041] [0.011] [0.008] [0.041] [0.041] [0.012] [0.034] [0.036] [0.008]
2.236 2.554 2.235 2.059 2.196 2.287 0.183 1.990 2.270
(1.032) (0.797) (0.653) (0.978) (1.015) (0.702) (1.609) (0.910) (0.687)
[0.062] [0.013] [0.008] [0.069] [0.063] [0.012] [0.912] [0.057] [0.009]
0.266 0.712 0.424 0.061 0.352 0.425 0.183 0.009 0.538
(1.644) (1.006) (0.938) (1.740) (1.703) (0.987) (1.609) (1.528) (0.956)
[0.875] [0.499] [0.662] [0.973] [0.842] [0.678] [0.912] [0.996] [0.587]
0.017 0.024
(0.029) (0.031)
[0.569] [0.470]
0.086 0.071 0.082 0.106
(0.116) (0.135) (0.146) (0.133)
[0.477] [0.611] [0.590] [0.449]
0.017 0.031
(0.065) (0.057)
[0.810] [0.596]
0.016 0.003 0.022
(0.053) (0.063) (0.051)
[0.773] [0.960] [0.672]
0.050 0.051 0.007
(0.139) (0.137) (0.123)
[0.724] [0.719] [0.954]
0.664 0.666 0.663 0.665 0.663 0.668 0.654 0.650 0.642
13 13 13 13 13 13 13 13 13
would be raised by 0.08 to 0.1 percentage points when increasing

by one standard deviation (4.8 percentage points) the percentage
of people aged 2565 who have at least tertiary education (TERP).
In turn, a one-standard-deviation change (1.04 percentage points)
of more education expenditure would increase the growth rate by
around 0.08 to 0.1 percentage points.
2. The coefcients on redistribution (RE) are ambiguous and sta-
tistically insignicant. When controlling for SECP they are positive,
but they are negative in all other cases. All coefcients on RE suggest
that it does not add much to explaining the cross-country variation
in growth rates.
216 Gunther Rehme
3. When controlling for initial income and fertility, the coefcients

on pre-tax and on post-tax income inequality are statistically insig-
nicant but they are always negative, no matter whether one also
controls for education (SECP, TERP, or GEDU). For instance, when
controlling for secondary education SECP in models (2)(5) a one-
standard-deviation change in pre-tax inequality (LIS.G) of 3.67
lowers the long-run growth rate by 0.28 percentage points. The same
change for post-tax inequality (LIS.N) amounts to 3.88 and lowers
the growth rate by 0.26. But when controlling for tertiary educa-
tion (TERP) or overall spending on education (GEDU), these effects
become smaller.
4. Pre-tax income inequality (LIS.G) appears to be more strongly,
negatively related to growth than post-tax income inequality
(LIS.N), as the coefcients for the latter are consistently closer to zero
than those found for LIS.G.
The ndings suggest that more inequality in gross incomes seems
to imply lower growth for the typical country in the sample. Then
the state intervenes by redistribution. That intervention does not
appear to affect growth very much in the typical country. However,
the resulting inequality in personal incomes after taxes is still nega-
tive. But that means that the state may not have intervened enough to
generate a situation where after-tax income inequality would not
negatively affect growth anymore.
The negative association between pre-tax or post-tax income in-
equality and growth also allows for an interpretation of results
based on some forms of unadjusted inequality measures.20
Lemma 1 If the true association between growth and pre-tax
income inequality, measured by the Gini coefcient, is negative, then
the estimated coefcients on unadjusted inequality measures based
on mixes of Gini coefcients for net and for gross incomes are most
likely to be biased upwards.
But then the estimated non-negative signs found on the coef-
cients for unadjusted inequality measures allow for another inter-
pretation: The true association between pre-tax inequality and
growth is negative. Post-tax inequality depends on pre-tax inequal-
ity and redistribution. If one uses the mix-generated unadjusted
inequality measure, one really runs a regression on a variable con-
taining information about pre-tax inequality and redistribution,
which are two different things. Thus, a positive coefcient in a

growth regression may also indicate that pre-tax inequality nega-
tively affects growth and redistribution strongly positively affects
growth.
Table 7.3 presents the effects of some widely used determinant
factors on income inequality and redistribution.
These regressions indicate that initial income and fertility are neg-
atively related to equality and redistribution. In turn, government
spending on overall education (GEDU) is consistently negatively
associated with pre-tax and post-tax inequality, but it is positively
related to redistribution.
7.3.3 The Role of the Schooling Technology
A common measure of the (internal) inefciency in schooling is the

dropout rate. Recently, the OECD has provided data on dropout
rates for students enrolled at the university-level tertiary education,
covering many OECD countries for the 1990s.21 Differences in these
rates are taken to reect structural differences that have not changed
much across countries and over a long time horizon. Thus, the results
below are only suggestive.
With this variable, called DROP, regressions have been run which
control for policy responses by means of interaction terms. The
regression results that are available on request indicate that most of
the relationships are statistically insignicant. The point estimates in
turn suggest that when controlling for the dropout rate, DROP, and
for policy reactions (e.g., GEDU*DROP), more education and more
spending on it may raise growth and reduce income inequality
but also redistribution. However, when looking at the total effect of
education expenditure GEDU (tertiary education, TERP) around the
sample mean, it turns out that more education expenditure GEDU
(tertiary education, TERP) seems to lower income inequality but
raises redistribution and growth.
The coefcients on the dropout rate DROP appear to conrm that
countries where it is more difcult to generate education have lower
inequality. But they also seem to be those that redistribute less. But
the latter is only true if one controls for the interaction terms. The
total effect is that around the sample mean an increase in DROP
raises pre-tax and post-tax inequality but lowers redistribution and
growth.
Table 7.3
218
Determinants of inequality
Dependent (1) (2) (3) (4) (5) (6) (7) (8) (9)
variable LIS.G LIS.G LIS.G LIS.N LIS.N LIS.N RE RE RE
Const. 87.562 79.199 43.246 104.390 92.394 59.485 14.814 11.109 13.734
(36.683) (35.545) (42.709) (41.330) (37.240) (45.991) (17.153) (16.805) (22.422)
[0.038] [0.053] [0.341] [0.030] [0.035] [0.232] [0.408] [0.525] [0.557]
LY70 10.903 10.627 7.789 12.268 11.872 9.275 1.158 1.035 1.243
(3.687) (3.527) (3.939) (4.154) (3.695) (4.241) (1.724) (1.668) (2.068)
[0.014] [0.015] [0.083] [0.014] [0.011] [0.060] [0.517] [0.550] [0.565]
LAFERT 23.036 21.910 18.214 23.728 22.114 18.730 0.525 0.026 0.296
(5.236) (5.066) (5.519) (5.899) (5.307) (5.943) (2.448) (2.395) (2.897)
[0.001] [0.002] [0.011] [0.002] [0.002] [0.014] [0.835] [0.992] [0.921]
CVLIB 3.163 2.896 0.231
(2.289) (2.465) (1.202)
[0.204] [0.274] [0.852]
GEDU 0.876 1.438 1.256 1.771 0.388 0.347
(0.626) (0.722) (0.655) (0.777) (0.296) (0.379)
[0.195] [0.081] [0.088] [0.052] [0.222] [0.387]
R2 0.660 0.720 0.774 0.620 0.730 0.770 0.051 0.204 0.207
Obs. 13 13 13 13 13 13 13 13 13
Note: Estimation by OLS. Standard errors in parentheses and t-probabilities in square brackets.
Gunther Rehme
In terms of the model this suggests that c and may not be inde-
pendent of one another. Furthermore, one may argue that they are
also related to b. For instance, Walde (2000) shows that the structure
of education systems (elitist vs. egalitarian) may inuence the skill
intensity in production. In that sense the estimates here would sup-
port such an argument.
7.4 Concluding Remarks
This chapter argues that education directly affects growth and

income inequality. In the model pre-tax and post-tax income
inequalitymeasured by the Gini coefcientas well as growth are
rst increasing and then decreasing in education. It is shown that a
less efcient schooling system would require more redistributive
education expenditure for a given level of education.
The data, which are based on consistent concepts for the measure-
ment of inequality, provide some evidence that, when controlling for
various factors such as initial income or fertility, long-run growth is
higher for countries that (1) spend more on education, and (2) have
lower pre-tax and lower post-tax income inequality. The data also
suggest that countries with a more productive, public education
technology exhibit lower income inequality and higher growth. The
chapter does not nd an indication that higher pre-tax or post-tax
income inequality is good for growth in rich countries.
The consistent negative relation between pre-tax and post-tax
inequality and growth in the data suggests that mixing Gini coef-
cients based on gross or on net incomes is most likely to produce an
upward bias when measuring the effect of unadjusted measures of
inequality on growth. It also indicates that a rich country could raise
growth and lower pre-tax and post-tax inequality, given its educa-
tion technology, if it spent more on public education. However, a
more precise disentanglement of the interaction between education
policy and education technology, and their effects on growth and
inequality appears to be called for and should be an interesting topic
for further research.
Notes
I am grateful for helpful comments made by Theo Eicher, Danny Quah, Cecilia Garca-
Penalosa, and the participants at the conference Growth and Inequality: Issues and
Policy Implications at CESifo, Munich, and Schloss Elmau, Bavaria, in 2001/2, and the
220 Gunther Rehme
Development Conference on Growth and Poverty at UNU/WIDER in Helsinki, May

2001. I have also beneted from useful conversations with Ingo Barens, Volker Cas-
pari, and Rafael Gerke. Furthermore, I owe a special thanks to Rafael Domenech and
an anonymous referee for their insightful comments and suggestions. Of course, all
errors are my own.
1. Banerjee and Duo (2000) have recently attempted to reconcile the two conclusions,
but they ended up with a negative result. Based on Deininger and Squires data and
the use of unadjusted inequality measures, they show that the data cannot really tell us
very much about the relationship between inequality and growth, especially once one
accounts for possible nonlinearities.
2. This literature is surveyed by, for example, Benabou (1996), Bertola (2000), Aghion,
Caroli, and Garca-Penalosa (1999), or Zweimuller (2000).
3. For instance, Barro (2000) nds for his data and for unadjusted inequality measures
that inequality is positively associated with growth when looking at a subsample of
rich countries and when including fertility as an additional control variable.
4. A constant b implies that the diversity in it across countries is structurally xed for
a long time. Thus, the paper abstracts from skill-biased technical change and should,
therefore, be viewed as complementary to research along the lines of, for example,
Galor and Tsiddon (1997), Acemoglu (1998), or Caselli (1999).
5. There is evidence (available on request) that across countries capital income taxes
are indeed signicantly positively related to expenditure on education. Constancy of
the tax rate is imposed in order to focus on long-run, time-consistent equilibria with
steady-state, balanced growth. For a discussion of private vs. public education, see, for
instance, Glomm and Ravikumar (1992) or Fernandez and Rogerson (1998).
6. In the model, agents are endowed by the same basic ability and receive basic train-
ing that is produced and provided costlessly. Education always means higher educa-
tion. Ex ante everybody is a candidate for receiving (higher) education and once
chosen to be in the education process will complete the degree. Thus, the education
sector is characterized by continuous excess demand due to rationing, which seems
realistic for most education systems. Furthermore, the education process is taken to be
sufciently productive in converting no skills into high skills. The model ignores
problems arising from the time spent receiving education by assuming that education
is provided as a public good and that all people spend the same time in school, but
attend different courses leading to different degrees. Opportunity costs of education
might easily be introduced into the model by subtracting a xed amount of happiness
from a high-skilled person for having spent time in school. The chapters results
would not change in that case.
7. Of course, redistribution may take other forms in reality. For instance, suppose that
in contrast to this models assumptions education is privately costly. Public expendi-
tures on higher education might then be regressive if higher-income families have
better access to educational opportunities. Redistribution in the form of social trans-
fers might in fact have a growth impact by loosening liquidity constraints that prevent
individuals in poor families from taking advantage of educational opportunities. In
that sense redistribution corrects for ex ante (wealth or income) inequality in the
chapter, because education is provided costlessly and ex ante the family background
of a student does not matter in the model. Second, there is redistribution ex post that
corrects for inequality in income after education has taken place.
8. Underlying this equation is the description of an education sector with spillovers

from, for instance, high-skilled to new high-skilled people or where the capital equip-
ment such as computers makes the education technology very productive. A more
formal, OLG-based justication of the setup is available on request.
9. Here the assumption is that regardless of the source of new ideas or blueprints
production is undertaken so that all agents are affected relatively equally from
knowledge spillovers. The results would not change if the externality depended on the
entire capital stock instead.
10. On this see, for instance, Bound and Johnson (1992), Katz and Murphy (1992), or
Autor, Krueger, and Katz (1998).
11. The idea that skill-biased technological change within rms with its corresponding
demand for high-skilled labor and the (education systems) supply of high-skilled
people are in a run determining wage inequality over time can, e.g., be found in
Tinbergen (1975).
12. This follows because dg=dx 1=nD1 D2 , where D1 c=x 1=1 1 a
A1 bx a and D2 1 cx 1= f1 aAabx a1 . For x ! 0 we have dg=dx y.
When x ! 1 we have dg=dx 1=c1 aA1 b 1 cf1 aAab < 0. Fur-
thermore, dD1 =dx > 0 and dD2 =dx < 0 imply d 2 g=dx 2 < 0. These properties capture that
expanding education may lead to lower growth under some especially congestive cir-
cumstances. See, for example, Temple (1999, 140).
13. In an economy with income growth such as the one modelled here this property of
the Gini coefcients often follows by construction. See Fields (1987) or Amiel and
Cowell (1999).
14. Recall that the result is conditional on (observable) x. Given that policy may react
to a change in fundamentals there is nothing to preclude the possibility that a change
in them produces different effects in total. Notice also that the result applies only when
income inequality is measured by the Gini coefcient. For instance, a higher may
increase income inequality (given t) when measuring it by the concept of Generalized
Lorenz Curve Dominance. See Shorrocks (1983). Thus, the results depend on which
measurement concept one uses.
15. See OECD (1998), Table A1.2A.
16. For a critical assessment of other frequently employed data sources measuring
human capital see de la Fuente and Domenech (2000).
17. As a sensitivity check, I have also used data from the World Income Inequality
Database (WIID) in Helsinki and found that the papers results are robust for samples
with more countries when using slightly less rigorous consistency requirements. These
results are available on request.
18. The estimate for b 3 was 0.082 with a standard error of 0.378 and a t-probability of
0.832.
19. Bearing in mind that due to the consistency requirement the sample size is small
and although given statistical insignicance the focus of this chapter is on the point
estimates and their economic signicance, but not on inferential statistics. On the dis-
tinction between statistical and economic signicance see, for example, McCloskey
(1985) or McCloskey and Ziliak (1996).
222 Gunther Rehme
20. Suppose one mixes Gini coefcients for net and gross incomes to get an unadjusted
inequality index I, which is then used in growth regressions.
g
The model really run: giu a bGi cGin i ;
g
The model run: giu a 0 bi0 IGi ; Gin i0 ;
where giu is the residual of the regression of gi on a vector of control variables. For
simplicity assume that the control variables are not correlated or even orthogonal to
the distributional variables. Commonly giu is called the unexplained part of the growth
rate. The latter is attempted to be explained by I in the models above. In general,
g
Gi Gin RiA for country i. This is true by the denition of RiA as a measure of redis-
g
tribution. Not every observation in Gi may necessarily have a corresponding match-
ing value of Gi in the data set. However, values of Gin are included by many NCV
n
proponents in order to boost the number of observations. Now what is basically done
g g g ~ A where G~ g and R
is to create Ii Gi Gin . But this means Ii Gi G~i R i i
~ A may be
i
g
unknown if there is no matching value for Gi available in the data set. This implies
that their estimate for b 0 , call it ^b 0 , would contain information of the distribution of
gross incomes, but also information on redistribution. Thus, the model they run is
g g
giu a bGi cG~i cR
~ A i which puts a restriction on the effect that R
i
~ A exerts.
i
g
Relaxing it and noting that G~i and R ~ A are unknown, when there are no matching
i
g
values in Gi , the regression is really run on values of known Gini coefcients for gross
incomes, some unknown Gini values for gross income, and some unknown values for
redistribution. But consequently their true model is
g g
g u a bG cG~ dR
i i i
~ A i ;
i
where d c. But this suggests the following argument: The estimates for b and c
have often been found to be negative in earlier contributions. Furthermore, there is an
empirical literature that shows that the estimate for d is likely to be positive. Thus, one
has an upward bias for the estimate of b 0 . This bias may be so strong that one may
even get a positive estimate for b 0 b c d, that is, b^0 b 0 is compatible with b^ < 0,
c^ < 0 and d^ > 0.
21. See OECD (2000), Table C4.1.
References
Acemoglu, D. 1998. Why do new technologies complement skills? Directed technical

change and wage inequality. Quarterly Journal of Economics 113: 10551089.
Aghion, P., E. Caroli, and C. Garca-Penalosa. 1999. Inequality and economic growth:
The perspective of the new growth theories. Journal of Economic Literature 37: 1615
1660.
Alesina, A., and D. Rodrik. 1994. Distributive politics and economic growth. Quarterly
Journal of Economics 109: 465490.
Amiel, Y., and F. A. Cowell. 1999. Thinking about Inequality. Cambridge, UK: Cam-
bridge University Press.
Atkinson, A. B. 1970. On the measurement of inequality. Journal of Economic Theory 2:

244263.
Atkinson, A. B. 1983. The Economics of Inequality, 2d ed. Oxford: Clarendon Press.

Atkinson, A. B., and A. Brandolini. 2001. Promise and pitfalls in the use of second-
ary data-sets: Income inequality in OECD countries as a case study. Journal of Eco-
nomic Literature 39: 771799.
Autor, D., A. Krueger, and L. Katz. 1998. Computing inequality: Have computers
changed the labour market. Quarterly Journal of Economics 63: 11691213.
Banerjee, A. V., and E. Duo. 2000. Inequality and growth: What can the data say?
Working Paper 7793, NBER.
Barro, R. J. 1997. Determinants of Economic Growth: A Cross-Country Empirical Study.
Barro, R. J. 2000. Inequality and growth in a panel of countries. Journal of Economic

Growth 5: 532.
Barro, R. J., and J.-W. Lee. 1994. Sources of economic growth. Carnegie Rochester Con-
ference Series on Public Policy 40: 146.
Benabou, R. 1996. Inequality and growth. In NBER Macroeconomics Annual 1996, ed. B.
S. Bernanke and J. J. Rotemberg, 1173. Cambridge: The MIT Press.
Bertola, G. 2000. Macroeconomics of distribution and growth. In Handbook of Income
Distribution, vol. 1, ed. A. B. Atkinson and F. Bourguignon, 477540. Amsterdam:
Elsevier.
Bound, J., and G. Johnson. 1992. Changes in the structure of wages in the 1980s: An
evaluation of alternative explanations. American Economic Review 82: 371392.
Caselli, F. 1999. Technological revolutions. American Economic Review 89: 78102.
Caselli, F., G. Esquivel, and F. Lefort. 1996. Reopening the convergence debate: A new
look at cross-country growth empirics. Journal of Economic Growth 1: 363389.
Chiu, W. H. 1998. Income inequality, human capital accumulation and economic per-
formance. Economic Journal 108: 4459.
Cowell, F. A. 1995. Measuring Inequality, 2d ed. Hemel Hempstead, UK: Prentice Hall/
Harvester Wheatsheaf.
de la Fuente, A., and R. Domenech. 2000. Human capital in growth regressions: How
much difference does data quality make? Working Paper 262, OECD.
Deininger, K., and L. Squire. 1996. A new data set measuring income inequality. World
Bank Economic Review 10: 565591.
Deininger, K., and L. Squire. 1998. New ways of looking at old issues: Inequality and
growth. Journal of Development Economics 57: 259287.
Easterly, W., and S. Rebelo. 1993. Fiscal policy and economic growth. An empirical
investigation. Journal of Monetary Economics 32: 417458.
Fernandez, R., and R. Rogerson. 1998. Public education and income distribution: A
dynamic quantative evaluation of education-nance reform. American Economic Review
88: 813833.
Fields, G. S. 1987. Measuring inequality change in an economy with income growth.
224 Gunther Rehme
Forbes, K. J. 2000. A reassessment of the relationship between inequality and growth.

American Economic Review 90: 869887.
Galor, O., and D. Tsiddon. 1997. Technological progress, mobility, and economic
growth. American Economic Review 87: 363382.
Galor, O., and J. Zeira. 1993. Income distribution and macroeconomics. Review of Eco-
Glomm, G., and B. Ravikumar. 1992. Public versus private investment in human
capital: Endogenous growth and income inequality. Journal of Political Economy 100:
818834.
Johnson, G. E. 1984. Subsidies for higher education. Journal of Labor Economics 2:

303318.
Katz, L. F., and K. M. Murphy. 1992. Changes in relative wages, 19631987: Supply
and demand factors. Quarterly Journal of Economics 107: 3578.
Kuznets, S. 1955. Economic growth and income inequality. American Economic Review
45: 128.
Lambert, P. J. 1993. The Distribution and Redistribution of Income: A Mathematical Analy-
sis, 2d ed. Manchester, UK: Manchester University Press.
Li, H., and H.-F. Zou. 1998. Income inequality is not harmful for growth: Theory and
evidence. Review of Development Economics 2: 318334.
McCloskey, D. N. 1985. The loss function has been mislaid: The rhetoric of signicance
tests. American Economic Review 75: 201205.
McCloskey, D. N., and S. T. Ziliak. 1996. The standard error of regressions. Journal of
Economic Literature 34: 97114.
OECD. 1998. Education at a Glance 1998. Paris: OECD.
OECD. 2000. Education at a Glance 2000. Paris: OECD.
Perotti, R. 1993. Political equilibrium, income distribution, and growth. Review of Eco-
Perotti, R. 1996. Growth, income distribution, and democracy: What the data say.
Persson, T., and G. Tabellini. 1994. Is inequality harmful for growth? American Eco-
nomic Review 84: 600621.
Rehme, G. 1999. Education, economic growth and personal income inequality across
countries. Working Paper ECO 99/42, European University Institute, Florence, Italy.
Romer, P. M. 1986. Increasing returns and long-run growth. Journal of Political Economy
94: 10021037.
Saint-Paul, G., and T. Verdier. 1996. Inequality, redistribution and growth: A chal-
lenge to the conventional political economy approach. European Economic Review 40:
719728.
Sala-i-Martin, X. 1996. A positive theory of social security. Journal of Economic Growth
1: 277304.
Shorrocks, A. F. 1983. Ranking income distributions. Economica 50: 117.

Summers, R., and A. Heston. 1991. The Penn World Table (Mark 5): An expanded set
of international comparisons, 19501988. Quarterly Journal of Economics 106: 327368.
Temple, J. 1999. The new growth evidence. Journal of Economic Literature 37: 112156.
Tinbergen, J. 1975. Income Distribution: Analysis and Policies. Amsterdam: North-

Holland.
Walde, K. 2000. Egalitarian and elitist education systems as the basis for international
differences in wage inequality. European Journal of Political Economy 16: 445468.
Zweimuller, J. 2000. Redistribution, inequality and growth. Empirica 27: 120.
8 The Impact of Tax Policy
on Inequality and Growth:
An Empirical and
Theoretical Investigation
Theo S. Eicher,
Stephen J. Turnovsky, and
Maria Carme Riera Prunera
8.1 Introduction
Vast literatures exist analyzing the relationship between economic

growth and income inequality, on the one hand, and between taxa-
tion and economic growth, on the other. This chapter combines these
strands of the growth literature and examines the effects of tax policy
on both income inequality and growth. This analysis requires an
explicitly specied sector of skill accumulation, since the growth and
inequality effects stemming from scal policy actions do not impinge
on all sectors of the economy equally. The key feature of our model
is that we introduce both raw (unskilled) labor, together with human
capital (skills) and physical capital, into a two-sector framework that
incorporates the accumulation of both skills and physical accumula-
tion. In addition, government expenditure is introduced as a pro-
ductive factor in the production of new output, together with a
distortionary income tax. The production characteristics of the two
sectors are sufciently general to permit our simulations to allow
for the diverse cross-country experiences with regards to taxation,
inequality, and growth.1 These features are integrated into a non-
scale growth model, the implications of which have been shown
to adhere closely to the economic performance of industrialized
countries, and which replicates the distinctly nonlinear nature of
inequality in the United States with relative ease ( Jones 1995).
The results derived in the chapter draw attention to the fact that
the nonscale growth model not only ts the U.S. data well for the
long run (as rst shown by Jones 1995), but also that our parameter-
ization of the model is able to track important aspects of the short-
run evolution of the economy. Specically, we show that the initial
phase of the transitional adjustment of the skill premium in response
228 Theo S. Eicher, Stephen J. Turnovsky, and Maria Carme Riera Prunera
to changes in the tax code is inherently nonlinear, a feature that

closely follows the U.S. skill premium experienced in the 1980s.
Our simulations also relate well to the literature on the macro-
economic effects of tax reforms. Studies examining tax reforms show
that higher taxes tend to discourage investment rates (Auerbach and
Hassett 1991; Cummins, Hassett, and Hubbard 1994, 1996) and that
income taxes seem to increase wealth inequality, compared to the
distributional neutrality of a consumption tax, as some authors have
pointed out (Perroni 1995; Felder 1997). The empirical studies show
that the impact of tax changes depends on the mechanism underly-
ing the production of knowledge, as well as on the complementarity
between knowledge and physical capital, a feature also shared by
our model.
We choose as our benchmark the dramatic changes in the U.S. tax
code in the early 1980s and the ensuing, signicant changes in the
skill premium. Any growth model that seeks to speak to the discus-
sion on policy, taxation, and inequality must be able to replicate
key events in the data to render its implications relevant. A key fea-
ture of the data in the 1980s was the nonmonotonic nature of the
response of the skill premium to changes in the tax code. This is a
transition that is difcult for a conventional one-sector growth model
or a two-sector endogenous growth model to explain. This is because
in either case the transitional adjustment path is a one-dimensional
locus and therefore can generate only monotonic adjustments. In
contrast, Eicher and Turnovsky (1999) have shown that the tran-
sitional path in a two-sector nonscale model can easily be non-
monotonic (depending upon the nature of the underlying shock),
and is thus capable of explaining the nonlinear characteristics of the
data. The original Eicher-Turnovsky model included knowledge, in
the form of a public good, and here we modify their original formu-
lation to include endogenous private skill formation.
In our simulations, using actual tax rates for the 1980s, we nd
that the model replicates both key steady-state variables of the
economy as well as tracking the short-run, nonlinear transition path
of the skill premium fairly closely. Most importantly we are able to
highlight that the speed of adjustment during the transition, and the
magnitude of the change in inequality is a function of the external-
ities in production and education sector.
The previous literature on the topic is separated into two strands:
the growth and inequality literature, and the growth and taxation
The Impact of Tax Policy on Inequality and Growth 229
models. Existing explanations for the observed patterns of inequality

have focused on sectoral mobility (from agriculture to industry),2
capital market imperfections,3 sociopolitical (in)stability and re-
distribution,4 or skill-biased technical change.5 In contrast, the tax
reform literature centered on macroeconomic variables, such as
employment and accumulation. These studies show how higher
taxes discourage investment rates (e.g., Auerbach and Hassett 1991)
or labor supply (e.g., Eissa 1996). Models that actually correlate
changes in the tax structure with changes in growth (e.g., Auerbach
1996) focus on the distorting effects of current tax structure, and not
on the productive services that the government might nanceto
the benet of economic growth.
The structure of the chapter is as follows. Section 8.2 presents the
nonscale model with endogenous skill formation. Section 8.3 describes
the general equilibrium and stationary states, and section 8.4 outlines
how tax reforms in the 1980s correlate both with the predictions of
the model and the evolution of real inequality. Section 8.5 concludes.
8.2 A Two-Sector Nonscale Economy
We begin by outlining the structure of a two-sector nonscale model

that features exogenous population growth and endogenously
growing physical and human capital. We focus on a decentralized
economy in which population, N, is assumed to grow at the steady
rate N_ =N n. The government plays a simple role. It taxes nal
income and uses the proceeds to nance the purchases of a produc-
tive input that increases productivity in the nal output sector.
8.2.1 Individual Agents
Individual i produces nal output, Yi , and gross additions to human

capital, Ji , in separate sectors, each subject to externalities according
to the Cobb-Douglas production functions:
Yi aF y bN cHi bH fKi bK K cK H cH GScG ; 1a
Ji aJ 1 y eN 1 cHi eH 1 fKi eK K fK H fH : 1b
Each individual is endowed with a unit of labor, y of which is allo-
cated to the production of new output and 1 y to the production
of new human capital. In addition, he allocates a fraction c of his
current human capital, Hi , to the production of nal output, and the

balance 1 c to the accumulation of further human capital. Like-
wise, he allocates a fraction f of his physical capital, Ki , to the pro-
duction of nal output and the rest 1 f to the human capital
sector. The production of new output is subject to positive external-
ities arising from the aggregate stock of physical capital, K, human
capital, H, as well as the services, GS , provided by government
spending, G, on a productive input. The public good is assumed to
be rival, but excludable, so that GS G=N. The constants aF , aJ rep-
resent exogenous technological shift factors to the production func-
tions. To allow for positive externalities from all factors in both
sectors, bi and ci represent the private productive elasticities and ei
and fi represent the externalities.
All agents are identical so that aggregate and individual quantities
are related by
Y NYi ; K NKi ; H NHi : 2
Government expenditure is proportional to aggregate output, while
government services derived by the individual agent are propor-
tional to individual output, in accordance with
G gY gNYi ; GS gYi ; 3
and substituting (3) into (1a) we can rewrite the production function
as
Yi aF y bN =1cG cHi bH =1cG fKi bK =1cG K cK =1cG H cH =1cG ; 1a 0
where aF 1 aF g cG 1=1cG .
Allowing for depreciation of both human and physical capital
(dK ; dH ), the rate at which the individual accumulates the two types
of capital is described by
K_ i 1 ty Yi Ci Ti dK nKi ; 4a
H_ i Ji dH nHi : 4b
According to (4a), income is taxed at the rate ty , and in addition we
allow for lump-sum taxation, Ti .
Agents maximize their intertemporal utility function
y
1
Ci 1g ert dt r > 0; g > 0; 5
1g 0
subject to the production functions (1a) and (1b), the accumulation

constraints (4a) and (4b), and the usual initial conditions. His deci-
sion variables are (1) the rate of consumption, Ci ; (2) the sectoral
allocation of labor, y; human capital, c, physical capital, f and (3) the
rates of accumulation of physical and human capital. The optimality
and transversality conditions to this problem can be summarized as
follows:
Cg
i ni ; 6a
Yi Ji
ni bN 1 ty mi eN ; 6b
y 1y
Yi Ji
ni bH 1 ty mi eH ; 6c
c 1c
Yi Ji
ni bK 1 ty mi eK ; 6d
f 1f
Yi m Ji n_
bK 1 ty dK n i e K r ; 6e
Ki ni Ki n
ni Yi Ji m_
bH 1 ty dH n eH r ; 6f
mi Ki Hi m
together with the transversality conditions
lim ni Ki ert lim mi Hi ert 0; 6g

t!y t!y
where ni ; mi are the respective shadow values of physical capital and

human capital.
Equation (6a) equates the marginal utility of consumption to the
shadow value of capital. Equations (6b), (6c), and (6d) determine the
sectoral allocations of labor, human capital, and physical capital
such that their respective after-tax marginal products are equated
across sectors. Equation (6e) equates the marginal return to investing
in an additional unit of physical capital to the return on consump-
tion, both measured in terms of units of nal output. Analogously,
(6f) equates the marginal return to human capital to the return on
consumption, both expressed in units of human capital. In both cases
the return to the asset reects the fact that the additional unit will be
allocated across the two sectors.
8.2.2 The Aggregate Economy
To derive the behavior of the aggregate economy we rst sum (1a 0 )

and (1b) over the N individuals in the economy. We may express the
resulting quantities in terms of the aggregates
Y aF y sN f sK c sH K sK H sH N sN ; 7a
J aJ 1 y eN 1 f eK 1 c eH K hK H hH N hN ; 7b
where6
bN bK bH
sN 1 ; sK 1 ; sH 1 ;
1 cG 1 cG 1 cG
bK cK bH cH bH bK
sK 1 ; sH 1 ; sN 1 1 ;
1 cG 1 cG 1 cG
hH 1 e H f H ; hK 1 eK fK ; hN 1 1 e H e K :
Next, we introduce the government and consider the aggregate
accumulation equations. We assume that the government nances its
expenditure in accordance with a balanced budget, which, aggre-
gated over N individuals, can be expressed as
ty NYi NTi gNYi ;
or, in terms of the aggregate quantities

ty Y T gY: 8
Summing (4a) and (4b) over the individuals in the economy, and
applying the government budget constraint (8), the aggregate rates
of capital accumulation can be expressed as
K_ 1 gY C dK K; 9a
H_ J dH H; 9b
where Y, J are dened in (7a) and (7b).
8.2.3 Balanced Growth Equilibrium
Before describing the dynamics, we characterize the balanced

growth equilibrium. This is dened to be a growth path along which
all variables grow at constant, but possibly different, rates. In accor-

dance with U.S. data, we assume that the output/capital ratio, Y=K,
is constant. A key feature of the nonscale model is that the equilib-
rium percentage growth rates of human and physical capital, H ^ and
^
K, respectively are determined entirely by production conditions.
Taking the differentials of the production functions (7a) and (7b),
and solving, we obtain

^ hN 1 sK sN hK
H n 1 bH n; 10a
1 hH 1 sK sH hK

^ ^ ^ sN 1 hH sH hN
KYC n 1 b K n; 10b
1 hH 1 sK sH hK
and thus the per capita growth rate of output (capital) is
^ n 1 hH sH sN sK 1 sH hH hN hK 1n :
Y 10c
1 hH 1 sK sH hK
It is evident from (10a) and (10b) that the magnitudes of the relative
sectoral growth rates depend upon the assumed production elastic-
ities in conjunction with the population growth rate. Equation (10c)
implies that countries converge to identical output per capita growth
rates if either their production technologies are identical, or their
production functions exhibit constant returns to scale. If production
technologies differ across countries, growth rates exhibit conditional
convergence. The fact that (10c) implies that the per capita growth
rate of output in general depends upon the population growth rate,
n, is a double-edged sword. The evidence on the correlation between
population growth and output growth is ambiguous (see, e.g., Barro
and Sala-i-Martin 1995). On the other hand, the growth rate has the
great advantage that it is no longer a function of the absolute size of
the population, N, as in previous R&D-based models, an implication
that is soundly rejected by the data.
Finally, the possibly differential equilibrium growth rates of phys-
ical capital and knowledge are reected in the growth rates of their
respective shadow values, n; m. Using the optimality conditions (6),
these can be shown to grow in accordance with
n^ m^ bH bK n; 10d
where since agents are identical we drop the subscript i.
8.3 Dynamics of a Two-Sector Model
To derive the equilibrium dynamics about the balanced growth path,

we dene the following stationary variables:
y 1 Y=N bK ; k 1 K=N bK ; c 1 C=N bK ; h 1 H=N bH ;
j 1 J=N bH ; q 1 n=mN bH bK :

For convenience, we refer to y, k, c, and h as scale-adjusted quantities.7
This allows us to rewrite scale-adjusted output and human capital as
y aF y sN f sK c sH k sK h sH ; 11a
j aJ 1 y eN 1 f eK 1 c eH k hK h hH : 11b
The optimality conditions then enable the dynamics to be ex-

pressed in terms of these scale-adjusted variables, as follows. First,
substituting (11a) and (11b) into the labor allocation condition, (6b),
the human capital allocation condition, (6c), and the physical capital
allocation condition, (6d) yields the three relationships:
1 ty aF qbN y sN 1 f sK c sH k sK h sH
aJ eN 1 y eN 1 1 f eK 1 c eH k hK h hH ; 12a
1 ty aF qbH y sN f sK c sH 1 k sK h sH
aJ eH 1 y eN 1 f eK 1 c eH 1 k hK h hH ; 12b
1 ty aF qbK y sN f sK 1 c sH k sK h sH
aJ eK 1 y eN 1 f eK1 1 c eH k hK h hH : 12c
In principle, we can solve these three relationships for the allocation

of labor, human capital, and physical capital across the production
and education sectors:
y yq1 ty ; h; k; qy=qq > 0; sgnqy=qh sgnsH hH ;
sgnqy=qk sgnsK hK ; 13a

c cq1 ty ; h; k; qc=qq > 0; sgnqc=qh sgnsH hH ;
sgnqc=qk sgnsK hK ; 13b
f fq1 ty ; h; k; qf=qq > 0; sgnqf=qh sgnsH hH ;

sgnqf=qk sgnsK hK : 13c
Intuitively, an increase in the relative value of physical capital, q,

attracts resources to the output (capital-producing) sector; labor,
human capital, and physical capital therefore move from human
capital production to nal output production. An increase in the
stock of either form of capital raises the productivity of both sectors
in proportion to an amount that depends upon the respective pro-
ductive elasticity. Resources will therefore move toward the sector
in which the input has the greater production elasticity (is more
productive).
Using the optimality conditions, the dynamics of the system can
be expressed in terms of the redened stationary variables by
h c i
k_ k 1 gaF y sN f sK c sH k sK 1 h sH dK bK n ; 14a
k
h_ haJ 1 y eN 1 f eK 1 c eH k sK h hH 1 dH bH n; 14b

eH
q_ q aJ 1 y eN 1 f eK 1 c eH k hK h hH 1
1c

bK
1 ty aF y sN f sK c sH k sK 1 h sH
f

dH dK b H bK n ; 14c

c bK
c_ 1 ty aF y sN f sK c sH k sK 1 h sH
g f

r dK g1 b K 1n ; 14d
where y:, c:, f: are determined by (13). To the extent that

we are interested in the per capita growth rates of physical and
_ b 1n and h_ =h b 1n,
human capital, they are given by k=k K H
respectively.
The steady state to this system, denoted by @ superscripts, can
be summarized by
1 g~
y c~
b K n dK ; 15a
~k ~k
~j
b H n dH ; 15b
~
h
~j eH y~ bK
dH b H n 1 ty dK b K n; 15c
h~ 1 c~ ~k f~
y~ bK
1 ty dK bK n r 1 g1 bK n; 15d
~k f~
together with the two production functions (11a, 11b), and the sec-
toral allocation conditions (13).
These nine equations determine the steady-state equilibrium in the
following sequential manner. First, equation (15b) yields the gross
equilibrium growth rate of knowledge, ~j=~h ~J =H ~ , in terms of the
returns to scale, bH , and the rates of population growth and depre-
ciation. Next, given ~j=~ h, equations (15c) and (15d) determine the
sectoral allocation of human capital, c~, such that the rate of return
on investing in human capital, given by the left hand side of (15c),
equals the rate of return on consumption, given by the right-hand
side of (15d). Having determined the allocation of human capital, c~,
dividing (12a) and (12b), respectively, by (12c) yields the corre-
sponding sectoral allocation of labor, y~, and physical capital, f~.
Having determined f~, (15c) and (15d) determine the output-capital
ratio such that the rate of return on physical capital equals the
(common) rate of return on consumption and human capital.
Then having obtained the output-capital ratio, (15a) determines the
consumption-capital ratio consistent with the growth rate of capital
necessary to equip the growing labor force and replace deprecia-
tion. Given y~, c~, f~, y~=~k, and ~j=~
h, the scale-adjusted production func-
tions determines the stocks of physical capital, ~k, and human capital,
~
h, and thus y~. Finally, having derived y~, c~, f~, ~h, ~k, any of the three
sectoral allocation conditions determine the long-run equilibrium
relative shadow value of the two assets, q~.
A more formal characterization of the transitional dynamics is
provided in the appendix. The dynamics underlying our analysis are
based on the linearization of the fourth-order system (14). In order
for the dynamics to describe a unique stable adjustment path, we
require that the number of unstable roots equal the number of jump
variables (2). Unfortunately, the system is too complex to yield in-
tuitive formal stability conditions that ensure well-behaved saddle-
point behavior. Nevertheless, we assume that this condition is
met, as indeed it is in all of our numerical simulations. For the pur-
poses of this chapter, the analysis of changes in tax regimes on
inequality, using simulation analysis, the formal stability conditions

are of only secondary importance.
8.3.1 The Impact of Public Policy on Inequality
Our concern is to analyze the dynamic response of the economy to

changes in income tax rates. We seek to go beyond qualitative results
and simulate variations in tax rates that reect the signicant policy
events that occurred in the United States during the 1980s. To un-
derstand and interpret the adjustments in the economy in response
to tax changes, it is useful to rst derive the qualitative nature of the
long-run equilibrium.
An immediate manifestation of the nonscale characteristic of the
model as seen from equations (10a) and (10b) is that the steady-state
equilibrium growth rates, of both capital goods are independent of
the tax rate ty (see Jones 1995; Eicher and Turnovsky 1999). This is a
major advantage of the model, since the evidence shows no lasting
impact of tax policy on long-run growth (Stokey and Rebelo 1995;
Easterly and Rebelo 1993; Jones 1995). The fact that long-run growth
is not affected by public policy does not rule out substantial effects of
tax policy on economic variables in the short and intermediate term.
The magnitude of such effects is depends on the speed of adjustment
in the economy. We address this point further in our simulations.
Since taxes do not inuence the long-run growth, the equilibrium
sectoral asset allocations, y~, c~, and f~, are also independent of the tax
rate. From (15d), we know that a lower tax rate reduces only the
output-capital ratio in the long run, which, in turn, leads to a lower
consumption-capital ratio in order for the equilibrium growth rate of
nal output to remain unchanged. Again, the sectoral allocations
may vary signicantly during the transition and impact the skill
premium, as we show in our simulations.
The impact on the steady-state scale adjusted stocks of physical
and human capital are, respectively,
q~k=qty hH 1
< 0; 16a
~k 1 ty 1 sK 1 hH sH hK
q~
h=qty hK
< 0: 16b
~
h 1 t y 1 sK 1 hH sH hK
Thus a decline in the tax rate will increase the long-run scale-
adjusted stocks of both physical and human capital. Assuming that
hK < 1 hH , so that human capital is relatively more important than
physical capital in the production of new human capital, then the tax
cut will have a relatively larger positive impact on physical capital.
This assumption, which is conventional throughout the two-sector
endogenous growth models, is equivalent here to hN > fK fH , thus
imposing an upper bound on the externalities in the human capital
sector.
Much of the focus of our simulations shall be on income inequal-
ity, which we measure by the skill premium
rH qY=cH bH y N
wR 1 1 1 1 > 1; 17
w qY=yN bN c H
where w is the marginal product of unskilled workers and, rH , the

marginal product of human capital, represents the return to skills.
Essentially, the relative skill premium is the base wage received for
raw labor plus the marginal product of skills in the nal goods sec-
tor, scaled by the return to raw labor.
Recalling the denition of h, we see that since a reduction in the
tax rate on nal output increases h (and therefore H ) in the long run,
it reduces the skill premium as dened by (17). The short-run
response may be quite different, however, and requires a detailed
simulation of the adjustment. This is because h is xed in the short
run, so that the short-run adjustments in the skill premium operate
entirely through the short-run allocations of raw labor y and human
capital c. Indeed, in the short run a decrease in ty (equivalent to an
increase in q) attracts resources to the output (physical-capital) pro-
ducing sector. If raw labor is more sectorally mobile than human
capital (i.e., y responds more intensely than c), then the tax cut will
be associated with a short-run rise in the skill premium, before ulti-
mately declining over time.
The differential adjustment in the long and short run of the skill
premium is novel to the literature. It requires a nonlinear transi-
tion path that is particular to this strand of growth models. In pre-
vious models, the transition speed was constant throughout the
entire adjustment. In our model, not only the speed of adjustment
of all variables changes over time, but their transition is also
nonmonotonic.
8.4 Response of Skill Premia to Tax Reforms
One key question of the growth literature is whether the theoreti-

cal policy guidance derived from models is empirically relevant.
Hence, the purpose of the simulations in this section is twofold.
First, we would like to conrm that reasonable parameter values
generate plausible steady-state values to key economic variables.
This would conrm whether this class of models is capable of gen-
erating growth rates of output and skills, as well as allocations of
resources to the two sectors that are reasonable, given observed real-
world quantities. This endeavor is of key concern since the Solow
model generates excessively large rates of convergence (see Barro
and Sala-i-Martin 1992), while the rst-generation growth models
(e.g., Romer 1990) generate key variables and convergence rates that
depend implausibly on the size of the population.
Our second purpose is to gain insights into the qualitative nature
of the transition path followed by the economy in response to a tax
shock. If our benchmark economy replicates transition paths that are
similar to those observed in the data, our simulations can be taken as
policy guidance. We ask how well these transition paths implied by
the model correlate with the actual observed changes in inequality
following specic changes in the U.S. tax code. As mentioned earlier,
our benchmark is the dramatic change in the tax code in the United
States in the 1980s.
Table 8.1 reports the values we employ for our fundamental
parameters. Insofar as possible, these are consistent with those sug-
gested by previous calibration exercises.8 Our benchmark economy
assumes that the individual production function and the individual
education function exhibit constant returns to scale in the private
Table 8.1
Benchmark parameters
Production function aF 1, bN 0:41, bK 0:31, bH 0:28
cG 0:1, cK 0, cH 0
Education sector aJ 1, eN 0:30, eK 0:20, eH 0:50
fK 0, fH 0
Preferences r 0:04, g 2:5
Depreciation/Population dK 0:05, dH 0:05, n 0:015
Fiscal policy g 0:1, ty 0:15152
factors of production. The private shares of raw labor, physical cap-

ital, and human capital0.41, 0.31, and 0.28, respectivelyin the
production function for output correspond to the estimates obtained
for non-oil-producing countries by Mankiw, Romer, and Weil (1992).
In the base calibration, we assume g 0:10 cG , which corresponds
to government expenditure ow being set at its socially optimal
level. Government services are external to the private individual,
which introduces increasing returns to scale in the aggregate pro-
duction function. The benchmark calibration abstracts from exter-
nalities in aggregate physical and human capital.
Information on the education function is sparse. We feel that the
most important input in augmenting the stock of human capital is
human capital, followed by raw labor, with physical capital being
the least important. Thus we set eH 0:50, eN 0:30, eK 0:20 as a
plausible benchmark, with the relative magnitudes of eH , eK being
consistent with the endogenous growth models of human capital.
Analogously, the benchmark calibrations assume the absence of
externalities in the two capital goods.
The key policy variable we consider is the income tax rate, and the
experiment we analyze and try to replicate are the short-run effects
of the accumulated tax cut that occurred in the United States under
the Reagan regime in 19801985. Following Mendoza, Razin, and
Tesar (1994) and Carey and Tchilinguirian (2000), we obtain a mea-
sure for an effective tax rate on total household income illustrated in
gure 8.1.9
We choose to focus on the U.S. economy because the United States
underwent signicant changes in its tax code during the Reagan
years, especially during the rst term 19801984. Figure 8.1 clearly
shows the legacy of the Reagan administration, with substantial,
successive tax cuts in 19801981 from 15.152 percent to 11.031 per-
cent in 1985. We choose the 19801981 tax rate as our benchmark
and analyze a reduction to 11.031 percent, its value in 1985. These
years capture the most signicant changes in the U.S. tax code in the
past fty years. The U.S. skill premium time series gure 8.2 reveals
an additional reason why this time period is especially attractive.
The relative wage hardly moved during the mid- to late-1970s,
until the Reagan tax reforms were instituted. After the rst, large tax
cut was instituted the relative wage dropped briey only to rise
precipitously until 1984. Clearly, the effect of taxes on relative wages
is a function of time lags and an array of interrelated economic fac-
Figure 8.1
Effective U.S. income taxes
Source: OECD (1984, 1997).
tors. However, the model developed earlier incorporates key eco-

nomic variables, such as investment in education and physical capi-
tal. Hence, we would like to explore how successful the model
would be in explaining the markedly nonlinear trend in the skill
premium, since it coincides with U.S. tax changes during the same
period of time.
A key aspect to note in gure 8.2 is that the relative wage move-
ment is nonmonotonic. Although taxes fell between 1980 and 1985, the
relative wage rst dips and then recovers. One interpretation is to
assume that factors external to the model account for such non-
monotonicities. An alternative explanation we explore here is to see
whether the growth model can replicate such nonmonotonic adjust-
ments as part of its intrinsic dynamics. Indeed, we know from Eicher
and Turnovsky (1999) that the transition path in nonscale economies
may be characterized by signicant nonmonotonicities in the state
variables. In particular, we have shown that both the initial state
of the economy and the size of the tax cut may inuence the
Figure 8.2
U.S. skill premium (college/high school)
Source: U.S. Census Bureau (2000). Mean earnings of workers 18 by educational
attainment.
actual transition to the new steady state. This result, established in

simulations for a model without human capital in Eicher and Turn-
ovsky (1999), is extended here for the case incorporating human
capital. This characteristic of the nonscale growth model stands in
sharp contrast with both the Solow model, as well as with the rst-
generation endogenous growth models (e.g., Romer 1990), in which
transitions are represented either by monotonic adjustments in all
key variables, or by instantaneous jumps to the balanced growth
path.
The benchmark parameters in table 8.1 yield a set of benchmark
values of key economic variables reported in table 8.2a. With the
only externality being in government expenditure, which is exclud-
able, the long-run growth rates of the per capita stocks of capital are
zero. The share of labor allocated to the production of nal output is
about 85 percent, the share of capital allocated to production is 86
Table 8.2a
Initial benchmark equilibrium values
d
Y=N d
H=N y~ j~ c~ g
Y=K g
C=Y
0 0 0.845 0.861 0.691 0.344 0.711
Table 8.2b
New steady-state equilibrium
d
Y=N d
H=N y~ j~ c~ g
Y=K g
C=Y
0 0 0.845 0.861 0.691 0.332 0.702
percent and about 31 percent of the skills are used in the human
capital sector. The implied equilibrium output-capital ratio is 0.34,
and the consumption-output ratio is 0.71. All of these simulated
equilibrium values coincide with those found in standard OECD
data. Although less information exists on the sectoral allocation of
assets, we feel the implied fractions are reasonable.
We can now analyze the dynamics predicted by the model fol-
lowing an accumulated reduction in the tax rate to 11.031 percent by
considering the new steady-state equilibrium as reported in table
8.2b. As discussed, the only long-run responses to the tax cut are
reductions in the ratios of output to capital and consumption to
output. The sectoral allocations of raw labor, physical capital, and
human capital remain unchanged. As mentioned earlier, this invari-
ance of the growth rate to tax policy is indeed a feature of the U.S.
long-run growth rate.
More interesting, however, is the determination of the extent to
which variations in the skill premium, as generated by the model,
correlate with U.S. data in the short run as described in gure 8.2.
During 19801985 the skill premium initially fell, then rose, and
subsequently fell for one period, before continuing to rise during the
next two periods. Figure 8.3A shows that the this nonscale model not
only replicates the steady-state variables well, but also that the sim-
ulations also mimic the short-run transition of the skill premium in
response to the Reagan tax cuts simulations quite successfully as
well. As panel (i) of gure 8.3A shows, on impact the skill premium
declines, it then increases for uniformly for approximately four
years, after which it begins to decline. The explanation for that can
be seen by considering panels (ii) and (iii).
Figure 8.3
Dynamic adjustments
The immediate impact of the reduction in the income tax rate is to

attract resources away from the human capital sector to the output
sector; y; c; f all increase on impact (see 13ac). In particular, human
capital is reallocated more rapidly than unskilled labor, hence the
ratio y=c falls. With the initial stock of human capital xed, this
implies a corresponding immediate reduction in the skill premium.
The higher fraction of labor and skills attracted to the nal output
sector reduces their respective shadow values in that sector relative
to the human capital sector, so that both y; c begin to fall. Further-
more, with the greater initial mobility of human capital, its shadow
value declines relative to that of raw labor so that as a reaction to the
initial fall in y=c is that y=c now begins to rise, and indeed it con-
tinues to do so monotonically. This causes the skill premium to begin
rising.
At the same time, the reduction in the tax rate raises the return on
physical capital relative to that of human capital, and thus physi-
cal capital begins to accumulate, while the rate of growth of hu-
man capital declines below the population growth rate, so that the
scale-adjusted human capital actually declines. With human capi-
tal increasing in relative scarcity, this increases the skill premium
further, thus accentuating the effect of the rising relative shares y=c.
Over time, the effects of the reallocation of factors become less
important, revert, and cause the skill premium to decline. First, as
raw labor and skills are gradually restored toward their respective
(unchanged) steady-state sectoral allocations, y=c increases at a
reduced rate, slowing down the rising skill premium. In addition,
as physical capital is accumulated, this raises the productivity of
human capital, slowing down its initial decline, and inducing its
accumulation as well. This is illustrated by the V-shape of the phase
diagram relating human and physical capital in panel (iii). As
human capital increases, this puts downward pressure on the skill
premium, and eventually this effect dominates, in which case the
skill premium begins to decline. This decline occurs after approxi-
mately four years.10
Figure 8.3B illustrates the same time paths in the case where the
human capital production function is subject to an externality from
aggregate human capital, of the amount fH 0:10. The steady state
with this externality is essentially unchanged from table 8.2b; how-
ever, the speeds of adjustment differ. While the qualitative charac-
teristics of the adjustments are essentially unchanged, this externality
decreases the magnitudes of both negative eigenvalues that deter-

mine the speed of convergence. Hence both the initial adjustment
and the adjustment to the long-run steady state are slowed. As a
consequence, the peak in the relative wage occurs after ve, rather
than three to four years.
We also conducted sensitivity analyses to gauge the effects of var-
iations in other externalities. In all cases the shapes are generally as
indicated in gure 8.3AB, but again, the speeds of adjustment are
affected. Introducing a human capital externality in the nal output
sector speeds up the adjustment during the early phases and reduces
it during the latter phases. As a consequence, the peak in the skill
premium occurs after only two periods. A physical capital external-
ity in either sector slows down the adjustment throughout the entire
transition, although its impact is more moderate than is that of a
comparable externality due to human capital. Finally, increasing the
externality due government expenditure has a similar effect to that
of human capital; it raises the speed of adjustment in the short term,
but reduces it during the latter stages of the transition. By contrast,
the speed of convergence is essentially invariant with respect to the
share of output claimed by the government. The sensitivity anal-
ysis thus reveals that how effective tax reductions are in foster-
ing increases in factor accumulation, or reductions in inequality will
depend on the nature of the economy. Those economies that have
fewer human capital externalities will see less of a rise in the relative
wage in the short run and a prolonged deviation from the steady
state. However, the larger the share of government purchases in
output, and the greater the externality associated with such services,
the more muted the relative wage response.
8.5 Conclusion
The purpose of the chapter has been to examine the relevance of the
most recent strand of growth models in predicting steady states and
transitions in response to policy experiments. We derive qualitative
results and simulate the model to examine specically how the
model predicts the transition of the skill premium to changes in the
income tax. The policy experiment that we have in mind is the his-
toric Reagan tax cuts of the early 1980s.
We show that the model simulates the steady states, convergence
speeds, and the actual nonlinear adjustment pattern of the skill pre-
mium exceedingly well. In addition, we highlight that how fast and

in which fashion the skill premium adjusts is a function of the exter-
nalities in production, either human, physical or of public services.
The implication for the policy effects is clear. Lower taxes change the
incentives for factor accumulation, and hence the return to all fac-
tors. While long-term adjustments may not be signicant, they may
not preclude dramatic changes in short run variations of the skill
premium.
8.6 Appendix: Characterization of Transitional Dynamics
Henceforth we assume that the stability properties are ensured so

that we can denote the two stable roots by m1 ; m2 , with m2 < m1 < 0.
The key variables of interest are physical capital, and technology.
The generic form of the stable solution for these variables is given
by
kt ~k B1 e m1 t B2 e m2 t ; A:1a
ht ~
h B1 n21 e m1 t B2 n22 e m2 t ; A:1b
where B1 ; B2 are constants and the vector
1 n2i n3i n4i 0 i 1; 2

(where the prime denotes vector transpose) is the normalized eigen-
vector associated with the stable eigenvalue, mi . The constants, B1 , B2 ,
appearing in the solution (10) are obtained from initial conditions
and depend upon the specic shocks. Thus suppose that the econ-
omy starts out with given initial stocks of capital and knowledge, k0 ,
a0 and through some policy shock converges to ~k, a~. Setting t 0
in (A.1a) and (A.1b) and letting d~k 1 ~k k0 , d~h 1 ~h h0 , B1 ; B2 are
given by
d~
h n22 d~k n21 d~k dh
~
B1 ; B2 : A:2
n22 n21 n22 n21
In studying the dynamics, we are interested in characterizing the

slope along the transitional path in h k space. In general, this is
given by
dh B1 n21 m1 e m1 t B2 n22 m2 e m2 t
A:3
dk B1 m1 e m1 t B2 m2 e m2 t
and is time varying. Note that since 0 > m1 > m2 , as t ! y this con-
verges to the new steady state along the direction dh=dkt!y n21 ,
for all shocks. The initial direction of motion is obtained by setting
t 0 in (12) and depends upon the source of the shock.
It is convenient to express the dynamics of the state variables in
phase-space form:
0 1
! m1 n22 m2 n21 m2 m1 !
B C
k_ B n22 n21 n22 n21 C k ~k
B C : A:4
h_ @ m2 m1 n21 n22 m2 n22 m1 n21 A h ~h
n22 n21 n22 n21
By construction, the trace of the matrix in (A.4) m1 m2 < 0 and the
determinant m1 m2 > 0, so that (A.4) describes a stable node. The
dynamics expressed in (A.1) and (A.4) are in terms of the scale
adjusted quantities, from which the growth rates of per capita capital
and knowledge can be derived.11
Equations (A.1a) and (A.1b) highlight the fact that with the tran-
sition path in k and h being governed by two stable eigenvalues, the
speeds of adjustment for physical capital and human capital are nei-
ther constant nor equal over time. In addition, with output being
determined by the two types of capital, the transition of output is
also not constant over time, but a simple composite of the transition
characteristics of h and k as determined in (A.4).
Notes
We thank Uwe Walz, Danny Quah, Walter Fisher, and participants at the Munich
Inequality and Growth Conferences in May 2001 and January 2002 for helpful com-
ments on earlier drafts.
1. Cross-country evidence on the effect of growth on inequality and on that of in-
equality on growth is inconclusive (see Anand and Kanbur 1993; Deininger and Squire
1998; and Forbes 2000), while historical and recent time series shows a diversity of
experiences (Williamson 1991, 1999; Gottschalk and Smeeding 1997). For a review see,
Aghion, Caroli, and Garca-Penalosa (1999).
2. See Kuznets (1955).
3. See Galor and Zeira (1993).
4. See Bertola (1993), Alesina and Rodrik (1994), Persson and Tabellini (1994), and
Alesina and Perotti (1995).
5. See Eicher (1996), Garca-Penalosa (1994), and Galor and Tsiddon (1996).
6. The elasticity sN reects the assumption that the government good is rival but
excludable. If it were a pure public good sN 1 sK sH =1 cG .
7. Under constant returns to scale, these scale-adjusted quantities are just regular per
capita quantities.
8. See Lucas (1988), Jones (1995), Ortigueira and Santos (1997), and Ogaki and Rein-
hart (1998).
9. ty taxes on income, prot, and capital gains divided by the sum of (1) unin-
corporated surplus of private unincorporated enterprises, (2) households property
and entrepreneurial income, and (3) wages and salaries of dependent employment.
Our tax data are OECD Revenue Statistics and OECD National Accounts.
10. In the more distant future (around 50 years) the decline in the skill premium is
reversed and it convergesalbeit slowlyto its long-run steady state, which lies
below its original starting point. However, this is not relevant for the impact of the
policy shock we are considering.
11. Note that the representation of the transitional dynamics in k h space takes full
account of the feedbacks arising from the jump variables, q and c; these are incorpo-
rated in the two eigenvalues.
References
Aghion, P., E. Caroli, and C. Garca-Penalosa. 1999. Inequality and Economic

Growth: the Perspective of the New Growth Theories. Journal of Economic Literature
37(4): 16151660.
Alesina, A., and D. Rodrik. 1994. Distributive Politics and Economic Growth. Quar-
terly Journal of Economics 109(2): 465490.
Alesina, A., and R. Perotti. 1995. Taxation and Redistribution in an Open Economy.
European Economic Review 39(5): 961979.
Anand, S., and S. M. R. Kanbur. 1993. Inequality and Development: A Critique.
Journal of Development Economics 41(1): 1943.
Auerbach, A. 1996. Tax Reform, Capital Allocation, Efciency and Growth. In Eco-
nomic Effects of Fundamental Tax Reform, ed. H. Aaron and W. Gale. Washington DC:
The Brookings Institution.
Auerbach, A., and K. Hassett. 1991. Recent U.S. Investment Behavior and the Tax
Reform Act of 1986: A Disaggregate View. Carnegie-Rochester Conference Series on
Public Policy 35: 185215.
Barro, R. J., and X. Sala-i-Martin. 1992. Convergence. Journal of Political Economy 100:
223251.
Barro, R. J., and X. Sala-i-Martin. 1995. Economic Growth. New York: McGraw-Hill.
Bertola, G. 1993. Factor Shares and Savings in Endogenous Growth. American Eco-
nomic Review 83(5): 11841198.
Carey, D., and H. Tchilinguirian. 2000. Average Effective Tax Rates on Capital,
Labour and Consumption. OECD Working Paper no. 31.
Cummins, J., K. Hassett, and R. G. Hubbard. 1994. A Reconsideration of Investment

Behavior Using Tax Reforms as Natural Experiments. Brookings Papers on Economic
Activity 2: 159.
Cummins, J., K. Hassett, and R. G. Hubbard. 1996. Tax Reforms and Investment:
A Cross-Country Comparison. Journal of Public Economics 62(12): 237273.
Deininger, K., and Lyn Squire. 1998. New Ways of Looking at Old Issues: Inequality
and Growth. Journal of Development Economics 57: 259287.
Easterly, W., and S. Rebelo. 1993. Fiscal Policy and Growth: An Empirical Investiga-
tion. Journal of Monetary Economics 32: 417458.
Eicher, T. 1996. Interaction between Endogenous Human Capital with Technological

Change. Review of Economic Studies 63: 127144.
Eicher, T., and S. Turnovsky. 1999. Convergence in a Two-Sector Nonscale Growth

Model. Journal of Economic Growth 4: 413428.
Eissa, N. 1996. Tax Reforms and Labor Supply. In Tax Policy and the Economy, vol.
10, ed. J. Poterba, 119151. Cambridge: The MIT Press for the National Bureau of
Economic Research.
Felder, S. 1997. Tax Reform in a Two-Class Growth Model. Empirical Economics
22(2): 273291.
Forbes, K. 2000. A Reassessment of the Relationship between Inequality and
Growth. American Economic Review 90(4): 869887.
Galor, O., and D. Tsiddon. 1996. Income Distribution and Growth: The Kuznets
Hypothesis Revisited. Economica 63(250): S103S117.
Galor, O., and J. Zeira. 1993. Income Distribution and Macroeconomics. Review of
Economic Studies 60(1): 3552.
Garca-Penalosa, C. 1994. Inequality and Growth: A Note on Recent Theories.

Investigaciones Economicas 18(1): 97116.
Gottschalk, P., and T. M. Smeeding. 1997. Cross-National Comparisons of Earnings

and Income Inequality. Journal of Economic Literature 35(2): 633687.
Jones, C. I. 1995. Time-Series Tests of Endogenous Growth Models. Quarterly Journal
of Economics 110: 495525.
Kuznets, S. 1955. Economic Growth and Income Inequality. American Economic
Review 45(1): 128.
Lucas, R. E. 1988. On the Mechanics of Economic Development. Journal of Monetary
Economics 22: 342.
Mankiw, N. Gregory, David Romer, and David N. Weil. 1992. A Contribution to the
Empirics of Economic Growth. Quarterly Journal of Economics 107(5): 407437.
Mendoza, E. G., A. Razin, and L. L. Tesar. 1994. Effective Tax Rates in Macro-
economics Cross-Country Estimates of Tax Rates on Factor Incomes and Consump-
tion. Journal of Monetary Economics 34: 297323.
OECD. Various years. Main Aggregates. Statistical Bulletins.
Ogaki, M., and C. Reinhart. 1998. Measuring Intertemporal Substitution: The Role of
Durable Goods. Journal of Political Economy 106: 10781098.
Ortigueira, S., and M. S. Santos. 1997. On the Speed of Convergence in Endogenous
Growth Models. American Economic Review 87: 383399.
Perroni, C. 1995. Assessing the Dynamic Efciency Gains of Tax Reform when
Human Capital is Endogenous. International Economic Review 36(4): 907925.
Persson, T., and G. Tabellini. 1994. Is Inequality Harmful for Growth? Theory and
Evidence. American Economic Review 84: 600624.
Romer, P. 1990. Endogenous Technical Change. Journal of Political Economy 98:
S71S103.
Stokey, N. L., and S. Rebelo. 1995. Growth Effects of Flat-Rate Taxes. Journal of
Political Economy 103(3): 519550.
U.S. Census Bureau. 2000. Available online at hhttp://www.census.gov/population/
socdemo/education/tableA-3.txti. 1/25/2001.
Williamson, J. G. 1991. Inequality, Poverty and History. Oxford: Blackwell.

Williamson, J. G. 1999. Globalization and the Labor Market: Using History to Inform
Policy. In Growth, Inequality and Globalization, P. Aghion and J. G. Williamson. Cam-
bridge: Cambridge University Press.
V Technology and Natural
Resources and Inequality
9 Inequality and Economic
Growth: Do Natural
Resources Matter?
Thorvaldur Gylfason and

Gyl Zoega
9.1 Introduction
For a long time, many economists were of the view that economic
efciency and social equality were essentially incompatible, almost
like oil and water. The perceived but poorly documented trade-off
between efciency and equality was commonly regarded as one of
the main tenets of modern welfare economics. One of the key
ideas behind this perception was that increased inequality could
increase private as well as social returns to investing in education
and exerting effort in the hope of attaining a higher standard of
life. Redistributive policies were supposed to thwart these tenden-
cies and blunt incentives by penalizing the well-off through taxation
and by rewarding the poor. Economic efciencyboth static and
dynamicwas bound to suffer in the process, or so the argument
went.
More often than not in recent empirical work, measures of income
inequality have turned out to have a negative effect on economic
growth across countries. Thus Alesina and Rodrik (1994), Persson
and Tabellini (1994), and Perotti (1996) report that inequality hurts
growth. Barro (2000) assesses the relationship between economic
growth and inequality in a panel of countries over the period from
1965 to 1995 and ndsby studying the interaction of the Gini index
and the initial level of income in a growth regressionthat increased
inequality tends to retard growth in poor countries and boost
growth in richer countries.1 However, Barro nds no support for a
relationship between inequality and growth in his sample as a
whole. Forbes (2000) nds that the relationship between inequality
and growth becomes positive in a pooled regression when country
effects are included. She claims that country-specic, time-invariant,
256 Thorvaldur Gylfason and Gyl Zoega
omitted variables generate a signicant negative bias in the esti-

mated coefcients reecting the effects of inequality on growth in
pure cross sections and mentions corruption and the level of public
education as two candidates in this regard. Banerjee and Duo
(2000b) claim that this result is misleading, and arises from imposing
a linear structure on highly nonlinear data.
The previously mentioned empirical resultsshowing, by and
large, that rapid economic growth tends to go along with less, not
more, inequalitycall for an explanation. Thus far, the explana-
tions on offer involve showing how inequality affects growth either
directly or indirectly through its effects on public policy, includ-
ing taxes and transfers and education expenditures. We now briey
describe some of these theories before returning to our proposed
thesis, which involves natural resources as a joint determinant of
both inequality and growth.
First, large inequalities of income and wealth may trigger political
demands for transfers and redistributive taxation. To the extent
that transfers and taxation distort incentives to work, save and
invest, inequality may impede growth. It is not clear, however, that
this type of political-cum-scal explanation necessarily implies an
inverse relationship between inequality and growth, for it is possible
that during the redistribution phase increased equality and a drop in
growth go hand in hand, especially in panel data that reect devel-
opments over time country by country as well as cross-sectional
patterns. Perotti (1996) nds little empirical support for this type of
explanation. Moreover, in democratic countries with an unequal
distribution of income and with many poor people, the electorate
may vote for more and better education as well as higher taxes and
transfers (Saint-Paul and Verdier 1993, 1996), thus obscuring the
relationship between inequality and growth. Absent democracy,
dictators may still nd it in their own interest to redistribute incomes
and reform education in order to promote social peace and
strengthen their own hold on political power (Alesina and Rodrik
1994). Easterly and Rebelo (1993) report empirical results that sug-
gest that increased inequality is associated with both higher taxes
and more public expenditure on education in a large sample of
countries in the period 19701988.
In second place, the initial extent of inequality probably makes
a difference. An equalization of incomes and wealth in countries
with gross inequities, such as Brazil where the Gini index is sixty,
Inequality and Economic Growth 257
would seem likely to foster social cohesion and peace and thus to
strengthen incentives rather than weaken them, whereas in places
like Denmark and Sweden, where the Gini index is 25 and incomes
and wealth are thus already quite equitably distributed by world
standards, further equalization might well have the opposite effect.
Excessive inequality may be socially divisive and hence inefcient: It
may motivate the poor to engage in illegal activities and riots, or at
least to divert resources from productive uses, both the resources of
the poor and those of the state. Social conict over the distribution of
income, land, or other assets can take place through labor unrest, for
instance, or rent seeking that can hinder investment and growth
(Benhabib and Rustichini 1996).2 Alesina and Perotti (1996) report
empirical evidence of an inverse relationship between inequality and
growth through sociopolitical instability.3
Third, national saving may be affected by inequality if the rich
have a higher propensity to save than the poor (Kaldor 1956). In this
case inequality may be good for growth in that the greater the level
of inequality, the higher is the saving rate and hence also investment
and economic growth. Against this Todaro (1997) suggests that the
rich may invest in an unproductive mannercount their yachts and
expensive cars. Barro (2000) nds no empirical evidence of a link
between inequality and investment.
Fourth, increased inequality may hurt education rather than help-
ing it as suggested by the political-economy literature referred to at
the beginning of this brief discussion. If so, increased inequality may
hinder economic growth through education. Galor and Zeira (1993)
and Aghion (1998) argue that this outcome is likely in the presence
of imperfect capital markets. If each member of society has a xed
number of investment opportunities, imperfect access to credit, and
a different endowment of inherited wealth, the rich would end up
using many of their investment opportunities while the poor could
only use a few. Therefore, the marginal return from the last in-
vestment opportunity of the rich would be much lower than the
marginal return of the last investment opportunity of the poor.
Redistribution of wealth from the rich to the poor would increase
output because the poor would then invest in more productive proj-
ects at the margin. This argument can also be applied to investment
in human capital if we assume diminishing returns to education. In
this case, taking away the last few quarters of the university educa-
tion of the elite and adding time to the more elementary education
of the poor would raise output and perhaps also long-run growth,
other things being the same. Income redistribution would reverse the
decline in investment in human capital resulting from the credit-
market failure.4
The distribution of income and wealth may also affect the amount
of public and private investment in education. When a large part of
the population is poor, it may be more likely that the majority of
voters will support expenditures on public education aimed at the
poor, as argued by Saint-Paul and Verdier (1993, 1996) and corrobo-
rated empirically by Easterly and Rebelo (1993), but the effect could
also, in principle, go the other way. If so, the more deprived and
detached from the mainstream population is the poorer segment, the
less likely the poor are to participate in or affect the outcome of
elections. As a result the general level of education may suffer
the more so, the more capital-constrained is the poorer segment of
the population. A virtuous circle may arise when redistribution of
income leads to an increase or improvement in human capital, which
then induces voters to prefer higher expenditures on education,
which again pulls more workers out of poverty, and so on. At an
empirical level, we would expect increased equality to enhance eco-
nomic growth through its effect on education, and vice versa. That is,
more and better general education may be expected to reduce public
tolerance against extreme inequality and thus to reduce inequality
through the political process, thereby stimulating economic growth.
These processes can be mutually reinforcing; that is, if increased
social equality encourages education and economic growth, this does
not mean that more and better education cannot similarly, and
simultaneously, enhance equality and growth.
The previous models all have the same basic structure: Inequality
affects some unknown intermediate variable X which, in turn, makes
a difference for economic growth. In this chapter we take a different
approach: we view both economic growth and inequality of incomes
as well as of educational attainment and of land as endogenous var-
iables and argue that the inverse relationship between inequality and
growth does not imply causality one way or the other. We propose
an explanation which, in contrast to the ones surveyed in the litera-
ture reviewed briey earlier, involves a variable that is exogenous to
most economic models. This variable is the abundance of, or rather
dependence on, natural resources, which we measure by the amount
of natural capital per person and the share of natural capital in
national wealth, respectively. We will argue, on theoretical grounds

as well as empirically, that a direct relationship between natural
resource intensity and inequality, on the one hand, and between
natural resource intensity and growth, on the other hand, can help
account for the inverse cross-sectional relationship between inequal-
ity and growth that is observed in the data, assuming that natural
resources are given. The rst relationshipbetween natural resource
intensity and inequalitywas documented by Bourguignon and
Morrison (1990) in a sample of thirty-ve developing countries in
1970, while the second relationshipbetween natural resource in-
tensity and growthhas been scrutinized by a number of authors in
recent years, beginning with Sachs and Warner (1995). Moreover, we
assume that the ownership of natural resources tends to be less
equally distributed than other assets within as well as across coun-
tries. To the extent that this is not the case at the outset, we assume
that rent seeking and other forces, frequently compounded by a lack
of democracy, will see to it that the natural resources end up in the
hands of a relatively small minoritya military regime, say, or a
royal family.
The chapter proceeds as follows. In section 9.2, we set out our
view of the way in which natural resources can affect inequality and
growth. In section 9.3, we describe the data that we use to measure
income inequality and also gender inequality in education; we also
discuss inequality in the distribution of land. In section 9.4, we
present simple cross-country correlations between three different
measures of education, three different measures of inequality and
economic growth, and thus allow the data to speak for themselves.
In section 9.5, we attempt to dig a little deeper and report the results
of cross-sectional multiple regression analysis where growth is
traced to natural resource intensity, education and inequality as well
as to other factors commonly used in growth regression analysis
(investment and initial income), and where some of the determinants
of growth, including education and inequality, are explicitly mod-
eled as endogenous variables. Section 9.6 concludes the discussion.
9.2 Resources, Distribution, and Growth
An important potential weakness of the many stories purporting to

explain the relationship between inequality and growth is that both
of these variables are endogenous. This leaves open the possibility
that a third, exogenous variable is affecting both, thus giving rise to

the inverse correlation between the two. Specically, a countrys
abundance of, or dependence on, natural resources can under many
circumstances be viewed as exogenous to models of economic
growth and also to models attempting to explain the extent of
income inequality. But even if we treat natural resources as ex-
ogenous, we are aware that both natural resource extraction and
reserves can respond to economic forces; for example, oil prices
can inuence oil production as well as oil exploration. We do not
address this problem in this paper, but we acknowledge its potential
importance; at some point, this problem will need to be addressed.
Here we want to let it sufce to explore the possibility that natural
resource ownership impinges on both inequality and growth and
thus illuminates the inverse relationship between inequality and
growth that has been observed in cross-sectional data.
We will now show how natural resource dependence is inversely
related to both equality and growth in a standard growth model.
Thereafter, we will test this prediction empirically in a sample of
eighty-seven industrial and developing countries in the period 1965
1998. Our theoretical model can be summarized as follows: Workers
can earn a living either by working in the primary sector extracting
natural resources from the soil or the sea or through paid employ-
ment in the manufacturing sector, including services. Because hu-
man capital is equally spread across the population, wage income in
manufacturing is the same for all workers. However, due to the
whims of nature, or the competition for the rent generated by the
natural resource, earnings in the primary sector are unequal at each
point in time. It follows that the more time workers devote to natural
resource extraction, the more unequal the distribution of income.
And growth is also affected. If we assume, quite plausibly, that the
manufacturing industry provides greater opportunities for learning
and innovation, it follows that the more time workers spend in the
primary sector, the lower will be the rate of growth. Hence, abun-
dant natural resources cause both inequality and slow growth by
tempting workers away from industries where technology and out-
put are more likely to progress and grow and where earnings are
more equally shared. Elsewhere (Gylfason and Zoega 2001b) we
show how saving and investmentand hence also growthcan de-
pend inversely on natural resources. The intuition is again straight-
forward: When physical capital is less important in the production
technology, the optimal rate of saving is lower. Therefore, the

optimal level of steady-state capital is lower. If we now postulate
learning-by-investing (as in Romer 1986), the rate of technological
progress and the rate of growth of output per capita will conse-
quently both be lower.
Our hypothesis has the advantage that here we have an exogenous
variable that affects the two endogenous variables in a predictable
way, and this makes any empirical testing of the theory more robust.
We will show how the relationship between inequality and growth
can arise in the presence of natural resources. If natural resources
affect both inequality and growth, then this could shed new light on
the statistical relationship between inequality and growth. But to do
this we need to identify, on theoretical grounds as well as empiri-
cally, the relationship between natural resources and inequality, on
the one hand, and between natural resources and growth, on the
other hand. It is to this task that we now turn.
9.2.1 Allocation of Time
Imagine a world in which natural resources generate a constant ow

of riches. All one has to do is go out and pick the fruits of nature,
be they diamonds, sh, or oil. This could involve passively stand-
ing beneath an apple tree or a coconut palm and picking up the
fruits that fall to the ground or one could have to exert oneself
looking for fruits, diamonds or sh, to take a few examples. The
value of each bundle of the natural resource is equal to R and the
likelihood of nding a bundle increases with the time spent search-
ing. Now imagine that amid the bounties of nature there is a manu-
facturing industry that uses labor and capital to produce output
without using or depending in any way on the natural resource.
Assume, crucially for our argument, that workers face a more chal-
lenging and stimulating work environment in the manufacturing
industry, because manufacturing is more likely to foster learning
and innovation. In particular, assume that there is learning-by-
investing in manufacturing.
Workers have a choice when it comes to their work effort: They
can spend part of or all of their time trying their luck picking fruits
or they can take a paid job in industry. Each individual has to decide
how much time to spend picking fruits and how much time to
spend in paid employment. We denote the fraction of time spent in
productive employment by b and the fraction spent picking fruits by

1 b.
Now assume that the discovery of a bundle of natural resources
valued R is a random event and follows a Poisson distribution.
Denote the number of such discoveries by the random variable N.
The random event is then dened as a worker nds a bundle of
the natural resource during a unit of time and has the following
density:
eg1b g1 b N
f N for N 0; 1; 2; . . . ; 1
N!
where the mean arrival ratethat is, the expected number of dis-
coveries by a given worker or, equivalently, the probability that a
discovery will be made by the worker within a unit of timeis
EN g1 b. The expected number of discoveries for the repre-
sentative individual is thus a linear function of the fraction of time
spent searching. The larger the share of time spent in nature, the
more bundles will be discovered. The parameter g measures search
effectiveness.
There are L individuals (identical by assumption) spending part of
their time searching. The aggregate income from the natural resource
is then
Y n NLR: 2
The expected value and the variance of N given by the Poisson dis-
tribution are both equal to g1 b. Since all individuals are identi-
cal, it follows that the variance across the population in the number
of discoveries of the natural resource bundles per unit of time is also
equal to g1 b. We now have the following result: The variance of
the distribution of income emanating from the natural resource is an
increasing function of the time devoted by each worker to the natu-
ral resourcebased sectorprimary sector, for short. Dene income
per capita by lower-case letters. We then have
E y n g1 bR; vary n g1 bR 3
The expected per capita income or rent from the natural resource as
well as the variance of this per capita income across the population
of workers is an increasing function of the abundance of the resource
R and also an increasing function of the time spent procuring it
1 b.
We now turn to the manufacturing industry, which offers workers

an alternative to wandering around nature. This industry uses capi-
tal and labor to produce output and offers opportunities for learning
and innovation. The production function is
Yi qKi a KbLi 1a 4
Here q denotes the quality of capital and takes a value between zero
and one,5 Ki and Li denote the capital and labor used by rm i and K
is the aggregate capital stock in the manufacturing sector. As in
Romer (1986) the aggregate capital stock is a proxy for the accumu-
lated knowledge that has been generated in the past through invest-
ment at all rms. This is what sets manufacturing apart from the
primary sector; it uses capital and the installation of new units of
capital generates a ow of ideas that raises productivity in a labor-
augmenting fashion. In contrast, the primary sector does not offer
similar opportunities for learning and innovation.
We assume a perfectly competitive market for labor and capital.
Assuming symmetric equilibrium, so that K kL, gives the follow-
ing rst-order conditions for maximum prot, and also for equilib-
rium in the two factor markets:
dYi
1 akq a bL 1a bw 5
dLi
dYi
aq a bL 1a r d; 6
dKi
where w is the real wage, r is the real interest rate and d is the rate at
which installed capital loses its usefulness over time, as a result of
economic obsolescence as well as physical wear and tear (Scott
1989).6
The representative worker/consumer has to make two decisions
each moment of his innite life. He has to decide how much to con-
sume and save and how much time to spend working in the manu-
facturing sector rather than trying his luck in the primary sector. We
assume that he cannot do both at the same time. Hence a decision to
spend more time in the primary sector causes him to spend less time
in paid employment making manufactures. Moreover, we assume
that time spent in the primary sector is costly: A direct cost h is in-
curred for each moment spent. Finally, there is a tax on wages tw and
also a tax on income from the natural resources tn .
The worker maximizes the discounted sum of future utility from

consumption:
y
max logct ert dt; 7
b; c 0
where r is the discount rate, subject to

a_ t rat bw1 tw g1 bR1 tn g1 bh ct :
By assumption, the worker does not gain any utility (or suffer dis-
utility) in the primary sector, nor from being employed. The worker
has assets a, which he accumulates if his earnings exceed expendi-
tures (henceforth, we omit time subscripts). His earnings come from
three sources: There is interest income on assets ra, which is tax-free,
there is wage income from employment bw, taxed at tw , and there is
the value of the primary goods he picks or produces 1 bR, taxed
at tn . The worker then incurs the direct cost h and consumes c per
unit of time. A necessary condition for optimal consumption is
1
c ; 8
l
where l denotes the shadow price of wealth. Consumption is at an

optimum when the marginal utility of consumption is equal to the
shadow price of wealth at each instant. More interesting is equation
(9), which helps determine the optimal allocation of time:
w1 tw gR1 tn h: 9
The left-hand side of equation (9) shows the marginal benet from
working longer in manufacturing net of taxes, while the right-hand
side shows the marginal benet from fruit picking, also net of taxes.
While each worker takes wages w as given, wages do nevertheless
respond to market forces. Combining equations (5) and (9) gives the
following equation:
1 akq a L 1a b a 1 tw gR1 tn h: 10
Solving for b gives
1=a
1 akq a L 1a 1 tw
b : 11
gR1 tn h
The time spent in industrial employment b is decreasing in the value

of the natural resource R and search effectiveness g as well as in taxes
on wage income tw , and increasing in the accumulated knowledge in
the manufacturing industry kL K, the productivity of capital q,
taxes on natural resources tn and the cost of utilizing the natural
resource h.
9.2.2 Work and Growth
We can now describe the various ways in which natural resources

affect the allocation of labor in our model.
0
The discovery of natural resources R raises the reward to pro-
ducing primary output and reduces the optimal time spent in
manufacturing.
0
A decrease in the cost of producing primary output h and an
increase in search effectiveness g have an effect identical to that of
a resource discovery: labor leaves manufacturing for the primary
sector.
0
The structure of the tax system affects the allocation of time. The
higher are taxes on wages tw and the lower are taxes on income or
rent from natural resource extraction tn , the more time is devoted to
producing primary goods.
0
History matters because past learning-by-investing in the indus-
trial sector determines current knowledge as reected by k and hence
also real wages. The more advanced the manufacturing sector, the
higher the wages it can afford to pay and the more time workers
spend in manufacturing.
The last point explains why natural resource abundance and

dependence do not have to go together. Abundance of natural
resources is a signicant impediment to growth only if productivity
and wages in the manufacturing sector are low, that is, if there is
little accumulated knowledge and expertise in the sector. But the
presence of abundant natural resources can prevent manufacturing
from taking off, thereby preventing innovation and learning from
taking place:
0
When R is sufciently high, or when productivity in the manu-
facturing sector is sufciently low, it can be optimal not to spend any
time in manufacturing. In this case, growth never takes off.
Education provides a possible solution to this dilemma by in-

creasing labor productivity in manufacturing:
0
Education can increase knowledge, and thereby also labor produc-
tivity which, like past learning-by-doing, lifts wages and draws
workers to the manufacturing industry from the primary sector.
We now turn to the remaining necessary condition for maximum
utility, the Euler equation giving optimal growth of consumption:
c_
r r: 12
c
Equations (6) and (12) give the optimal rate of growth of consump-
tion and output:
g aq a bL 1a d r: 13
Growth is an increasing function of b, the share of time spent pro-
ducing manufactures rather than primary goods.
There are two market failures in the model. The rst is the stan-
dard one that rms, when investing, neglect the gains from learning
and knowledge spillovers to other rms. In contrast, a social planner
uses the average product of capitalnot the private marginal prod-
uctto measure the cost of capital. Second, workers compare the
current benet from spending time in the two sectors but ignore
the growth effect of industrial employment: by spending more time
in the manufacturing industry they, collectively, would raise the
marginal product of capital, the interest rate and economic growth.
This makes their wages grow more rapidly. By withdrawing labor
from the primary sector, workers would invest in a higher future
wage. However, each worker has only a very small effect on growth
imparting an external benet to others.
We can now summarize the relationship between natural
resources and growth.
0
A rise in the natural resource rent R attracts more people to the
primary sector in the hope of securing a piece of the action. These
people leave the manufacturing sector, thereby lowering the private
marginal product of capital, the rate of interest and the rate of
growth of consumption and output. This is the Dutch disease work-
ing through the labor market (Paldam 1997).
0
When abundant natural resources reduce the incentive to provide
good education (Gylfason 2001), this reduces labor productivity and
wages, hence reinforcing the incentive to stay in the natural resource
sector. An abundant natural resourcea high value of Rattracts
workers and this effect is reinforced by bad education which drives
people away from industrial employment.
0
If natural resources reduce the quality of societys institutions, this
could manifest itself in a reduction in the private cost of rent seeking
h. Moreover, less developed capital markets are likely to generate a
lower quality capital stock q, which depreciates at a higher rate d
(Gylfason and Zoega 2001a, 2001b).
We can now combine these insights with the earlier result showing
that the variance of income or rent emanating from the natural re-
source sector is
vary n g1 bR: 14
Equations (13) and (14) show that while economic growth is

increasing in b, inequalitymeasured by the variance of income
is decreasing in b. According to our thesis, any variable that
increases the value of b is likely to stimulate growth and reduce
inequality. Equation (11) shows that an abundance of natural
resourcesrelative to the level of technological know-howwill
lower the value of b. In contrast, any variable that raises labor
productivity and wages in the manufacturing sector will raise b,
increase growth and reduce inequality. The knowledge that has been
generated through past investment and production is one such fac-
tor. Another factor is the level of education. Education that raises
productivity and wages in industry will discourage workers from
spending time in the natural resource sector and hence raise growth
and reduce inequality. At last, the tax system can affect growth and
equality: A high tax on natural resource rents and a low tax on
wages increases the value of b, hence raising growth and reducing
inequality.
9.3 Measuring Inequality
In what follows, we make use of three different measures of in-

equality. Take income inequality rst. The Gini index measures the
Figure 9.1
The Gini index and the 20-20 ratio
extent to which income (or, in some cases, consumption) among

individuals or households within an economy deviates from a per-
fectly equal distribution. A Gini index of 0 represents perfect equal-
ity, while a Gini index of 100 means perfect inequality. As gure 9.1
shows, the Gini index is closely correlated with the log of the ratio of
the income or consumption of the 20 percent of households with the
highest incomes to the income or consumption of the 20 percent of
households with the lowest incomes (the 20/20 ratio). In our sam-
ple, the 20/20 ratio is lowest (2.6, Gini 19.5) in the Slovak Republic
and highest in Sierra Leone (57.6, Gini 62.9). The regression line
through the scatter in gure 9.1 shows that each ten-point increase in
the Gini index goes along with roughly a doubling of the 20/20 ratio.
Thus, for example, the Nordic countries have a Gini index of 25 and
a 20/20 ratio of 3 whereas the United Kingdom has a Gini index of
35 and a 20/20 ratio of 6. The corresponding gures are 30 and 4 for
Germany and 40 and 8 for the United States as well as for China and
Russia. The data come from nationally representative household
surveys and refer to different years between 19831985 and 1998
1999 (World Bank 2000, Table 2.8). The data refer to either (a) per-
Figure 9.2
Income inequality and gender inequality
sonal or household incomes before taxes and transfers or (b) consump-

tion expenditures and, hence, implicitly incomes after taxes and
transfers. Whenever possible, consumption was used rather than in-
come (World Bank 2000). The Gini index of income inequality is
available for seventy-ve of the eighty-seven countries in our sample.
Our second inequality measure is intended to reect one aspect of
social inequality, that is, the unequal access of males and females to
education. We take the difference between the average secondary-
school enrollment rates of males and females in 19801997 to repre-
sent gender inequality in education. In a majority of cases where the
rates are different, more males than females go to secondary school.
In some cases, however, more females than males attend secondary
schools. Even so, we use the arithmetic rather than absolute differ-
ence between male and female enrolment rates as our inequality
measure. This means that we view a change from a situation where,
say, the secondary-school enrollment rate for males is 17 percentage
points higher than that for females (as in Egypt) to a situation where
the secondary-school enrollment rate for females is 17 percentage
points higher than that for males (as in Finland) as a decrease in
gender inequality. Surprisingly, gure 9.2 shows that there is in
Figure 9.3
Income inequality and initial income: The Kuznets curve
our sample no discernible correlation between income inequality as

measured by the Gini index and gender inequality of education as
measured by the excess of male over female secondary-school
enrollment. Thus economic and social inequality, as measured here,
do not necessarily go hand in hand.
Our cross-country data support the notion of a Kuznets curve:
Income inequality tends to increase with income at low levels of
income and to decrease with income at higher levels of income, as
shown in gure 9.3. Galor and Moav (1999) suggest the following
interpretation of the Kuznets curve: in early stages of development,
when investment in physical capital is the main engine of economic
growth, inequality spurs growth by directing resources toward
those who save and invest the most, whereas in more mature
economies human capital accumulation takes the place of physical
capital accumulation as the main source of growth, and inequality
impedes growth by hurting education because poor people cannot
fully nance their education in imperfect credit markets. On the other
hand, the gender inequality of education varies inversely and linearly
with initial income, without any visible tendency for gender in-
equality to increase with income at low levels of income (gure 9.4).
Figure 9.4
Gender inequality and initial income
The third measure of inequality that we will use is the Gini index
for the distribution of land. This measure is taken from Deininger
and Olinto (2000), and covers fty of the eighty-seven countries in
our sample. Figure 9.5 shows that, almost without exception, land is
less equally distributed than income in our sample. Spearmans rank
correlation between the two measures is 0.57.
9.4 Cross-Country Patterns in the Data
In this section, we allow the data to speak for themselves in the form
of a series of bivariate cross-sectional correlations. We rst take a
look at the correlations between our three measures of inequality
and economic growth, all of which are unambiguously negative in
our data: Greater inequality in the distribution of income and land as
well as in access to education tends to go together with lower rates
of growth. We then move on to show that two of the three measures
of inequality increase from country to country in tandem with the
share of natural capital in national wealth. This opens up the possi-
bility that it is the variation in natural capital in the sample that
Figure 9.5
Distribution of income and land
generates the apparent relationship between inequality and growth:

When natural resources become more important, inequality rises and
growth recedes. This was the prediction of our model in section 9.2.
At last, we also show that income inequality and three different
measures of education are inversely related, while education and
growth are positively correlated. This nding accords with earlier
research indicating that education, by reducing inequality and fos-
tering growth, can help clarify the inverse relationship between
inequality and growth that is observed in the data. Unlike natural
resource abundance, however, education is probably best viewed as
an endogenous variable, a possibility that we address explicitly in
the regression analysis presented in section 9.5.
9.4.1 Inequality and Growth
Let us now begin by looking at the cross-country pattern of income

inequality and economic growth. Figure 9.6 shows a scatterplot of
the annual rate of growth of gross national product (GNP) per capita
from 1965 to 1998 (World Bank 2000, Table 1.4) and the inequality
Figure 9.6
Income inequality and economic growth
of income or consumption as measured by the Gini index (World

Bank 2000, Table 2.8). The growth rate has been adjusted for initial
income: the variable on the vertical axis is that part of economic
growth that is not explained by the countrys initial stage of devel-
opment, obtained as a residual from a regression of growth during
19651998 on initial GNP per head (i.e., in 1965) as well as the share
of natural capital in national wealth, taken from World Bank (1997).
The seventy-ve countries shown in the gure are represented by
one observation each.7 The regression line through the scatterplot
suggests that an increase of about twelve points on the Gini scale
from one country to another is associated with a decrease in per
capita growth by 1 percentage point per year on average. Twelve
points on the Gini scale correspond roughly to the difference be-
tween income inequality in the United Kingdom (Gini 36) and in
Sweden and Japan (Gini 25). The relationship in gure 9.6 is
statistically signicant (Spearmans rank correlation is 0.50). If
rich countries and poor are viewed separately, a similar pattern is
observed in both groups (not shown). Shaving one percentage point
off any countrys annual growth rate is a serious matter because the
(weighted) average rate of per capita growth in the world economy
Figure 9.7
Gender inequality and economic growth
since 1965 has been about 1.25 percent per year. We see no signs of
the positive cross-sectional relationship between inequality and
growth in rich countries reported by Barro (2000), nor do we see any
evidence of the nonlinearity in the panel relationship documented by
Banerjee and Duo (2000a,b).
Figures 9.7 and 9.8 tell a similar story. Here we see the cross-
country pattern of per capita growth as measured in Figure 9.6
and gender inequality of education (gure 9.7) and land inequality
(gure 9.8). The pattern is not as clear as in gure 9.6, but it is still
statistically signicant (Spearmans rank correlation is 0.32 and
0.37, respectively). The number of countries is seventy-ve and
fty in the two gures. All countries for which the requisite data
are available are included in all the gures in the paper, without
exception.
9.4.2 Natural Resources, Inequality, and Growth
In gure 9.9, we measure natural resource dependence by the share

of natural capital in national wealth in 1994namely, the share of
natural capital in total capital, which comprises physical, human,
Figure 9.8
Land inequality and economic growth
Figure 9.9
Income inequality and natural capital
and natural capital (though not social capital; see World Bank 1997).
The natural capital variable is intended to come closer to a direct
measurement of the intensity of natural resources across countries
than the various proxies that have been used in earlier studies,
mainly the share of primary (i.e., nonmanufacturing) exports in total
exports or in gross domestic product (GDP) and the share of the
primary sector in employment or the labor force. The latter proxies
may be prone to bias due to product and labor market distortions.
Figure 9.9 shows that the share of natural capital in national
wealth is positively correlated with income inequality as measured
by the Gini index. Spearmans rank correlation is 0.41. Notice the
cluster of ve countries (Niger, Guinea-Bissau, Madagascar, Mali,
and Zambia, in descending order) in the northeast corner of the
gure with a natural capital share above 35 and Gini above 45. Even
if this cluster is removed from the sample, the pattern remains sta-
tistically signicant. Notice, further, the two countries (Sierra Leone
and the Central African Republic) with a natural capital share of
around 30 and Gini above 60. If this pair of observations is omitted,
the pattern remains signicant. If, however, both clusters (i.e., all
seven countries) are removed from the sample, the remaining pat-
tern becomes insignicant in a statistical sense. In this sense, this
group of seven African countries in the northeast corner of the gure
explains the inverse correlation. Even so, we are inclined to keep
these African countries in our sample. We nd it instructive that no
country with a natural capital share above 25 has a Gini coefcient
below 45.
Figure 9.10 shows that the natural capital share is also positively
correlated with gender inequality as measured by the male minus
female secondary-school enrollment rate. Spearmans rank correla-
tion is 0.32. The pattern observed is statistically signicant with or
without the seven African countries mentioned above. Moreover,
there is a positive albeit insignicant correlation between land
inequality and natural capital in our sample (not shown); Spear-
mans rank correlation is 0.19.
From gures 9.9 and 9.10 combined with gure 9.11, which shows
that the natural capital share varies inversely with per capita eco-
nomic growth from 1965 to 1998 across the same group of countries,
we conclude that these ndings may help explain the inverse cross-
sectional relationship between inequality and growth shown in
gures 9.6 and 9.7. In gure 9.11, the rank correlation between nat-
Figure 9.10
Gender inequality and natural capital
Figure 9.11
Natural capital and economic growth
Figure 9.12
Natural capital and expenditure on education
ural capital and growth (r 0:64) is statistically signicant, and

remains so even if the two clusters in the southwest corner and the
northeast corner of the gure are excluded from the sample (see
Gylfason and Zoega 2001b).
At last, gure 9.12 shows that, in our sample, natural capital is
also inversely and signicantly correlated with public expenditure
on education (r 0:32). Natural capital is also inversely and sig-
nicantly related to years of schooling for girls and secondary-school
enrollment for both genders (not shown).
9.4.3 Inequality and Education
Let us now consider the three measures of education inputs, out-

comes, and participation and how they vary with inequality and
economic growth. Figure 9.13 shows a scatterplot of public expendi-
ture on education from 1980 to 1997 as reported by UNESCO (see
World Bank 2000, Table 2.9) and income inequality. Public expendi-
ture on education varies a great deal from country to country. In
the 1990s, some countries spent as little as 1 percent of their GNP
Figure 9.13
Expenditure on education and income inequality
on education (Haiti, Indonesia, Myanmar, Nigeria, and Sudan).

Others have spent between 8 percent and 10 percent of their GNP on
education, including St. Lucia, Namibia, Botswana, and Jordan, in
descending order. Public expenditure is admittedly an imperfect
measure of a nations commitment to education, not least because
some nations spend more on private education than others. More-
over, public expenditure on education may be supply-led and of
mediocre quality, and may thus fail to foster efciency, equality, and
growth, in contrast to private expenditure on education, which is
generally demand-led and thus, perhaps, likely to be of a higher
quality. Even so, this yardstick should reect at least to some extent
the governments commitment to education. The regression line
through the seventy-four observations in gure 9.13 suggests that an
increase in public expenditure on education by one percent of GNP
from one country to the next is associated with a decrease of 2.3
points in the Gini index. The relationship is statistically signicant
(r 0:36).
Figure 9.14 shows scatterplots of the expected number of years of
schooling for females from 1980 to 1997 and income inequality. This
Figure 9.14
Years of schooling and income inequality
indicator of schooling is intended to reect the total education

resources, measured in school years, that a girl will acquire over her
lifetime in school or as an indicator of an education systems overall
state of development. In gure 9.14, the regression line through the
forty-six observations, one per country, suggests that an increase by
one year of the schooling that an average girl at the age of school
entry can expect to receive is associated with a decrease in the Gini
index, namely, increased equality, by almost one point. The rela-
tionship is statistically signicant (r 0:49). Unlike the relationship
in gure 9.13, the one in gure 9.14 is signicantly nonlinear (not
shown), suggesting that the marginal effect of increased education
on equality is rising in the level of educationthat is, there may be
increasing returns to schooling in terms of equality. Sen (1999),
among others, emphasizes the importance of educating girls in
developing countries. The corresponding relationship for males (not
shown) is virtually the same as for females.
In gure 9.15, we present a scatterplot of secondary-school enrol-
ment and income inequality. The pattern is clear: an increase in
secondary-school enrollment by 5 percent of each cohort goes hand
in hand with a decrease in the Gini index by one point. The data
Figure 9.15
School enrollment and income inequality
exhibit a similar, albeit not quite as strong, relationship between

secondary-school enrollment and gender inequality (not shown). The
same applies to gures 9.13 and 9.14: Public expenditure on educa-
tion and years of schooling for girls are also inversely related to
gender inequality (not shown). All three measure of education are
positively correlated with economic growth (not shown).
These patterns seem to suggest that more and better educa-
tion goes along with less inequality as well as more rapid growth
and that human capital, like natural capital, thus can perhaps
help explain the inverse relationship between inequality and growth
that we observe in the data. To nd out, we need to dig a little
deeper.
9.5 Regression Analysis
Table 9.1 reports seemingly unrelated regression (SUR) estimates of

a system of ve equations for the eighty-seven countries in our sam-
ple for the years 19651998. The equations reveal how natural capi-
tal intensity can affect growth through various channels: through
investment, education and inequality, as well as directly.
9.5.1 The Model and Estimation
The rst equation shows how economic growth depends on (a) the
logarithm of initial per capita income (i.e., in 1965), dened as
income in 1998 divided by an appropriate growth factor, (b) the
share of natural capital in national wealth (which comprises physi-
cal, human and natural capital), (c) the share of gross domestic
investment in GDP in 19651998, (d) the log of the secondary-school
enrollment rate (the log in order to capture diminishing returns to
education), (e) the Gini index, and (f) gender inequality of education
as measured by the difference between male and female secondary-
school enrollment rates in 19801997. This equation can be inter-
preted either as a description of endogenous long-run growth or of
medium-term growth in the neoclassical model where economic
growth is exogenous in the long run. Initial income is intended to
capture conditional convergence. Natural capital is another exoge-
nous determinant of growth. Investment and education are intended
to capture the contribution of physical and human capital accumu-
lation to growth. The inequality measures reect the hypothesized
effects of income and gender inequality on growth.
The second equation shows the relationship between the invest-
ment rate and the natural capital share (as spelled out in Gylfason
and Zoega 2001b; the underlying explanation is that increased
dependence on natural resources reduces the share of physical capi-
tal in GDP and thereby weakens the incentive to save and invest by
our extension of the Golden Rule).
The third equation shows how the enrolment rate depends on
initial income (because wealthy countries can afford to spend more
on education) as well as on natural capital (as in Gylfason 2001;
Gylfason and Zoega 2001b; the idea behind this formulation is that
the natural-resource-intensive sector may nd it protable to use
workers with fewer skills than the manufacturing sector).
The fourth equation shows the relationship among the Gini index,
initial income (i.e., the Kuznets curve), and the natural capital share
that we documented in section 9.4. The fth and last equation
shows the relationship between gender inequality and the natural
capital share. To recapitulate, our hypothesis from section 9.2 is that
because natural resource ownership tends to be less equally dis-
tributed than other assets, countries that depend heavily on their
natural resources tend to have a less equal distribution of income,

education, and land than countries that are less dependent on their
natural wealth.
The recursive nature of the system and the conceivable correlation
of the error terms in the four equations make SUR an appropriate
estimation procedure (Lahiri and Schmidt 1978). However, the fact
that ordinary least squares (OLS) estimates of the system (not
shown) are almost identical to the SUR estimates shown in table 9.1
indicates that the correlation of error terms across equations is of
minor consequence. In our data, each country is represented by a
single observation. This is because our data on natural resources are
limited to a single year, 1994. In view of this, our analysis is conned
to a cross section of countries, even if panel data on income distri-
bution have recently become available (Deininger and Squire 1996).
An extension of our analysis to panels must await richer data on
natural capital. This may be important because some writers (e.g.,
Forbes 2000) have reported panel regression results on inequality
and growth that seem to go against some of the results that have
been obtained from cross-sectional studies (but see Banerjee and
Duo 2000b, who disagree with Forbes 2000 and also Benabou 1996).
9.5.2 Empirical Results
All the coefcient estimates shown in table 9.1 are economically and
statistically signicant, with one exception. The coefcient on initial
income in the growth equation indicates a conditional convergence
speed of 1.3 percent per year. The direct effect of the natural capital
share on growth is 0.05 and the indirect effects through investment
and education are 0:20 0:11 0:022 and 0:03 1:08 0:032.
The additional indirect effect of the natural capital share on growth
via the Gini index is 0:30 0:04 0:012. The total effect of natu-
ral capital on growth is, therefore, about 0.12 (for given initial
income). Hence, the income distribution channel accounts for about
one-tenth of the total effect of natural capital intensity on growth.
Of additional interest here are the effects of education and
inequality on growth. The rst equation in the table shows the direct
effect of education on growth to be 1.08/E 0.025, evaluated at the
mean value of the secondary-school enrollment rate, E 0.43; this
means that an increase in the enrollment rate by ten percentage
Table 9.1
284
Regression results
Initial Natural
Initial income capital Investment Enrollment Gender
Dependent variable income squared share rate rate (log) Gini index inequality R2 Countries
Economic growth 1.26 0.05 0.11 1.08 0.04 0.01 0.68 74
(6.05) (5.19) (3.82) (3.88) (2.84) (0.76)
Investment rate 0.20 0.15 87
(3.98)
Enrollment rate 0.54 0.03 0.70 87
(11.31) (6.29)
Gini index 48.88 3.20 0.30 0.31 74
(3.54) (3.69) (2.84)
Gender inequality 0.25 0.09 87
(2.98)
Note: Estimation method: SUR. t-ratios are shown within parentheses. Constant terms are not reported to conserve space.
Thorvaldur Gylfason and Gyl Zoega
points from one country to another increases growth by 0.25 of a

percentage point. The direct effect of increased income inequality on
growth is also rather strong: an increase in the Gini index by fteen
points, which corresponds to the difference between Norway
(Gini 26) and the United States (Gini 41), from one place to
another is associated with a decrease in growth by 0.6 percentage
points which, in turn, is about a half of the average per capita
growth in our sample over the period under review. On the other
hand, we do not nd signicant evidence of a negative effect of
increased gender inequality of education and economic growth; the
coefcient reported in the top line of table 9.1 is negative, true, but
small and insignicant. Even so, an increase in the natural capital
share increases both types of inequality signicantly and substan-
tially. Thus, an increase in the natural capital share by 10 percentage
points from one country to another increases the Gini index by three
points and the difference between male and female secondary-school
enrollment rates by 2.5 percentage points; the latter type of increased
inequality, however, does not signicantly hamper growth.
It is interesting to note that the inclusion of the natural capital
share and the secondary-school enrollment rate in the growth equa-
tion does not reverse the sign of the estimated coefcient of the Gini
index. In particular, the relationship between growth and inequality
remains negative, in contrast to the results of Forbes (2000). How-
ever, the size of the income distribution effect is reduced by about a
half by the inclusion of the natural capital and school enrollment
variables. This seems to suggest that in growth equations without
natural capital and education, the income distribution variable picks
up a good part of the inuence of the omitted variables. Our cross-
sectional results bear out a long-term relationship between inequal-
ity and growth while the pooled estimation of Forbes (2000) reects
short- to medium-term relationships by her own reckoning. It is also
possible that the inclusion of omitted, country-specic variables
other than natural capital and education could reverse the sign of the
coefcient of the Gini index.
Notice, at last, that the data support the notion of a Kuznets curve
relating income inequality and initial income. There is, however, no
comparable nonlinear relationship between gender inequality and
initial income. In our data, initial income has no signicant effect on
investment across countries.
9.5.3 Other Possibilities
We have experimented with several variations of the model speci-

cation in table 9.1.
First, we added natural capital per person as a proxy for natural
resource abundance in order to distinguish between natural resource
abundance and natural resource intensity (as in Gylfason and Zoega
2001b). By intensity, or dependence, we mean the importance of
natural resources to the national economy, while abundance refers to
the supply (per capita) of the natural resources. Some countries
Australia, Canada, and the United States, to name a fewhave
abundant natural resources but are not particularly dependent upon
them, not any more. Our argument has been that it is natural
resource dependence that matters for inequality and growth. We do
not expect Australia, Canada, or the United States to suffer from
their abundance of natural resources, far from it. When we add nat-
ural capital per person as an independent explanatory variable to
each equation in table 9.1, it turns out that natural resource abun-
dance encourages economic growth, investment and education and
reduces gender inequality, but has no effect on income inequality. In
other respects, the results remain virtually the same as in table 9.1.
This means that increased dependence on natural resources hurts
growth, as we hypothesized, while increased abundance helps (for
more, see Gylfason and Zoega 2001b).
Next, we entered the natural capital share and the Gini index
of income inequality multiplicatively rather than additively in our
growth equation in order to study the interaction between the two
variables. Now the coefcient of the multiple is 0.0011 (with
t 3.72). This means that the negative effect of natural resource
dependence on growth varies directly with income inequality: The
more unequal the distribution of income, the greater is the adverse
effect of natural resource dependence on growth. Evaluated at the
mean value of the Gini index in our sample (42), the effect of the
natural capital share on growth is 0.05 as in table 9.1. This new
specication also means that the negative effect of income inequality
on growth varies directly with natural resource dependence: The
greater the natural capital share, the greater is the adverse effect of
income inequality on growth. Evaluated at the mean value of
the natural capital share in our sample (12), the effect of income
inequality on economic growth is 0.013, which is smaller than the
coefcient of the Gini index in the rst equation in table 9.1. When
we replace the Gini index of income inequality in the above experi-
ment with our measure of gender inequality or of land inequality,
we obtain the same results: The greater the natural capital share, the
greater is the adverse effect of increased inequality on growth.
Third, we replaced our gender inequality measure (the arithmetic
difference between male and female secondary-school enrollment
rates) by the absolute difference between male and female enrolment
rates. The new measure means that a change from a situation where
more boys than girls go to school to one where more girls than boys
go to school leaves gender inequality unchanged if the numbers are
the same. When we reestimate our system using this new measure,
increased gender inequality reduces economic growth directly: The
coefcient on gender inequality in the rst equation in table 9.1 is
now 0.05 with t 2.09. In this case, however, the effect of the nat-
ural capital share on gender inequality becomes small and statisti-
cally insignicant (the coefcient is 0.08 with t 1.47). In other
respects, the regression results (not shown) are very similar to those
reported in table 9.1.
Our fourth and last experiment involves Africa and Latin America.
When we add a dummy variable for Africa to each equation in our
model, in case Africa might be different from other regions, as some
studies have shown, the dummy coefcient has the expected sign
everywhere, but it is statistically signicant only in the equations for
education and the Gini index. The annual rate of per capita growth
in Africa is thus 0.75 of a percentage point smaller than elsewhere
according to our results (not shown), but the difference is not signif-
icant (t 1.73). The investment rate is almost 2 percentage points
lower in Africa than elsewhere, but again the difference is insigni-
cant (t 1.38). The secondary-school enrollment rate is 15 percent-
age points lower in Africa than elsewhere (evaluated at the sample
mean), and this difference is signicant (t 3.23). Gender inequality
in education is also signicantly greater in Africa than elsewhere, by
almost ve percentage points (t 2.15). There is, on the other hand,
no signicant difference between the Gini index in Africa and the
rest of our sample. All the estimates shown in table 9.1 remain
essentially intact in the presence of the African dummy. When
we add a dummy variable for Latin America (with or without Cen-
tral America) rather than for Africa, the dummy has no effect on
growth, investment, or education, but it does matter for distribution;
specically, the Latin dummy reduces gender inequality by 7.5 per-

centage points (t 2.48) and increases the Gini index of income in-
equality by ten points (t 3.20). Again, our estimates in table 9.1
remain unchanged. We conclude that the specication of our model
in table 9.1 is sufciently broad to render the inclusion of regional
dummy variables superuous.
9.6 Conclusion
The inverse empirical relationship between inequality and economic

growth across countries that has emerged from several recent studies
has spurred several authors to suggest various potential theoretical
explanations for the relationship. These explanations have generally
been of the following kind: Inequality is bad for some variable X
for example, educationand X is good for growth, so increased
inequality hurts growth by hurting X. We approach this issue from a
different angle: We argue that a countrys dependence on its natural
resources inuences both inequality and growth. We showboth
theoretically and empiricallyhow variations in the share of natural
resources in national wealth can help explain the inverse relationship
between inequality and economic growth across countries.
The essence of our story is this: If the distribution of ownership of
natural resources is more unequal than the distribution of other
forms of wealth, the inequality of the distribution of income, educa-
tion, or land is directly related to the share of natural resources in
national income. Specically, we showin the context of an endog-
enous growth model of the simplest kindhow natural resources
can reduce growth and increase inequality by attracting workers
away from higher technology industries. Our data appear to conrm
this prediction: they suggest that the Gini index of income inequality
as well as gender inequality varies directly with the share of natural
capital in national wealth. The data also bear out an inverse rela-
tionship between economic growth and the share of natural capital
in national wealth.
Differences in human capital across countries appear also to help
explain the inverse cross-country correlation between economic
growth and inequality. More and better educationmeasured by
secondary-school enrollment, years of schooling, or public expendi-
ture on educationis associated with less inequality and more rapid
growth in our data. This suggests a clear role for public policy in
combating the potentially adverse effects of excessive dependence

on natural resources on income inequality and growth. In addition,
tax policy can be used to combat the adverse effect of natural
resources on inequality and growth. When income or rent from nat-
ural resource extraction is taxed at a higher rate than wage income,
this discourages workers from spending time in the natural resource
sector, raises the marginal product of capital in manufacturing,
increases the real rate of interest and thereby also the rate of growth
of output and consumption per capita.
Our regression results suggest that natural capital intensity
reduces growth directly as well as indirectly by reducing equality,
secondary-school enrollment rates and investment rates. This leaves
an important role for public policy, which can be used to encour-
age growth by enhancing equality, among other things. We con-
clude that the trade-off between equality and (dynamic) efciency is
affected by both natural and human capital, as well as by tax policy.
Notes
Thorvaldur Gylfason gratefully acknowledges nancial support from Jan Wallanders

och Tom Hedelius Stiftelse in Sweden. Gyl Zoega acknowledges with thanks helpful
comments from Theo Eicher, Stephen Turnovsky, and other participants of the CESifo
conferences on Growth and Inequality: Issues and Policy Implications.
1. This empirical nding does not support the claim of Garca-Penalosa (1995) that in
rich countries increased inequality discourages education and growth by increasing
the number of poor people who cannot afford education whereas in poor countries
increased inequality encourages education and growth by increasing the number of
rich people who can afford education.
2. Further, Aghion (1998) suggests that excessive inequality may be associated with
macroeconomic volatility through credit cycles because of unequal access to credit and
thus to investment opportunities, and that this may hurt investment and growth.
3. See also Aghion, Caroli, and Garca-Penalosa (1999).
4. For a further discussion of recent empirical literature on inequality and growth, see
Benabou (1996).
5. Like Scott (1989), we distinguish between quantity and quality. If some investment
projects miss the mark and fail to add commensurately to the capital stock, we have
q < 1. There are three ways to interpret q: (a) as an indicator of distortions in the allo-
cation of installed capital due to a poorly developed nancial system, trade restric-
tions, or government subsidies that attract capital to unproductive uses in protected
industries or in state-owned enterprises where capital may be less productive than in
the private sector (Gylfason, Herbertsson, and Zoega 2001); (b) as the ratio of the eco-
nomic cost (i.e., minimum achievable cost) of creating new capital to the actual cost of
investment (Pritchett 2000)that is, K is then measured on the basis of actual costs,
which may overstate its productivity; or (c) as a consequence of aging: the larger the
share of old capital in the capital stock currently in operation, namely, the higher the
average age of capital in use, the lower is its overall quality (Gylfason and Zoega
2001a). For our purposes, the three interpretations are analytically equivalent. How-
ever, we assume that the quality of capital has remained constant in the past, which
means that all units of capital are of the same quality. In other words, we are not
interested here in the implications of having different vintages of capital.
6. The parameters q and d can both be modeled as endogenous choice parameters (as
in Gylfason and Zoega 2001a), but here we treat them as exogenous magnitudes for
simplicity, even if we acknowledge that depreciation may depend on quality, through
obsolescence.
7. All countries for which the requisite data are available are included in all the
gures in the chapter, without exception.
References
Aghion, Philippe, with Eve Caroli and Cecilia Garca-Penalosa. 1998. Inequality and
Economic Growth. Part I in Growth, Inequality and Globalization: Theory, History and
Policy, Philippe Aghion and Jeffrey G. Williamson. Cambridge: Cambridge University
Press.
nomic Growth: The Perspective of the New Growth Theories. Journal of Economic Lit-
erature 37: 16151660, December.
Alesina, Alberto, and Roberto Perotti. 1996. Income Distribution, Political Instability,
and Investment. European Economic Review 40: 12031228, June.
Alesina, Alberto, and Dani Rodrik. 1994. Distributive Politics and Economic
Growth. Quarterly Journal of Economics 109: 165190, May.
Banerjee, Abhijit V., and Esther Duo. 2000a. Inequality and Growth: What Can the
Data Say? Working Paper No. 0009, Department of Economics, MIT.
Banerjee, Abhijit V., and Esther Duo. 2000b. A Reassessment of the Relationship
Between Inequality and Growth. Unpublished manuscript, Department of Econom-
ics, MIT.
Barro, Robert J. 2000. Inequality and Growth in a Panel of Countries. Journal of Eco-
nomic Growth 5: 532, March.
Benabou, Roland. 1996. Inequality and Growth. NBER Macroeconomics Annual.
Benhabib, Jess, and Aldo Rustichini. 1996. Social Conict and Growth. Journal of
Economic Growth 1: 124142, March.
Bourguignon, Francois, and C. Morrison. 1990. Income Distribution, Development
and Foreign Trade. European Economic Review 34: 11131132, September.
Bruno, Michael. 1984. Raw Materials, Prots, and the Productivity Slowdown.
Quarterly Journal of Economics 99: 128, February.
Deininger, Klaus, and Pedro Olinto. 2000. Asset Distribution, Inequality, and
Growth. Working Paper No. 2375, World Bank, Development Research Group.
Deininger, Klaus, and Lyn Squire. 1996. A New Data Set Measuring Income Inequal-
ity. World Bank Economic Review 10: 565591, September.
Easterly, William, and Sergio Rebelo. 1993. Fiscal Policy and Growth: An Empirical
Investigation. Journal of Monetary Economics 32: 417458, December.
Forbes, Kristin J. 2000. A Reassessment of the Relationship Between Inequality and
Growth. American Economic Review 90: 869887, September.
Galor, Oded, and Omer Moav. 1999. From Physical to Human Capital: Inequality in
the Process of Development. CEPR Discussion Paper No. 2307, December.
Galor, Oded, and Joseph Zeira. 1993. Income Distribution and Macroeconomics.
Review of Economic Studies 60: 3552, January.
Garca-Penalosa, Cecilia. 1995. The Paradox of Education or the Good Side of
Inequality. Oxford Economic Papers 47(2): 265285.
Gylfason, Thorvaldur. 2001. Natural Resources, Education, and Economic Develop-
ment. European Economic Review 45: 847859, May.
Gylfason, Thorvaldur, and Gyl Zoega. 2001a. Obsolescence. CEPR Discussion

Paper No. 2833, June.
Gylfason, Thorvaldur, and Gyl Zoega. 2001b. Natural Resources and Economic
Growth: The Role of Investment. CEPR Discussion Paper No. 2743, March.
Gylfason, Thorvaldur, Tryggvi Thor Herbertsson, and Gyl Zoega. 2001. Ownership
and Growth. World Bank Economic Review 15: 431449, October.
Kaldor, Nicholas. 1956. Alternative Theories of Distribution. Review of Economic
Studies 23: 83100.
Lahiri, Kajal, and Peter Schmidt. 1978. On the Estimation of Triangular Structural
Systems. Econometrica 46: 12171221.
Paldam, Martin. 1997. Dutch Disease and Rent Seeking: The Greenland Model.
European Journal of Political Economy 13: 591614, August.
Perotti, Robert. 1996. Growth, Income Distribution, and Democracy: What the Data
Say. Journal of Economic Growth 5: 149187, June.
Persson, Torsten, and Guido Tabellini. 1994. Is Inequality Harmful for Growth?
American Economic Review 84: 600621, June.
Pritchett, Lant. 2000. The Tyranny of Concepts: CUDIE (Cumulated, Depreciated,
Investment Effort) Is Not Capital. Journal of Economic Growth 5: 361384, December.
Romer, Paul M. 1986. Increasing Returns and Long-Run Growth. Journal of Political
Economy 94: 10021037, October.
Sachs, Jeffrey D., and Andrew M. Warner. 1995. Natural Resource Abundance and
Economic Growth. NBER Working Paper 5398, Cambridge. Revised 1997, 1999.
Saint-Paul, Gilles, and Thierry Verdier. 1993. Education, Democracy and Growth.
Saint-Paul, Gilles, and Thierry Verdier. 1996. Inequality, Redistribution and Growth:
A Challenge to the Conventional Political Economy Approach. European Economic
Review 40: 719728.
Scott, Maurice Fitzgerald. 1989. A New View of Economic Growth. Oxford, UK: Claren-
don Press.
Sen, Amartya. 1999. Development as Freedom. Oxford: University Press.
Todaro, Michael P. 1997. Economic Development. London: Longman.
World Bank. 1997. Expanding the Measure of Wealth: Indicators of Environmentally

Sustainable Development. Environmentally Sustainable Development Studies and
Monographs Series No. 17, World Bank, Washington, DC.
World Bank. 2000. World Development Indicators 2000. Washington, DC: World Bank.
10 Wage Inequality and the
Effort Incentive Effects of
Technical Change
Campbell Leith,
Chol-Won Li, and
Cecilia Garca-Penalosa
10.1 Introduction
The recent increase in earnings inequality in a number of industrial-

ized countries is by now a well-documented event. Countries have
differed in their experiences, with the most pronounced increases
taking place in the United Kingdom and the United States. An
important component of this increase in inequality has been the rise
in the educational wage differential. Between 1980 and 1988, the
wage ratio of university graduates to workers with no qualication
increased by almost 8 percent in the United Kingdom, and the wage
ratio of college to high school graduates rose by some 25 percent in
the United States over the period 19791995.1
The main explanation for the upsurge in wage inequality has been
the hypothesis of an acceleration in skill-biased technological change. In
particular, it has been argued that the development of new informa-
tion technologies has resulted in a shift in the relative demand for
labor in favor of those with greater skills (see Berman, Bound, and
Griliches 1994).2 The aim of this chapter is twofold. First, it con-
tributes to the theoretical literature on the relationship between
technological progress and relative wages by examining how, in
the presence of imperfect information in the labor market, techni-
cal change can affect not only demand but also the effective supply
of skills. Second, imperfect information will generate equilibrium
unemployment, and will allow us to account for a fact that has
largely been ignored by previous explanations of the rise in the skill
premiumnamely, that the increase in the relative wage has been
accompanied by an increase in unemployment rates for both skilled
and unskilled workers (see the discussion in section 10.2).
Our argument is based on the efciency wage model of Shapiro
and Stiglitz (1984), whereby imperfect information on the part of
294 Campbell Leith, Chol-Won Li, and Cecilia Garca-Penalosa
rms about whether or not employees are shirking forces the former
to pay wages above the market clearing level, which in turn leads
to unemployment. The combination of high wages and the risk of
remaining unemployed if found shirking and red, induces optimal
effort on the part of workers. We introduce technological progress
into this framework. We stress that an important feature of new
technologies is that they not only create new jobs, they also destroy
old ones. When an innovation arrives, some workers retain their jobs
but others are reallocated between jobs or made redundant.3 This
process affects the effort incentives of workers, and hence the effec-
tive labor supply. That is, changes in the rate of technical change
alter the trade-off between pay and unemployment that rms face,
and will affect equilibrium wages and employment.
In our model, the net impact of technical change on wages is
ambiguous. Faster technical change increases the discounted wage
ow but, since it also raises turnover, it reduces the probability
of remaining with the current employer. Hence it may increase
or decrease the present value of being employed, depending on
parameter values. We consider two types of workers, skilled and
unskilled, and assume that it is easier to monitor the effort levels of
the unskilled, and that it is easier for a skilled worker that has lost
her job to immediately nd a new one, as her transferable skills
make her more adaptable to the new technology than an unskilled
worker. These differences imply that the incentives of the two types
of workers will not be affected in the same way by a change in the
rate of technical progress, and that consequently the relative (effec-
tive) supply of workers will shift.
A number of results emerge. First, if technical change is biased, in
the sense that it increases the demand for skilled workers relative to
that for the unskilled, then the incentive mechanism may strengthen
or partially offset the impact of demand on relative wages. A more
surprising nding is that if technical change is skill-neutral, in the
sense that it leaves the relative demand for labor unchanged, an
increase or fall in the rate of technical change will still change the
relative wage. Third, we show that a reduction in the rate of tech-
nical change can generate an increase in the skill premium, which is
consistent with the productivity slowdown experienced during the
1970s and 1980s.
The model also generates patterns of unemployment that t the
data. As we see in detail in section 10.2, previous work cannot
explain the increase in both the skilled and unskilled unemployment
Wage Inequality and Technical Change 295
rates alongside the increase in relative wages. The efciency wage

model implies that there is equilibrium unemployment in all labor
markets. Moreover, changes in the rate of technical change will affect
both wages and employment. In this context it is possible for a
decline in the rate of technological progress to increase the wages of
the skilled relative to the unskilled and reduce the level of employ-
ment for both types of workers.
This chapter contributes to two recent strands in the growth liter-
ature, both of which have not received as much attention as they
merit. The rst one is concerned with the relationship between
unemployment and technological progress, pioneered by Aghion
and Howitt (1994); the second is the literature on growth and
imperfect information in the labor market, where the cost of train-
ing a worker is assumed to depend on her ability, which is not
known to the rm (see Eicher and Kalaitzidakis 1997; Eicher 1999).
Our modeling strategy differs from the former in that we consider
how technical change affects the supply rather than the demand for
labor, and from the latter in the type of information asymmetry
considered.
The chapter proceeds as follows. Section 10.2 discusses the exist-
ing literature on technical change and the skill premium and argues
that there are a number of empirical regularities that they have dif-
culty explaining. Section 10.3 outlines the model and considers the
incentive effects of technological progress. In section 10.4 we use
the model to analyze the impact of technical progress on the skill-
premium and show that both skill-biased and skill-neutral techno-
logical progress affect the relative wage. We then show in section
10.5 how our framework can generate a simultaneous increase in
the relative wage and in the unemployment rates of both types of
workers and argue that differences in unemployment benets and
employment protection legislation can explain the different experi-
ences of continental Europe relative to the United States and the
United Kingdom. We summarize our results and discuss further
policy implications in section 10.6.
10.2 Problems with Existing Explanations of the Increase in

Relative Wages
The early empirical literature on the increase in relative wages in the

1980s found little support for the role of supply or international
trade as potential explanations.4 Theoretical work has consequently
concentrated on modeling the way in which new technologies shift

the relative demand for labor. Recent empirical work has, however,
documented the importance of both international trade and changes
in the supply of skills. Feenstra and Hanson (1999) nd that when
we measure trade by the degree of outsourcing, increased compe-
tition from newly industrializing countries can account for a large
fraction of the change in the relative wage in the United States. Sup-
ply effects have been documented by Card and Lemieux (2001), who
show that, once the labor force is decomposed into cohorts, there
has been a slowdown in the rate of growth of educational attainment
of those cohorts that have experienced the greatest increase in the
education premium. Still, they nd that there has been an increase in
the returns to education for all cohorts which may well be due to
technological change.
The existing theoretical literature has, however, encountered two
problems when trying to t the evidence. The rst one is the pro-
ductivity slowdown. The 1970s and 1980s witnessed a sharp reduc-
tion in rates of total factor productivity (TFP) growth, with TFP
growth in the United States falling from 3 percent in the mid-
1960s to around 1 percent by the late 1980s, and by even more in the
United Kingdom, France, and Germany.5 Yet most work explaining
wage inequality through technical change relies on an increase in
the rate of technological progress.6 The reason is that this approach
is based on the hypothesis, rst put forward by Nelson and
Phelps (1966), that the main difference between educated and non-
educated workers is the greater capacity of the former to absorb
and implement new technologies. The relative demand for skilled
labor will then only increase if there is faster technical change that
forces rms to employ skilled workers to implement the new tech-
nologies.
A second question that explanations of the increase in the skill
premium must answer is: Why is it that changes that should have
affected all industrial economies in roughly the same waysuch as
competition from newly industrializing countries or biased technical
changehave not had similar effects on relative wages? The stan-
dard explanation is that in the United States and the United King-
dom, exible labor markets permitted an adjustment of wages in
response to the increase in the relative demand for skills; in Europe,
wage rigidities left the skill premium constant, leading, instead, to
an increase in unskilled unemployment.
As was rst pointed out by Nickell and Bell (1995, 1996), unem-
ployment rates were much higher in the 1980s than in the 1970s for
both skilled and unskilled workers. This increase in unemployment
took place in both the North American and the European economies.
Neither the demand-based explanations nor the hypothesis of an
increase in the supply of unskilled labor are capable of accounting
for the simultaneous shift in relative wages and the increase in
unemployment for both types of workers. An exception is the search
model developed by Acemoglu (1999). Yet, in his setup, the mecha-
nism that triggers the changes is an acceleration of the rate of skill-
biased technical change. By using a supply-side approach based on
an efciency-wage model we are capable of providing a framework
in which productivity, relative wages, and unemployment can move
in a way consistent with the evidence.
10.3 The Model
10.3.1 Features of the Economy
10.3.1.1 Workers
Time is continuous and denoted by t. There are H skilled and L
unskilled workers, and Ei t and Ui t, i H; L denote the number of
workers employed and unemployed, respectively. This means
H EH t UH t and L EL t UL t. All workers have identical
preferences, are risk-neutral, and their intertemporal utility function
is time-additive.7 This implies that the real rate of interest is given
by the rate of time preference, r, which is common to all consumers.
We assume that agents consume all their labor income, wi t, as they
receive it. They also decide whether or not to exert effort when
employed. The instantaneous utility function when employed is
wi t eTi t, where eTi t is the disutility of effort and e can take
values of either 0 or 1. Ti t is an index of the level of technology
which is specic to each type of labor since, as we will see later,
skilled and unskilled workers operate different technologies.
We assume that technological progress is the only way in which
workers are separated from rms in equilibrium. There is a proba-
bility hi that, following a technological innovation which destroys
her job, a worker immediately nds a job elsewhere. This assump-
tion captures the observation of Davis and Haltiwanger (1992) that
job-to-job reallocation represents a substantial fraction of worker
turnover. In what follows we assume hH > hL , that is, that the rate of
job-to-job reallocation is larger for skilled than for unskilled workers,
reecting the fact that the former have more transferable skills mak-
ing adoption of new technologies easier.
The return to a worker from being employed and not shirking,
denoted by ViN t, is dened by the following asset equation,
where ViU t is the value of being unemployed. The interest rate, r,
times asset value ViN t must equal the ow benets
rViN t wi t eTi t bi tViU t ViN t V_ iN t 1

of being an employed nonshirker, which consists of the real wage
wi t, the disutility of effort, eTi t, and capital gains/losses. The rate
of worker dislocation, bi t, is endogenous and as we will see below
results from the fact that technological progress destroys jobs. With
probability bi t the worker suffers the capital loss associated with
moving from employment to unemployment, ViU t ViN t. The
term V_ iN t captures the capital gains/losses arising from changes in
wages due to productivity growth and the dynamics of employment
adjustment.
The value of being an employed shirker, denoted by ViS t, follows
a similar recursive equation,
rViS t wi t bi t si ViU t ViS t V_ iS t; 2
where the probability of entering unemployment is increased by si ,

the probability of being found shirking. This probability is specic to
each category of worker and, in line with the literature on worker
monitoring, we assume sL > sH .
Last, the value of being unemployed is
rViU t zTi t ai tViN t ViU t V_ iU t: 3
zTi t is unemployment benet and ai t the probability of exiting
unemployment. Since in equilibrium no worker shirks, the only way
the worker can reenter employment is if an innovation creates new
jobs.
10.3.1.2 Production
We assume that we are in a small open economy, where the prices of
the N varieties of good are exogenously determined in world mar-
kets. A particular variety is produced by one type of labor only. Let
nH be the number of varieties produced by skilled workers and nL

the number produced by unskilled workers, with N nL nH . Sup-
posing that all unskilled-produced goods have the same price, PL ,
and that all skill-produced varieties have price PH 1, we can write
aggregate output as
nL nH
Y PL QL j dj QH nL j dj: 4
0 0
The production of nal goods takes place according to a Cobb-

Douglas technology in which only labor is used,
QL j AxLa j AxLa ; 5a
QH j AxHa j AxHa ; 5b
where 0 < a < 1, A is a scale parameter, and x j is employment

in the production of good j. Firms maximize prots by equating
the marginal product of labor to the real wage, which implies
the inverse labor demand functions wi aAPi =xi1a . We dene the
index of technical progress as Ti ni1a , and let oi 1 wi =Ti denote the
productivity-adjusted wage. We can then express the demand func-
tions as
aAPi
oi ; 6a
ni x i 1a
aAPi
; 6b
Ei1a
where Ei ni x i is total employment of type i workers.
10.3.1.3 Technical Change

Technical change is exogenous and takes the form of expanding
variety. The rate of growth of unskilled-produced and skilled-
produced varieties are, respectively,
n_ L n_ H
gL and gH : 7
nL nH
We say that (1) technical change is neutral whenever gH gL g,

and (2) technical change is skill-biased (deskilling) whenever
gH > gL gH < gL . To understand these denitions, recall the relative
demand for labor. From equation (6b) we have

EHt wLt =PLt 1=1a nH0 gH gL t
e ; 8
ELt wHt nL0
where ni0 is the initial number of type i varieties. When the rates of
growth of the two types of varieties are the same, the two labor
demand functions shift proportionally, leaving the relative demand
for skills unchanged. This is what we term skill-neutral technical
change. A faster (slower) rate of growth of skill-produced varieties
implies that the relative demand for skills increases (decreases) over
time, namely, results in skill-biased (deskilling) technical change.
10.3.2 The Incentive Effects of Technological Progress
10.3.2.1 Labor Reallocation

A key feature of technological progress is that new jobs are created
as old jobs are destroyed. To understand how these effects work in
our model, consider the labor demand functions (6a). The number of
workers used to produce a given variety depends on the number of
varieties of intermediate goods and on the equilibrium wage. Log-
differentiating equation (6a) and using (7), we obtain

1 o_ i
x_ i x i gi : 9
1 a oi
The number of jobs lost in a given variety in a unit time interval, x_ i ,
is proportional to the number of jobs that existed with a coefcient
determined by the rate of increase in real wages and by the rate of
technological progress. If all workers who are separated from rms
could not nd jobs elsewhere, x_ i would be equivalent to the num-
ber of individuals becoming unemployed in a given variety. How-
ever, recall that we have assumed that a fraction hi of workers who
are separated from rms are immediately recruited by a new rm.
Therefore, the number of workers joining the unemployment pool
from a given variety is 1 hi x_ i , and the probability of a given
worker becoming unemployed is bi 1 hi x_ i =x i . We then have

1 o_ i
bi 1 hi gi
1 a oi
!
E_ i
1 hi gi : 10
Ei
The total number of workers of type i becoming unemployed

during a unit time interval is ni bi x i . The number of unemployed
workers who nd jobs is ai Ui . Therefore, changes in employment are
E_ i ai Ui ni x i bi , which upon rearrangement gives
hi E_ i 1 hi gi Ei
ai : 11
Ui
10.3.2.2 Incentive Effects

We can now examine the impact of technological progress on
workers effort incentives and its effect on the wage-employment
trade-off. Firms ensure that workers do not shirk by setting
ViN ViS , which using equations (1) and (2) can be solved for
productivity-adjusted wages oi . Let viU 1 ViU =Ti and viN 1 ViN =Ti
be the productivity-adjusted values of being unemployed and
employed respectively. The individual no-shirking condition (NSC)
can then be expressed as

e gi E_ i =Ei
oi r 1 agi viU e 1 hi e v_ iN ; 12
si si
where viU is to be determined.

Using (1) and (2), the condition ViN ViS can also be expressed as
vi viU e=si . Differentiating, we have v_ iN v_ iU , which together
N
with equations (3) to (11) implies

e hi E_ i 1 hi gi Ei
z v_ iN
si Ui
viU : 13
r 1 agi
These two equations together determine the combinations of

wages and unemployment that ensure that workers do not shirk.
Before obtaining the equilibrium NSC it is worth examining in detail
the incentive effects of technical change. Equation (12) gives the
combinations of oi and Ei that prevent shirking for a given viU , and
shows how they are affected by technical change. On the one hand,
technological progress results in increased returns to employment,
implying that workers lose more if they are found to be shirking.
It therefore tends to strengthen the disciplinary effect of unem-
ployment, allowing rms to reduce the wage for a given level of
employment. We call this the employment capitalization effect of pro-
ductivity growth.8
On the other, there is what we call a job destruction effect. Recall

that bi 1 hi gi E_ i =Ei is the probability of a worker becom-
ing unemployed, and its inverse, 1=bi , is the average duration of
employment. As gi increases, employment duration falls, weakening
the disciplinary effect of unemployment. Firms are required to raise
oi in order to extract effort from workers. The strength of the job
destruction effect depends on the extent of job-to-job reallocation. If
the latter is high, the expected duration of employment is long, and
the impact of job destruction weakens.
Technical change also affects the employment-wage trade-off
through viU , as the greater the value of being unemployed, the higher
the wage needed to prevent shirking. From equation (13), a higher gi
reduces the effective discount rate at which the unemployed capital-
ize future benets and makes unemployment a more attractive
option, tending to raise oi . We call this the unemployment capitaliza-
tion effect. The last effect operates through the job acquisition rate ai .
Its inverse 1=ai is the average duration of unemployment. As gi rises,
duration falls and the disciplinary effect of unemployment weakens.
This is termed the job creation effect. Note that as more jobs are cre-
ated, real wages rise.
Equations (12) and (13) together yield the equilibrium NSC

si 1 hi i
oi z 1 a gi r s i
e i Ei
E_ i Ei : 14
hi
1 i
i Ei
In steady state, where E_ i 0, this condition reduces to

e 1 hi
oi z e r 1 agi gi : 15
si 1 Ei =i
Firms need to use a combination of higher wages and unemploy-

ment to prevent shirking. The steady state NSC thus implies an
upward-sloping relationship between the wage and the level of
employment. The four incentive effects we discussed earlier com-
bine into two competing tendencies. First, the term r 1 agi
is the effective discount rate and captures the employment and
unemployment capitalization effects. Since the ow benets from
unemployment are necessarily less than the ow benets from
employment, the employment capitalization effect dominates, and
Figure 10.1
Equilibrium in the labour market
a higher gi reduces the effective discount rate and hence the no-
shirking wage. Second, the job destruction and job creation effects
are also combined in a single term capturing the probabilities of
entering and exiting unemployment, as ai bi 1 hi gi =1 Ei =i.
Through these probabilities faster technical change increases job
turnover, implying that workers have less incentive to avoid shirk-
ing and rms need to increase the wage. Either of these two net
effects may dominate.
10.3.3 Equilibrium and Comparative Statics
The equilibrium wage and employment level are given by the inter-
section of the demand function with the steady state NSC, which are,
respectively,
aAPi
oi ; DD
Ei1a

e e 1 hi
oi z e r 1 a gi : NSC
si si 1 Ei =i
As depicted in gure 10.1, the demand function is monotonically

decreasing and the NSC monotonically increasing, yielding a unique
equilibrium, Ei ; oi . Note from equation (14) that whenever the
wage is greater than oi , then E_ i > 0, while for oi < oi , E_ i < 0. This
implies that the equilibrium is stable, with rms moving along the
demand function until the equilibrium is reached.
We can now examine the effect of technological change on the
equilibrium. Differentiating the steady state NSC with respect to gi ,
we have
Figure 10.2
The impact of a productivity slowdown
8
> 0 for Ei > Eî
doi < h a
0 for Ei Eî ; where Eî i i < i: 16
dgi : 1a
< 0 for Ei < Eî
Whether a change in g increases or decreases the productivity-

adjusted wage depends on the equilibrium level of employment
relative to a critical value, Eî . Figure 10.2 illustrates the case of a
reduction in the rate of technological change. A lower gi pivots the
NSC curve around Eî , from the solid to the dotted curve. If employ-
ment is initially above Eî , then the wage falls and employment
rises, while if employment is initially below Eî , oi increases and Ei
decreases. The intuition for this result is simple. For high levels
of employment, the duration of any unemployment is likely to be
small. A productivity slowdown then reduces job turnover, weakens
the incentives to shirk, and allows rms to cut the real wage. When
employment is low, the net capitalization effect dominates, and
wages rise following a decrease in gi .
Note that the job-to-job reallocation rate plays a crucial role in
shaping the relationship between technological progress and wages,
as it determines the threshold level of employment Eî . For a given
level of employment, the larger hi is, the weaker the job creation
destruction effect is, and hence the more likely it is that an increase
in the rate of technical change reduces the equilibrium wage.
The rest of the comparative statics are straightforward. A higher
unemployment benet shifts the NSC upward, increasing the wage
and reducing employment; while an increase in either the probabil-
ity of being caught shirking, the rate of job-to-job reallocation, or the
supply of labor tend to reduce the wage and raise employment. A
lower Pi shifts the demand function leftward, resulting in a lower

equilibrium wage and level of employment.
10.4 Relative Wages
We are now in a position to examine the effect of technical change on

the skill premium. Let Wt 1 wHt =wLt be the relative wage at time t.
Using the demand functions (6), the skill premium can be expressed
as

nH0 1a ELt 1a 1 1agH gL t
Wt e : 17
nL0 EHt PLt
In the absence of incentive effects or rigidities, the levels of employ-

ment are simply equal to the supplies of the two types of labor.
Equation (17) then encompasses the three hypotheses that have been
put forward to explain the recent increase in the skill premium: the
relative supply of skills; the effect of international trade, captured by
a change in the relative price of unskilled-produced goods PLt ; and
skill-biased technical change, as reected in any difference in the rate
of innovation of the two types of goods, gH gL .
Introducing incentive considerations implies that wages will
depart from their market-clearing levels, and adds a further mecha-
nism through which technological progress can affect relative wages.
There are two important ways in which the supply-side effect differs
from the demand-side impact of technical change. First, in contrast
to the existing literature, an increase in the skill premium can be
consistent with a reduction in the rate of technological change. Sec-
ond, as we will see in what follows, technical change may increase
the skill premium even if it is skill-neutral.
10.4.1 Biased Technical Change and the Productivity Slowdown
That skill-biased technical change increases the relative wage in our

model will come as no surprise. Still, it is worth examining how the
supply-side effects interact with the standard demand-side impact.
Because of the importance of the productivity slowdown during
the 1980s, let us consider the effect of a fall in the rate of technologi-
cal change. Suppose, more precisely, that we start from a situation
of neutral technical change, with gL gH , and technical progress
becomes skill-biased through a reduction in gL to gL0 , while gH re-

mains constant. As a result of this slowdown in technical progress,
the economy experiences a productivity slowdown.
Now consider what happens to wages. The skilled labor market
remains unchanged, employment remains constant and the real
wages of skilled workers keep growing at rate 1 agH . In the
unskilled labor market, the NSC curve pivots. As we saw before, two
situations are possible. If the initial level of employment is above
(below) the threshold E^L , then the productivity adjusted wage oL
falls (rises) and, as a result, the skill premium, oH =oL , rises (falls).
This creates the possibility that a similar productivity slow-
down has resulted in a larger increase in the skill-premium in
the United States and the United Kingdom relative to continental
Europe because E^L is higher in Europe than in the United States/
United Kingdom, ceteris paribus. A recent study by Boeri (1999),
nds evidence that there is more job-to-job reallocation in Europe
than in the United States (which suggests that hL , and therefore E^L , is
indeed higher in Europe than the United States/United Kingdom).
His argument is that, as a result of tighter labor protection regulation
in Europe, there is an intermediate labor market status between
employment and unemployment. In other words, workers that are
about to be red can remain in their jobs for a period of time, which
will give them the chance of nding a new position, thus moving
from one job to another without a spell of unemployment in be-
tween. In the United States, weak employment protection implies
that workers enter unemployment as soon as they are given notice of
termination.
Note that for EL > E^L , the real unskilled wage, wLt
oL nL0 1a e1agL t , may actually fall when gL falls.9 If the fall in gL is
large enough, then the reduction in the productivity adjusted wage
could, for a period of time, offset the effect of improving productivity
and we would simultaneously observe an increase in the skill pre-
mium, a productivity slowdown, and a reduction in the real wages
of the unskilled.
10.4.2 Skill-Neutral Technical Change
Now suppose that the number of the two types of varieties increase
at the same rate, gH gL g, and that there is a reduction in the rate
of technical change. What would be the impact on the relative wage?
The rst thing to note is that there is no demand effect as the

demand for both types of workers shifts proportionally. The only
impact stems from the impact of a lower g on the effective supply,
and hence on productivity-adjusted wages.
Consider the ratio of the productivity-adjusted wages, oH =oL .
Differentiating, we have

doH =oL 1 doH oH doL
; 18
dg oL dg oL dg
where
doi 1 aEi Eî

:
dg si i Ei 1 hi g iEi

e 1 a oi i Ei
Equation (18) implies that skill-neutral technical change can affect

inequality by having a different impact on the effort incentives of the
two types of workers. As we saw in section 10.4.1, slower techno-
logical progress may increase or decrease productivity-adjusted wages
depending on whether the net capitalization or the job creation-
destruction effect dominates. Which effect dominates depends cru-
cially on the value of Eî which, in turn, depends upon, hi .
In order to look at possible patterns of wage inequality, we cali-
brate the model and obtain numerical examples. We choose the fol-
lowing parameter values:
Preferences: r 0:04, e 1 Labor market: L 1, H 1

Technology: a 0:6, A 4:1 sH 0:02, sL 0:2
Prices: PH 1, PL 0:66 z 0, hL 0:75,
hH 0:99
The values of r and a are standard, corresponding to a rate of time

preference of 4 percent and labor income share of 60 percent. The
cost of effort and the scale parameter A have been arbitrarily chosen.
The price of the skill-produced good is used as a numeraire, and it is
assumed to be about 50 percent higher than that of the unskilled
good in world markets. There are no unemployment benets. The
probability of a shirker being caught is assumed to be ten times as
large for unskilled than for skilled workers.
The choice of job-to-job reallocation parameters is not obvious,

as evidence is sparse. Evidence on transfers following job destruc-
tion suggests that in Germany 32 percent of all separations result
in reemployment within one week, and in Canada 53 percent of
workers were in a new job within three weeks.10 Because we are
using annual values in the calibrations, the corresponding rates of
job-to-job reallocation should be much higher. We can obtain an
indirect estimate from the evidence presented by Davis and Halti-
wanger (1992). They nd that, in the United States, total worker
reallocation in a yearnamely, the proportion of workers that
change employers or transit from employment to joblessness during
a yearwas 36.8 percent over the period 19721986. We can then
use unemployment rates to estimate the proportion of those sepa-
rated from an employer who have another job within a year. Our
estimates give a value of hH between 0.944 and 0.994, and of hL
between 0.77 and 0.977. In our benchmark calibration we use the
values hH 0:99 and hL 0:75, representing the greatest difference
between the two categories of workers implied by these estimates.11
Values for the benchmark economy are depicted in the rst three
columns of table 10.1. We consider the effect of a reduction in the
rate of productivity growth from 5 percent to 1 percent. For the
benchmark economy, a reduction in g increases oH and reduces oL .
A high value of hH implies that the job creation destruction effect
almost disappears in the skilled labor market. The net capitaliza-
tion effect then leads to a higher equilibrium skilled wage. In the
unskilled labor market, low rates of job-to-job reallocation imply
that the job creation-destruction effect dominates, resulting in a
lower unskilled wage. That is, oH increases and oL falls, leading to
a higher-skill premium accompanied by a reduction in the real
unskilled wage.
In the next two columns, we raise the rate of unskilled job-to-job
reallocation, hL , from 0.75 to 0.8. This signicantly reduces the level
of inequality at all levels of technical progress. This conrms the idea
that labor market policies in Europe which raise the rate of job-to-job
reallocation for unskilled workers may help explain the smaller
increases in inequality observed in Europe.
Finally, we reduce the rate of shirker detection for unskilled
workers, sL , from 0.2 to 0.1. As can be seen in the nal two columns
of table 10.1, this also serves to reduce the skilled wage premium
at all rates of technical progress. The intuition is straightforward.
Table 10.1
Inequality and the productivity slowdown
hL 0:75 and hL 0:8 and hL 0:75 and
sL 0:2 sL 0:2 sL 0:1
oH oL oH =oL oL oH =oL oL oH =oL
g 0:05 2.67 1.70 1.57 1.68 1.49 1.79 1.40
g 0:03 2.73 1.67 1.63 1.66 1.54 1.74 1.47
g 0:01 2.88 1.64 1.75 1.64 1.72 1.68 1.68
A high sL implies a weaker trade-off between wages and unem-

ployment (i.e., a atter NSC curve), and consequently the fall in g
results in a small wage change. Again, to the extent that employers
operating in European labor markets are constrained by employ-
ment protection legislation from ring shirking workers, this may
account from the different trends in the skill premium in Europe rel-
ative to the United States and the United Kingdom.
10.5 The Wage-Employment Puzzle
As we have already argued, one of the problems of existing explan-

ations of the increase in relative wages is that they have difculty
accounting for the increases in unemployment rates for both skilled
and unskilled workers during the 1980s. In contrast, our framework
can generate a simultaneous increase in the wage premium and in
unemployment for both types of workers. There are in fact several
circumstances under which this may happen. One possibility is that
the changes in employment and wages are only due to a technologi-
cal slowdown. Suppose that we are in a situation in which EH < E^H
and EL < E^L . A reduction in the rate of technical change will pivot
the NSC in the two labor markets, leading to an increase in the
productivity-adjusted wages and a reduction of employment levels
in both markets. If the skilled wage increases more sharply than the
unskilled wage, then the skill premium and unemployment rates
will increase together.
However, we saw in section 10.2 that there is strong evidence
supporting the role of increased competition in the market for low-
skill manufactures from newly industrializing countries in explain-
ing the increase in the skill premium. In this section we argue that a
reduction in the rate of skill-neutral technological progress, together
with this effect can simultaneously generate the observed changes in

employment and relative wages.
10.5.1 Replicating the U.S. Data
Consider the following scenario. Suppose that hH is large and hL is

small, so that we are initially in an equilibrium where EH < E^H and
EL > E^L . A reduction in the rate of technical change will pivot the
NSC in the two labor markets. For skilled workers, the extent of job
creation destruction is relatively small, and the fall in g will have
the effect of increasing productivity-adjusted wages and reducing
employment levels. In the unskilled market, the opposite will hap-
pen, oL falls and EL rises. As we saw in section 10.4.1, this reduction
in the productivity-adjusted wage may well result in a lower real
unskilled wage. Suppose that at the same time there is a fall in PL ,
which will reduce the demand for unskilled workers. As we can see
in gure 10.3, if the demand shift is sufciently large this will result
in a fall in unskilled employment. That is, the combination of the
productivity slowdown and the reduction in the world price of low-
skill manufactures, results in an increase in the skill premium and an
increase in both skilled and unskilled unemployment. The reasons
for the increase in unemployment are, however, different in the two
labor markets. Skilled employment falls because the strength of the
net capitalization effect requires the disciplining of workers through
unemployment, while unskilled unemployment is the direct effect of
a lower price for their output.
To illustrate these effects, we calibrate the model to match U.S.
data. Productivity growth in the United States fell from 3 percent to
Figure 10.3
The unskilled labor market
1 percent between 1970 and 1990. We choose parameters so that such

a reduction in g together with a fall in the price of the unskilled good
replicate that the data on unemployment and the skill premium. We
x the following parameters: a, r, z, e, PH , and sL , and let the data
determine the rest. This yields
Preferences: r 0:04, e 1 Labor market: L 1, H 1,

Technology: a 0:6, A 3:0387 sH 0:045
Prices: PH 1, PL 0:660 sL 0:7, z 0,
hH 0:995
hL 0:74
Table 10.2 reports the results. The rst row replicates the situation
in 1970, with a skill premium of 1.49 and low rates of unemployment
of 1.7 and 5.27 for skilled and unskilled workers, respectively.12 The
second row considers the effect on wages and employment of a fall
in the price of the unskilled-produced good from PL 0:660 to
PL0 0:568. (i.e., by 16%). The next row adds to that a fall in g. For
the unskilled, a low reallocation rate implies that the productivity
slowdown partly offsets the increase in unskilled unemployment,
while it reinforces the increase in the skill premium. For the skilled, a
lower g results in a higher wage and unemployment rate. A demand
shift due to increased import penetration together with a fall in the
rate of (skill-neutral) technical progress can thus account for the
simultaneous increase in the skill premium and the unemployment
rates for both educated and noneducated workers.
10.5.2 Unemployment Benets and Job-to-Job Reallocation
In this section we argue that labor market policies can indeed help
explain differences in inequality between Europe and the United
Table 10.2
Simulating unemployment and wages in the United States
oH 1-EH oL 1-EL oH =oL
g 0:03, PL 0:660 1.836 1.7 1.232 5.27 1.49

g 0:03, PL 0:568 1.836 1.7 1.109 15.85 1.65
g 0:01, PL 0:568 1.844 2.8 1.085 11.00 1.70
Table 10.3
Changes in unemployment benets and job-to-job reallocation
Labor market Technical change
parameters and prices oH 1-EH oL 1-EL oH =oL
z 0:05 g 0:03, PL 0:660 1.839 2.2 1.240 7.3 1.49
hL 0:7435 g 0:01, PL 0:568 1.870 6.2 1.121 18.1 1.67
z 0:12 g 0:03, PL 0:660 1.849 3.42 1.260 10.9 1.47
hL 0:7435 g 0:01, PL 0:568 1.930 13.3 1.180 28.5 1.63
z 0:12 g 0:03, PL 0:660 1.849 3.42 1.233 5.9 1.50
hL 0:90 g 0:01, PL 0:568 1.930 13.3 1.175 27.1 1.63
States, although we need the more complex depiction of the labor

market used in this chapter. A common argument is that a major
difference between the U.S. and European labor markets is the level
of unemployment benets. In the late 1980s, benet-replacement
ratios were 12 percent in the United States, 19 percent in the United
Kingdom, and 28 percent in both Germany and Sweden (OECD
1994). In our model, benet-replacement ratios, captured by the
parameter z, play an important role in determining the position
of the NSC curve relative to the demand curve, and hence the
magnitude of the impact of changes in other parameters. Table 10.3
considers an economy identical to that in table 10.2, except that
unemployment benet is now greater. In the rst two rows, we con-
sider a situation where z 0:05 (rather than 0, as in our simulated
U.S. economy), and we examine what happens to wages and
employment as the rate of productivity growth falls from 3 percent
to 1 percent, and the price of unskilled-produced goods drops from
0.660 to 0.568. The initial relative wage is still 1.49. However, fol-
lowing the shocks, there is a smaller increase in inequality (to 1.67
rather than 1.70), and a greater increase in unemployment for both
types of workers. Rows 3 and 4 consider an economy with an even
higher replacement ratio, z 0:12, and show that following the
shocks there is an even smaller increase in the skill premium (11%
compared to 14% in the simulated U.S. economy) and a greater
increase in unemployment rates.
To explore the possibility that a package of European labor market
policies may account for the smaller increase the wage premium in
Europe, the last two rows of table 10.3 consider an economy with a
higher benet replacement ratio, z 0:12, and a higher job-to-job
reallocation rate for unskilled workers, hL 0:90,13 than our cali-
brated U.S. economy. This change implies that in response to the

technology and price shock there is an even weaker increase in the
relative wage and an even greater increase in unemployment than in
the previous case. If we compare this economy with the one in table
10.2, the differences are large: the skill premium increases by 9 per-
cent rather than 14 percent, while skilled and unskilled unemploy-
ment rates rise by 9.9 percent and 21 percent, respectively, rather
than by 1.1 percent and 5.73 percent.
The intuition for the effect of changes in either z or h is the same.
The higher the unemployment benet or the greater the probability
of being immediately reallocated to a new job, the weaker the disci-
plinary effects of unemployment. Consequently, large changes in
employment are required in order to maintain workers on the NSC
curve after a change in the rate of productivity growth.
10.6 Summary and Further Policy Implications
In contrast to most of the literature linking inequality and technical

progress, this chapter shifts the emphasis to the supply side, arguing
that incorporating job ows arising from technical progress into a
model of efciency wages allows us to model the impact of technical
change on workers effort incentives. The impact of technical change
on wages is ambiguous and depends crucially on two competing
factorstechnical progress not only increases expected labor income
through what we call the employment capitalization effect (and
thereby reduces incentives to shirk), it also introduces the possibility
that workers will become unemployed as a result of technical even if
they do not shirk (thus increasing incentives to shirk through the job
destruction-creation effect). To the extent that technical change is
more likely to result in unemployment for unskilled workers (as
hH > hL ), the productivity slowdown reduces their incentives to
shirk, allowing rms to cut unskilled wages without compromising
workers effort levels. This is the case even when technical progress
is skill-neutral.
Our analysis suggests that labor market policies which affect
the costs of unemployment and the ease of job-to-job transfer can
account for a different response in Europe and the United States/
United Kingdom to a similar productivity slowdown. The role of hi ,
the rate of job-to-job reallocation, has been shown to be crucial.
Consistent with the study by Boeri (1999), we have argued that
European employment protection legislation, by implicitly allowing

workers time to search for a new job when their existing job is lost
due to technical change, implies that unskilled workers in Europe
are less likely to enter unemployment than their American counter-
parts, even if unskilled unemployment rates are higher in Europe.
Additionally, higher replacement ratios in Europe further reduce
the expected costs of unemployment for the unskilled. As a result
unskilled wages in Europe have to be maintained to prevent shirking
behavior. In the United States, workers are more sensitive to changes
in labor turnover due to technical change, hence rms can use
the productivity slowdown as an opportunity to reduce unskilled
wages.
It is also possible to interpret the rate of job-to-job transfer, h, as a
parameter capturing the relative effectiveness of job search of the
employed relative to the unemployed since, in steady state, it mea-
sures the proportion of new jobs lled directly by workers leaving a
job rather than exiting unemployment. Seen in this light, policies
which raise the job search skills of the unemployed, may actually
raise inequality by reducing h.
Any attempt to reduce unemployment by improving the job
search skills of the unemployed, will make existing workers more
vulnerable to competition from the unemployed and allow rms to
reduce wages, ceteris paribus. Since the unemployment rates are
higher for unskilled workers, this will have a disproportionate affect
on the unskilled. However, policies that seek to reduce inequality by
slowing the rate of labor turnover will not only increase unemploy-
ment, but by reducing new job opportunities to the unemployed are
also likely to increase the duration of unemployment.
This highlights the fact that policies that tend to reduce inequality
have a cost in terms of unemployment: High job-to-job reallocation
rates (due to job protection legislation and high replacement ratios)
result in less inequality at the expense of greater changes in employ-
ment. The argument that there is a trade-off between inequality and
unemployment, and that differences in labor market regulations
imply high inequality and low unemployment in the United States
and low inequality and high unemployment in Europe, is not new.
An essential feature of our approach is that lower inequality is often
obtained at the expense of higher rates of unemployment for both
skilled and unskilled workers, rather than reducing only unskilled
employment (see table 10.3). This has important implications for the
policymaker and may affect the chances of generating support for

particular policy measures.
Notes
We would like to thank participants at the CESifo conferences on Growth and

Inequality: Issues and Policy Implications for helpful comments on an earlier draft of
this paper, which was originally published in discussion paper form as Leith and Li
(2001). We would also like to thank Theo Eicher, Raji Jataraman, Nobuhiro Kiyotaki,
and seminar participants at the Universities of Stirling, Otaru and Hokkaido, the
Kansai Macroeconomics Workshop, and the Bank of Japan for their comments.
1. See Acemoglu (2000).
2. This hypothesis has been theoretically explored by Eicher (1996), Galor and Tsid-
don (1997), Greenwood and Yorukoglu (1997), Acemoglu (1998), and Caselli (1999),
among others.
3. The importance of this process has been documented by Davis and Haltiwanger
(1992). They nd that in the United States, between one-third and one-half of total
worker reallocation (between employers or from employment to joblessness) is due to
shifts in employment opportunities across rms.
4. See Murphy and Welch (1992) and Berman, Bound, and Griliches (1994).
5. See OECD (2001).

6. A number of papers have attempted to reconcile an acceleration technical progress
with the observed productivity slowdown, arguing that it takes time to implement
new technologies (see Greenwood and Yorukogklu 1997; Aghion and Howitt 1998,
chap. 8; and Galor and Tsiddon 1997, respectively). Yet, productivity has fallen over a
twenty-year period, which seems too long as an experimentation period.
7. The assumption of risk neutrality is made for reasons of tractability. We conjecture
that introducing risk aversion would raise the costs of unemployment such that the
job ows induced by technological change would have an even greater impact on
inequality through the channels we describe here.
8. It is analogous to what Aghion and Howitt (1994) call the capitalization effect of
growth on labor demand, which increases the return of creating a new job and makes
it protable for rms to hire more workers.
9. A fall in the real wage of the lowest decile of the earnings distribution was
observed in the United States during the 1980s (OECD 1993).
10. See OECD (1996).

11. During this period the unemployment rates for high-education and low-education
workers were 2 percent and 7.8 percent, respectively (Nickell and Bell 1995). The ow
into unemployment of type i workers can be expressed as fi 0:368 Ei 1 hi ,
assuming the same reallocation rate for skilled and unskilled workers. The stock of
unemployment is then Ui fi =ai , where ai is the probability of exiting unemployment
in a year. Assuming ai lies between 1 and 0.1 (i.e., between 10% and 100% of workers
nd a job within a year), we get the preceding values.
12. We use observations for the average unemployment rates over the period 1971
1974 from Nickell and Bell (1995).
13. A higher rate of job-to-job reallocation in Europe than in the United States is con-
sistent with the evidence in Boeri (1999).
References
Acemoglu, D. 1998. Why Do New Technologies Complement Skills? Directed Tech-

nical Change and Wage Inequality. Quarterly Journal of Economics 113: 10551090.
Acemoglu, D. 1999. Changes in Unemployment and Wage Inequality: An Alternative
Theory and Some Evidence. American Economic Review 89: 12591278.
Acemoglu, D. 2000. Technical Change, Inequality, and the Labour Market. NBER
Working Paper No. 7800.
Aghion, P., and P. Howitt. 1994. Growth and Unemployment. Review of Economic
Studies 61: 477494.
Aghion, P., and P. Howitt. 1998. Endogenous Growth Theory. Cambridge: The MIT
Press.
Berman, E., J. Bound, and Z. Griliches. 1994. Changes in the Demand for Skilled
Labor within U.S. Manufacturing Industries: Evidence from the Annual Survey of
Manufactures. Quarterly Journal of Economics 109: 367365.
Boeri, T. 1999. Enforcement of Employment Security Regulations, On-the-Job Search
and Unemployment Duration. European Economic Review 43: 6589.
Card, D., and T. Lemieux. 2001. Can Falling Supply Explain the Rising Return to
College for Younger Men? A Cohort-Based Analysis. Quarterly Journal of Economics
116: 705746.
Caselli, Francesco. 1999. Technological Revolutions. American Economic Review 89:
78102.
Davis, S. J., and J. Haltiwanger. 1992. Gross Job Creation, Gross Job Destruction and
Employment Reallocation. Quarterly Journal of Economics 107: 819863.
Eicher, T. S. 1996. Interactions between endogenous human capital and technological
change. Review of Economic Studiess 63: 127144.
Eicher, T. S. 1999. Training, Adeverse Selection and Appropriate Technology. Journal
of Economic Dynamics and Control 23: 272246.
Eicher, T. S., and P. G. Kalaitzidakis. 1997. The Human Capital Dymension to Foreign
Direct Investment: Training, Adverse Selection, and Firm Location. In Trade, Dynam-
ics, and Growth, ed. B. Jensen and K. Y. Wong, 337364. Ann Arbor: Michigan Univer-
sity Press.
Feenstra, R. C., and G. H. Hanson. 1999. The Impact of Outsourcing and High-
technology Capital on Wages: Estimates for the United States, 19791990. Quarterly
Journal of Economics 114(3): 907940.
Galor, Oded, and Daniel Tsiddon. 1997. Technological Progress, Mobility, and Eco-
nomic Growth. American Economic Review 87: 363382.
Greenwood, Jeremy, and Mehmet Yorukoglu. 1997. 1974. Carnegie-Rochester Confer-

ence Series on Public Policy 46: 4995.
Leith, C., and C.-W. Li. 2001. Wage Inequality and the Effort Incentive Effects of
Technological Progress. CESifo Working Paper No. 513.
Murphy, K. M., and K. Welch. 1992. The Structure of Wages. Quarterly Journal of
Economics 107: 285326.
Nelson, R., and E. Phelps. 1966. Investments in Humanns, Technological Diffusion,
and Econmic Growth. American Economic Review 56: 6975.
Nickell, Stephen, and Brian Bell. 1995. The Collapse in the Demand for the Unskilled
and Unemployment across OECD Countries. Oxford Review of Economic Policy 11:
4062.
Nickell, Stephen, and Brian Bell. 1996. Changes in the Distribution of Wages and
Unemployment in OECD Countries. The American Economic Review 86(2): 302309.
OECD. 1993. Employment Outlook. Paris, July.

Shapiro, C., and J. E. Stiglitz. 1984. Equilibrium Unemployment as a Worker Disci-
pline Device. American Economic Review 74: 433444.
Index
Acemoglu, D., 297 Benhabib, Jess, 257

Aghion, Philippe, 28, 121, 122, 144, 145, Berman, E., 293
156, 202, 257, 295 Bertola, Giuseppe, 28
Agricultural sector employment, 163 Birdsall, N., 156
164, 261265 Blanchard, Olivier, 121
Ahituv, Avner, 62 Boeri, T., 313
Ahluwalia, M. S., 159 Bolton, Patrick, 145
Albania, 122 Botswana, 279
Alesina, Alberto, 28, 145, 156, 201, 202, Bound, J., 293
255, 256, 257 Bourguignon, Francois, 30, 168, 259
Algeria, 23t Boyd, R. L., 162
Allocation of time, 261265 Brandolini, Andrea, 31, 179, 202
Anand, S., 159 Brazil, 6, 10, 14, 23t, 256257
Angrist, J. D., 63 Bruno, M., 168
Anti-globalization, 2728 Budget constraints on child rearing, 6768
Arends, M., 63 Bulgaria, 161t, 189t
Armenia, 160t, 164, 188t Burnhill, P., 92
Atkinson, Anthony B., 31, 92, 101, 132,
179, 202, 212 Canada, 111, 213, 286, 308
Attacking Poverty, World Development Card, D., 63, 296
Report 2000/2001, 3 Cardak, B. A., 93
Auerbach, A., 228, 229 Carey, D., 240
Australia, 286 Caroli, Eve, 28, 144, 202
Azariadis, C., 91 Caselli, F., 212
Azerbaijan, 160t, 164, 188t Central African Republic, 276
Chen, Shaohua, 3, 6, 30, 47
Baltagi, B. H., 169 Children. See also Fertility
Banerjee, Abhijit V., 28, 29, 145, 256, 274, budget constraints on raising, 6768
283 capital accumulation effect of raising,
Bangladesh, 23t 7274
Barro, Robert J., 29, 170, 201, 212, 233, and the formation of human capital, 66
239, 255, 257, 274 67, 7072, 9395
Becker, G. S., 63, 64, 67, 91, 93 individual preferences regarding
Belarus, 160t, 188t raising, 66
Belgium, 38 optimization of capital transfers to, 68
Belka, Marek, 122 69
Bell, Brian, 297 parental involvement in education of,
Benabou, Roland, 28, 31, 145, 283 9397
320 Index
Chile, 23t and equilibrium between capital and

China labor, 97100, 234237
growth spells in, 23t and evolution of education, 6972
inequality of income in, 3538, 268 and fertility rates, 7482
per capita income growth in, 3435 in the former Soviet Union countries
population, 35, 36t, 37 (FSU), 157159
poverty in, 3641 and human capital, 6674, 9193
Chiu, W. H., 202 and income distribution, 814, 2831,
Cohen, D., 64 100106, 257258
Colombia, 23t and inequality of income, 3134, 9293,
Commander, Simon, 122, 156 122125, 142146, 155159, 255259,
Consumer price index (CPI), 126, 165 272274
Conventional Consensus View (CCV) in and ination, 162
growth and income inequality, 201 long-run, 97100
Corneo, G., 92 measurement across countries, 3841
Costa Rica, 23t model, 6574
Cote dIvoire, 6, 14, 23t and natural resource abundance, 286
Cowell, Frank, 46, 212 and natural resource dependence, 259
Croatia, 161t, 165, 189t 267
Cummins, J., 228 policies, 2122
Cumulative Liberalization Index, 166, 176 and population growth, 233
Czech Republic, 122, 161t, 189t poverty reduction through, 37, 814
and production, 6566, 9091, 9597,
Datt, G., 6, 8, 10 104106, 112, 242246
Davis, S. J., 297, 308 and redistribution, 201203, 257258
Deininger, Klaus, 14, 28, 2930, 38, 55, relationship to taxation, 227229
155, 156, 159, 165, 168, 170, 201, 202, research on, 2731
212, 270, 283 taxation effect on, 237238
De Janvry, A., 6 in transitional countries, 157164
De Melo, M., 166 in two-sector nonscale economies, 232
Denizer, C., 166 233
Denmark, 257 Ecuador, 23t
Dewatripont, Mathias, 121, 124, 145 Education. See also Human capital
Dollar, David, 6, 22, 30 and adult literacy, 109110
Dominican Republic, 23t costs of, 9395, 102103
Drazen, A., 91 dropout rates, 217219
Dreze, Jean, 31 effect on economic growth, 9293, 203,
Dropout rates, schooling, 217219 212219, 257258
Duo, Esther, 28, 29, 256, 274, 283 effect on inequality, 278288
Duryea, S., 63 effect on production, 112, 203205, 267
evolution of, 6972
Easterly, William, 156, 202, 237, 256, 258 and gender, 6465, 269270, 276, 278,
Eckstein, Z., 91, 94 279288
Economic growth high-skill versus low-skill individuals
causality of, 2831, 4142 and, 203205
Conventional Consensus View (CCV) home, 8991, 9397, 104
on, 201 majority voting effect on, 107109
in Eastern Europe (EE), 157159 and market growth equilibrium, 206
effect of education on, 6972, 9293, 208
203, 257258 and natural resources, 265267
empirical evidence linking poverty and poverty, 6365, 7274, 258
reduction and, 1520 private, 258
Index 321
and private-sector enterprises, 205206 closed economy model of homogenous

public, 8991, 9397, 102103, 258, 279 individuals and, 6574
281 diversity in levels of, 61
rates of return, 6365 and economic growth, 7482
relationship to fertility levels, 6163, and education rates of return, 6365
7274, 7879 and the formation of human capital, 66
steady states of, 7274, 234237 67
taxation and, 9395, 102103, 204211, and income levels, 6162
212213 and optimization of capital transfers to
technological change effect on, 104106, children, 6869
110111 public policies on, 8283
and wages, 205208 rates in specic countries, 7482
Egypt, 23t, 269 relationship to educational levels, 61
Eicher, Theo S., 91, 92, 228, 237, 241242, 63, 7274, 7879
295 Finland, 269
Eissa, N., 229 Fischer, R. D., 91, 92
Elasticity of poverty, growth, 1120 Forbes, Kristin J., 29, 93, 168, 201, 255,
El Salvador, 23t 283, 285
Employment. See also Human capital; Foster, J., 14
Wages France, 111, 213, 296
agricultural sector, 163164, 261265
capitalization effect, 301302 Galor, Oded, 28, 67, 91, 93, 144, 202, 257,
equilibrium levels, 303305 270
high-skilled versus low-skilled, 203 Garca-Penalosa, Cecilia, 28, 144, 202
211, 293315 Garner, C., 92
incentive effects of technological change Garner, T. I., 156
on, 300303 Gelb, A., 166
industrial sector, 163164, 166, 261265 Gender
market equilibrium in, 206208, 265 and education, 6465, 269270, 276, 278,
267 279288
private-sector, 205206 inequality and natural capital share,
steady states of, 303305 282288
unemployment, 162, 165, 173175, 294 Georgia, 160t, 164, 179, 188t
295, 297298, 302, 311313 Germany, 213, 268, 296, 308, 312
Engel, E., 129 Ghana, 23t
Esquivel, G., 212 Gini coefcients
Estonia, 23t, 160t, 188t applied to China and India, 30, 31, 33,
Ethiopia, 23t, 61 3538, 42, 46, 4950
Europe, Eastern (EE). See Transitional applied to Poland, 122124, 128, 129
countries 134, 140142
European Bank for Reconstruction and applied to transitional economies, 164
Development (EBRD), 122, 125, 146, 165, 167176
166 and high-skilled versus low-skilled
Evans, David S., 144 labor, 208211
and human capital formation, 106, 111
Feenstra, R. C., 296 in measuring inequality, 267271
Felder, S., 228 and natural resources, 282288
Fernandez, R., 91, 93 and poverty reduction, 1520
Ferreira, F. H. G., 156, 163, 175 and redistribution of income, 208211,
Fertility. See also Children 256257
and budget constraints for raising sensitivity analysis of, 177185
children, 6768 Glaeser, E. L., 92
322 Index
Glomm, G., 91, 93 estimating individual distributions of,

Gomulka, Stanislaw, 140 4446
Gradstein, M., 164, 176 and fertility levels, 6162, 7582
Granger, C. W. J., 168 and gender, 269270
Granger causality tests, 168 and Gini coefcients, 1520, 30, 31, 33,
Greene, W. H., 168, 169 3538, 42, 46, 4950, 99, 106, 111, 122
Greer, J., 14 134, 140142, 164185, 208211, 256
Griliches, Z., 293 257, 267271, 282288
Guatemala, 23t growth and poverty reduction, 37, 8
Guinea-Bissau, 276 14
Gustafsson, B., 162, 164 growth spells in specic countries, 22
Gylfason, Thorvaldur, 260, 267, 278, 282, 24, 2831, 157159
286 high-skilled versus low-skilled, 208
211, 293297, 308311
Haiti, 279 household, 122124, 126129, 164
Haltiwanger, J., 297, 308 human capital accumulation effect on,
Hansen, Lars Peter, 45 9293, 9597, 101106
Hanson, G. H., 296 industrial-sector, 163164, 166, 261265
Hassett, K., 228, 229 ination effect on, 162, 173, 179
Heston, Alan, 30, 35, 55, 212 and land distribution, 270
Honduras, 23t, 38 lognormal distribution and, 5153
Howitt, P., 295 in pensioner-headed households, 135
Hubbard, R. G., 228 139, 147
Human capital. See also Education; in Poland, 122142, 157159
Employment and political freedom, 167, 176, 186187
accumulation and fertility, 6674 probability models, 3134
and allocation of time, 261265 research, 2731, 9193, 203, 211212
effect of taxation on, 239246 in Slovenia, 157159
effect on income, 9293, 9597, 101106 and social transfers, 132, 136149
formation, 9397 targeting of transfers of, 136139, 147
high-skilled versus low-skilled, 203211 148
and long-run economic growth and unemployment, 162, 165
equilibrium, 97100 Index of Political Freedom (IPF), 167
and public education, 104106 India, 6, 10, 14, 23t
research, 9193, 211219 inequality of income in, 3538
technological change that affects, 104 population, 35, 36t, 37
106, 110111 poverty in, 3641
in two-sector nonscale economies, 229 Indonesia, 23t, 279
237 Industrial-sector employment, 163164,
Hungary, 122, 142144, 161t, 189t 166, 261265
Huppi, M., 6 Inequality, income. See also Income
and the agricultural sector, 163164
Income. See also Inequality, income; causality of, 2831, 4142, 167170,
Redistribution, income; Wages 185187
agricultural, 163164, 261265 in China and India, 3538
and budget constraints on child rearing, Conventional Consensus View (CCV)
6768 on, 201
in China, 3435 cross-country measurements, 3841,
distribution, 6, 814, 2122, 2834, 43 111, 132134, 142144, 155157, 164
55, 100106, 144146, 155157, 167 170, 256257, 267272
170 and economic growth, 3538, 9293,
and elasticity of poverty, 1520 142146, 255259, 272274
Index 323
education effect on, 278288 Lahiri, Kajal, 283

xed-effects model of, 168169 Lam, D., 63
and gender, 269270 Land distribution, 270
Granger causality test for, 168 Latvia, 160t, 188t
between high-skilled and low-skilled Lavy, V., 63
labor, 208211, 293297, 308311 Lee, J.-W., 212
and the industrial sector, 163164 Lefort, F., 212
measuring, 267271 Lemieux, T., 296
and natural resource dependence, 259 Lesotho, 23t
267 Levy, F., 164
New Challenge View (NCV) on, 201202 Lewis, H. G., 63, 67
in Poland, 124125, 129142, 157159 Li, Hongyi, 29, 38, 170, 201
and political freedom, 167, 186187 Liberalization, economic, 166, 176
potential determinants of rising, 159 Litcheld, J., 175
164 Literacy, 109110
public policy impact on, 237238 Lithuania, 160t, 188t
random-effects model of, 169 Living standards, 78, 30, 61
relationship to per capital GDP, 170185 Lognormal distribution and income, 51
research data on, 2731, 156157, 164 53
167 Lorenz curve (LC), 208209
and the saving rate, 257 Loury, G., 91, 93
in Slovenia, 157159 Lucas, R., 91
taxation effect on, 228, 237238, 246247 Luxembourg Income Study (LIS), 203,
in transitional countries, 157164 212
in the United States, 111, 132, 212213
Ination, 162, 173, 179 Macedonia, 161t, 165, 189t
International Monetary Fund, 140 Madagascar, 23t, 276
International trade and wages, 296 Malaysia, 23t
Italy, 61, 75 Mali, 276
Mankiw, N. Gregory, 240
Jamaica, 23t Manski, Charles F., 45
Japan, 273 Mauritania, 23t
Jeanne, O., 92 McPherson, A., 92
Johansson, M., 164 Meier, T., 173
Jones, C. I., 227, 237 Mendoza, E. G., 240
Jordan, 23t, 279 Metcalf, C., 173
Jovanovic, Boyan, 91, 144 Mexico, 23t
Milanovic, Branko, 30, 47, 49, 134, 156,
Kakwani, N., 6, 8 162, 163, 164, 173, 176
Kalaitzidakis, P. G., 295 Moav, Omer, 62, 65, 91, 270
Kaldor, Nicholas, 257 Moldova, 160t, 164, 188t
Kanbur, S. M. R., 159 Morocco, 23t
Karni, E., 112115 Morrison, C., 168, 259
Kazakhstan, 23t, 160t, 189t Murnane, R. J., 164
Keane, Michael P., 126, 129, 135 Murphy, H. G., 63, 64
Kenya, 23t Myanmar, 279
Korea, 31, 75
Kraay, Art, 6, 22, 30 Namibia, 279
Krajewski, Stefan, 122 Natural resources
Krueger, A. B., 63 abundance and economic growth, 286
Kuznets, Simon, 28, 144, 159, 201 and allocation of time of individuals,
Kyrgyz Republic, 23t, 160t, 189t 261265
324 Index
Natural resources (cont.) overall inequality in, 129134, 161t

cross-country patterns of inequality and, pensioner-headed households in, 135
271272 139, 147
dependence and economic growth, 259 policy reform effects in, 135136, 144
267 149
and education, 265267 price controls in, 121
and gender inequality, 282288 private enterprises in, 125, 145146,
measuring dependence on, 274278 148
and measuring income inequality, 267 privatization of state-owned enterprises
271 in, 122
Nelson, R. R., 63, 296 social transfers in, 132, 136149
Nepal, 23t targeting of transfers in, 136139, 147
New Challenge View (NCV) on growth 148
and income inequality, 201202 transition period inequality in, 129134
Newman, Andrew F., 145 unemployment in, 122
New Zealand, 111 Policy, public
Nickell, Stephen, 297 impact on inequality, 237238
Niger, 276 natural resource dependence and, 288
Nigeria, 23t, 279 289
Nyarko, Y., 91 poverty reduction, 37, 2122, 156
privatization, 144149
Olinto, Pedro, 270 public education, 109112
Optimization of capital transfers, 6869 related to fertility rates, 8283
taxation, 239246
Pakistan, 23t technological change, 313315
Paldam, Martin, 266 in transitional countries, 162164
Palmer, E., 162 Political freedom, 167, 176, 186187
Panama, 24t Population growth and production
Paraguay, 24t elasticity, 233
Pareto distribution and income, 5355 Poverty
Paukert, F., 159 based on standards of living, 78
Penn World Table, 212 changes over time, 67
Pensioner-headed households, 135139, in China, 3641
147 dening, 78
Perotti, Roberto, 28, 142, 145, 201, 202, and distribution of income, 67, 814,
255, 256, 257 2122, 156157
Perroni, C., 228 and education levels, 6365, 7274, 258
Persson, Torsten, 28, 93, 144, 156, 201, empirical evidence linking growth and,
202, 255 1520
Peru, 24t evolution across growth spells, 1520,
Phelps, E., 296 2931
Philippines, 24t, 31 and fertility levels, 6162, 7582
Physical capital, 239246, 260261, 282. gap index, 48
See also Production growth elasticity of, 1120
Pinto, Brian, 122 headcounts, 1520, 4748
Poland, 177 incomes of those in, 30
Household Budget Surveys (HBS) of, in India, 3641
122124, 126129, 136139 reduction strategies, 37, 2122, 3942
income surveys in, 122124, 126129 Prasad, Eswar S., 126, 129, 135
industrial sector in, 163 Price controls, 121
inequality in income in, 129144, 157 Private-sector enterprises, 122, 186, 205
159, 189t 206
Index 325
Probability models for income Senegal, 14, 24t

distribution dynamics, 3134 Serra, P. J., 91, 92
Production Shaked, M., 112113
economic growth and, 6566 Shanthikumar, J. G., 113
education effect on, 112, 203205, 267 Sierra Leone, 276
equilibrium in, 9597 Silverman, Bernard W., 49
and fertility rates, 9091 Slovak Republic, 122, 161t, 163, 189t
natural resource dependence and, 261 Slovenia, 122, 142144, 161t, 165, 189t
265 income inequality and growth in, 157
and population growth, 233 159
taxation effect on, 242246 Smeeding, Timothy M., 132
technological change effect on, 104106, Social transfers, 132, 136149
298299, 305306, 310311 Soviet Union countries, former (FSU). See
and total factor productivity (TFP) Transitional countries
growth, 296 Spain, 75
Psacharopoulos, G., 63 Squire, Lyn, 14, 28, 2930, 38, 55, 155,
156, 159, 165, 168, 170, 201, 202, 212,
Quah, Danny, 28, 30, 38, 41, 64, 170 283
Sri Lanka, 24t
Rainwater, Lee, 132 St. Lucia, 279
Ravallion, Martin, 3, 6, 8, 10, 21, 30, 47, Stack, S., 162
159, 168 Standards of living, 78, 30, 61
Ravikumar, B., 91, 93 State-owned enterprises (SOEs), 122
Razin, A., 240 Stokey, N. L., 237
Reagan, Ronald, 240241, 243 Sudan, 279
Rebelo, Sergio, 202, 237, 256, 258 Summers, Robert, 30, 35, 55, 212
Redistribution, income. See also Income Sweden, 213, 257, 273, 312
and education level, 204219 Sylwester, K., 156
effects on growth, 201203, 257258
from high-skilled to low-skilled labor, Tabellini, Guido, 28, 93, 144, 156, 201,
208211 202, 255
in the United States, 212213 Tajikistan, 160t, 164, 189t
Rodrik, Dani, 28, 145, 156, 164, 201, 202, Tamura, R. F., 63, 64, 91, 92
255, 256 Taxation
Rogerson, R., 91, 93 and costs of human capital
Roland, Gerard, 121, 124, 145 accumulation, 9395
Romania, 24t, 161t, 189t and education levels, 204219
Romer, David, 239, 240, 242 effect on inequality, 228, 237238, 246
Romer, Paul M., 206, 261 247
Rosenzweig, M. R., 67 effect on production, 9597, 242246
Ross, D., 156 and ination, 162
Russia, 24t, 160t, 164, 179, 188t and long-run economic growth, 97100
Rustichini, Aldo, 257 majority voting effect on, 107109
Rwanda, 61 and public education, 102103, 109112
and redistribution of income, 202, 212
Sabot, R., 156 213
Sachs, Jeffrey D., 259 relationship to economic growth, 227
Sadoulet, E., 6 229
Saint-Paul, Gilles, 202, 256, 258 research, 228229
Sala-i-Martin, Xavier, 30, 202, 233, 239 in transitional countries, 162
Saving rates, 257 in two-sector nonscale economies, 229
Schmidt, Peter, 283 231
Sen, Amartya, 31 in the United States, 228, 239246
326 Index
Tchilinguirian, H., 240 Turnovsky, Stephen J., 228, 237, 241242

Technological change, 104106, 110111 Two-sector nonscale economy model
biased, 305306 aggregate economy behavior in, 232
and demand for labor, 299300 balanced growth equilibrium in, 232
effect on production, 298299, 305306, 233
310311 equilibrium dynamics of, 234237
effect on wages, 293297, 305 government expenditures in, 230
effect on workers, 297298 impact of public policy on inequality in,
incentive effects of, 300303 237238
labor reallocation from, 300302, 303 individual agents in, 229231
305, 308, 311313, 314 taxation effect in, 237238
public policy, 313315
skill-biased, 293295, 305306 Uganda, 24t
skill-neutral, 306309 Ukraine, 24t, 160t, 175, 188t
and unemployment, 294295, 297298, Unemployment, 162, 165, 173175
311313, 314 benets and job-to-job reallocation, 311
Temple, J., 170, 212 313
Terrell, K., 156 capitalization effect, 302
Tesar, L. L., 240 and technological change, 294295, 297
Thailand, 24t 298, 314
Thorbecke, E., 14 United Kingdom, 111, 132, 213, 268, 273
Thurow, L., 173 technological change and wage
Todaro, Michael P., 257 inequality in, 293315
Tomes, N., 91, 93 unemployment benets in, 311313
Total factor productivity (TFP) growth, United Nations Statistical Yearbook, 74
296 United States, the
Transitional countries. See also Poland entrepreneurship in, 144
causality of income inequality in, 185 fertility levels, 61
187 income inequality in, 111, 132, 212213,
economic liberalization in, 166, 176 268, 293
government consumption in, 166, 175 natural resources of, 286
government policies in, 162164, 186 tax code changes, 228, 239246
income inequality and GDP growth in, technical change and wage inequality
157159, 164176 in, 293315
inequality measurements, 164167, 177 unemployment benets in, 311313
185 UNU (United Nations University), 55,
ination in, 162, 173, 179 164
models and estimations, 167170 Uzbekistan, 160t, 189t
political freedom in, 167, 176, 186187
potential determinants of rising Valenzuela, J., 63
inequality in, 159164 Van Marrewijk, C., 91
privatization in, 144149, 166, 175176, Venezuela, 24t
186 Verdier, Thierry, 202, 256, 258
regression results for models, 170176 Viaene, J.-M., 91, 93, 110
relationship between GDP and income Voting, majority, 107109
inequality in, 170176
unemployment in, 162, 165, 173175 Wages
Transition Report, 122, 146 and biased technological change, 305
Trinidad and Tobago, 24t 306
Tunisia, 24t equilibrium levels, 303305
Turkey, 24t explanations for increases in, 295297
Turkmenistan, 160t, 189t and international trade, 296
Index 327
of skilled workers, 203211, 293297,

308311
and skill-neutral technological change,
306309
of unskilled workers, 308311
Walde, K., 219
Warner, Andrew M., 259
Weil, David M., 67, 240
West Germany, 61
Wolpin, K. I., 67
Women
education of, 6465, 269270, 276, 278,
279288
fertility of, 6165, 7482
unemployed, 162
World Bank, 4, 47, 132, 140, 156, 164167,
179, 268, 269, 272273, 276
World Development Indicators (WDI),
165
World Income Inequality Database
(WIID), 164
Yemen, 24t
Yemtsov, R., 156
Ying, Y., 164, 176
Zambia, 14, 24t, 276

Zeira, Joseph, 28, 93, 144, 202, 257
Zilcha, I., 91, 93, 94, 110, 112115
Zoega, Gyl, 260, 267, 278, 282, 286
Zou, Hengfu, 29, 38, 170, 201

Inequality and Growth-Theory and Policy Implications-Theo S. Eicher, Stephen J. Turnovsky

Uploaded by

Copyright:

Available Formats

Inequality and Growth-Theory and Policy Implications-Theo S. Eicher, Stephen J. Turnovsky

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Inequality and Growth-Theory and Policy Implications-Theo S. Eicher, Stephen J. Turnovsky

Uploaded by

Copyright:

Available Formats

Inequality and Growth

The CESifo Seminar Series

Inequality and Growth: Theory and Policy Implications, ed. Theo S.

Theo S. Eicher and

The MIT Press

Library of Congress Cataloging-in-Publication Data

Series Foreword vii

I Measuring the Impact of Growth on Inequality 1

1 The Growth Elasticity of Poverty Reduction: Explaining

2 One Third of the Worlds Growth and Inequality 27

II Population, Education, and Inequality 59

3 Fertility Clubs and Economic Growth 61

4 Human Capital Formation, Income Inequality, and

III European Transition and Inequality 119

5 Social Transfers and Inequality during the Polish

6 Growth and Inequality: Evidence from Transitional Economies

IV The Political Economy of Inequality 199

7 (Re-)Distribution of Personal Incomes, Education, and

8 The Impact of Tax Policy on Inequality and Growth: An

V Technology and Natural Resources and Inequality 253

9 Inequality and Economic Growth: Do Natural Resources

10 Wage Inequality and the Effort Incentive Effects of Technical

This book is part of the CESifo Seminar Series in Economic Policy,

Economists and policymakers alike are inherently interested in the

empirical, and policy issues involved. The ultimate goal is to provide

authors provide both theory and empirical evidence to support the

this venture possible and worthwhile. Finally we would like to

Part of the ongoing debate on poverty reduction strategies bears on

the availability of fully comparable household surveys at the two

The second source of heterogeneity is of course the change in the

in each country or region, it may seem cumbersome. Instead, this

1.2 The Arithmetic of Distributional and Poverty Changes

Any poverty index may be seen as a statistics dened on all indi-

countries, for instance, the familiar $1 or $2 a day after correction for

To show the contribution of growth to the change in poverty, it is

intermediate step, which is the horizontal translation of the initial

This expression is the direct application in the case of the head-

The distribution effect is more difcult to translate in terms of elas-

where P is the cumulative distribution function of the standard

where l stands for the ratio of the density to the cumulative

In that expression, the growth elasticity of poverty appears ex-

whereas the elasticity with respect to the standard deviation of the

where P and p are respectively the cumulative distribution and

1.3 Revisiting the Empirical Evidence on the Growth-Poverty

The identity linking growth, poverty reduction, and distributional

nal model relies on the lognormal approximation discussed earlier.

derived in the preceding section under the assumption that the

pothesis that cannot be tested independently of the lognormality is

(dependent variable percentage change in poverty gap during growth spell)

1.4 Some Implications

actual determinants of the theoretical value of the growth-poverty

4. Figure 1.2 provides a very handy general representation of distributional changes

Three concerns underlie all research on income inequality and eco-

This chapter is part of a body of research that argues that the

2.2 Related Literature

A conventional wisdom recently emerged from empirical research

from reducing inequality through redistribution. Deininger and

Atkinson and Brandolini (2001) or Dreze and Sen (1995). My calcu-

2.3 Probability Models for Income Distribution Dynamics

At xed t, dene the estimator y^t for yt as