Burnside and Dollar - Aid, Policies and Growth

8/3/2019 Burnside and Dollar - Aid, Policies and Growth

1/23

American Economic Association

Aid, Policies, and GrowthAuthor(s): Craig Burnside and David DollarSource: The American Economic Review, Vol. 90, No. 4 (Sep., 2000), pp. 847-868Published by: American Economic AssociationStable URL: http://www.jstor.org/stable/117311 .

Accessed: 18/10/2011 05:56

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 formsof scholarship. For more information about JSTOR, please contact [email protected].

American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The

American Economic Review.

http://www.jstor.org
http://www.jstor.org/action/showPublisher?publisherCode=aeahttp://www.jstor.org/stable/117311?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/117311?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=aea


2/23

Aid, Policies, and GrowthBy CRAIG BURNSIDEAND DAVID DOLLAR*

Thispaper uses a new database onforeign aid to examine the relationshipsamongforeign aid, economicpolicies, and growth of per capita GDP. We ind that aid hasa positive impacton growth in developingcountries withgoodfiscal, monetary,andtradepolicies but has little effect in the presence of poor policies. Goodpolicies areones that are themselves mportantforgrowth.Thequalityof policy has onlya smallimpacton the allocation of aid. Our resultssuggest that aid wouldbe moreeffectiveif it were more systematicallyconditionedon good policy. (JELF350, 0230, 0400)

Growthof developingeconomies dependstoa large extent on their own economic policies:this findinghas been established n a wide rangeof recent studies.' On the other hand, foreignaid has not raised growth rates in the typicalpoor country,according o recent workby PeterBoone (1995, 1996). We investigate a new hy-pothesis about aid: that it does affect growth,but that its impact is conditional on the samepolicies that affect growth.Poor countries withsound economic policies benefit directly fromthe policies, and in this environmentaid accel-erates growth. In highly distorted economies,however, aid is dissipated n unproductivegov-ernmentexpenditure.A modifiedneoclassicalgrowthmodelprovidestheanalyticalramework or thisinvestigation.Tothe extent that international apital marketsareimperfect, oreign aid can have an importantm-pacton a poor country.One interpretationf for-eign aid is that t acts as an income transfer.Thisincome transfermay or may not producegrowth.The outcomedependson how aid is used: is itinvested,so thatdomesticoutputcanincrease,or

is it consumed?To the extent that it is invested,aid will be effective. Both the incentive o investaid and its subsequentproductivity s capital areaffected by various policy distortionsthat canlower thereturn o capital. t. s straightforwardoshow, in a neoclassicalmodel,that the impactofaid will be greaterwhen there are fewer distor-tions. Ingeneral,developingcountrygrowthrateswill dependon initial income, institutionalandpolicy distortions,aid, and aid interactedwithdistortions.2To investigateour hypothesisempiricallyweuse a new databaseon foreign aid developed bythe World Bank. The grantcomponentsof con-cessional oans have beenadded o outright rantsto yield a truerestimateof foreign aid.We drawon the recentempiricalgrowthliteratureo de-velop a model of growthwith a rangeof institu-tionaland policydistortions, nd we estimate hismodelusing a panelof 56 countriesandsix four-year time periodsfrom 1970-1973 until 1990-1993. Aside from the institutionaland politicalvariables, the policies that have considerableweight n thisequationare thebudgetsurplus,heinflation ate,andthe opennessdummydevelopedby SachsandWamer 1995).We forman index ofthesethreepolicies to interactt withforeign aid.Once we enterforeignaid into our empiricalmodel, we find that it has a positive effect ongrowth na goodpolicyenvironment. heresult srobust o a varietyof specificationsn which out-liers are ncludedorexcluded,andmiddle-incomecountries re ncludedorexcluded.Thisfinding s

* WorldBank, Washington,DC 20433. Views expressedare those of the authors and do not necessarilyrepresentofficial opinionsof the WorldBank. Helpful comments onthis work have been provided by an anonymous referee,Alberto Alesina, Paul Collier,William Easterly,and LantPritchett.We thankCharlesChangand ElianaLa Ferrara orexcellent research assistance. We gratefully acknowledgefinancial assistance from the World Bank Research Com-mittee (RPO 681-70).1 The particularpapers n the literature hat we focus onare William R. Easterly and Sergio T. Rebelo (1993), Stan-ley Fischer (1993), and Jeffrey D. Sachs and Andrew M.Warner 1995).2 These results are established n an Appendixavailablefrom the authorsupon request.

847


3/23

848 THEAMERICANECONOMICREVIEW SEPTEMBER 000consistentwith Boone's work n that he estimatedimpactof aidforacountrywithaveragepolicies szero.Countrieswith good policies and significantamountsof aid, on the otherhand, performverywell, better than can be explainedby the othervariables n the growthregression.Turning to allocation issues, we estimate anequation to explain aid receipts (as a share ofGDP). Donors direct their aid to low-incomecountries,but are also influencedby population(small countriesget more) andby variablesthatreflect their own strategicinterests. After con-trolling for these other influences, we find notendency to allocate more aid to countries withgood policies, as measuredby our index. Whenwe distinguishbetween bilateral and multilat-eralaid, we findthatit is the former hatis mostinfluencedby donorinterests,whereas the latteris largely a function of income level, popula-tion, andpolicy.We also estimatean equation or governmentconsumption as a share of GDP. We treat thisvariable separatelyfrom the other policy vari-ables because it has no robustassociation withgrowth. We find thatbilateralaid, in particular,has a strong positive impact on governmentconsumption. This result is consistent withother evidence that aid is fungible andtendstoincrease government spendingproportionately,not just in the sectorthat donors thinkthey arefinancing. That aid tends to increase govern-ment consumption,which in turnhas no posi-tive effect on growth, provides some insightinto why aid is not promoting growth in theaveragerecipient country.In ourwork we considered he possibility hatthepolicyindex shouldbe treatedas endogenous.In an earlierdraftof the paper we estimatedanequationfor policy and found that exogenouschanges in aid had no systematiceffect on theindex of policies. For simplicity,here we treatpolicy as exogenous and present the resultsofspecificationests to justifythis assumption.Overall, our results indicate that aid mighthave more impacton growthin the developingworldif it were systematicallyallocatedtowardgood policy environments.Up through he mid-1990's, however, donors were not favoringgood policy environments n their allocations.One caution about this conclusion is that, ifdonors change their allocation rule, then thequantity of aid may begin to affect policies.

Intuitively,one would thinkthat aidconditionedon good policy might have a positive effect onpolicy. Empirically, this is an interesting andopen area for furtherresearch.The remainderof the paper is organized asfollows: in the first section we describe themodel to be estimated, ourempirical methodol-ogy, the identifying assumptionswe make, andthe data used in the analysis. In the secondsection we describe the results concerning theimpact of aid on growth.In the third section wedescribe the determinantsof aid. The fourthsection examines the impact of aid on govern-ment consumption. The fifth section containsconcluding remarks.I. EmpiricalModel andData Sources

Our empirical work attempts to answer twokey questions:(1) Is the effect of aid on growthconditional on economic policies? and (2) Dodonorgovernmentsand agencies allocate moreaid to countries with good policies? More gen-erally we ask what other factors affect growthand aid flows.We investigatethese questionsby estimatingvariantsof the following equations:(1) git = Yit3),+ ait3a + Pijt3p

+ a. pif1t + zftpz + gt + 4,t(2) ait = yityy+ p'typ + Zty/z + at + 4awhere i indexes countries,t indexes time, git isper capitareal GDP growth,Yit s the logarithmof initial realper capitaGDP, ait is aid receiptsrelative to GDP, pit is a P X 1 vector ofpolicies thataffectgrowth,zit is a K X 1 vectorof other exogenous variablesthat might affectgrowth and the allocation of aid, gt and a, arefixed-timeeffects, andsg andea aremean zeroscalars.We include fixed-timeeffects to capturethe impactof worldwidebusiness cycles.Theway in whichaid andthe policy variablesenter equation (1) can be derived from a neo-classical growth model. For example, a lump-sum gift of aid shouldhave a positive effect ongrowth,which would be transitoryf therewerediminishing returns to capital. If there werepolicies that affected growth, however, they


4/23

VOL. 90 NO. 4 BURNSIDEAND DOLLAR:AID, POLICIES,AND GROWTH 849would also affect the extent to which a gift ofaid is used productively.Hence, if aid is addedto the growth equation,it should be interactedwith policies, as in equation (1).Earlierwork on aid and growthestimatedanequation uchas (1) without he interaction f aidand policy. For example, Keith Griffin(1970),ThomasE. Weisskopf(1972), Hollis B. Cheneryand Moises Syrquin (1975), Paul Mosley et al.(1987), and VictorLevy (1988) have previouslyattemptedomeasurehe impactof aid on savings,investment,and growthin developingcountries.The conclusionsreachedby the authorsof thesepapershave differedwidely,andtheyhave facednumerouseconometricdifficulties, n particularthe fact that the error ermsin equations 1) and(2) are likely to be correlated.RecentpapersbyBoone (1995, 1996)have used instrumentalari-able techniquesand have concluded hat aid hasno significantpositiveimpacton growth.We re-visit thatwork, ntroducinghehypothesis hat heimpactof aid is conditionalon policy.To estimateequation 1) we use both ordinaryleastsquares OLS) anda two-stage east-squares(2SLS)procedure ecause he errorerms n equa-tions (1) and(2) maybe correlated.The directionof correlation s not obvious. The errortermswould have a negativecorrelation o the extentthatdonorsrespond o negativegrowthshocksbyprovidingmoreassistance.But thereareplausiblereasons why the errorsmay have a positivecor-relation.Oneconclusionof earlier tudiesandourown work is that aid is not given only for devel-opmentalpurposes; t may servethe strategicorcommercial nterestsof donors.In that case acountry njoyinga commodityboom,or any pos-itive shock to growth,may receive specialfavorfrom some donors, ntroducing positivecorrela-tion betweenthe error erms.

Our strategyfor achieving identificationofthe system is as follows: we build the specifi-cation of the growth equationdrawing on thelarge empiricalliteratureon growth. Then wedevelop the specification of the aid equationdrawing on the literature on aid allocation.These literatures uggestthat there are variablesthatbelong in the aid equation hatdo not affectgrowth, and vice versa, allowing us to achieveidentificationby using zero restrictionson ,and -y,. We provide the details of these exclu-sion restrictions n the following subsections.Having achieved identificationby excluding

some of the exogenous variablesfrom each ofthe equations, we estimate them by 2SLS andpresent summary statistics from our first-stageregressions to indicate the relevance of our in-struments.The equationsare estimated using apanel across six four-yearperiods from 1970-1973 through 1990-1993. Thus,an observationis a country' performance averaged over afour-yearperiod.

A. The GrowthEquationThe recent empirical growth literaturepro-vides guidanceconcerningthe institutionalandpolitical factors and economic policies that af-

fect growth, and we follow this literatureinbuilding up the base specification.3The generalstrategy s to account for a rangeof institutionaland policy distortions that can help to explainthe growth performanceof poor countries, toensurethatanyinferences abouttherelationshipbetween aid and growthare robust.As is standard n the empirical growthliter-ature,to captureconvergenceeffects we allowgrowth during period t to depend on yit, thelogarithmof real per capita GDP at the begin-ning of the period. Since we are interestedinassessing the effectiveness of foreign aid, ourgrowth equation ncludesai,i, the level of aid, asa fraction of GDP, received by country i inperiod t.We also want to know how macroeconomicpolicies affect growth.As indicators of macro-economic policy we include the following vari-ables as elements of pi,. Firstwe use a dummyvariablefor tradeopenness developed by Sachsand Warner 1995). Closed economies are onesthat have average tariffs on machinery andmaterials above 40 percent, or a black-marketpremiumabove 20 percent, or pervasive gov-ernment control of key tradables. FollowingFischer (1993), we take inflation as a measureof monetarypolicy. Finally, we considered twofiscal variables suggested by Easterly andRebelo (1993), the budget surplusand govern-ment consumption,both relative to GDP. Thebudget surplus variable has foreign grants in-cluded in revenue and aid-financed projects

3See Ross Levine and David Renelt (1992) for a reviewof alternativespecificationsof empirical growth equations.


5/23

850 THEAMERICANECONOMICREVIEW SEPTEMBER 000included in expenditures, so that there is nonecessary relationship between aid and thismeasure of the budget surplus.The budget sur-plus is quite strongly negatively correlatedwithgovernment consumption. In regressions thatincluded both variables,we generally foundthebudget surplus to be marginally significant,whereas governmentconsumptionwas not. Forthis reason we dropped governmentconsump-tion from our analysis. Our results were notsensitive to this choice.In the previous section we argued that theeffectiveness of foreign aid would depend onthe nature of economic policies, so our growthequation ncludesnot only measures of aid andpolicies, but also their interaction.Our growth equationalso includes a subset ofthe K X 1 vector of exogenous variables zjt,which we assume are not affectedby shocks togrowth or the level of aid. These variables areincluded to capturevarious institutionalandpo-litical factors that might affect growth. In par-ticular, with reference to Stephen Knack andPhillip Keefer (1995) we use a measure of in-stitutionalqualitythatcapturessecurityof prop-erty rights and efficiency of the governmentbureaucracy.Since this variable is not widelyavailable before 1980 we use each country's1980 figure throughout,on the assumption hatinstitutional factors change slowly over time.Anothervariable hat does not changeovertimein our data set is the ethnolinguisticfractional-ization variable used by Easterly and Levine(1997), who findthatethnic fractionalization scorrelated with bad policies and with poorgrowth performanceafter controlling for poli-cies. Thus the institutionalqualityand the eth-nic fractionalization ariablescapture ong-termcharacteristicsof countriesthataffect bothpol-icies and growth.We also include the assassinationsvariableusedby severalstudies o capture ivil unrest,andan interactive erm between ethnic fractionaliza-tion andassassinations. hefinal nstitutional ari-ableis the level of broadmoney (M2)overGDP,whichproxies orthedevelopment f the financialsystem (RobertG. King and Levine, 1993). Be-causeof concernovertheendogeneity f thelattervariablewe lag it one period.We consideredsome othervariables hathavebeen used in the literature,in particulartheeducation variables developed by Robert J.

Barroand Jong-WhaLee (1993). We found thatthese variables had little explanatorypower (t-statistics well below 1.0), but their inclusionsignificantlyreducedthe numberof countries nthe sample, so we did not include them.Finally, we include regional dummy vari-ables for sub-SaharanAfrica and East Asia inthe growth equation.

B. The Aid EquationThere is a significant literatureon the deter-minants of aid, a few examples of which areRobert D. McKinlay and Richard Little (1978,1979), Alfred Maizels and Machiko K. Nis-

sanke (1984), Bruno S. Frey and FriedrichSchneider 1986), and William N. TrumbullandHowardJ. Wall (1994). In generalthis literaturehas found that donors' strategic nterestsplay animportant ole in the allocation of aid, whereascommercial interests have not been as impor-tant.Furthermore,more aid is given to countrieswith low income, and aid relative to GDP ismuch higher for countries with small popula-tions. Frey and Schneider(1986) find evidencethat commitment of World Bank assistance isassociated with good policies such as low infla-tion, but no one has examined whethertotal aidis allocatedin favor of good policies.Ourspecification f theaidequation 2) buildson this literature. t includesthe logarithmof ini-tial incomeyi,. It also includesa numberof othervariables:helogarithm f population nda groupof variables hat capturedonors' strategic nter-ests. For these we use dummyvariables or sub-SaharanAfrica (to which most Europeanaid isdirected),he Franczone(whichgets special reat-ment from France),Egypt (an important lly ofthe United States), and CentralAmericancoun-tries(also nthe U.S. sphereof influence).Wealsouse a measure of arms importsrelative to totalimports aggedoneperiod.Toexplorewhetheraidis allocated n favor of good policy we also in-clude our policy variables n the aid allocationequation.

C. Constructinga Policy IndexIn practice, we found it difficult to obtainprecise estimates, even in OLS regressions,ofthe vector of coefficients I,3on the three inter-

actions terms in equation (1). In addition, in


6/23

VOL. 90 NO. 4 BURNSIDEAND DOLLAR:AID, POLICIES,AND GROWTH 851termsof expositionandsimplicitywe thought twould be useful if we had one overall measureof economic policy rather than three separatevariables. We considereda numberof alterna-tive methods. The first method we consideredwas a simple principalcomponents approach,that is, using the first principal component inour analysis rather than all three policy vari-ables. Unfortunately,n oursample the firsttwoprincipal componentsare almost perfectlycor-related with openness and inflation, respec-tively. Thus, theprincipalcomponentsapproachdid not lead us to anatural ingle indexmeasureof policy. Instead it effectively suggested thatwe drop the budgetsurplusvariableand includeboth openness and inflationin our regressions.This turned out not to solve our problem withprecisionin estimating nteraction erms,so weproceededto an alternativemethod.Our model suggests that it is the distortionsthat affect growththat will determine he effec-tiveness of aid. Therefore,we thought it wasnatural hat our policy index should weight thepolicies accordingto theirimpacton growth,afeaturethatis absent fromthe principalcompo-nents analysis. This would allow us to discussthe effectiveness of aid in "good" and "bad"policy environments,where "good"and "bad"would have a precise meaning. Thus, the keyfeatureof ourpolicy index is thatit weights thepolicy variables accordingto their correlationwith growth.To be more precise, we use an OLS regres-sion of the growth equationwith no aid terms(3) git yi-py + ? zitP + g, + _-to fix the values of the coefficients that deter-mine the policy index. That is, we let pit =Pitbp, where bp is the OLS estimate of ,ep inequation(3). Then,rather hanestimatingequa-tions (1) and (2) we estimate(4) git- yity + aita + pitOp aitpit0O

+ Z'itf + gt + 4tand(5) ait = yityy+ Pitp + zytyz + at + a

Documents

Burnside and Dollar - Aid, Policies and Growth