You are on page 1of 40

POLICY RESEARCH WORKING PAPER 1563

Equity and Growth There is no intrinsic tradeoff


between long-run aggregate
in Developing Countries economic
growth andoverall
equity. Policies aimed at
Old and New Perspectives helpingthe pooraccumulate
productive assets- especially
on the PolicyIssues policiesto improveschooling,
health, and nutrition when
Michael Bruno adoptedin a relatively
Martin Ravallion nondistortedframework,are
Lyn Squire importantinstrumentsfor
achieving higher growth.

The World Bank


Office of the Vice President
Development Economics
and
Policy Research Department
Poverty and Human Resources Division
and Office of the Director
January 1996
POLICYRESEARCHWORKINGPAPER1563

Summary findings
The "stylized fact" that distribution must get worse with below $1 a day moved in opposite directions.
economic growth in poor countries before it can get The gains to poor people from a distribution-neutral
better turns out not to be a fact at all. Growth's effects growth process will tend to be lower, the higher the
on inequality can go either way and are contingent on extent of initial inequality. A smaller share of total
several other factors. income must imply a smaller absolute gain from a given
Bruno, Ravallion, and Squire found no sign in the new increment to total income. Compensatory direct
cross-country data they assembled that growth has any interventions can be important, provided they are
systematic impact on inequality. Possibly measurement integrated into a framework of fiscal and monetary
errors confound the true relationship, but they think it discipline.
more likely that the relationship between growth and The evidence does not suggest that growth is always
distribution is not as simple as some theories have held. distribution-neutral, and it would be wrong to conclude
Since distribution does not worsen, growth reduces that changes in distribution are of little consequence. The
absolute poverty. Indeed, absolute poverty measures point is not that distribution is irrelevant or that it never
typically respond quite elastically to growth, and the changes, but that its changes are roughly uncorrelated
benefits are certainly not confined to those near typical with economic growth.
poverty lines. Trhere is no intrinsic tradeoff between long-run
Of course, one cannot say that growth always benefits aggregate efficiency and overall equity. Policies aimed at
the poor or that none of the poor lose from pro-growth helping the poor accumulate productive assets-
policy reform Only aggregate effects are studied. But for especially policies to improve schooling, health, and
17 of the 20 countries for which they assemble quite nutrition - when adopted in a relatively nondistorted
good data (from at least two surveys since the mid- framework. are important instruments for achieving
1980s), the mean and the proportion of people living higher growth.

This paper-a product of the Office of the Vice President, Development Economics, and the Poverty and Human Resources
Division and Office of the Director, Policy Research Department - was prepared for the IMF Conference on Income
Distribution and Sustainable Growth, June 1-2, 1995. Copies of this paper are available free from the World Bank, 1818
H Street NW, Washington, DC 20433. Please contact Patricia Sader, room N8-040, telephone 202-473-3902, fax 202-
522-1153, Internet address mravallion(a@worldbank.org. January 1996. (30 pages)

The PolicyResearchWorkingPaperSeriesdisseminatesthe findingsof work in progressto encouragethe exchangeof ideasabout


arelessthanfully polished.The
developmentissues.An objectiveof theseriesis toget the findingsout quickly,evenif thepresentations
paperscarrythe names of the authorsand should be used and citedaccordingly.The findings, and conclusionsarcthe
interpretations,
authors'own and shouldnot he attributedto the WorldBank,its ExecutiveBoardof Directors,or any of its membercountries.

Produced by the Policy Research Dissemination Center


EQUrrY ANDGRowTH IN DEVELOPINGCouNTIEs:

OLD ANDNEW PERSPECTIVESON THEPOLICYISSUES*

MichaelBrmno,Martin Ravallion and Lyn Squire

Prepared for the IMF Conferenceon IncomeDistributionand SusainableGrowth,June 1-2, 1995. For their helpfl
commets, the aitborsare gateftdlto the dissanss, Lou Jiwei, Jacob M. Mwanz, and DaniRodrikl,as wel as other
comrene pardcWas,andto KL Ddninger,Peter LAnjouw,Andr6sSolimmno, Jack vanHlostPelckun, Domique
van de Wafe, HolgerWoaf, id Shklcnw Yii.bai.
Table of Contents

pape

Introduction .. 1.....................

I. How Does Growth Affect Distribution?. 2

1.1 The KuznetsHypothesis.3


1.2 Cross-CountryStudies .................. ; 4
1.3 Further IntertemporalEvidence. 5
1.4 Other Lessonsfrom Tests of the Kuznets Hypothesis. 8
1.. The Impactof Growth on AbsolutePoverty. 9

H. Do Pro-growthReformsHave Adverse DistributionalEffects? ....... . 11

2.1 AdjustmentandTransition .11


2.2 Evidencefrom Three Regions .13

m. How DoesDistributionAffect Growth?.16

3.1 Credit, Distributionand Growth .16


3.2 The Political Economy.18
3.3 The Evidence.20

Iv. Condusions ...... . .....................................21

References .. 26
Introduction
Do the poor lose-either absolutely or relatively-from policies that promote aggregate
economucgrowth? Doesthe answerdifferbetweenmiddle-incomenewlyindustrializedeconomies
and low-incomedevelopingcountries? These questions are not new and were very much at the
center of the developmentdebate some twenty years ago in the discussionof how to achieve
'redistribution with growth' (Chenery et al., 1974). They have recently achieved renewed
prominenceas manycountriesadjustfrom the growthcrises of the last two decades, and as others
switch from centrally-plannedsystems to market-based ones. The claim has been made that
growth-oriented reform policies of the kind usually advocated by the InternationalFinancial
Institutionshave worsenedthe lot of the poor.
The first section of this paper reviews recent evidence indicatingthat while income
inequalitydifferssignificantlyacrosscountries,there is no discernablesystematicimpact over time
of growth on inequality. Though there are exceptions, as a general rule sustainableeconomic
growthbenefitsall layers of societyroughly in proportion to their initiallevels of living. Based
on the evidence of the last three decades, there seems to oe no crediblesupport for the Kuznets
Hypothesis. And there have been few cases of immiserizinggrowth.
In the second section we switch from long-run growth to issues of adjustment and
iransition. Here we argue that the key componentslinking growth, as a necessaryconditionfor
sustainedpoverty reduction, and adjustment(stabilizationplus structuralreform) as a necessary
conditionfor aggregategrowthrepovery,come out strengthenedfrom the recentgrowth crises and
associatedreformefforts. Obviouslynecessityis not sufficiencyand we do not argue that growth
alwtaysbenefitsthe poor, or that none of the poor lose from any pro-growthpolicy reform. But
we do contendthat macroeconomicadjustmentand structural reform are essentialfor sustainable
growth recovery whichin turn is necessaryfor a sustainedreductionin aggregatepoverty.
The firsttwo sectionsof the paper support and strengthenthe case for policies conducive
to broad-basedeconomicgrowthas part of a comprehensivepovertyreductionstrategy, as argued
in the WorldDevelopmentReporton poverty(World Bank, 1990), andthe associatedPolicy Paper
(World Bank,1991)on AssistanceStrategiesto Reduce Poverty. But a macro-policyenviromnent
conducive to growth is not enough. The second part of the poverty reductionstrategy outlined
in World Bank (1990)-namely promotinguniversal access to basic education,health and social
2

infrastructure(as well as the adoptionof social safety nets, particularlyin the process of recovery
from a low-levelgrowthcrisis)-has in recentyears received added supportfrom new research on
the reverse linkagefrom initial distribution of assets and income to subsequentgrowth. In the
third section we review the evidence that high inequality countries, such as a number in Latin
America and Africa, have lower growth and remain inegalitarian, whereas low inequality
countries,suchas manyin East Asia, remainegalitarianand achieverapid povertyreduction from
the process of growth.
The theoreticalunderpinningsof this reverse linkage are only graduallybeing understood.
Somelines of argumentoriginatefrom politicaleconomyconsiderations:concentrationof wealth,
such as in land or humancapital,lead to policiesthat protectsectarianinterestsand impede growth
for the rest of society;inequalitymay also enhance political instability. Anotherargument has to
do with credit-market imperfections, whereby investment in human and physical capital is
confinedto the ownersof initial wealth. The policy implicationis that reducinginequality, such
as through securing wide access to basic education and health, not only benefits the poor
immediatelybut will benefitall through higher growth.
The final sectionof the paper draws out implicationsfor domesticpolicy and the IFIs.
I. How DoesGrowth Affect Distribution?
Recognizingthat we are concernedabouthow the benefits of growth in aggregate incomes
are distributed, the questionarises as to whether there is any systematictendencyfor inequality
to change in theprocessof risingaverageaffluerce.' This is a long standingissue in development
economics. A still widely held view is that economic growth in low-incomecountries will
necessarily be inequitable, and this view has had considerable influence on thinkiDngabout
developmentpolicy (amongstboth advocates and critics of redistributiveinterventions). By this
view "...the rich are usually the first to reap the benefits of national income growth" (Watkins,

There are tumnrousmeasures of inequality that might be considered compelling, and in principle they can diverge
greatly in their assessments of whether distribution has improved. In practice, however, for many of the purposes of
measurement there appears to be considerable congruence amongst a number of these measures. We will rely heavily
here on the most widely used summary statistic on distribution, namely the Gini index. There is also the question:
inequality of what? Here we focus mainly on current income or consumption inequality; both may diverge from other
measures that might be compellingsuch as inequality in life-time utility or inequality in 'capabilities' (on the latIer see
Sen, 1992).
3

1995, p.34). Here we reviewthe theories and evidence, and providenew results on more recent
and improveddata.
1.1 The KlunetgHypntheSis
An influentialargumentas to why we might expect inequitablegrowth in poor countries
was sketched by Kuznets(1955). This claims that inequalitywill increasein the early stages of
growth in a developingcountry and then-after some point-it will begin to fall i.e., the
relationship betweeninequality(on the vertical axis) and average income(horizontal)will trace
out an invertedU. Kunets did not set out a formaltheoryof why thismighthappen, but sketched
an argument which has subsequentlybeen formalized. As typicallypresented, the "Kuznets
Hypothesis"assumesthat the economycomprisesa low-inequalityand low-meanrural sector, and
a richer urban sectorwith higher inequality. Growth occursby rural labor shiftingto the urban
sector, such that a representative
slice of the rural distributionis transformedinto a representative
slice of the urban distnbution.Thus (by assumption)distributionis unchangedwithin each sector.
Startingwith all thepopulationin the mral sector, when the first workermovesto the urban sector
inequality must increase. And when the last rural worker leaves, it must clearly fall again.
Between theseextremes,the relationshipbetween inequalityand averageincome will follow an
inverted U.2
Kuznets himself was tentative about the hypothesis. Yet it has found many supporters
since, to the point of being de6med "fully confirmed" by Oshima(1970), a "stylizedfact" by
Ahluwalia (1976), and an "economic law" by Robinson (1976). Claims of support for the
hypothesiscan be foundin a literaturespanning25 years.3 We shall argue that the evidence from
cross-countrydatasetshas been misleadingbecauseof omittedcountry-leveleffects. New studies
using panel data and within-countrytime-series data do not supportthe hypothesis.

2 See A nd andKanbur(1993)for a more precise formulation,and necessaryand sufficientconditionsfor the


inverted U for six possible inequalitymeasures.

3 ADi_flneia earlyexample was Adelman and Morris (1971). At the time of writing the most recent example
we know of is Ram (1995).
4

1.2 Crnsm-CountryStudies
There have been enumerabletests of the Kuznets Hypothesison cross-countrydata sets,
by regressing a measureof inequalityagainst a suitable functionof averageincome, and seeing
if that function followsan inverted U. We shall not review the earlier literaturehere and only
note that these tests have typically been ad hoc, with no clear link to the assumptionsof the
hypothesis. Instead, we focus on a nagging concern about all the tests using cross-country
data-namely, that theremay be importantcountry-leveldeterminantsof inequality(includingpast
inequality) which are correlated with current income levels, and so lead to biased estimates.
Indeed, suchbiasescouldarisesolely from differences in the type of data. For example, income
is a more common measurefor inequalityin many middle-incomedevelopingcountries, notably
in Latin America,whereasconsumptionis more common elsewhere,includingamong the Asian
economies-manyof whichwerecloser to the bottom of the incomeladder20-30 years ago when
the dataused to test thehypothesiswere set up. And since consumptioninequalityis bound to be
lower than incomeinequalitydue to consumption smoothing,thesedifferencesalone would tend
to yield an inverted U relationshipeven if none existedusing the same welfaremeasure. With
strong latentcountry-leveleffectsthere can be no guarantee that differencesat one point in time
will reveal how inequalitywill evolve with growth.
If suchcountry-leveleffe6tswere not in fact a problem, then one would expect to see the
inverted U reappearingin later country cross-sections. So what do data since the mid-1980s
suggest about the KuznetsHypothesis? Using data from 63 surveysspanning1981-92covering
44 countries, 4 we tried replicating a number of the specifications for testing the hypothesis
typically found in the literature.5 This was done for both levels and changes over time, to
eliminatethe country-levelfixed effect. In no case was there evidenceof an inverted U, and in
no case could one rejectthe null hypothesisthat the regressioncoefficientswere jointly zero. This

4 This is the same data set used in Chen, Datt and Ravallion (1994), which gives details.

5 We tried regressing the Gini index against a quadratic function of mean consumption(both linear and logs) as
well as the Anand and Kanbur (1993) specification in which the Gini is regressed on the mean and the reciprocal of the
mean. We also tried thespecificaionproposed by Ram (1995) in which a quadratic function of the mean is used but with
the interceptsuppressed;while thistest did suggest an inverted U, it appears to have very low power to reject the Kuznets
Hypothesis; indeed, on suppressing the intercept one will find an inverted U between any two independent random
variables with positive means (Ravallion, 1995b).
5

also confirmsearlier resultsfor smaller samples reported by World Bank (1990), Fields (1989),
and Ravallion (1995a).6
It appears then that the cross-country inverted U found in many earlier tests of the
hypothesis-mainly using compilationsof distributional data for the 1950sto early 1970s-may
well have becomeblurred,if notvanished,over time. This probablyreflectshow various omitted
variableshave evolved. Thenew data confurmearlier concerns that theseomittedvariables were
creatingan appearanceof a cross-countryinverted U which had littleto do with the hypothesis.
We would conjecturethat with the growth seen in much of Asia, and the lack of it in much of
Africa, the poor and low inequalitycountriesof 20-30 years ago have slit into two, blurring the
old inverted U but (quite possibly)better revealing the true relationship.
1.3 Further Intertemporal Fvidence
To avoid confusing the effects of independent country-specificcharacteristics(initial
conditions)withthoseof intertemporalchangesof policies or economicconditions,argumentsfor
or against the existenceof a Kuznets process should ideally be based on time-seriesevidence.
Here we report on two exercisesusing time-series data. The first draws on panel data covering
45 developedand developingcountriesfor the years 1947 to 1993. It contains486 observations
on Gini indices.7 And the secondmakes use of the most extensivetime-seriesdata for any single
developing country, namelyIndia.8
Table 1 gives decadeav6ragesof the Gini indices for each of the 45 countriesfor which
reasonably comparable estimates are available for four or more surveys. While there is clearly
variation over time (some of which could be differences between surveys and/or measurement
errors), the data suggest substantiallygreater variation in inequality across countries at given time
than over time for a given country. Indeed 92 percent of the variance in Gini indices by country
and date is accounted for by cross-country variation whereas only 7 percent is accounted for by
variation over time.

6 The latter paper allowsfor fixed country-leveleffects. Fields and Jakubson(1992)findthat the invertedU
"flips" to an ordinaryU whenone allowsfor fixedeffects, but our datado notconfirmthisfinding.

7 See Deininger,SquireandZhang(1995)for further details.

8 See Ravallionand Datt(1995).


6

Table 1: Gini Indices1960s-90s(Decadl Averages?

Country Observations 1960sb 1970s 1980s 1990s Trend'


Czechoslovakia 10 22.6 20.9 21.1
Bulgaria 25 22.1 21.9 23.0 27.3 0
Hungary 7 24.4 22.2 22.8 . 0
Poland 7 . . 25.2 . 0
Spain 6 . . 25.7 . 0
United Kingdom 31 25.0 24.3 27.3 32.4 +
Soviet Union 4 . . 26.0 . +
The Netherlands 9 . 28.1 28.6 . +
Taiwan 26 31.2 29.3 29.0 30.5 0
Finland 6 . 30.7 31.0 . 0
Canada .. 23 31.6 31.6 31.5 27.6 0
India 29 31.5 30.9 31.4 31.1 -
China 12 . . 31.5 36.2 +
New Zealand 11 31.4 34.1 * +
Sweden 14 . 33.1 33.7 32.3 0
Indonesia 7 . 36.6 33.4 33.1 0
Pakistan 6 . 35.5 33.4 . 0
Norway 7 36.8 35.3 31.0
Korea 10 31.5 36.1 35.6 . 0
Japan 22 35.6 34.1 34.4 35.0 -
Italy 15 . 37.4 33.4 32.2 -
Bangladesh 9 33.5 34.8 37.3 . 0
USA 45 34.6 34.5 36.9 37.9 +
Australia 10 32.0 36.7 36.2 32.5 0
Belgium 8 36.4 42.0 29.6 35.8 0
Portugal 4 . 40.6 36.8 36.2 0
Germany, F 6 . 36.0 35.8 45.4 +
CoteD'lvoire 5 . . 39.1 41.4 0
Singapore 6 . 39.0 40.7 . 0
Venezuela 4 . 41.5
Sri Lanka 7 46.0 38.8 43.7 . 0
Tunisia 5 42.3 44.0 43.0 41.0 0
Philippines 6 42.9 45.3 40.0 . 0
Hong Kong 10 47.5 41.9 41.4 45.0 0
France 7 48.0 41.6 37.8
Thailand 8 42.0 41.7 37.8 50.2 +
Baharas 11 . 48.2 44.4 43.0 -
Trinidadand Tobago 4 . 48.5 41.7 . 0
Costa Rica 5 52.6 46.1 45.1 . 0
Malaysia 5 . 51.5 48.0 . 0
Colombia 5 . 52.1 51.2 . 0
Mexico 4 55.3 49.7
Honduras S . . 54.0 52.7 0
Chile 13 . . 54.8 53.1 0
Brazil 7 . 59.0 55.6 . 0

a. The table includesall countrieswithfour or more observations,based on householdsurvey data withnationalcoverage. All
Giniindicesare measuredfor the same indicator (eitherconsumptionor income)over timefor a givencountry,though it varies
betweencountries. This accountsfor some of the cross-countydifferences,thoughon addingdummyvariablesfor the type of
data in a pooled modelone stillfindsthat the bulk of the variation is between countriesrather thn overtime.
b. Rankcorrelationsof inequaliybetweendecades: 1960s-70s:0.909; 1970s-80s:0.863; 1980s-90s:0.849; 1960s-80s: 0.850.
c. The signs indicatethe significanceof the Gini timetrends ("0" indicatesno significanttrend).
7

The inequalityrankingsof countriesare thus highly stable over the decades;betweenthe


1960s and 1980sthe rank correlation coefficient is 0.85 (Table 1). The last column of Table 1
also gives the direction of the trend;9 a "0" in the final column indicatesthat the coefficienton
time is not significantlydifferentfrom zero at the 5 percent level, while a "+ " ("-") indicatesthat
it is significantly positive (negative). Only 17 countries out of 45 have a significanttrend in
inequality one way or another, and in 12 of the cases its value is small(+ or - 0.4 a year).
It is plainfrom Table 1 that there are strong countryeffects in inequalitywhich could well
entail appreciablebiasesin standardtests of the Kuznets Hypothesis.' 0 For example,if (as Table
1 suggests)past inequalityis an important predictor of current inequality,and (as the arguments
and evidencereviewedin part III will suggest) past inequalityinfluencescurrentincomes, then
the standardcross-countryregressionsused to test the inverted U will be biased.
All this lendssupportto the viewthat failureto allow for countryeffectscould be serious.
The search for a general law linking growth and inequalitymust confrontthe fact that the vast
bulk of the variationonefindsis amongstcountries,not over time. Further statisticaltests on the
data set confirmthispoint. If one allowsfor country-specificeffects, none of the countriesin the
samnpleappearto follow the predictions made by Kuznets hypothesis(as mentionedin Deininger
and Squire, 1995).
It is worth reviewingthe data for India in more detail because it is one of the most
extensive and reliable series, and because it bears on subsequentdiscussion. At the time of
writing we couldconstructdistributionsof real householdconsumptionexpendituresper person
in India from 33 nationallyrepresentativeand reasonably,comparablehouseholdsurveysspanning
the period 1951to 1992." Figure 1 plots India's Gini index and net domesticproductper person
from 1951to 1991. There was a trend decreasein inequalityup to about the mid-1960s,but no

9 These are based on ordinary least squares estimates of the coefficient on time.

10 Ihese effects may entaileither an omitted dynamic effect of past inequality or some other omitted country-level
fixed effect in the error term; either will bias standard tests on cross-sections of country data.

i The surveyswere done by India's National Sample Survey Organization. To form the national distributions of
real consumptionfrom the NSO tabulationsof nominal expenditure distributions, an allowance was made for urban-rural
cost-of-livingdifferences, and for differences in the rate of inflation between urban and rural areas; for details see Datt
(1995).
Figure 1: Inequalityand average income in India

40
0
- 0.7 11
to

C , Gini index Log average - 0.6


C
o
,", (left axis) income
0-35 - (right axis) 0.5
E._
a)
35 - . a

. 0.4 U)_
col
0 ~~~~~~~~~~~~~~~
ci

U)
a
E o 0*3.2 E
I~~~~~~~~~~~~~~~~~~
a)

030~~~~~~~~~~~~~~~~~~~~~~~~0 ~~~~~~~~~~~~~~
X
25Ureydt)
0. V
190 15V90 16 97 95 18 95 19
8

trend in either directionafter that. There is no sign that growth increasedinequality, including
during the period of higher growth in the 1980s. On running the Anand-Kanburtest equation
appropriateto the Gini indexone obtains, not an inverted U, but an ordinaryU, though for most
of the rangeof the datainequalityfalls as average income increases. However, if one takes fit
differencesof the aboveequation(so that it is the changein the Gini index betweensurveys which
is regressedon the changein average income and the changein its inverse)then the relationship
vanishes. Thereis no signin thesedatathat higher growth rates in India put any upward pressure
on overall inequality.
1.4 Other LeRsns from Te.ctc of the KXmnets H=ytheqis
The fact that thereare suchstrong country-level effectsin distributiondoes not mean that
distributionis unchangeable.Some of the observed variation(acrosscountriesand over ime) is
clearly due to differencesin the underlying data and measurementerrors. But the literature on
testing the Kuznets Hypothesishas also suggested a number of other factors which appear to
influence inequality, and explain at least some of the omitted country-leveleffects identified
above. Kuznets(1966)speculateson a number of those factors, includingshiftinginter-sectoral
inequalities, a decliningshare of (unequally distributed) property income, and policy changes
concemingsocialsecurityandemployment. But, on all these, the data base for testingwas weak
at the time Kuznetswas writing. That has changed.
Higher (primaryand secondary)school enrolmentrates tend to be associatedwith lower
inequality,and the significanceof the incomevariables tends to diminishwhen education is taken
into account. The quantitativeimportanceof this effect suggeststhat it may be policy-relevant:
A one-percentincreasein the percentage of the labor force that has at least secondaryeducation
increasesthe shareof incomereceived by the bottom 40 percent or 60 percentby between 6 and
15 percent(Bourguignonand Morrisson, 1990). Papanek and Kyn (1986)find that primary and
secondary schoolenrollmenthas a quantitatively importanteffect on the income-sharereceived
by the poorest 40 percent. By contrast, it is significant but of low quantitativeimportancein
reducing inequality(as measuredby the Gini index). Human capital in primary and secondary
education had a significanteffect on reducing the Gini coefficient in Korea, and increasingthe
9

share of the bottom20 percent, whereas university education slightlyincreasedthe Gini and did
not significantlyaffect the bottom share of the income distribution(Jung, 1992).
Mineraland agriculturalexportswould be expectedto increaseinequalityto the degree that
they produce concentratedrents. This is confirmed for developingcountrieswhere a sizable
(greater than 5 percent)contributionof mineralexports to GDP was associatedwith a 4-6 percent
decreases of the bottom 40 percent income share (Bourguignonand Morrisson, 1990). High
importance(greaterthan 5 percentof GDP) of agriculturalexports leadsto greater inequality only
if such exports are producedon large, rather than on small and mediumfarms. By contrast,
Papanek and Kyn (1986), using data for developing as well as developedcountries, fail to find
significant effects, presumablysince failure to correct for protectiondoes not allow inferences
concerning the internationalcompetitivenessof such exports.
Trade theorywouldpredictthat protectionlowers the rewardfor the most abundant(most
equallydistributed)factor of productionand increases returns to scarcefactors, which are likely
to be the more inequitablydistributed. Presence of protectionindeedseemsto worsen income
distribution (Bourguignonand Morrisson, 1990).
There is evidencefor Indiathat the sectoralcompositionof growthhas playeda role in the
evolution of distribution. Recall that at the aggregate level, the data revealed little affect of
growthon inequality.Whichof the Kuznets'assumptionsdo not holdfor India? At any one date,
both mean consumptionandinequalityare higher in urban areas as he postulated. But the radical
departure from the assumptionsof Kuznets is in the nature of India's growthprocess. Growth
under the KuzmetsHypothesisis driven by rural to urban migration,assumingthat the means and
distributionsremainthe samewithin each sector. However, Ravallionand Datt (1995) find that
this process has been only a minor source of growth in India, the bulk of whichhas come from
intra-sectoralgrowth;between 1970 and 1990, the Kuznets growthprocessaccountedfor only 6
percent of total consumptiongrowth, while growth within the urban and mrualsectors accounted
for 20 percent and 74 percentrespectively.
1.5 The Tmpactnf G.rnwthon Ahsohite Poverty
The still quite widely-heldpessimism about the scope for reducing poverty through
economic growthhas restedin large part on the belief that growthwould be inequitablein poor
10

countries. We have surveyed past and new evidence on this view, and rejected it as a
generalization;yes, therehave been cases in which growth was associatedwith rising inequality,
but there have been at leastas many cases of falling inequality. There does not appear to be any
systematictendencyfor distnbutionto improveor worsen with growth. On averagethen absolute
poverty will fall. This is confirmedby the results of a number of recent studies(Fields, 1989;
World Bank, 1990, 1995; Squire, 1993; Ravallion, 1995a; Lipton and Ravallion, 1995).
How responsiveis poverty to economic growth? Regressingthe rates of change in the
proportion of the populationliving on less than $1 per day againstthe rate of change in the real
4.

value of the survey mean for the 20 countries spanning 1984-93 we obtained a regression
coefficient of -2.12 (with a t-ratio of -4.67); thus a 10 percent increase in the mean can be
expectedto resultin roughlya 20 percent drop in the proportionof peopleliving on less than $1
per day.' 2 This reflectsin large part the density of people living around$1 per day. But if we
also consider "higher-order"measures of poverty the effect is even stronger; for the squared
povertygap indexproposedby Foster et al., (1984) the correspondingelasticityis even higher at
-3.46 (t=-2.98). '3 This indicatesthat the gains are not confined to those near the poverty line.
These results confirm thoseof Ravallion (1995a) on a smaller data set.
Somewhatsmallerelasticities,but broadly similar results, are obtainedif we look at the
evolution of povertyover 40 years in India. Over 33 household surveys,the elasticity of the
proportion of the populationbelow India's official poverty line and mean consumptionis -1.33
(t=15.19). For the squaredpoverty gap, the elasticity is -2.26 (t=10.22).'4
But growth is only one of the factors that has influencedprogressin reducing poverty,
albeit an importantone. The aboveregressions for rates of povertyreductionstill leave a sizable

12 is pardyspurious,sinceboth the surveymeanandthe povertyindexwere


It mightbe arguedtdatthiscormlation
estimatedfromthe samedata. If insteadwe use an Instrunental VariablesEstimator,usingthe growthrate in GDP per
capitabetweenthe surveydatesas theinstrumentthenwe get a very similarresult, namelyan elasticityof -2.15 (t-ratio=-
3.24). Since the nationalaccountsand census are largely independentof the householdsurveysour estirate of the
elasticityappearsto be robust.

3 The correspondingInstrmental VariablesEstimateis 4.11 (t=-2.36).

14 Usingtherate ofgrowthin consumption per personfromthe nationalaccountsas an instrument,the Instrumental


variablesEstimatesare -1.47(t=6.51) for the head-countindex and -2.51 (t=4.50) for the squaredpoverty gap.
11

share of the variance in country performance unaccounted for by growth. Some of this is
measurement error. But, measured changes in inequality do have a strong independent
explanatory power; indeed, rates of poverty reduction respond even moreelasticallyto rates of
change in the Gini index than they do to the mean. Regressingthe change in the log of the
proportionof the populationliving on less than $1 per day on the changein the log of the survey
mean and the change in the log of the Gini index across 20 countries with two reasonably
comparableobservationsin the period 1984-92one obtains an elasticityto the mean of -2.28 (t=-
6.07) while the elasticityto the Gini is 3.86 (t=3.20).'5 So even seeminglymodest changes in
overall inequalitycan entail sizable changes in the incidenceof poverty. Whencombined with
the tests of "augmented"KuznetsHypothesesdiscussedabove, we can postulatea numberof other
factors that matter through their influence on inequality, includingeducation,the trade regime,
and the sectoral compositionof growth. Later we will see whether some or all of these factors
might also matter to the poor via their impact on growth.
II. Do Pro-growth Reforms Have Adverse Distributional Effects?
So far we have arguedthat the rate of overall economic growthhas no systematicimpact
on inequality. Yet, it has been argued that some of the policy changesadvocatedto promote
growth increase inequality. For example, real devaluationscan promote growth, but they also
impact on inequality,thoughthe directionof that effect is not obviouson a priori grounds. Here
we look more closelyat therole played by economy-widepolicy changes. In particular, we ask:
Do the economy-widefactors (including macroeconomicpolicy changes)which are likely to
increase the overall rate of economicgrowth also have,distributionalimplications?
2.1 Adju-tmernt and Trangition
For much of thedevelopingworld, the 1980swas a periodof rapidlyrisingservicingcosts
on foreign debt, externalterms-of-tradeshocks, and fiscal and external imbalancesentailing an
unsustainableexcess of aggregate demand over supply. Adjustmentprogramswere introduced
to help restore macroeconomicbalance, by combining fiscal contraction-cutting government

is Ihe elasticityto changesin the Gini indexis evenhigher if one uses a measureof povertywhichbetterreflects
distributionamongstthe poor; usingthe 'squared povertygap' index, the elasticityto the Gini risesto 8.07 (t=2.49),
whilethe elasticityto the meanis -3.79 (t=-3.61).
12

spendingand/or raisingtaxes-with supply-sidemeasures aimed at reducinginefficiency,suchas


by cuttingtradedistortionsor wastefulparastatals. Unless there is an exceptionallyrapid supply-
sideresponse,somebody'sconsumptionmust fall. The distributionof the burden of adjustment
has been one of the most debatedissues in developmentstudies over the last decade. The issue
has been of evengreatersignificancein the centrally planned economiesthat are now privatizing
and placing much greater reliance on market solutions. What can we say about the impactof
adjustmentand transitionon the poor?
Many countrieswerenot well equippedwith relevant householdlevel data for monitoring
..

welfareimpactsof policy reform at the time that adjustmentbegan in the early 1980s. This has
improved since. Yet even with good data, it can be difficult to isolate the role played by
adjustment. Poverty may have risen during an adjustmentperiod; but it may have risen even
fiurtherwithoutadjustment.Much of the criticismof adjustmentpoliciesmay have to do with the
observationof real hardshipsthat are temporarilyincurred at the stabilizationstage, yet would in
all probabilitybe much greater were the crisis allowed to deepenfurther.
One of the few clear patterns to emerge from the new household-levelevidence on the
evolutionof povertyindicatorsduring adjustmentis that the povertymeasurestend to move with
the mean consumptionor income of households, increasingin recessionand falling in recovery
(Lipton and Ravallion, 1995, section5.3).
Whathappensto the rate-f growthin the adjustmentto a deep crisis is thereforeof crucial
importance. In this respect, an important link to likely outcomesfor the poor can be found in
recentfindingsof relativelyspeedygrowthrecovery(in GDP though not in investment)after deep
inflation (and growth) crisis (Bruno and Easterly, 1995). The medianper capitagrowth rate in
a group of 13 successfulstabilizersfrom more than 40 percent inflationshifted from -4 percent
in the years up to and includingthe first year after stabilizationto a positive 1.5 percent in the
secondyear andcloseto 4 percentin the third and beyond. Even when aggregategrowth remains
temporarilynegative as inflationalready falls, it is not at all clear which way the distributional
13

outcomegoes, as incomegroupswhose nominal income is not tied to inflation, or whose income


taxes are withheldat the source, will gain in relative terms as inflationfalls drasticaly. 16
The incomedistributionalimpacts of adjustment depend heavilyon the economy's initial
conditions, includingits openness, and the extent of flexibilityin its output and factor markets,
thus pointing to the importanceof market reforms as an important conditioningenvironment.
Actualexperiencesin distributionalshifts during adjustmenthave been diverse. For example, in
the Philippines adversedistributionaleffects resulted in higher povertydespite(modest) growth
in the late 1980s(Balisacan,1993). A small irnprovementin distributionhelpedthe poor during
adjustment in Indonesiaduring the mid-1980s (Ravallionand Huppi, 1991). Dorosh and Sahn
(1993) argue that the distributionaleffects of real devaluationswill tend to be pro-poor in a
7
number of Africancountries, since the rural poor tend to be net producersof tradable goods.1
The diversity of initial conditionswarns against generalizationson the distributionalimpacts of
adjustment.
A common presumptionis that countries under shock face a dynamictrade-oft; living
standardsmay fall in the shortterm during adjustment (relativeto non-adjustment),but they will
rise in the longer term. However, this trade-off could well be overstated. For example, Peru
initially avoided adjustment,and poverty rose sharply in 1985-90 (Glewwe and Hall, 1994). Yet
the subsequent period of more orthodox reforms quickly saw positive growth and falling poverty
measures in 1991-94 (Favaro and MacIsaac, 1995).
2.2 Fyidence from Three Regions
In this subsection,we review the evidence now available for Sub-Saharan Africa and Latin
America, the two regions most closely associated with adjustment, and then turn to the evidence
for the transitional economies of Eastern Europe and the Former Soviet Union.

16 This, for example, was the case for wage earners as against profit earners in the Israeli stabilization of 1985.
Measurement is complicatedby the fact that inflation during the household survey period will generally put an upward
bias on inequality measures defined on nominal incomes (Kakwani, 1987); conversely, stabilizationwill
impart a downward bias.

1 Lipton and Ravallion (1995) review other recent arguments and evidence on the impacts of adjustment on the
poor.
14

New encouragingresults have been recorded for some countriesin Africa. Demery and
Squire(1995)use householdsurveydata at two points of time in the mid-eightiesto early nineties
to assess the change in poverty in the six African countries for whichsuch data are available.
They find that the five countriesexperiencing improvementin an indexmeasuringperformance
in fiscal, exchangerate, andmonetarypolicies also saw povertydecline,whereasthe one country
in the samplethat witnesseda deteriorating policy performancealso suffered increasedpoverty
(Table 2).

Table2: AlacroeconomicPolicyand Povertyin Africa

Country Changein PercentPoor Changein MacroPolicy


(percentage
pointsper annwn) (weightedscoreof
macro-policyvariables)

Cote d'lvoire +5.30 -1.65


Kenya -0.28 +0.45
Nigeria -1.44 +1.79
Tanzania -1.83 +2.76
Ghana -1.95 +1.35
Ethiopia -3.60 +0.55

Source: Demeryand Squire(1995).

These results cannotbe extrapolated to the rest of the continent;policy implementation


8 Nor can it be concluded
varied widely and on balancepoverty has almost certainly increased.'
that all the poor benefittedin the countries that saw decliningpovertyon average; the surveys
reveal that somneamong the poor suffered greater deprivation. And it cannotbe claimed that
causalityfrom macroeconomnic
policy to poverty has been established. Nevertheless,the data do
confirm that improvements
in macroeconomicpolicy are consistentwith declinesin poverty even
in the short run. This is in turn consistent with evidenceon growth; povertyfell where growth
was positiveand increasedwheregrowthwas negative. Indeed,Demeryand Squireshow that the
change in poverty was primarily determined by the change in mean income with changes in

18 See Chenet al., (1994),whoalso show that countries withoutadequatepovertydata tendalso to have worse
macro performnance,
so compilations of availablepoverty data may wellunder-statethe problem.
15

inequalityplayinga secondaryrole and, at least in 1s sample, working in the oppositedirection


to growth as far as the poor are concerned.
In the Latin Americancontext Morley (1994) has likewiserecorded a close relationship
between growth and outcomesfor the poor in the adjustmentprocess. Reviewingperiods of
recession(fallingper capitaincomefor at least two years) and periods of recovery,he finds that
povertyincreasedin 55 of the 58 cases of recession, and in 22 out of 32 recoveriespoverty fell.
Contrary to the resultsfromthe sampleof six Africancountries,Morleyfmds that recessionswere
accompaniedby risinginequality(thepoor suffereddoubly)while recoverieswere associated with
falling inequality (the poor benefited doubly). But, as in the African sample, the changes in
poverty could be attributedmainly to changes in mean income.
Evidenceis also now appearingfor the transitional economiesof EasternEurope and the
Former SovietUnionwhichagainpointsto the importanceof changesin aggregateGDP but also
to a systematictrend towardsgreater inequality. As might be expectedthe large drops in GDP
in these countries have been reflected in substantiallyhigher levels of poverty. What is more
interestingis the tendencytowardsgreaterinequality. These countriesbegan the period with some
of the lowest Gini coefficientsin the world. The transition has entailedconsistentassociation
between growth and inequality:they have both deteriorated (Milanovic,1995).
Thus we findevidenceof a systematicworseningof inequalityin the transitionaleconomies
as GDP has declined, but obsefveno simple relationship between growthand inequalityin the
adjusting countries, althoughthe shifts in the Gini coefficient at least in Africa appear to have
been larger duringthe adjustmentphase than during periods of stable growth. We conclude front
this discussionthat successfuladjustmentusually leads to growth recoverywhichwill in general
also reduce poverty. We end this section with two qualifications.
First, it is important to stress that the detailed policy response, particularly in the
.ompoSitinnlof public expenditure cuts, can greatly affect the poverty outcomes of adjustment.
In some cases, aggregate budget contraction has been combined with rising shares (and
occasionally rising absolutelevels) of public spending in the social sectors, includingtargeted
transfers(Ribeet al. 1990;WorldBank, 1990, Ch.7; Selowsky, 1991). In Indonesia,the careful
mix of public spendingcuts during adjustment, and the rapid currency devaluations,helped
16

mitigate the short-term consequences for the poor of declining growth (Thorbecke, 1991).
Maintainingpublicinfrastructurecan alsobe crucial to the successof reformprograms. The fiscal
"crunch" often tempts governmentsto cut these infrastructuralsectors. There is another lesson
here for the nature of fiscal retrenchmentduring stabilization.
And second, we have said nothingabout other dimensionsof poverty, includinghuman
development,whichmay not be adequatelyreflected in income or consumption-basedmeasures
(Sen, 1992). It is beyondour scope to go deeplyinto the non-incomedimensionsof welfare. But
there is evidence to support two claims: I) that progress in reducing income poverty is
instrumentallycrucial to progress against most non-incomedimensionsof poverty; and ii) that
incomes are not all that matter, and indeed, for some non-incomedimensionscommandover
market goods may well be secondary to command over key publicly-providedsocial services,
notably access to basic health care and schooling.19 Cuts in key categoriesof social spending
during adjustmentcan entail heavy burdens on poor people, both in the short-runand long-run.
m. How Does Distribution Affect Growth?
So far we have looked at how growth might alter distribution. We now consider the
possibilityof a reversecausation. Thereare a number of ways in whichthis couldhappen.20We
focuson two: credit constraintsand politicaleconomy. Both have potentialimplicationsfor the
accumulationof capital,especiallyhumancapital, and growth. The first affectsthe access of the
poor to educationwhile the second affects incentivesand the returnsto education.
3.1 Credit Distrihbution
and (Growth
By preventing the poor from making productive investments(such as schooling)credit
constraintsarisingfrom asymmetricinformationperpetuate a low and inequitablegrowthprocess.
Furthermore,the more inequitablethe initial distribution (and, hence, the greaterthe numberof
poor and typicallycredit constrainedpeople) the more severe this effect will be. A number of
authors have examined credit market imperfectionsin general equilibriummodels with lumpy

" On these issuessee Anandand Ravallion(1993)and Bidaniand Ravallion(1995).

2 Quitegenerally,whenmarketsare incomplete therewil be efficiencyimplicationsof changesin distribution(Hoff,


1993). Somespecificexamplesin the literatureare reviewedin Liptonand Ravallion(1995,section5.1). The following
discussiondrawsin part on Deiningerand Squire(1995).
17

investment(Banerjeeand Newman1991, Tsiddon 1992, Saint-Pauland Verdier 1992, Galor and


Zeira 1993). The main result is that, where credit market constraintsprevent the poor from
making productive indivisible investments, inequalities in the wealth distribution can have
significant negative impacts on growth. What can policy do? Here we review three possible
actions: provisionof credit, redistributionof assets, and tax-subsidyinterventions.
Interventionin credit markets aimed at channelingcredit directlyto rationed groups via
subsidized interest rates may well reduce growth even further. In a dynamicperspective such
interventionsare likelyto cause efficiency-decreasingdistortionsand rent-seekingbehavior, thus
further reducing efficiencyand equity (Bencivengaand Smith, 1991).
An alternative approachentails equalizing the distributionof assets both to increase the
poor's ownershipof capitaldirectlyand to increasetheiraccessto credit markets. A large number
of analytical modelshave stressed the importance of the initial distributionof endowments,and
the potentiallylargeincreasesin social welfare that could be gainedby an initial redistributionof
assets (including Banerjee and Newman 1993, and Chatterjee 1991). Evidence from Asian
countries (Japan,Taiwan, and Korea)-where externallyimposedland reform was followed by
high growth-appears to supportthe hypothesis. But in many situationssuch redistributionmay
be possibleonly withfull compensation. Whether, and under what circumstances,such schemes
2" There are often less
will then pass the scrutinyof careful evaluation has yet to be determined.
ambitiousbut stillpotentiallyimportant opportunitiesfor giving poor farmersgreater security of
tenure in places where land rights are ill defined.
If the informationalimperfections that cause. credit rationing cannot be eliminated,
governments can seek ways around them by subsidizing education and taxing future wages.
Assumingthat highereducationis reflected in higher lifetime earnings, governmentscan provide
subsidiesto schoolingand finance them through a tax on future earnings, withouthaving to deal
with the problems involved in identifying individual ability (Hoff and Lyon, 1994). It can be
shown that policies mandatingcompulsory schooling, financed by a proportionaltax on wage

21 OngoingBankinvolvement land reform operationsin SouthAfricaand Colombiawouldprovide


in market-assisted
an opportunityto test thisempirically.
18

income, increase economicgrowth and, by distributing from agents with high human capital
* endowment to those with less, make the intragenerationaldistributionof income more equal
(Ecksteinand Zilcha, 1994). Whereit is very difficult to identifythe "type"of individualagents
ex ante, or if accessto credit markets is highly unequal, such policiescan be desirable.
3.2 Political Fcnnnmy
The preceding discussion suggests that amongst economies characterized by credit
rationing,those witha moreequaldistributionof wealth will accumulatemore human capital and
grow fasterthan thosemarkedby a more inegalitariandistribution.Highinequalitywill also make
it easier to adoptdistortionarypoliciesthat will negativelyaffectindividuals'investmentdecisions,
stifle growth, and conceivablygenerate political instability.
The most common mechanism used to establish a link between political forces and
economic outcomesis the notion of the median voter. Accordingto this argument,the median
voter's distance from the averagecapital endowmentin the economywill increase with wealth
inequality,thus leadinghim or her to supporta capitaltax rate that is higher the more unequal the
distributionof wealth. Thisin turn would reduceincentivesfor investmentin physicaland human
capital resulting in lower growth.
However, the median-votermodel may not be a plausibledescriptionof the political
in most developingcountries. An alternativemechanismrelies
processgoverningdecision-making
on lobbying. Greater wealthaflows the rich to spend more resourceson lobbyingactivitiesto
obtain differential treatment. In the extreme form, the ability to lobby would be directly
proportionalto the amountof economicassetsowned by.an individual. A modelthat utilizes this
assumption is providedby Persson and Tabellini (1992) who draw a connectionbetween high
concentrationof land, landowners'ability to successfullylobby governmentfor preferential tax-
treatmentof this asset, and the ensuing over-investment in land. Such disproportionatetaxation
of non-landowninggroupsleads to increasing inequalityover time and to slowergrowth.
Inequalityof assetownershipis also at the root of the manymodels that relate inflation to
inequalityof the distributionof income. The key idea is that inflationimposeslosses on certain
groups and that such losses are distributed very inequitably.While inflationtaxes holders of
moneyassets(i.e., the rich),access to foreign currency and capitalflight allowsthem to shift the
19

burden of inflationto the poor(Verdierand Saint-Paul, 1993). This opensnot only the possibility
for the rich to 'park" theirassetsabroad and then approve inflationarypolicies(as these would be
financed by the poor), it could also form the basis for strategicbehaviorof the rich (in support
of 'populist' policies)that could give rise to the typical "stop-andgo" policycyclesobserved in
many Latin-Americancountries(Labanand Sturzenegger, 1992). Similarly,Ozler and Tabellini
(1991) model the "class struggle"between workers, capital, and the governmentand-based on
the capitalists' ability to investin a risk-free foreign asset at the worldinterestrate-show there
is a broad range of situationswhere domestic investment and growth would be negatively
associated with inequalityin the distributionof assets.
In contrastto median-votermodels, lobbying models can incorporatedynamiceffects and
strategicbehavior. If politiciansare self-interested,the ability of the rich to offer high bribes, and
the inability of the poor to resist taxation, can lead to path-dependentequilibria(Brainardand
Verdier,1994);for example,industriesaffectedby a negative shock may choosewhetherto adjust
or to lobby for protectiondepending on the type of politician in power. Adjustmentwill be
slower, the more responsivepoliticiansare to lobbying, in which case growth-reducingpolicy
interventionswouldbe expectedto increase with overall wealth-inequality.
IRecentmodelshave emphasizedthat major policy decisions,in particularthe adoption of
macro-economicstabilizationmeasures,can be understoodin the frameworkof a bargaininggame
between different social groups. While many factors beyond the distributionof income can
influence bargaining power,income distribution plays an importantrole. Modelsthat describe
economic stabilization as a strategic game between the rich and the poor show that
stabilization-being associatedwith an increase in aggregate productivity-is more likely to be
delayed, the greater the inequalityof the income distribution (Alesinaand Drazen 1991). The
reason is that an unequaldistributionof income (or differentialaccessto "fmancialtechnology"
that could be used to diversifyrisk) implies that waiting reduces the utility of the rich only
marginallywhileimposinglarge costs on the poor. This would, in turn, increasethe probability
that in the end the poor will give in and shoulder all of the cost of adjustment. The model can
alsobe used to show that even if (often under external pressure or acute fiscal crisis) adjustment
measures are adopted, the lack of social consensus or the perceptionby some groups that they
20

have to pay a disproportionateshare of adjustmentcosts may lead to backslidingas soon as the


external pressures subside(Labanand Sturzenegger,1994).
3.3 nhe vidence
The argumentsreviewedabove suggest that greater income inequalitywill lead to lower
investmentin physicalandhumancapitaland hence to slowergrowth. There havebeen a number
of recent attempts to test this hypothesis. Data quality is unusually worryinghere. While
household surveymethodshave improvedgreatly in the last 10-15years, a largequestionmark
must be attached to the quality and comparabilityof the historical data on distributionin the
1950s, 1960sand 1970swhichhave beenused to test the impactof initial distributionon growth.2
And (unlikethe testsof the KuznetsHypothesis),the noisy inequalityvariableis now on the right
hand side, so theremust be a generalpresumptionthat standardestimatorswill give biased results.
While these and other issuesof data and econometric specificationshouldnot be under-rated,23
they take us beyondour present scope.
The tests that have been reported in recent literatureconfirm a negativeimpactof initial
inequality on growth, both in developedas well as developingcountries(Alesinaand Rodrik,
1994; Persson and Tabellini, 1994; Clarke, 1994). For a sample of nine OECD countries,
analysisof 20-yeargrowthrates startingfrom 1830 show that the income share of the top quintile
is negativelyrelatedto growth;it explainsabout 20 percent of the varianceof growthrates across
countriesand an increaseof onestandarddeviation in this share decreasesthe growthrate by half
a percentage point (Perssonand Tabellini, 1994). For a sample of developingand developed
countries,Clarke(1993)showsa robust, though quantitativelysmall, relationshipbetweeninitial
inequalityand growththat holdsfor differenteconometricspecificationsand independentlyof the
regime type. In addition, the empirical prediction that high inequalityin landownershipis
associated with lower capitalaccumulationand growth is confirmed for samplesof 35 and 50

22 Some of the 'data points"in these older compilationswere not evenbasedon householdsurveys,but were
synthoticestimates-let alonethat the qualityof the surveydata sets used was highlyvariable;for an overviewof these
issues see Fields(1994). Recentcompilationshave gone some way toward eliminatingthese problems(Chenet al.,
1994).

23 The inchlsionof theinitialaverageincomevariableon the righthand sideof theseequationsexplainingthe rate


of growthalso raisesconcernsaboutbias in the estimatorswidelyused in this literature.
21

countries (Persson and Tabellini, 1994). There is also evidence from a cross-sectionof 70
countriesfor the period1960-85that economicinequalityincreasespoliticalinstabilityand reduces
physical capitalinvestment(Alesina and Perotti, 1993).
But the verdictis not yet in on how strong or robust is the impact of initial inequalityon
future growth. For example, in one test, Fishlow (1995)reports no significanteffect once one
controlsfor Latin America(with simultaneouslyhigh inequalityand low growthfor much of the
period).2" As in tests of the Kuznets hypothesis discussed in section 1.2, there may well be
regional effects in the cross-sectionalrelationship, though that fact alone does not mean the
relationship is spurious-like the Kuznets Hypothesis, the real test will be in how the regional
effects evolveover time. Further empirical work is clearly needed, and the better distributional
data now availableshouldstimulatefuture research into the role of initial inequality.
IV. Conclusions
The "stylized fact" that distribution must get worse in poor countriesbefore it can get
better turns out not to be a fact at all. Effects of growth on inequalitycan go either way and are
contingent on a number of other factors. There is little sign in the new cross-countrydata we
assembled of any systematicimpact of growth on inequality.Possiblymeasurementerrors are
confounding the true relationship. But we think it more likely that the relationship between
growth anddistributionis by no means as simpleas sometheories in developmenteconomics have
postulated.
If distributionis unchangedthen growth will reduce absolutepoverty. Indeed, absolute
povertymeasurestypicallyrespond quite elasticallyto growth, and the benefitsare certainly not
confined to those near the poverty line.
One should be clear about what can and cannotbe concludedfrom our results. Let us
reiterate that it would not be correct to say that growthalways benefitsthe poor, or that none of
the poor lose from pro-growthpolicy reforms. Here we are only lookingat broad aggregates.
Cases of sufficientlyadversedistributionalimpacts to wipe out the aggregategains to the poor do
appear to be unusual; indeed, for 17 of the 20 countriesfor which we can assemblequite good

24 Clarkereportsthat the inequalityeffect on growthis robust to this andother changesin specification.


hbough
22

data from at leasttwo surveyssince the mid-1980s, the mean and the proportionof people living
below $1 per day moved in opposite direction. But there can be large differences between
countriesin the extentto whicheven a distribution-neutralgrowth processwill impact on absolute
poverty. The gains to poor people from such a growth process will tend to be lower the higher
the extentof initialinequality;under distribution-neutrality,a smallershare of total income must
implya smallerabsolutegain from a givenincrementto total income. And even in countrieswith
initiallylow inequalityand a growth process which brings rapid and sizablegains to most of the
poor, some will not be in a position to take advantageof the new opportunities,and some may
well lose. There can be an important role here for compensatorydirect interventions,providing
they are well integratedinto the general policy framework, in keeping with overall fiscal and
monetary discipline.
Nor does the evidencesuggest that growth is alwaysdistribution-neutral.And it would
alsobe wrongto conchlude
that changesin distribution are of little consequence.Indeed, we find
that povertymeasuresrespondquite elasticallyto changesin distribution;our cross-countrydata
suggest that a 10 percent increase in the Gini index is typically associatedwith roughly a 40
percent increase in the proportionof the population living on less than $1 per day (holding the
mean incomeconstant). The point is not that distributionis irrelevant,or that it never changes,
but rather that its changesare roughly orthogonal to economicgrowth.
Whilethereis littleconvificingevidencein our view that growthtends to alter distribution
in a systematicway,there are more compelling (theoreticaland empirical)argumentsas to why
initial distributionmattersto the extent and nature of subsequentgrowth. This link can operate
throughcredit marketconstraints,by limitingthe abilityof the poor to invest. The negative effect
on growth is strengthenedif distortionary policy-interventionsin favor of the rich further
undermine the poor's incentivesto invest.
Thusthereis no intrinsictrade-off between long-runefficiencyand equity. In particular,
policies aimed at facilitatingaccumulationof productive assets by the poor-when adopted in a
relativelynon-distortedframework-are also important instrumentsfor achievinghigher growth.
The problem shouldnot be posed as that of choosing betweengrowthand redistribution.
23

When one putsthesetwohalvestogether-one on the impactof growthon distribution, and


one on the reverse causation-we can begin to see the structure and some of the details of a joint
model of distributionand growth, and hence of poverty. The extent to which this is a truly
simultaneousmodel is a mootpoint; distributionappearsto affectgrowthmore than growth affects
distribution, though this interrelationshipis still being researched. There is also a dynamic
structureto thisjoint model,in whichinitialconditions (of averageincomes,inequality,and other
factors) do matter. Withinthis structure, a common set of policy-relevantexplanatoryvariables
can be identified,of whichbasiceducationis one of the more robust predictorsof both variables;
higher proportionsof men and women with good basic schoolingentailsa better distribution of
a larger total income.
Countrieswhichgivepriorityto basichuman capabilitiesin schooling,healthand nutrition
not only directly enhance well-being, but are also more likely to see improving income
distributions and higher average incomes over the longer term. There are often also ways in
which governmentscan help relieve the credit constraints facing the poor, though even means-
tested credit subsidieswill typically not be the best way; reducingtransactioncosts and helping
people organize themselveshave often proved to be better approaches. A more equitable
distributionof physicalassets, notably land, can also help greatly (both directlyand by relieving
credit constraints on investmentby poor people), though the policy implicationsare not as
straightforwardas with healthand education. The sectoral compositionof economicgrowthhas
also been emphasizedas an importantfactor. Sectoral biases againstthe rural sector in pricing,
exchange rates, and public investmentare not in the interests of either highergrowth or better
policies appear to be essentialfor sustainedgrowth, and
distribution. And soundmacroeconomic
either have no systematiceffecton distribution,or have potentiallyadverseshort-termimpacts but
which are typically not strong enough to outweigh the gains to the poor from growth. Paying
of publicexpendituresin the adjustmentprogramand to the inclusion
attentionto the cornposition
of effectivesafetynets for the poor will help improve the distributionaloutcomein the transition
to a pro-poor growth recovery.
Some of the key factors in achieving an equitable growthpath, suchas better schooling,
also raise the current living standards of poor people, in both "income"and "non-income"
24

dimensions. The nature of the dynamic interaction of initial conditionswith future growth and
distributional change can also have important policy implications. Countrieswith poor initial
conditions(due in part to past policies)will tend to diverge from the rest. It may only be possible
to overcomethis if the laggingcountriescan get a large enough "jumpstart" and here there may
be an especiallyimportantrole for internationaldevelopmentassistance,as privatecapital flows
usuallycome in onlyat a laterstage of the reform process. There will undoubtedlyremain areas
of socialpolicyor infrastructurein which private capital will not participateeven after successful
reform.
The upshotof all that one knows at the present juncture is that promotinggrowth is good
because it is a potentially and (in most case) actually important vehicle for improving the living
standards at all levels, and we now have a better idea about the policies that lead to growth,
ranging from the fundamentalinstitutional and market incentivesto the promotion of macro
stability. While thesepoliciesshould be pursued in all countries, we now suspectthat these will
be less effective and/or less well implemented in high-inequalitycountries. Thus reducing
inequalityis goodbecauseit willbenefitthe poor both immediatelyand in the longerterm through
higher growth.
Apart from the detailsof structural and macro-policyinterventionsthat have already been
mentioned, there are two major aspects that our analysishighlightsfor the changingrole of the
InternationalFinancialInstitutions. First, there is an important implicationin the area of greater
selectivity among countries. Obviously the IFIs should support growth-promoting policies in all
countries. Butthe focusshouldbe on countriesthat are clearly committedto reform. It appears
that low-inequalitycountriesmay well be more likely to be responsiveto the need for reforms,
and more able to implementthem in a shared-growthfashion. Testingcommitmentin high-
inequality countrieswould seem especially important. Up-front actionsthat are both growth-
promoting and equity-enhancingmay be the only realistic solution but even this-experience
shows-does not necessarilyguarantee sustainabiity. The second importantimplicationcomes
from the externality that appearsto be associated with improvementin the distributionof assets
and income;futuregenerationsbenefitbecausefuture growthwill tend to be higher through better
policies and better access to credit markets. If further research establishesthe strength and
25

robustnessof this result, then it has an important policy implication:the IFIs should be willing
to subsidizeactionsthatencourageredistibution, especiallymvestmentin basic education and land
reform.
26

References
Adelman, Irma and CynthiaTaft Morris (1973) Economic Growthand SocialEquity in
DevelopingCountries,Stanford: Stanford UniversityPress.
Adelman, Irma and ShermanRobinson(1988) 'Income distributionand development'in
Chenery,H., and T.N. Srinivasan(eds.) Handbook of DevelopmentEconomics:
lblume 1, Rotterdam:North Holland.
Ahluwalia, Montek S. (1976) 'Income distributionand development:some stylizedfacts,'
AmericanEconomicReview Papersand Proceedings66: 128-135.
Alesina, Albertoand Allan Drazen. (1991) 'Why are stabilizationsdelayed?' The American
EconomicReview,81: 1170-1188.
Alesina, Albertoand RobertoPerotti. (1993) 'Income distribution,politicalinstability,and
investment.' NBER %brkingPaper. Series No. 4486. Cambridge,MA: National
Bureau of EconomicResearch.
Alesina, Albertoand Dani Rodrik (1994) 'Distributive politics and economicgrowth,'
QuarterlyJournal of Economics 108: 465-90.
Anand, Sudhirand S.M. Ravi Kanbur (1993) 'The KuznetsProcessand the Inequality-
DevelopmentRelationship,'Journal of DevelopmentEconomics40:25-52.
Anand, Sudhir,and Martin Ravallion(1993) 'Human developmentin poor countries:On the
role of privateincomes fiadpublic services,' Journal of EconomicPerspectives,7: 133-
150.
Balisacan, ArsenioM. (1993) 'Anatomy of poverty during adjustment:the case of the
Philippines,' mimeo, University of the Philippines,QuezonCity, Philippines.
Banerjee, AbhijitV., and Andrew F. Newman(1993) 'Occupationalchoiceand the process of
development,'Journal of PoliticalEconony 101:274-99.
Bencivenga,Valerieand Bruce Smith (1991) 'Financial Intermediationand Endogenous
Growth,' Reviewof EconomicStudies. 58: 195-209.
Bidani, Benu and Martin Ravallion(1995) 'Decomposingsocial indicatorsusing distributional
data,' Journal of Econometrics,forthcoming.
27

Bourguignon,Franqoisand C. Morrison (1990) 'Income distribution,developmentand


i6 foreign trade: A cross-sectionalanalysis.' EuropeanEconomicReview. 34: 1113-
1132.
Brainard, S. Lael and ThierryVerdier(1994) 'Lobbying and adjustmentin declining
industries.' EuropeanEconomicReview 38: 586-595.
Bruno, Michaeland WilliamEasterly (1995) 'Inflation crises and long-rungrowth,' mimeo,
WorldBank, WashingtonDC.
Chatterjee, S. (1991) 'The effect of transitional dynamicson the distributionof wealth in a
neoclassicalcapitalaccumulationmodel,' Federal ReserveBank of Philadelphia,
WorkingPaper.
Chen, Shaohua,GauravDatt, and Martin Ravallion (1994) 'Is povertyincreasingor
decreasingin the developingworld?' Review of Income and Vkalth, 40: 359-376.
Chenery, H., M. Ahluwalia,C. Bell, J. Duloy and R. Jolly (1974)Redistributionwith
Growth,Oxford: OxfordUniversityPress.
Clarke, George R.G. (1993)'More evidence on income distributionand growth,' mimeo,
Departmentof Economics,Universityof Rochester.
Datt, Gaurav (1995) 'Povertyin India: 1951-1991,' mxnimeo,
PolicyResearchDepartment,
World Bank.
Datt, Gaurav and Martin Ravalffon(1992) 'Growth and redistributioncomponentsof changes
in povertymeasures:a decompositionwith applicationsto Braziland India in the
1980s,' Journal of DevelopmentEconomics 38:.275-295.
Deininger, Klaus and Lyn Squire (1995) 'Income distributionand development,'mimeo,
Policy ResearchDepartnent, World Bank.
Deininger, Klaus, Lyn Squireand Tao Zhang (1995) 'A new data base on internationalincome
distribution,' mimeo, PolicyResearch Department,WorldBank.
Demery, Lionel and Lyn Squire(1995) 'Poverty in Africa: An emergingpicture,' miimeo,
World Bank, Washington,DC.
28

Dorosh, Paul A., and DavidE. Sahn (1993) 'A general equilibriumanalysisof the effect of
macroeconomicadjustmenton poverty in Africa,' mimeo, CornellUniversityFood and
NutritionPolicyProgram.
Eckstein, Zvi and Itzhak Zilcha (1994) 'The effects of compulsoryschoolingon growth
income distributionand welfare.' Journal of PublicEconomics54:339-359.
Favaro, Edgardoand DonnaMacIsaac(1995) 'Who benefited from Peru's reformprogram?
Poverty note,' mimeo, Latin America3 Country Department,WorldBank,
Washington,DC.
Fields, Gary (1989) 'Changesin Povertyand Inequality in DevelopingCountries,' Wrld
Bank ResearchObserver,4:167-186.
(1994) 'Data for measuringpoverty and inequalitychangesin the developing
countries,' Journalof DevelopmentEconomics44: 87-102.
Fishlow, Albert (1995) 'Inequality,poverty and growth: Where do we stand?,' Paper
presented to the WorldBank's Annual Conferenceon DevelopmentEconomics.
Foster, James., J. Greer, and E. Thorbecke (1984) 'A class of decomposablepoverty
measures,' Econometrica52:761-765.
Galor, Oded and Joseph Zeira (1993) 'Income distributionand macroeconomics'Rview of
EconomicStudies 60: 35-52.
Glewwe, Paul and GilletteHall.1994) "Poverty, inequality,and living standardsduring
unorthodoxadjustment:The case of Peru,' EconomicDevelopmentand Cultural
Change42: 689-717.
Hoff, Karla (1993) 'The secondtheorem of the second best,' STICERDWorkingPaper,
London Schoolof Economics(Journalof PublicEconomics,forthcoming).
Hoff, Karla and AndrewB. Lyon (1994) 'Non-Leaky Buckets: OptimalRedistributive
Taxationand AgencyCosts.' NOER W,rking PaperSeries No. 4652. Cambridge,
MA: NationalBureauof EconomicResearch.
Jung, Jin Hwa (1992) 'Personal incomedistribution in Korea, 1962-1986:A Human Capital
Approach,' Journal of Asian Economics3: 57-72.
29

Kakwani, Nanak (1987) 'Inequality of incomederived from surveydata during the inflationary
period,' EconomicsLetters 23: 387-388.
Kuznets, Simon (1955) 'Economicgrowthand income inequality,' AmericanEconomicReview
45:1-28.
(1966)ModernEconomicGrowth, New Haven: YaleUniversityPress.
Laban, Raul and FedericoSturzenegger(1992) 'La EconomiaPoliticaDe Los Programas De
Estabilizacion.' ColeccionEstudios CIEPLAN No. 36 (December):pages 41-66.
Li, Hongyi, Lyn Squire and Heng-fu Zou (1995) 'Income inequality,humancapital
accumulationand politicaleconomy,' mimeo.
Lipton, Michael and Martin Ravallion(1995) 'Poverty and policy.' In Jere Behrmanand T.N.
Srinivasan(eds)Handbookof DevelopmentEconomicsW,lume3 Amsterdam:North-
Holland.
Milanovic,Branko (1995) 'Poverty,inequalityand social policy in transitioneconomies,'
TransitionEconomicsDivision, ResearchPaper 9, Washington,DC,The WorldBank.
Morley, SamuelA. (1994) 'Povertyand inequalityin Latin America: Past evidence,future
prospects,' Policy Essay 13. OverseasDevelopmentCouncil, WashingtonDC.
Oshima, H. (1970) 'Income inequalityand economic growth: The post-warexperienceof
Asian countries,' MalayanEconomicReview, 15: 741.
Ozler, S. and G. Tabellini(199r) 'External debt and political instability'NBER VMrkingPaper
No. 3772. Cambridge,MA: NationalBureau of EconomicResearch.
Papanek, Gustavand OldrichKyn (1986) 'The effect on income distributionof development,
the growthrate and economic strategy,' Journal of DevelopmentEconomics,23:55-65.
Persson, Torstenand Guido Tabellini(1994) 'Is inequalityharmful for growth?'American
EconomicReview 84: 600-621.
Ram, Rati (1995) 'Economicdevelopmentand inequality: An overlookedregression
constraint,' EconomicDevelopmentand CulturalChange, 3:425-434.
Ravallion,Martin (1995a) 'Growth and poverty: evidencefor the developingworld,'
EconomicsLetters, forthcoming.
30

(1995b) 'On Rati Ram's test of the Kuznets hypothesis,' mimeo, World Bank,
WashingtonDC.
Ravallion,Martin and GauravDatt (1995). 'How important to India's poor is the sectoral
compositionof growth?' V,rld Bank EconomicReview, forthcoming.
Ravallion,Martin and Monika Huppi (1991) 'Measuring changesin poverty:a methodological
case study of Indonesiaduring an adjustmenltperiod,' Wrld Bank EconomicReview
5 :57-84.
Ribe, Helena, S. Carvalho,R. Liebenthal, P. Nicholas, and E. Zuckerman(1990)'How
adjustmentprograms can help the poor,' WorldBank DiscussionPaper 71, World
Bank, Washington,DC.
Robinson, Sherman(1976) 'A note on the U-hypothesisrelatingincome inequalityand
economicdevelopment,'American EconomicReview66:437-440.
Saint-Paul, Gillesand Thierry Verdier(1992) 'Historical accidentsand the persistenceof
distributionalconflicts,' Journal of the Japaneseand InternationalEconomies6: pp.
406-422.
Selowsky,Marcelo (1991) 'Protecting nutrition status in adjustmentprograms:recentWorld
Bank activitiesand projects in Latin America,' Food and NutritionBulletin 13:293-
302.
Sen, Amartya (1992)Inequality"Re-Examined.Oxford: Oxford UniversityPress.
Squire, Lyn (1993) 'Fighting Poverty,' American EconomicReview 83(2): 377-382.
Thorbecke, Erik (1991) 'Adjustment, growth and incomedistributionin Indonesia,' Vlbrld
Development19:1595-1614.
Tsiddon, Daniel(1992) 'A moral hazard trap to growth.' InternationalEconomicReview,33:
299-321.
Watkins, Kevin(1995) The OXFAMPovertyReport. Oxford: OXFAM.
World Bank (1990) Wrld DevelopmentReport. New York:Oxford UniversityPress.
(1991) Assistance Strategies to Reduce Poverty. A World Bank Policy
Paper, The WorldBank, WashingtonDC.
(1993) The EastAsian Miracle, New York: Oxford University Press.
Policy Research Working Paper Series

Contact
Title Author Date for paper

WPS1545 International Commodity Control: Christopher L. Gilbert November 1995 G. Ilogon


Retrospect and Prospect 33732

WPS1546 Concessions of Busways to the Jorge M. Rebelo November 1995 A. Turner


Private Sector: The Sao Paulo Pedro P. Benvenuto 30933
Metropolitan Region Experience

WPS1547 Testing the Induced Innovation Colin Thirtle November 1995 M. Williams
Hypothesis in South African Robert Townsend 87297
Agriculture (An Error Correction Johan van Zyl
Approach

WPS1548 The Relationship Between Farm Size Johan van Zyl November 1995 M. Williams
and Efficiency in South African Hans Binswanger 87297
Agriculture Colin Thirtle

WPS1549 The Forgotten Rationale for Policy Jonathan Isham November 1995 S. Torres
Reform: The Productivity of Daniel Kaufman 39012
Investment Projects

WPS1550 Governance and Returns on Jonathan Isham November 1995 S. Fallon


Investment: An Empirical Daniel Kaufman 38009
Investigation Lant Pritchett

WPS1551 Sequencing Social Security, Dimitri Vittas December 1995 P. Infante


Pension, and Insurance Reform 37642

WPS1552 Unemployment Insurance and Milan Vodopivec December 1995 J. Walker


Duration of Unemployment: Evidence 37466
from Slovenia's Transition

WPS1553 Spatial Correlations in Panel Data John Driscoll December 1995 R. Martin
Aart Kraay 39065

WPS1554 Rationing Can Backfire: The "Day Gunnar S. Eskeland December 1995 C. Bernardo
without a Car" in Mexico City Tarhan Feyzioglu 37699

WPS1555 Capital Markets, Financial Roland Egerer December 1995 D. Brown


Intermediaries, and Corporate 33542
Governance: An Empirical Assessment
of the Top Ten Voucher Funds in the
Czech Republic

WPS1556 Workers in Transition Michal Rutkowski December 1995 WDR


31393

WPS1557 Electricity Demand in Asia and Masayasu Ishiguro December 1995 G. llogon
the Effects on Energy Supply and Takamasa Akiyama 33732
the Investment Environment
PolicyResearchWorkingPaperSeries

Contact
Title Author Date for paper

WPS1558 In Searchof PriceRigidities JacquesMorisset December1995 N. Cuellar


(RecentSectorEvidencefrom 37892
Argentina

WPS1559 HaveTransportCostsContributed AzitaAmjadi December1995 S. Lipscomb


to the RelativeDeclineof Sub- AlexanderJ. Yeats 33718
SaharanAfricanExports?Some
PreliminaryEmpiricalEvidence

WPS1560Trade and Fluctuations Aart Kraay December1995 R. Martin


JaumeVentura 39065

WPS1561 IncomeInequalityand Aggregate KlausSchmidt-Hebbel January1996 E. Khine


Saving:The Cross-CountryEvidence Luis Serven 37471

WPS1562 CatchingUpwith EasternEurope? BernardHoekman January 1996 F. Hatab


The EuropeanUnion'sMediterraneanSimeonDjankov 35835
Free Trade Initiative

WPS1563 Equityand Growthin Developing MichaelBruno January1996 P. Sader


Countries:Old and NewPespectives MartinRavallion 33902
on the Policy Issues Lyn Squire

You might also like