Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Springer is collaborating with JSTOR to digitize, preserve and extend access to Journal of Economic Growth.
http://www.jstor.org
ESTHER DUFUO
MITandNBER
ofEconomics,
Department
Thispaperdescribes thecorrelations
between inequality andthegrowth data.Usingnon-
ratesincross-country
parametricmethods, we showthatthegrowth U-shapedfunction
rateis an inverted ofnetchangesininequality:
changesin inequality areassociatedwithreducedgrowth
(in anydirection) in thenextperiod.The estimated
is robustto variations
relationship in controlvariablesand estimation methods.This inverted U-curveis
witha simplepolitical
consistent economy modelbutitcouldalsoreflect thenatureofmeasurement errors,
and,
ingeneral,efforts thisevidencecausallyrunintodifficult
tointerpret identification Weshowthatthis
problems.
toexplainwhyprevious
is sufficient
non-linearity estimates between
oftherelationship and
thelevelofinequality
growth areso differentfromoneanother.
Keywords:
inequality, cross-country
growth, regressions
Ol 1,O15
JEL classification:
1. Introduction
2. Specifications
in theLiterature
The standard
procedureforestimatingtherelationship
betweeninequality
andgrowthin
theliterature
is toassumea simplelinearrelationship
between
inequalityandsubsequent
growth:
Table1. Relationship
between andchangesinGini,linearspecifications.
growth
Variable:(y(t+ a) -y(i))la
Dependent
Perotti
Specification BarroSpecification
3. The Inequality-Growth
Relationship
3.1. PoliticalEconomyArguments
Politicaleconomy models,intheirsimplest
version,startwiththepremise
thatinequality
leadstoredistribution
andthenitis arguedthatredistribution
hurts Sinceourgoal
growth.5
is to illustrate
whatcan happenin thisclass of models,we presenta versionof the
argument thatminimizes
institutional
detail.
livedorhaveshorthorizons.Whentheydecidewhether
or notto resist,
theyignorethe
itwillhaveon outputin future
effect periods.
3.1.2. ResultsandImplications
forEmpiricalWork
iyit^~yit) =
ay, + XJ + k(git- git.a) + v,.+ sin (5)
As g goesup,thefirst goesdownbutthesecondgoesup,making
termofthisexpression it
difficultto predict the sign of the relationship.However, as long as
maxja^} + max{a#}< 1, there exist values of g satisfyingmax{afl}< g < 1
- max{a,4},and forsuch intermediatevalues of g, thereare no plannedchangesin
There
inequality. are in
plannedchanges inequality forg < aB,to theextent
of
3.2. WealthEffectArguments
32.1. A Model
Diagram1 A+l
3.2.2. ResultsandImplications
forEmpiricalWork
^"-^
+XJ + k(git- git_a)+ h{git)+ v,+ £,-,.
= ocyit (9)
This of courseassumesthatwe have not eliminatedthe convergence effectby
adequately controllingfor mean wealth (or mean income). In fact most that
specifications
are estimated do tryto controlfortheconvergence effect, as is standard in growth
regressions, byincluding a linearfunction ofthemeanlevelofincomeatthebeginning of
theperiod(as inequations (1) and(9)). In first onecontrols
differences, forpastgrowth (as
inequation(3)). Formostfunctions/(wnp) however, theconvergence term doesnotenter
linearly.Moreover, itseemsplausiblethatdifferent economieswillhavedifferent Asand
thereforewillconverge atdifferent rates.Therefore, controlling linearlyforpastlevel(in
thelevelequation)orpastgrowth (in thefirstdifferences equation)willnotnecessarily
in the of
help solving non-monotonicity relationshipthe between growth andchangesin
inequality.
Themodelalso tellsus thatwhileinitialdistribution matters forthegrowth rate,itonly
matters in theshortrun.Overa longenoughperiod,twoeconomiesstarting at thesame
meanwealthlevelwillexhibit thesameaveragegrowth rate,sincetheybothwouldhave
gonefrom theinitialmeanwealthtoa meanwealthofw*.In otherwords, thelength ofthe
timeperiodoverwhichgrowth is measuredwillaffectthestrength of therelationship
betweeninequality andgrowth.15
Notethatall thisis stillinthecontext ofwhatis,moreorless,thebest-behaved model
wecouldcomeupwith.Thereis,forexample, noverygoodreasontoassumethath(k)y the
production function intheaboveexample,is globallyconcave - mostmachines, forone,
comeina fewdiscrete sizes.16Consider a simplevariantofthemodelabovewherethere is
^ - I- r1
- It*- h 'h- 5r
^. ^ w*-Aw2 w - w*+Aw,
Diagram 2 ^+1
relatedto country
characteristics
whicharethemselves relatedto changesin thegrowth
rate.Even in thesimplestmodel,controlling forconvergence effectslinearlyis not
and
adequate, relationships such as equations(1) and (3) cannotbe derivedfromthe
model. In particular,
one would expect strongnon-monotonicity in the observed
relationshipbetween (and changes inequality) growtheven if the
inequality in and
underlying mechanism a
implies negative betweeninequality
relationship andgrowth.
ErrorArguments
3.3. Measurement
3.4. Summary
Table2. Countries
withlargechangesin ginicoefficients.
Decreasein ginicoefficient
largerthan Increasein ginicoefficient
largerthan
3 percentagepoints 3 percentagepoints
Source:Deininger
andSquire(highqualitysample).
4. Estimation
and Results
Table3. Relationship
between andchangesin inequality
inequality andgrowth.
Variable:(y(t)-y(t-a))/a
Dependent
Perotti
Specification BarroSpecification
(p valuein
parentheses)
obtainedusingrandom
Note:Coefficient effect specifications.
errors
Standard in parentheses;a is equalto 5 (5-yearperiods).
variables:see noteto Table 1.
Control
between
Figure1. Relationship andlaggedginigrowth:
incomegrowth linearmodel(Perotti
partially variables).
Figure2. Relationship
between
incomegrowth
andlaggedginigrowth: linearmodel(Barrovariables).
partially
alsoranthesespecifications
usingtheBarroexpanded dataset,and10-yearlagsinsteadof
5-yearlags,and we findthesameinverted U-shapedrelationshipbetweenchangesin
andgrowth,
inequality albeitestimated whichis notsurprising
withlessprecision, given
thatwe areleftwithonly78 observations not
(results reported).
Onbalance,thereis nostrong ofinequality
evidenceofa directcorrelation ongrowth in
theshortrun(overa 5-yearlag period),butthereseemsto be an associationbetween
changesin inequalityand growth. Changesin inequality,
whatever are
theirdirection,
associatedwithlowergrowth in thenextperiod.We discussat theend of thissection
whether anycausalinterpretation butbeforethatwe report
canbe giventothisresult, the
fromourreducedformestimates.
results
43. TheEffectofLaggedInequality
Table4. Estimation
ofthereducedformmodel.
Variable:(y{t+ a) -y(t))/a
Dependent
Perotti Barro
Variable:Changein GiniCoefficient
Dependent
Perotti Barro
g(t) - g(t- a) (g(t) - g(t - a))2 g(t) - g(t- a) (g(t) - g(t- a))2
Note:Coefficientobtainedusingrandomeffect specifications.
Standarderrorsin parentheses;
a is equal to 5 (5-yearperiods).
Controlvariables: X{t) standsforcontrolvariablenotlagged.
X(t- a) standsforcontrol variableslaggedone period(5 years).
Fora listof control
variablessee noteto Table 1.
Figure3. Relationship
between
giniandsquareofginichanges.
Figure4. Reducedform,
withPerotti
variables.
withtheLiterature
5. Relationship
5.1. Non-linearity
As we notedinSection2,all theapproachesbasedondifferencingthedatarelyheavilyon
thelinearityofequation(1) andtheexclusionofthedifferenced term.If eitherofthese
conditionsare violated,thefixedeffectand first
differenceestimatesof y willnotbe
andbothwillbe different
identical, fromtheOLS estimate ofequation(1) evenifall the
otherconditions oftheOLS estimate
forthevalidity Itwillthenbe important
aresatisfied.
eachofthesecoefficients.
to be verycarefulin interpreting
The resultsin theprevioussectionsuggestthatchangesin inequality
werenegatively
correlatedwithsubsequentgrowth.Assumingthe relationship betweenthe level of
inequalityand growthis indeedlinear(h(g)= yg),and differencing equation(9), one
obtains:
or
Figure5. Relationship
between andlaggedGinigrowth:
incomegrowth linearmodel.
partially
- - -
yit+a y» = (*<*+ i)(y* ?*-„) + <*(** x»-a)P
- 8u-a) + <«* - ^a-m
+ <><t>(git 0 *)
Table5. Non-linearity
oftherelationship changeinginiandgrowth
between inmodelsbasedonfirst
differences.
Variable
Dependent
B. Piecewiselinearassumption:
OLS coefficients
of (gini(/)-gini(/-a))
if gini(0-gini(/-a)<0 0.79 0.39 0.69 0.4
(0.30) (0.13) (0.38) (0.13)
if (gini(0-gini(/-a))>0 -0.3 -0.13 -0.49 -0.11
(0.35) (0.11) (0.38) (0.14)
C. Quarticspecification
F-testfornon-linearterms
jointlysignificant 2.21 3.37 2.55 3.3
(0.09) (0.02) (0.059) (0.02)
D. Quadraticspecification
gini(/)-gini(r-fl) 0.23 0.13 0.311 0.15
(0.18) (0.067) (0.19) (0.66)
(gini(f)- gini(r- a))2 -5.88 -3.24 -5.94 -3.28
(3.39) (1.26) (3.43) (1.23)
Numberofobservations 128 98 128 98
- y*)* = - -
(yu+a y»+a y» («« + 1)0^ - yu-a) - <*{*» - **-*)& (12)
wherea and ftare the values of a and /?obtainedby estimatingequation (1) using the
Arellanoand Bond estimator.If theassumptionsnecessaryforthevalidityof each method
are satisfied,a and P will be estimatedconsistently.
Then,accordingto equation(3), the
relationshipbetween(yit+a- yit)*and git- git_ashouldbe linear.
The next step is to make sure that the estimates of y obtained if we regress
-
(yit+a yu)*la linearlyon thedifference git- git_aare similarto thoseobtainedusinga
fixed-effectstype estimator.OLS estimatesare presentedin panel A of Table 5. They are
alternativeestimatesof y, consistentif equation (3) is correctlyspecifiedand if the
innovationin inequalityis notcorrelatedwiththe innovationin theerrorterm.They are
not identicalto the estimateof y reportedin Table 1, since theyuse different estimation
methods.However,theyare also positiveand significant, and theirmagnitudeis similarto
thatof the fixedeffectand Arellano and Bond estimates.In otherwords,as long as we
impose linearity, theresultsare verysimilarto whattheliterature finds.
Figure6. Relationship
between
incomegrowth
andlaggedGinigrowth:
usingArellanoandBondcoefficients.
UsingtheArellanoandBondTechnique
5.2.2. Estimation
Appendix
TableAl. Descriptive
statistics in thesample.
andcountries
in theSample
Countries
Australia Japan
Bangladesh Korea,Republicof
Belgium Malaysia
Brazil Mexico
Bulgaria Netherlands
Canada NewZealand
Chile Norway
China Pakistan
Colombia Peru
CostaRica Philippines
Denmark Poland
Dominican Republic Portugal
Finland Singapore
France Spain
Germany Sri Lanka
Greece Sweden
HongKong Thailand
Hungary TrinidadandTobago
India Tunisia
Indonesia Turkey
Ireland UnitedKingdom
Italy UnitedStates
Venezuela
Means(standard
deviation)
andHeston)
Log(GDPpercapita)in 1980dollars(Summers
1965 8.03(0.86)
1975 8.37(0.85)
1985 8.58(0.82)
1995 8.82(0.79)
Ginicoefficient
1965 0.38
1970 0.4
1975 0.4
1980 0.38
1985 0.37
1990 0.38
of thesampleand variable,see
and Squire"highqualitysample.*'Fortheconstruction
Source:Deininger
text.
6. Conclusion
Notes
References
Acemoglu, D., and F. Zilibotti. (1997) "Was PrometheusUnbound by Chance? Risk, Diversification,and
Growth," Journalof Political Economy 105(4), 709-751.
Acemoglu, D., and J. Robinson. (2000). "Why did the West Extend the Franchise?Democracy,Inequalityand
Growthin HistoricalPerspective," QuarterlyJournalof Economics 115(4), 1167-1 199.
Aghion,et al. (1999). "Dualism and Macro-EconomicStability,"QuarterlyJournalofEconomics 114(4), 1359-
1397.
Aghion,P., and P. Bolton. (1997). "A Trickle-DownTheoryof Growthand DevelopmentwithDebt Overhang,"
Review of Economic Studies 64(2), 151-172.
Alesina, A., and D. Rodrik. (1994). "Distributive Politics and Economic Growth," QuarterlyJournal of
Economics 109(2), 465-490.
Alesina, A., and R. Perotti.(1996). "Income Distribution,Political Instabilityand Investment,"European
Economic Review 8 1(5), 1170-1 189.
Angrist,J.D., and A. B. Krueger.(1999). "Empirical Strategiesin Labor Economics," In OrleyAshenfelter and
David Cards, eds., Handbook of Labour Economics, Vol. 3A. Amsterdam:NorthHolland: 1277-1366.
Arellano,M., and S. Bond. (1991). "Some Tests of SpecificationforPanel Data: Monte Carlo Evidence and an
Applicationto EmploymentEquations," Review ofEconomic Studies 58, 277-297.
Atkinson,A. B., and A. Brandolini.(2001). "Promise and Pitfallsin the use of 'Secondary' Data-Sets: Income
Inequalityin OECD Countriesas a Case Study," Journalof Economic Literature39(3), 771-799.
Banerjee, A., and A. Newman. (1993). "Occupational Choice and the Process of Development," Journalof
Political Economy 101(2), 274-298.
Banerjee, A., and A. Newman. (1994). "Poverty,Incentives,and Development," AmericanEconomic Review
84(2), 211-215.
Banerjee, et al. (2001). "Inequality, ControlRights and Rent Seeking: Sugar Cooperatives in Maharashtra,"
Journalof Political Economy 109(1), 138-190.
Barro,R. J. (2000). "Inequality and Growthin a Panel of Countries,"Journalof Economic Growth5(1), 5-32.
Benabou, R. (2000). "Unequal Socities: Income Distributionand the Social Contract,"AmericanEconomic
Review 90(1), 96-129.
Benhabib,J.,and A. Rustichini.(1998). "Social Conflictand Growth,"JournalofEconomic Growth1(1), 143-
158.
Benhabib,J.,and M. M. Spiegel. (1998). "Cross-CountrvGrowthRegressions." Mimeo.
Caselli, et al. (1996). "Reopening the ConvergenceDebate: A New Look at Cross-CountryGrowthEmpirics,"
Journal of Economic Growth1, 363-389.
Dasgupta, P., and D. Ray. (1986). "Inequality as a Determinantof Malnutritionand Unemployment:Theory,"
The Economic Journal96(384), 1011-1034.
Deininger,K., and L. Squire. (1996). "A New Data Set MeasuringIncome Inequality," WorldBank Economic
Review 10,565-591.