Theory, General Equilibrium, and Political Economy in Development Economics
Journal of Economic Perspectives—Volume 24, Number 3—Summer 2010—Pages 17–32
Theory, General Equilibrium, and Political Economy in Development Economics
Daron Acemoglu
eevelopmentvelopment eeconomicsconomics investigatesinvestigates tthehe ccausesauses ooff ppovertyoverty aandnd llowow iincomesncomes aaroundround tthehe wworldorld aandnd sseekseeks ttoo mmakeake pprogressrogress iinn ddesigningesigning ppoliciesolicies tthathat D ccouldould hhelpelp iindividuals,ndividuals, rregions,egions, andand countriescountries toto achieveachieve greatergreater economiceconomic pprosperity.rosperity. EconomicEconomic ttheoryheory playsplays a ccrucialrucial rroleole iinn tthishis eendeavor,ndeavor, nnotot oonlynly bbecauseecause iitt hhelpselps usus focusfocus onon thethe mostmost iimportantmportant eeconomicconomic mmechanisms,echanisms, bbutut aalsolso bbecauseecause iitt pprovidesrovides gguidanceuidance oonn tthehe external validity ofof eeconometricconometric eestimates,stimates, mmeaningeaning tthathat iitt clarificlarifi eses howhow wwee ccanan llearnearn ffromrom sspecifipecifi c empiricalempirical eexercisesxercises aaboutbout tthehe eeffectsffects ofof ssimilarimilar shocksshocks andand policiespolicies inin differentdifferent circumstancescircumstances aandnd wwhenhen implementedimplemented oonn ddifferentifferent scales.scales. GGeneraleneral equilibriumequilibrium andand politicalpolitical economyeconomy issuesissues oftenoften createcreate challengeschallenges fforor tthishis ttypeype ooff eexternalxternal vvalidity.alidity. GGeneraleneral eequilibriumquilibrium rrefersefers ttoo ffactorsactors thatthat bbecomeecome iimpor-mpor- ttantant whenwhen wewe considerconsider ccounterfactualsounterfactuals iinn wwhichhich largelarge changeschanges areare contemplated.contemplated. TThehe ddiffiiffi cultyculty lliesies iinn tthehe ffactact thatthat suchsuch counterfactualscounterfactuals willwill induceinduce changeschanges iinn ffactoractor ppricesrices aandnd ttechnology,echnology, wwhichhich wwee hholdold fi xedxed iinn ppartialartial eequilibriumquilibrium analysis,analysis, aandnd ccreatereate ddifferentifferent ccompositionomposition eeffectsffects thanthan inin partialpartial equilibrium.equilibrium. PoliticalPolitical economyeconomy rrefersefers ttoo tthehe ffactact thatthat thethe feasiblefeasible setset ooff iinterventionsnterventions iiss ooftenften ddeterminedetermined byby ppoliticalolitical ffactorsactors aandnd thatthat largelarge counterfactualscounterfactuals willwill induceinduce politicalpolitical responsesresponses fromfrom vvariousarious aactorsctors aandnd interestinterest ggroups.roups. GGeneraleneral eequilibriumquilibrium andand politicalpolitical economyeconomy considerationsconsiderations aarere iimportantmportant becausebecause partialpartial equilibriumequilibrium estimatesestimates thatthat ignoreignore responsesresponses fromfrom bbothoth ssourcesources willwill notnot givegive thethe appropriateappropriate answeranswer toto counterfactualcounterfactual exercises.exercises. IInn tthishis eessay,ssay, I fi rstrst explainexplain wwhyhy iitt iiss iimportantmportant ttoo tthinkhink ooff eexternalxternal validityvalidity iinn policypolicy analysis,analysis, particularlyparticularly inin developmentdevelopment economics,economics, andand I describedescribe thethe rolerole ooff economiceconomic theorytheory inin thisthis exercise.exercise. I thenthen illustrateillustrate thethe importanceimportance ofof generalgeneral
■ Daron Acemoglu is Charles P. Kindleberger Professor of Applied Economics, Massachusetts Institute of Technology, Cambridge, Massachusetts. His e-mail address is 〈[email protected]〉. doi=10.1257/jep.24.3.17 18 Journal of Economic Perspectives
eequilibriumquilibrium reasoningreasoning inin severalseveral majormajor problemsproblems inin developmentdevelopment economics.economics. FFinally,inally, I argueargue thatthat politicalpolitical economyeconomy considerationsconsiderations hhaveave ttoo bbee ccentralentral ttoo aanyny iinvestigationnvestigation ooff ddevelopmentevelopment pproblemsroblems aandnd tthathat iinferencesnferences tthathat iignoregnore politicalpolitical eeconomyconomy ccanan gogo wwrong.rong.
Why Development Economics Needs Theory
TTherehere isis nono generalgeneral aagreementgreement onon howhow muchmuch wewe shouldshould relyrely onon economiceconomic ttheoryheory inin motivatingmotivating empiricalempirical wworkork aandnd whetherwhether wewe shouldshould ttryry ttoo fformulateormulate aandnd eestimatestimate ““structuralstructural pparameters.”arameters.” I aarguergue tthathat tthehe aanswernswer iiss llargelyargely ““yes”yes” bbecauseecause ootherwisetherwise eeconometricconometric eestimatesstimates wwouldould llackack eexternalxternal vvalidity,alidity, iinn wwhichhich casecase theythey ccanan nneithereither iinformnform uuss aaboutbout wwhetherhether a pparticulararticular mmodelodel oorr ttheoryheory iiss a uusefulseful aapproximationpproximation toto reality,reality, nornor wouldwould theythey bebe usefuluseful inin providingproviding usus guidanceguidance onon wwhathat thethe effectseffects ofof similarsimilar shocksshocks andand policiespolicies wouldwould bebe inin differentdifferent circumstancescircumstances oorr ifif implementedimplemented iinn ddifferentifferent sscales.cales. I tthereforeherefore ddefiefi nene “structural“structural pparameters”arameters” aass tthosehose thatthat provideprovide eexternalxternal validityvalidity andand wouldwould thusthus bebe usefuluseful inin testingtesting theoriestheories oror iinn ppolicyolicy aanalysisnalysis bbeyondeyond tthehe sspecifipecifi c environmentenvironment aandnd ssampleample ffromrom wwhichhich ttheyhey aarere dderived.erived.1 EExternalxternal vvalidityalidity bbecomesecomes a particularlyparticularly challengingchallenging tasktask iinn tthehe ppresenceresence ooff generalgeneral eequilibriumquilibrium andand politicalpolitical economyeconomy considerations,considerations, aandnd a majormajor rroleole ooff economiceconomic theorytheory iiss iinn hhelpingelping uuss oovercomevercome tthesehese pproblemsroblems oorr aatt tthehe vveryery lleasteast aalertinglerting uuss ttoo ttheirheir iimportance.mportance. TToo illustrateillustrate thesethese points,points, considerconsider tthehe rrelationshipelationship bbetweenetween thethe costcost ooff sschoolingchooling andand schoolingschooling decisions.decisions. WeWe cancan describedescribe thisthis relationshiprelationship ppurelyurely aass a ddescriptiveescriptive one,one, focusingfocusing oonn a ssampleample aandnd llookingooking aatt tthehe ccorrelationorrelation oorr tthehe oordi-rdi- nnaryary lleasteast squaressquares relationshiprelationship bbetweenetween thesethese twotwo variables.variables. ForFor example,example, wewe couldcould sspecifypecify thethe followingfollowing reduced-formreduced-form relationship:relationship:
log (si ) = X ′i β – α log (ci ) + εi , where i denotes an individual in the sample, si is years of schooling, ci denotes the cost of schooling to the individual resulting, for example, from foregone earn- ings and actual costs of attending schools, Xi is a vector of characteristics of this individual for which we may wish to control, and β iiss a vvectorector ofof pparameters.arameters. TThehe pparameterarameter ooff iinterestnterest iiss α. WWee ccanan tthenhen uusese oordinaryrdinary lleasteast ssquaresquares ttoo eestimatestimate β aandnd α.
1 See Shadish, Cook, and Campbell (2002) on internal and external validity. The notion of external validity, in particular the emphasis on counterfactual exercises, as the defi ning characteristic of a structural parameter is closely related to Marschak’s (1953) defi nition, which distinguishes between structural parameters that provide “useful knowledge” for understanding the effects of policy within a given sample and/or in new environments. It also clearly presupposes that the empirical strategy has been successful in estimating “causal” effects (for example, as defi ned in Angrist, Imbens, and Rubin, 1996). Daron Acemoglu 19
AAlternatively,lternatively, wwee ccouldould sstarttart wwithith aann eeconomicconomic mmodel.odel. IInn ffact,act, ssomeome ssimpleimple ttheoriesheories willwill leadlead toto exactlyexactly tthishis eequation.quation. Suppose,Suppose, forfor example,example, thatthat thethe humanhuman ccapitalapital ooff aann iindividualndividual iiss a ffunctionunction ooff tthehe llevelevel ooff sschoolingchooling fforor tthathat iindividual.ndividual. IInn 1– σ pparticular,articular, supposesuppose thatthat thethe humanhuman capitalcapital ofof iindividualndividual i iiss givengiven byby hi = s i fforor ssomeome parameterparameter σ bbetweenetween 0 andand 1 andand si ddenotesenotes tthehe pperson’serson’s llevelevel ooff sschooling.chooling. TThehe iindividualndividual ccanan tthenhen eearnarn iincomencome eequalqual toto yi = whi , wwherehere w iiss tthehe mmarketarket wwageage pperer unitunit ofof humanhuman capital.capital. IInn aaddition,ddition, individualindividual i hhasas a costcost ooff sschoolingchooling ggiveniven bbyy
ζi c i s i , wwherehere ζi iiss anan unobservedunobserved nonmonetarynonmonetary costcost componentcomponent (for(for exampleexample duedue ttoo differentialdifferential discountingdiscounting oorr bborrowingorrowing cconstraints),onstraints), aandnd ci iiss thethe monetarymonetary costcost ooff sschoolingchooling fforor tthishis iindividual.ndividual. SSupposeuppose thatthat iindividualsndividuals mmaximizeaximize nnetet iincome,ncome, ssoo thatthat iindividualndividual i wwillill cchoosehoose sschoolingchooling ttoo mmaximizeaximize iincomencome nnetet ooff tthehe ccostost ooff 1 – σ sschooling—thatchooling—that iiss w s i – ζi c i s i . AAfterfter wworkingorking tthroughhrough thethe maximizationmaximization problem,problem, tthishis mmodelodel impliesimplies a rrelationshipelationship iidenticaldentical ttoo tthehe rreduced-formeduced-form eequationquation withwith wwhichhich wewe started,started, butbut nownow thethe parameterparameter α ccorrespondsorresponds toto 1/1/σ. 2 OOncence tthishis eequa-qua- ttionion isis dderived,erived, estimationestimation isis alsoalso straightforwardstraightforward andand cancan bebe performedperformed againagain byby oordinaryrdinary lleasteast ssquares.quares. NNextext comescomes thethe harderharder part.part. WeWe havehave seenseen thatthat tthehe ssameame eequationquation cancan bebe ppositedosited asas a reduced-formreduced-form relationship,relationship, oror itit cancan bebe derivedderived fromfrom anan economiceconomic mmodel.odel. BButut aatt tthehe eend,nd, iitt iiss tthehe ssameame eequation,quation, aandnd iitt ccanan bbee eestimatedstimated iinn tthehe ssameame mmanner.anner. SoSo inin whatwhat sensesense cancan wewe thinkthink ofof itit asas a “structural“structural relationship”?relationship”? TheThe aanswernswer iiss rrelatedelated ttoo tthehe nnotionotion ooff eexternalxternal vvalidityalidity iintroducedntroduced aabove.bove. SSupposeuppose wwee nnowow aasksk tthehe qquestion:uestion: wwhathat wwouldould bbee tthehe eeffectsffects ofof subsidiessubsidies ttoo rreduceeduce thethe costcost ooff sschooling,chooling, ci , forfor a setset ofof individuals?individuals? ThisThis counterfactualcounterfactual experimentexperiment ccouldould bbee mmotivatedotivated bbyy a ppotentialotential ppolicyolicy tthathat iiss bbeingeing ccontemplatedontemplated oorr iitt mmayay bbee uusedsed forfor uunderstandingnderstanding andand testingtesting tthehe iimplicationsmplications ooff oourur ttheory.heory. TThehe qquestionuestion mmightight bebe fforor tthehe ssameame ssampleample oonn wwhichhich thethe initialinitial estimationestimation waswas performedperformed oorr iitt ccouldould bbee fforor aann eentirelyntirely ddifferentifferent ssampleample oorr ppopulation.opulation. IInn eeitherither ccase,ase, oonene aanswernswer ttoo tthehe aabovebove qquestionuestion rreadilyeadily followsfollows fromfrom uusingsing tthehe eestimatesstimates ooff α ttoo ccomputeompute tthehe iincreasencrease iinn tthehe yyearsears ooff sschoolingchooling fforor iindividualsndividuals wwhosehose ccostost ooff sschoolingchooling hhasas ddeclined.eclined. BButut ccanan wwee ttrustrust tthishis aanswer?nswer? IIff α iiss indeedindeed a sstructuraltructural pparameter,arameter, tthenhen wwee sshouldhould ttrustrust tthishis aanswernswer ((obvi-obvi- oously,usly, ssubjectubject toto standardstandard errors),errors), butbut notnot otherwise.otherwise.3 ToTo illustrateillustrate whatwhat mightmight ggoo wwrongrong wwhenhen α ddoesoes nnotot ccorrespondorrespond ttoo a sstructuraltructural pparameter,arameter, iimagine,magine, fforor eexample,xample, tthathat yyearsears ooff sschoolingchooling aarere cconstrainedonstrained bbyy sschoolchool eenrollments,nrollments, wwhichhich aarere iinn tturnurn cconstrainedonstrained bbyy tthehe ssizesizes ooff sschools.chools. IInn tthishis ssetting,etting, lletet uuss ffurtherurther aassumessume tthathat individualsindividuals withwith lowlow costcost ooff schoolingschooling getget proportionatelyproportionately moremore ofof thethe availableavailable
2 – 1/σ 1/σ Specifi cally, the optimal choice of individual i is si = K(ζi ci ) , where in this case K = ((1 – σ)w) . After taking logs and defi ning εi = – log ζi /σ and α = 1/σ, this gives the reduced-form equation above. 3 Many empirical equations that do not correspond to structural relationships may nonetheless contain useful information; they just cannot be used for counterfactual policy analysis. We might simply be interested in uncovering correlations, which may help us distinguish between theories, since many relevant theories will have implications about what these correlations should look like. This suggests that it is often useful to estimate reduced-form relationships that do not have structural interpreta- tions, but when doing so, we should be explicit about how they should be (and not be) interpreted. 20 Journal of Economic Perspectives
sschoolchool rresourcesesources ((forfor iinstance,nstance, ddueue ttoo ssomeome ttypeype ooff eeffiffi ccientient rrationing).ationing).4 InIn thisthis eexample,xample, wwee ccanan sstilltill eestimatestimate thethe relationshiprelationship bbetweenetween s aandnd c, andand wewe willwill obtainobtain a mmeaningful-lookingeaningful-looking eestimatestimate ooff α. However,However, thethe estimateestimate willwill lacklack eexternalxternal vvalidity.alidity. CConsideronsider a ppolicyolicy ooff eexpandingxpanding tthehe ssubsidyubsidy fforor schoolingschooling toto individualsindividuals thatthat doesdoes nnotot cchangehange tthehe cconstraintonstraint tthathat ttotalotal yyearsears ooff sschoolingchooling aarere ddeterminedetermined bbyy tthehe ssizesizes ooff schools.schools. TThenhen tthehe eestimatestimate ooff α fromfrom tthehe ppre-subsidyre-subsidy rregimeegime wwillill nnotot nnecessarilyecessarily iinformnform usus aboutabout thethe post-subsidypost-subsidy relationshiprelationship bbetweenetween costcost ooff sschoolingchooling aandnd yyearsears ooff sschoolingchooling aandnd wwillill nnotot ggiveive uuss aaccurateccurate ppredictionsredictions aaboutbout tthehe eeffectffect ooff tthehe ppolicy.olicy. TThehe problemproblem describeddescribed herehere isis ofof coursecourse a versionversion ofof thethe LucasLucas (1976)(1976) critiquecritique tthathat rreduced-formeduced-form rrelationshipselationships wwillill nnotot bbee sstabletable iinn tthehe ffaceace ooff ppolicyolicy iinterventions.nterventions. HHowever,owever, tthehe ddiscussioniscussion aalsolso hhighlightsighlights tthathat tthishis pproblemroblem iiss nnotot ssimplyimply ccircum-ircum- vventedented byby derivingderiving tthehe rrelationshipelationship ooff iinterestnterest ffromrom aann eeconomicconomic mmodel,odel, uunlessnless tthishis mmodelodel iincorporatesncorporates tthehe rrelevantelevant cconstraintsonstraints aandnd mmarginsargins ooff cchoice.hoice. IInn tthehe aabovebove eexample,xample, nnoo mmodelodel tthathat ffailsails ttoo iincorporatencorporate tthehe cconstraintonstraint oonn ttotalotal eenrollmentsnrollments wwillill bebe informativeinformative aaboutbout counterfactualscounterfactuals iinvolvingnvolving llarge-scalearge-scale iinterventions.nterventions. TThus,hus, oourur cconfionfi dencedence inin thethe impliedimplied answersanswers toto policypolicy experimentsexperiments cruciallycrucially ddependsepends onon oourur cconfionfi dencedence inin havinghaving ccapturedaptured thethe appropriateappropriate structuralstructural relationshiprelationship withwith thethe mmodelodel wewe areare estimating.estimating. HHowow ddoo wwee cconvinceonvince oothersthers aandnd oourselvesurselves tthathat oourur eestimatesstimates hhaveave eexternalxternal vvalidityalidity andand ccanan bbee uusedsed forfor policypolicy analysisanalysis oror forfor testingtesting ttheories?heories? TThishis iiss wwherehere eeconomicconomic ttheoryheory bbecomesecomes pparticularlyarticularly uuseful.seful. AsAs a fi rstrst step,step, wewe havehave toto defend—defend— uusingsing eeconomicconomic ttheory,heory, commoncommon sense,sense, andand evidence—thatevidence—that kkeyey ffactorsactors ppotentiallyotentially aaffectingffecting thethe responseresponse toto thethe relevantrelevant counterfactualcounterfactual areare accountedaccounted forfor andand thatthat tthehe mmodelodel aandnd tthehe ffunctionalunctional fformorm wwee cchosehose iindeedndeed capturecapture thethe salientsalient aspectsaspects ofof thethe rrealityeality andand areare a goodgood approximationapproximation toto thatthat rreality.eality. ThisThis inin turnturn involvesinvolves arguingarguing tthathat thethe functionalfunctional formform isis stablestable overover timetime andand acrossacross relevantrelevant samples,samples, thatthat varia-varia- ttionion acrossacross individualsindividuals notnot capturedcaptured byby thethe covariatescovariates andand thethe costcost ofof sschoolingchooling ccanan bbee incorporatedincorporated intointo thethe errorerror termterm εi , andand thatthat tthishis eerrorrror ttermerm ccanan bbee mmodeledodeled aass aadditivedditive aandnd oorthogonalrthogonal toto (that(that is,is, uuncorrelatedncorrelated with)with) thethe otherother variablesvariables includedincluded iinn tthehe eequation.quation. UsingUsing eeconomicconomic ttheoryheory iiss ooftenften tthehe bbestest wwayay ooff cclarifyinglarifying wwhetherhether kkeyey ffactorsactors havehave beenbeen omittedomitted aandnd wwhetherhether thethe underlyingunderlying aassumptionsssumptions ccanan bbee ddefendedefended andand whetherwhether theythey provideprovide a goodgood approximationapproximation toto reality.reality. HHowever,owever, thethe previousprevious discussiondiscussion alsoalso highlightshighlights tthathat sspecifyingpecifying a mmodelodel tthathat jjustifiustifi eses a specifispecifi c estimatingestimating equationequation isis typicallytypically notnot diffidiffi cult,cult, andand maymay notnot solvesolve tthehe underlyingunderlying pproblem.roblem. ForFor example,example, wwee ssawaw hhowow wwee ccouldould derivederive exactlyexactly tthehe samesame eestimatingstimating eequationquation fromfrom a mmodelodel ooff iindividualndividual sschoolingchooling cchoice;hoice; bbutut iiff iinn rrealityeality
4 More specifi cally,_ the constraint _on school enrollments might imply that total years of schooling –1/σ should be equal to S that is, ∑i si = S . Suppose that the economic relationship si = K(ζi c i ) still holds at the individual level (i.e., individuals with low cost of schooling get proportionately more of the avail- able school resources). But it must do so with_ a different value of K than in footnote 2. In particular, the – 1/σ – 1 constraint on total schooling implies K = S (∑i ∈ (ζi c i ) ) . When the cost of schooling is subsidized, the underlying economic relationship with the new defi nition of K given here remains unchanged, but the reduced-form relationship captured by our estimating equation above changes (exactly as shown by the above formula for K). Theory, General Equilibrium, and Political Economy in Development Economics 21
yyearsears ooff sschoolingchooling aarere cconstrainedonstrained bbyy tthehe ssizesizes ofof sschools,chools, tthehe eestimatesstimates ooff α wwillill sstilltill nnotot bbee uusefulseful forfor understandingunderstanding thethe implicationsimplications ooff a llarge-scalearge-scale ssubsidyubsidy fforor sschooling.chooling. TThehe problemproblem ofof coursecourse isis thatthat forfor studyingstudying thethe implicationsimplications ofof thisthis typetype ofof ppolicy,olicy, tthehe cconstraintsonstraints rresultingesulting fromfrom tthehe ssizesizes ooff sschoolschools aarere ccentral,entral, aandnd aanyny mmodelodel tthathat ddoesoes nnotot rrecognizeecognize thesethese constraintsconstraints willwill notnot bebe helpfulhelpful inin suchsuch a study.study. ThisThis eemphasizesmphasizes tthathat tthehe pproperroper uusese ooff eeconomicconomic ttheoryheory ddoesoes nnotot mmeanean wwritingriting ddownown a sspecifipecifi c model;model; iinstead,nstead, itit requiresrequires thatthat wewe incorporateincorporate thethe appropriateappropriate constraintsconstraints aandnd marginsmargins ofof adjustments,adjustments, tthathat wwee ddevelopevelop tthehe ccasease tthathat eeconomicconomic ttheoryheory robustly lleadseads ttoo tthehe eestimatingstimating eequationquation iinn qquestion,uestion, aandnd tthathat wwee cclarifylarify wwhichhich iimportantmportant eeconomicconomic mmechanismsechanisms aandnd eeffectsffects areare beingbeing excludedexcluded fromfrom tthehe mmodel.odel.5 AAnothernother aadvantagedvantage ooff sstructuraltructural rreasoningeasoning bbasedased oonn ttheoryheory isis thatthat onceonce wewe gogo tthroughhrough thethe processprocess ofof explicitlyexplicitly justifyingjustifying tthehe eequationquation wewe areare estimating,estimating, eeitherither uusingsing eeconomicconomic ttheoryheory oror otherother ttheoreticalheoretical oror empiricalempirical aarguments,rguments, wwee mmayay rrealizeealize tthathat suchsuch anan equationequation cannotcannot easilyeasily bebe defended.defended. InIn suchsuch cases,cases, itit hashas toto bebe inter-inter- ppretedreted wwithith greatergreater caution,caution, oror perhapsperhaps itit hashas toto bebe modifimodifi eedd oorr aabandoned.bandoned. TThishis aadvantagedvantage bbecomesecomes pparticularlyarticularly iimportantmportant iinn ccontextsontexts wwherehere generalgeneral eequilibriumquilibrium aandnd ppoliticalolitical eeconomyconomy eeffectsffects areare present.present. Finally,Finally, economiceconomic theorytheory providesprovides thethe bbestest wwayay ooff iinterpretingnterpreting wwhathat tthehe eestimatesstimates ffromrom aann eequation,quation, ssuchuch aass tthehe oonene wwee sstartedtarted wwith,ith, mmean.ean. FForor eexample,xample, wwhenhen tthishis eequationquation iiss dderivederived fromfrom tthehe eeconomicconomic mmodelodel above,above, wwee understandunderstand thatthat α = 11//σ iiss a functionfunction ofof tthehe eelasticitylasticity ooff tthehe hhumanuman capitalcapital pproductionroduction ffunction.unction. TThehe structuralstructural aapproachpproach alsoalso facesfaces majormajor cchallenges,hallenges, hhowever.owever. FFirst,irst, aass aalreadylready eemphasized,mphasized, writingwriting downdown a modelmodel likelike thethe oneone describeddescribed aboveabove isis clearlyclearly notnot ssuffiuffi cientcient fforor aachievingchieving eexternalxternal vvalidity.alidity. TThathat mmodelodel iitselftself mmadeade sseveraleveral aassump-ssump- ttionsions whichwhich areare restrictiverestrictive andand maymay notnot provideprovide a goodgood aapproximationpproximation ttoo tthehe eeconomicconomic pphenomenahenomena iinn wwhichhich wwee aarere iinterested.nterested. TThishis iiss aagaingain iillustratedllustrated byby thethe aabovebove example,example, wwhichhich showedshowed thatthat oneone mightmight endend upup derivingderiving thethe samesame estimatingestimating eequationquation fromfrom a ttheoreticalheoretical mmodel,odel, aandnd tthushus rreacheach tthehe ssameame cconclusionsonclusions aaboutbout tthehe implicationsimplications ooff a ccounterfactualounterfactual ppolicyolicy cchange,hange, aass oonene mmightight havehave donedone byby justjust sspecifyingpecifying a rreduced-formeduced-form eequation.quation. SSecond,econd, wwee mmayay iinn ffactact questionquestion whetherwhether therethere isis anyany groundground forfor assumingassuming a cconstantonstant elasticityelasticity α bbetweenetween yearsyears ofof schoolingschooling andand costscosts ofof sschooling.chooling. AAfterfter aall,ll, wwee knowknow thatthat aallll ttheoriesheories areare abstractionsabstractions andand approximations,approximations, ssoo ttherehere iiss llittleittle rreasoneason toto believebelieve thatthat a parameterparameter suchsuch asas α——oror thethe intertemporalintertemporal elasticityelasticity ofof
5 The online appendix available with this paper at ⟨http://www.e-jep.org⟩ discusses some issues that arise in thinking about how we could develop such robust predictions and how we could try to map them to data. This discussion also highlights that in certain cases one could achieve counterfactual validity without much theory. For example, we need only the most basic theory in interpreting a controlled experiment designed to evaluate the effectiveness of a drug. In this case, we can say that common sense and a very limited amount of medical theory are suffi cient to interpret the results of the controlled experiment and decide whether they are informative about the effectiveness of the drug in question beyond the experimental setting. It should also be noted that the evaluation of the effective- ness of a drug in this example has a clear parallel to “modeling individual behavior” in economics. As further discussed below, the role of economic theory becomes even more central when our focus shifts to “modeling equilibrium behavior.” 22 Journal of Economic Perspectives
ssubstitution,ubstitution, oror thethe FrischFrisch elasticityelasticity ofof laborlabor supply,supply, oror thethe eelasticitylasticity ooff ssubstitutionubstitution bbetweenetween twotwo ffactors,actors, oror anyany ootherther MMarschakianarschakian ppreferencereference oror technologytechnology pparam-aram- eeter—shouldter—should bbee rreallyeally constant.constant. ButBut withoutwithout suchsuch constancy,constancy, therethere aarere ssevereevere llimitsimits ttoo externalexternal vvalidity.alidity. FFinally,inally, oonene maymay eveneven questionquestion thethe existenceexistence oorr uusefulnesssefulness ooff ““structuralstructural pparameters”arameters” altogether.altogether. WWhathat wwee ttakeake aass a sstructuraltructural pparameterarameter fforor oonene ttheoryheory wwillill naturallynaturally becomebecome anan endogenousendogenous oobjectbject iinn aanother.nother. SSoo a pparticulararticular mmodelodel ccanan sserveerve usus wellwell asas anan abstractionabstraction forfor a seriesseries ofof counterfactualcounterfactual experiments,experiments, bbutut ttherehere wwillill existexist ootherther eexperimentsxperiments fforor wwhichhich itit willwill bebe muchmuch lessless informative.informative. ForFor example,example, aann elasticityelasticity ofof substitutionsubstitution oror certaincertain technologytechnology parametersparameters maymay bebe constantconstant withwith rrespectespect toto certaincertain variations,variations, butbut wouldwould changechange inin responseresponse toto others.others. ThisThis isis almostalmost bbyy nnecessity:ecessity: a ppreconditionrecondition forfor externalexternal validityvalidity iiss tthathat kkeyey ffactorsactors rrelevantelevant fforor tthehe ooutcomeutcome ooff tthehe ccounterfactualounterfactual sshouldhould bbee iincludedncluded iinn tthehe model,model, andand modelsmodels asas aabstractionsbstractions havehave toto excludeexclude sseveraleveral rrelevantelevant ffactors,actors, ssoo nnoo ssingleingle mmodelodel ccanan iincludenclude aallll ofof tthehe relevantrelevant factorsfactors forfor allall possiblepossible counterfactualcounterfactual exercises.exercises. TThesehese cchallengeshallenges nnotwithstanding,otwithstanding, iitt iiss cclearlear tthathat eeconomistsconomists ooftenften hhaveave ttoo ttakeake a ppositionosition aboutabout thethe parametersparameters beingbeing estimatedestimated correspondingcorresponding toto structuralstructural pparametersarameters ((atat lleasteast fforor a wwell-defiell-defi nedned thoughthough perhapsperhaps limitedlimited setset ofof variationsvariations iinn eenvironmentnvironment aandnd ppolicy).olicy). OOtherwise,therwise, wwee wwillill hhaveave nnoo wwayay ooff pperformingerforming ccounter-ounter- ffactualactual exercisesexercises aandnd mmakingaking ppredictionsredictions aaboutbout policypolicy changeschanges ((Imbens,Imbens, 22009).009). BButut tthishis nnecessitatesecessitates a cclaimlaim ttoo eexternalxternal vvalidityalidity ((eveneven iiff iitt iiss oonlynly implicit),implicit), aandnd eeconomicconomic ttheoryheory isis ourour bestbest guideguide fforor fformulatingormulating tthehe aappropriateppropriate mmodelsodels aandnd jjustifyingustifying ssuchuch cclaimslaims ttoo eexternalxternal vvalidity.alidity. TThesehese iissuesssues bbecomeecome oonlynly mmoreore ccentralentral iinn thethe ppresenceresence ooff ggeneraleneral eequilibriumquilibrium eeffectsffects andand ppoliticalolitical eeconomyconomy ffactors,actors, ttoo wwhichhich I tturnurn nnext.ext.
The Centrality of General Equilibrium
TThehe bulkbulk ofof empiricalempirical workwork usingusing microdata,microdata, particularlyparticularly inin developmentdevelopment eeconomics,conomics, eengagesngages iinn ppartialartial eequilibriumquilibrium ccomparisons.omparisons. DDependingepending oonn mmagni-agni- ttudesudes ooff vvariousarious eeffects,ffects, generalgeneral eequilibriumquilibrium interactionsinteractions cancan offsetoffset oror eveneven reversereverse ssensibleensible partialpartial equilibriumequilibrium conclusions.conclusions. However,However, mostmost empiricalempirical strategiesstrategies ddoo nnotot ddirectlyirectly estimateestimate generalgeneral eequilibriumquilibrium eeffects.ffects.6 EEconomicconomic ttheoryheory nnonethelessonetheless pprovidesrovides ssomeome gguidanceuidance iinn aassessingssessing tthehe iimpor-mpor- ttanceance ooff ggeneraleneral eequilibriumquilibrium eeffects.ffects. TThreehree ttypesypes ooff ggeneraleneral eequilibriumquilibrium eeffects,ffects, wwhichhich areare usuallyusually notnot estimatedestimated inin partialpartial equilibriumequilibrium comparisons,comparisons, areare potentiallypotentially iimportant.mportant. FFirst,irst, iinn rresponseesponse ttoo llargearge ppolicyolicy iinterventionsnterventions oorr sshocks,hocks, iimperfectmperfect ssubstitutionubstitution betweenbetween factorsfactors andand diminishingdiminishing returnsreturns implyimply tthathat ffactoractor pproductivitiesroductivities
6 See also Townsend (forthcoming) for a complementary discussion of the role of general equilibrium analysis in development economics, with special emphasis on credit market issues; Heckman, Lochner, and Taber (1998) for a discussion of general equilibrium issues in the analysis of the effects of tech- nology on wage inequality; and Dufl o (2004a) for a discussion of other diffi culties in “scaling up” policy interventions evaluated using microdata. Daron Acemoglu 23
aandnd pricesprices willwill change.change. Second,Second, thethe samesame policypolicy interventionsinterventions oror shocksshocks ccanan leadlead toto eendogenousndogenous ttechnologyechnology rresponses.esponses. TThird,hird, ttherehere mmayay bbee ccompositionomposition eeffectsffects rresultingesulting ffromrom eequilibriumquilibrium ssubstitutionubstitution ooff ssomeome ffactorsactors oorr pproductsroducts fforor oothersthers ((wherebywhereby tthehe ccompositionomposition ofof mmicroicro unitsunits changeschanges ddifferentlyifferently iinn rresponseesponse ttoo ddifferentifferent ttypesypes ooff iinter-nter- vventions).entions). TTheoryheory generallygenerally impliesimplies tthathat tthehe fi rrstst aandnd tthehe tthirdhird eeffectsffects wwillill ttendend ttoo ppartiallyartially ooffsetffset oorr eevenven rreverseeverse ddirectirect ppartialartial eequilibriumquilibrium eeffects,ffects, whilewhile endogenousendogenous ttechnologyechnology responsesresponses couldcould eithereither ddampenampen oorr mmagnifyagnify tthemhem ((seesee AAcemoglu,cemoglu, 2007,2007, fforor ggeneraleneral ttheoreticalheoretical rresultsesults onon endogenousendogenous ttechnology).echnology). AAss anan exampleexample ooff ffactoractor ppricerice changes,changes, considerconsider tthehe pproblemroblem ofof estimatingestimating tthehe rreturnseturns ttoo sschooling.chooling. TThishis iiss ttypicallyypically ddoneone bbyy ffocusingocusing oonn a ssmallmall ggrouproup ooff iindividualsndividuals wwhoho areare inducedinduced toto remainremain inin schoolschool forfor longerlonger aandnd ccomparingomparing tthemhem ttoo otherother iindividualsndividuals iinn tthehe ssameame mmarketarket ((whowho tthushus ffaceace tthehe ssameame pprices)rices) wwhoho havehave ddroppedropped outout ofof school.school. TheThe implicitimplicit aassumptionssumption herehere isis thatthat alteringaltering sschoolingchooling ddecisionsecisions willwill notnot generategenerate changeschanges iinn mmarketarket pprices.rices. BButut forfor manymany ofof tthehe qquestionsuestions rrelevantelevant forfor developmentdevelopment eeconomics,conomics, wwee wwishish ttoo tthinkhink ooff ccounterfactualsounterfactuals iinn wwhichhich a largelarge fractionfraction ofof thethe populationpopulation acquiresacquires moremore schooling.schooling. InIn thisthis case,case, itit isis nono llongeronger pplausiblelausible ttoo aassumessume tthathat ppricesrices wwillill nnecessarilyecessarily rremainemain cconstant.onstant. IImperfectmperfect ssubstitutionubstitution betweenbetween ddifferentifferent sskillkill llevelsevels wwillill ttypicallyypically iimplymply tthathat aann iincreasencrease iinn tthehe sschoolingchooling llevelevel ooff a ssignifiignifi cantcant fractionfraction ooff tthehe ppopulationopulation mmayay rreduceeduce thethe returnreturn ttoo sschooling.chooling. FForor example,example, AAngristngrist ((1995)1995) sshowshows tthathat tthehe largelarge school-buildingschool-building pprogramsrograms inin thethe PalestinianPalestinian territoriesterritories ledled toto a sharpsharp dropdrop inin thethe skillskill ppremium.remium. AAss anan exampleexample ooff eendogenousndogenous ttechnologyechnology rresponses,esponses, cconsideronsider tthehe llargearge iincreasencrease iinn tthehe rrelativeelative ssupplyupply ooff ccollege-educatedollege-educated wworkersorkers iinn tthehe UUnitednited SStatestates sstartingtarting iinn tthehe latelate 1960s.1960s. GivenGiven technology,technology, thisthis changechange inin relativerelative supplysupply shouldshould havehave reducedreduced tthehe collegecollege premium.premium. AsAs isis wellwell known,known, thethe oppositeopposite happenedhappened inin practice,practice, andand tthehe collegecollege premiumpremium increasedincreased sharplysharply ffromrom tthehe llateate 11970s970s oonwards.nwards. IInn AAcemoglucemoglu ((1998),1998), I argue,argue, forfor example,example, tthathat tthishis wwasas a cconsequenceonsequence ooff tthehe endogenousendogenous rresponseesponse ooff technologytechnology toto tthehe rrelativeelative aabundancebundance ooff mmoreore sskilledkilled wworkers.orkers. TThehe ssameame rreasoningeasoning iimpliesmplies tthathat iinn eevaluatingvaluating tthehe eeffectffect ooff ttraderade oopening,pening, oonene ccouldould nnotot ssimplyimply rrelyely oonn ppartialartial equilibriumequilibrium estimatesestimates derivedderived fromfrom fi rm-levelrm-level variationvariation iinn accessaccess ttoo fforeignoreign mmarkets,arkets, sincesince tradetrade openingopening isis a generalgeneral eequilibriumquilibrium cchangehange tthathat wwillill aalsolso aaffectffect ttechnologyechnology cchoiceshoices andand thethe ddirectionirection ofof technologicaltechnological change.change. AAss anan exampleexample ooff ccompositionomposition eeffects,ffects, considerconsider tthehe pproblemroblem ofof estimatingestimating tthehe iimportancemportance ooff ccreditredit mmarketarket iimperfections.mperfections. BBanerjeeanerjee aandnd DDuflufl o ((2005)2005) ssurveyurvey a llargearge bodybody ofof evidenceevidence thatthat smallsmall andand medium-sizedmedium-sized bbusinessesusinesses iinn lless-developedess-developed eeconomiesconomies aarere ccreditredit constrainedconstrained aandnd tthathat aann eextensionxtension ooff ccreditredit ttoo tthesehese bbusinessesusinesses wwillill makemake tthemhem iincreasencrease production.production. NNowow cconsideronsider tthehe eeffectffect ofof a llarge-scalearge-scale ppolicyolicy ooff ccreditredit expansionexpansion ttoo ssmall-mall- andand medium-sizedmedium-sized businesses.businesses. ThisThis policypolicy couldcould leadlead ttoo a ddifferentifferent ttypeype ooff ccompositionomposition eeffectffect tthanhan tthehe oonene ooperatingperating iinn ppartialartial eequilib-quilib- rrium.ium. ForFor example,example, iitt mmayay bbee tthehe ccasease tthathat iinn ppartialartial eequilibriumquilibrium eestimationstimation ffocusingocusing oonn fi rrm-levelm-level vvariationariation wwee ffoundound thatthat fi rmsrms withwith betterbetter accessaccess toto creditcredit expanded,expanded, bbutut tthishis wwasas at the expense ooff ootherther fi rrmsms tthathat ddidid nnotot hhaveave aaccessccess ttoo ccreditredit ((thatthat iis,s, ppartlyartly bbyy ““stealingstealing bbusiness”usiness” ffromrom oothers).thers). TThehe ssameame rresponseesponse ccannotannot ttakeake pplacelace iinn ggeneraleneral eequilibrium.quilibrium. AAss a cconsequence,onsequence, wwhenhen additionaladditional creditcredit becomesbecomes availableavailable ttoo 24 Journal of Economic Perspectives
a largelarge fractionfraction ofof fi rms,rms, totaltotal outputoutput maymay notnot increaseincrease byby asas muchmuch oror atat all.all. OneOne couldcould tthushus iimaginemagine a ssituationituation iinn wwhichhich partialpartial equilibriumequilibrium estimatesestimates ofof rrelaxingelaxing ccreditredit cconstraintsonstraints aarere llarge,arge, wwhilehile thethe generalgeneral eequilibriumquilibrium eeffectsffects wouldwould bebe small.small. I nnowow ffurtherurther eelaboratelaborate thethe fi rstrst ggeneraleneral eequilibriumquilibrium eeffect,ffect, wworkingorking tthroughhrough eendogenousndogenous ffactoractor ppricesrices aandnd ddiminishingiminishing rreturns,eturns, iinn tthehe ccontextontext ooff tthehe eeffectffect ooff lifelife expectancyexpectancy ((andand hhealth)ealth) oonn eeconomicconomic ggrowth.rowth. A llargearge mmicroeconometricicroeconometric lliteratureiterature showsshows tthathat hhealthierealthier iindividualsndividuals aarere mmoreore pproductive:roductive: ssee,ee, aamongmong oothers,thers, BBehrmanehrman aandnd RRosenzweigosenzweig (2004),(2004), SchultzSchultz (2002),(2002), andand SStrausstrauss andand ThomasThomas ((1998).1998). OOnn tthishis basis,basis, wwee wwouldould eexpectxpect aann iincreasencrease iinn tthehe llifeife eexpectancyxpectancy ooff tthehe wworkforceorkforce ttoo leadlead toto greatergreater aggregateaggregate productivity.productivity. ButBut oneone shouldshould ttakeake ggeneraleneral eequilibriumquilibrium eeffectsffects iintonto aaccount,ccount, ssinceince aann iincreasencrease iinn llifeife eexpectancyxpectancy aalsolso iincreasesncreases ppopula-opula- ttion,ion, andand becausebecause ofof diminishingdiminishing rreturnseturns ttoo ccapitalapital aandnd lland,and, iitt mmayay ddecreaseecrease llaborabor pproductivityroductivity aandnd mmayay iinn ffactact reducereduce incomeincome perper capita.capita. HowHow couldcould oneone investigateinvestigate wwhetherhether thesethese generalgeneral eequilibriumquilibrium eeffectsffects areare iimportant?mportant? OOnene aapproachpproach iiss ttoo uusese iinformationnformation ffromrom ootherther ssourcesources iinn oorderrder ttoo ““calibrate”calibrate” tthehe vvaluesalues ooff tthehe pparam-aram- eetersters aandnd tthenhen ccombineombine tthishis wwithith mmicroicro eestimatesstimates ooff tthehe eeffectffect ofof hhealthealth aandnd llifeife eexpectancyxpectancy onon individualindividual outcomes.outcomes.7 ThisThis approachapproach willwill bebe successfulsuccessful whenwhen wewe hhaveave conficonfi ddenceence inin thethe calibrationcalibration exercise.exercise. A ssecondecond aapproachpproach iiss ttoo uusese ccross-ross-c countrountry vvariation,ariation, eevenven tthoughhough ssuchuch vvariationariation wwillill bbee aaffectedffected byby severalseveral potentiallypotentially omittedomitted ffactors.actors. InIn AcemogluAcemoglu andand JohnsonJohnson ((2007),2007), mymy coauthorcoauthor aandnd I aadoptdopt tthishis aapproach.pproach. WWee dderiveerive tthehe ffollowingollowing llinearinear rrelationshipelationship bbetweenetween loglog lifelife expectancy,expectancy, xiitt , andand loglog incomeincome perper capita,capita, yiitt , fromfrom a neoclassicalneoclassical growthgrowth modelmodel andand thethe possibilitypossibility thatthat lifelife expectancyexpectancy mmightight hhaveave a ddirectirect positivepositive effecteffect onon technologytechnology andand onon humanhuman capital:capital:
yit = πxit + ζi + t + εit .
The parameter of interest, π, measures the relationship between log income per capita and log life expectancy. In addition, the ζi’s denote a full set of country fi xed effects capturing any time-invariant cross-country differences, the t’s denote a full set of time effects, and εit is an error term capturing all omitted factors. Though this equation can be estimated by ordinary least squares, this is likely to lead to biased estimates of π since societies that are successful in solving economic and institu- tional problems to achieve higher growth are also likely to provide better public health and other measures that improve life expectancy, and also the increase in income per capita is likely to lead to a mechanical improvement in life expectancy.
7 This is the approach advocated by Banerjee and Dufl o (2005) and used by Weil (2007) in the context of health and economic development and by Heckman, Lochner, and Taber (1998) in the context of the relationship between technology and wage inequality. Another approach not mentioned in the text, perhaps most promising, is to combine microdata with regional variation to estimate partial and general equilibrium effects simultaneously. This approach is adopted and developed in Acemoglu and Angrist (2000) to estimate human capital externalities exploiting individual-level differences in schooling together with state-wide differences in average schooling. See also Dufl o (2004b) for an application to Indonesian data and Acemoglu, Autor, and Lyle (2004), where we estimate the general equilibrium effects of increased female labor supply (on male and female wages). Theory, General Equilibrium, and Political Economy in Development Economics 25
TToo overcomeovercome tthishis pproblem,roblem, iinn AAcemoglucemoglu aandnd JJohnsonohnson ((2007),2007), wwee aadoptdopt aann iinstrumentalnstrumental vvariablesariables strategy,strategy, exploitingexploiting ggloballobal ddiscoveriesiscoveries aandnd diffusiondiffusion ofof mmajorajor ddrugs,rugs, chemicals,chemicals, andand publicpublic healthhealth technologies.technologies. TheThe ideaidea isis thatthat thesethese iimprovementsmprovements sshouldhould hhaveave rraisedaised llifeife eexpectancyxpectancy ddifferentiallyifferentially iinn ccountriesountries tthathat wwereere ssubjectubject toto differentdifferent typestypes ofof initialinitial diseasedisease burdens.burdens. ToTo implementimplement thisthis idea,idea, wwee constructconstruct a ““predictedpredicted mortality”mortality” variable,variable, Miitt , bbasedased oonn tthehe 1155 mmostost iinfectiousnfectious ddiseasesiseases iinn 11940.940. WWee ccomputeompute tthehe ppre-interventionre-intervention ((1940)1940) mmortalityortality ffromrom eeachach ooff tthesehese 1515 diseasesdiseases inin eacheach country.country. Then,Then, whenwhen a globalglobal healthhealth interventionintervention (techno-(techno- llogicalogical bbreakthrough)reakthrough) ttakesakes pplacelace fforor a ggiveniven ddisease,isease, ppredictedredicted mmortalityortality iinn eeachach ccountryountry fallsfalls toto a differentdifferent levellevel dependingdepending onon itsits pre-interventionpre-intervention mortalitymortality fromfrom tthathat disease.disease. MoreMore specifispecifi cally,cally, thethe ppredictedredicted mortalitymortality vvariableariable uusesses a ccountry’sountry’s iinitialnitial mmortalityortality rrateate fforor eeachach ofof thethe 1515 diseasesdiseases untiluntil therethere isis a globalglobal iintervention,ntervention, aandnd afterafter thethe globalglobal iintervention,ntervention, tthehe mmortalityortality rrateate fforor tthehe ddiseaseisease inin questionquestion 8 ddeclineseclines toto thethe frontierfrontier mortalitymortality rrate.ate. PPredictedredicted mmortality,ortality, Miitt , iiss thenthen usedused aass aann iinstrumentnstrument fforor llogog llifeife eexpectancyxpectancy iinn tthehe eestimationstimation ooff tthehe rrelationshipelationship bbetweenetween llogog incomeincome perper capitacapita aandnd llogog lifelife expectancy.expectancy. WWithith tthishis rreasoning,easoning, tthehe fi rst-stagerst-stage rrelationshipelationship iiss
~ ~ xit = ψMit + ζ i + t + uit ,
~ ~ where the ζ i’s again denote a full set of country fi xed effects, the t’s denote a full set of time effects, and uit is the error term. For this instrumental variables approach to be valid, the key exclusion restriction for the estimation strategy is that the covariance between the predicted mortality variable and the error term in the earlier income per capita equation, εit , must be zero (that is, the covari- ance between Mit and εit must equal zero). Note that both the second and the fi rst stages (the exclusion restriction) are motivated by theory. The second stage is derived from the neoclassical growth model. The fi rst stage (and thus the exclusion restriction that the covariance between Mit and εit must equal zero) is predicated on the theory that global intervention for a particular disease will affect mortality in a country in proportion with the number of initial deaths from the disease in question in that country, and more importantly, that baseline levels of mortality from different diseases do not have a direct effect on future income beyond their effect working through future life expectancy and health condi- tions. In Acemoglu and Johnson (2007), we provide evidence consistent with this exclusion restriction. For example, prior to 1940 predicted mortality does not
8 Mathematically, predicted mortality is defi ned as
M = ∑ ((1 – I ) M + I M ), it dt dit 0 dt dFt d∈ where Mdit denotes mortality in country i from disease d at time t, Idt is a dummy for intervention for disease d at time t (it is equal to 1 for all dates after the intervention), denotes the set of the 15 infec- tious diseases, M dit 0 refers to the pre-intervention (1940) mortality from disease d in the same units, and MdFt is the mortality rate from disease d at the health frontier of the world at time t. 26 Journal of Economic Perspectives
predict future income or population growth, which is consistent with the notion that past levels of life expectancy do not have a direct effect on future growth. The online appendix available with this paper at ⟨http://www.e-jep.org⟩ discusses why the specifi c instrumental variables strategy suggested here is only valid with certain formulations of the second-stage equation. Different theories for the relationship between health and growth, encapsulated in different second-stage relationships, may not be consistent with the same exclusion restriction, thus further emphasizing the role of theory in guiding the estimation strategy. TThehe surprisingsurprising fi nndingding inin AcemogluAcemoglu andand JohnsonJohnson (2007)(2007) isis that,that, despitedespite thethe wwell-establishedell-established ppositiveositive mmicroicro eestimatesstimates ooff tthehe eeffectffect ofof healthhealth onon productivity,productivity, inin ggeneraleneral eequilibriumquilibrium tthehe eeffectffect onon incomeincome perper capitacapita appearsappears toto bebe negative.negative. ThisThis rresultesult probablyprobably arisesarises bbecauseecause tthehe iimprovementsmprovements iinn llifeife eexpectancyxpectancy wwereere aassociatedssociated wwithith veryvery largelarge iincreasesncreases inin population.population. WWhilehile tthishis cconclusiononclusion comescomes withwith severalseveral ccaveats,aveats, nnotot lleasteast bbecauseecause thethe negativenegative estimatesestimates areare oftenoften qquiteuite largelarge andand comecome ffromrom a specifispecifi c eepisodepisode ((duringduring wwhichhich mmortalityortality rratesates mmayay hhaveave ddeclinedeclined uunusuallynusually rrapidlyapidly rrelativeelative ttoo mmorbidityorbidity rrates),ates), iitt iillustratesllustrates tthehe ppossibilityossibility tthathat ggeneraleneral eequilib-quilib- rriumium eempiricalmpirical cconclusionsonclusions cancan bebe quitequite differentdifferent fromfrom ppartialartial eequilibriumquilibrium oones.nes.9 IItt reiteratesreiterates tthehe iimportancemportance ooff iincorporatingncorporating ggeneraleneral eequilibriumquilibrium considerationsconsiderations wwhenhen conductingconducting ccounterfactualounterfactual eexercisesxercises cconcerningoncerning tthehe eeffectsffects ooff llargearge cchangeshanges iinn vvariablesariables ssuchuch aass sschooling,chooling, healthhealth conditions,conditions, oorr aaccessccess toto creditcredit onon incomeincome perper ccapitaapita oorr ootherther aaspectsspects ofof economiceconomic development.development.
No Development without Political Economy
TTherehere isis increasingincreasing recognitionrecognition thatthat institutionalinstitutional andand politicalpolitical economyeconomy ffactorsactors areare centralcentral toto economiceconomic development.development. ManyMany problemsproblems ofof developmentdevelopment rresultesult fromfrom bbarriersarriers ttoo tthehe aadoptiondoption ooff nnewew ttechnologies,echnologies, llackack ooff ppropertyroperty rrightsights ooverver lland,and, llabor,abor, aandnd bbusinesses,usinesses, aandnd ppoliciesolicies distortingdistorting pricesprices andand incentives.incentives. TThesehese institutionsinstitutions aandnd ppoliciesolicies aarere notnot inin placeplace eexclusively,xclusively, oorr eevenven pprimarily,rimarily, bbecauseecause ofof a lacklack ooff uunderstandingnderstanding ofof economiceconomic principlesprinciples oonn tthehe ppartart ooff ppolicy-olicy- mmakers.akers. TTypically,ypically, ppolicymakersolicymakers introduceintroduce oror maintainmaintain suchsuch policiespolicies toto rremainemain iinn power,power, oorr ttoo eenrichnrich tthemselves,hemselves, oorr bbecauseecause ppoliticallyolitically ppowerfulowerful eeliteslites oopposeppose tthehe entryentry ofof rrivals,ivals, tthehe iintroductionntroduction ooff nnewew ttechnologies,echnologies, oorr iimprovementsmprovements iinn tthehe ppropertyroperty rrightsights ooff theirtheir workersworkers oror competitorscompetitors (for(for example,example, AAcemoglu,cemoglu, Johnson,Johnson, aandnd RRobinson,obinson, 2005a).2005a). ButBut thisthis perspectiveperspective impliesimplies tthathat ttheoryheory aagaingain bbecomesecomes pparticularlyarticularly importantimportant inin evaluatingevaluating ((oror fframing)raming) ppossibleossible eeffectsffects ooff llarge-scalearge-scale ppolicyolicy interventions.interventions. CCounterfactualounterfactual aanalysesnalyses tthathat iignoregnore ppoliticalolitical eeconomyconomy ffactors,actors, likelike thosethose thatthat ddoo nnotot ttakeake aaccountccount ooff ggeneraleneral eequilibriumquilibrium eeffects,ffects, mmayay ggiveive misleadingmisleading answers.answers. IInn tthishis ccase,ase, cconvincingonvincing mmicroicro oorr eevenven mmacroacro ((generalgeneral
9 The conclusions may also depend on the fact that in Acemoglu and Johnson (2007), we focus on changes in health largely (though not solely) associated with mortality. Bleakley (2007), focusing on changes related to morbidity, obtains different results. Daron Acemoglu 27
eequilibrium)quilibrium) evidenceevidence aaboutbout tthehe eeffectsffects ofof a particularparticular policypolicy changechange onon economiceconomic ooutcomesutcomes iiss nnotot iinn itselfitself suffisuffi ccientient ttoo ggaugeauge whatwhat thethe implicationsimplications wwillill bbee wwhenhen ssuchuch a policypolicy isis encouragedencouraged oror implemented.implemented. TThehe experienceexperience ooff GGhanahana withwith exchangeexchange rrateate ppolicyolicy uundernder PrimePrime MinisterMinister KKofiofi BBusiausia inin 19711971 providesprovides a sharpsharp iillustration.llustration. BBusiausia pursuedpursued expansionaryexpansionary eeconomicconomic ppoliciesolicies afterafter ccomingoming ttoo ppowerower iinn 11969,969, aandnd mmaintainedaintained vvariousarious ppricerice ccontrolsontrols aandnd anan overvaluedovervalued exchangeexchange rrate.ate. BButut GGhanahana wwasas ssoonoon ssufferinguffering fromfrom a sserieseries ooff bbalancealance ofof paymentspayments crisescrises andand foreignforeign exchangeexchange sshortages.hortages. FFacedaced wwithith tthesehese ccrises,rises, BBusiausia signedsigned anan agreementagreement withwith thethe IMFIMF onon DecemberDecember 27,27, 1971,1971, wwhichhich includedincluded a massivemassive devaluationdevaluation ooff tthehe ccurrency.urrency. A ffewew ddaysays ffollowingollowing tthehe aannouncementnnouncement ooff tthehe ddevaluation,evaluation, BBusiausia wwasas ooverthrownverthrown bbyy tthehe mmilitaryilitary lleded bbyy LLt.t. CCol.ol. AAcheam-cheam- ppong,ong, wwhoho iimmediatelymmediately reversedreversed thethe devaluationdevaluation (see,(see, forfor example,example, HHerbst,erbst, 11993;993; BBoafo-Arthur,oafo-Arthur, 1999).1999). ThereThere waswas littlelittle doubtdoubt thatthat devaluationdevaluation waswas goodgood economicseconomics iinn Ghana.Ghana. ButBut itit waswas notnot goodgood politics.politics. StateState controlscontrols overover prices,prices, wages,wages, marketingmarketing bboards,oards, andand exchangeexchange rratesates wwereere aann iimportantmportant ppartart ooff tthehe ppatronageatronage nnetwork,etwork, aandnd aanyny politicianpolitician whowho lostlost tthehe ssupportupport ofof thisthis networknetwork waswas susceptiblesusceptible bbothoth aatt tthehe ppollsolls aandnd aagainstgainst tthehe mmilitary.ilitary. BBusiausia ssuffereduffered thisthis fate.fate. TThishis eepisodepisode iillustratesllustrates a ggeneraleneral ppoint:oint: wwhenhen ppoliticalolitical eeconomyconomy ffactorsactors aarere iimportant,mportant, eevidencevidence oonn tthehe eeconomicconomic eeffectsffects ofof large-scalelarge-scale policypolicy changeschanges uundernder a ggiveniven ssetet ooff ppoliticalolitical cconditionsonditions iiss nnotot ssuffiuffi ccientient toto forecastforecast ttheirheir eeffectffect onon thethe eeconomyconomy andand society.society. ThisThis principleprinciple doesdoes notnot justjust applyapply toto exchangeexchange rrateate ppolicy.olicy. FForor example,example, thethe ffactact tthathat iincreasingncreasing aavailabilityvailability ooff ccreditredit ttoo fi rmsrms wwouldould iincreasencrease aaggregateggregate outputoutput givengiven allall otherother policiespolicies doesdoes notnot implyimply tthathat aann aactualctual rreformeform ofof tthehe ccreditredit marketmarket willwill necessarilynecessarily work.work. CConsistentonsistent wwithith tthishis pperspective,erspective, HaberHaber aandnd PerottiPerotti (2008)(2008) argueargue andand provideprovide evidenceevidence tthathat limitinglimiting aaccessccess ttoo fi nancenance isis a powerfulpowerful tooltool inin thethe handshands ofof politicalpolitical andand economiceconomic eeliteslites fforor restrictingrestricting eentryntry iintonto lucrativelucrative bbusinesses.usinesses. TThus,hus, rreformseforms ofof ccreditredit marketsmarkets willwill oftenoften faceface politicalpolitical ooppositionpposition ffromrom ppowerfulowerful pparties,arties, aandnd eevenven wwhenhen ttheyhey aarere iimplemented,mplemented, tthishis iimplementationmplementation mmayay bbee iimperfectmperfect oror accompaniedaccompanied byby otherother policiespolicies aimedaimed atat nulli-nulli- ffyingying tthehe eeffectsffects ofof tthehe rreform.eform. IInn AAcemoglu,cemoglu, JJohnson,ohnson, RRobinson,obinson, andand QQuerubínuerubín ((2008),2008), wewe termterm aann eendogenousndogenous ppolicyolicy rresponseesponse ooff tthishis ttype,ype, uunderminingndermining thethe oobjectivesbjectives ofof a reform,reform, thethe “seesaw“seesaw effect”effect” andand provideprovide evidenceevidence tthathat tthehe rreformseforms aaimedimed atat rreducingeducing iinflnfl ationation byby grantinggranting iindependencendependence ttoo tthehe ccentralentral bbankank ttypicallyypically ddoo notnot workwork inin societiessocieties withwith weakweak institutionsinstitutions aandnd ssometimesometimes triggertrigger ootherther ppolicyolicy rresponses—foresponses—for eexample,xample, llargerarger ggovernmentovernment defidefi cits—tocits—to uundondo tthehe rreductioneduction inin thethe aabilitybility ofof thethe governmentgovernment toto provideprovide favorsfavors toto politicallypolitically powerfulpowerful groups.groups. TTherehere areare manymany parallelsparallels betweenbetween thethe implicationsimplications ooff ggeneraleneral eequilibriumquilibrium eeffectsffects forfor thethe interpretationinterpretation (and(and extrapolation)extrapolation) ofof ppartialartial eequilibriumquilibrium eestimatesstimates ddiscussediscussed inin tthehe previousprevious sectionsection andand thethe implicationsimplications ooff ppoliticalolitical eeconomyconomy ffactors.actors. EEvenven thoughthough iinn ggeneraleneral wwee hhaveave llessess ooff aann understandingunderstanding ofof thethe channelschannels ooff iinflnfl u-u- eencence ofof politicalpolitical economy,economy, a generalgeneral principleprinciple providesprovides a uusefulseful startingstarting ppoint:oint: llarge-scalearge-scale sshockshocks aandnd ppolicyolicy iinterventionsnterventions wwillill ccreatereate ppoliticalolitical eeconomyconomy rresponsesesponses ffromrom tthosehose wwhoho seesee theirtheir eeconomicconomic oorr ppoliticalolitical rrentsents tthreatenedhreatened oror fromfrom tthosehose tthathat sseeee nnewew ooptionsptions ttoo iincreasencrease thesethese rents.rents. SuchSuch responsesresponses aarere tthehe bbasisasis ofof aallll tthreehree 28 Journal of Economic Perspectives
eexamplesxamples mmentionedentioned ssoo ffar:ar: tthehe ooverthrowverthrow ooff BBusia,usia, ppotentialotential oobstaclesbstacles ttoo ccreditredit mmarketarket reform,reform, andand tthehe sseesaweesaw eeffect.ffect. TheThe diffidiffi cultyculty lliesies iinn tthehe ffactact tthathat wwhichhich ggroupsroups andand individualsindividuals willwill bebe ableable toto mobilizemobilize andand respondrespond toto thesethese changeschanges willwill vvaryary aacrosscross differentdifferent applications.applications. HHowow shouldshould eempiricalmpirical rresearchesearch iinn eeconomicconomic developmentdevelopment ttakeake ppoliticalolitical eeconomyconomy intointo account?account? A fi rrstst stepstep wouldwould bebe toto useuse empiricalempirical workwork toto betterbetter under-under- sstandtand tthehe rroleole ooff ppoliticalolitical eeconomyconomy ffactorsactors inin development.development. TThishis ttypeype ooff rresearchesearch oonn empiricalempirical ppoliticalolitical economyeconomy ofof developmentdevelopment iiss rrelativelyelatively nnew.ew. TThehe fi rstrst genera-genera- ttionion ofof wworkork ffocusedocused oonn ccross-countryross-country vvariationariation ((seesee tthehe ooverviewverview iinn AAcemoglu,cemoglu, JJohnson,ohnson, andand Robinson,Robinson, 2005a).2005a). AlthoughAlthough researchresearch inin thisthis areaarea isis expanding,expanding, ggiveniven tthehe importanceimportance ofof politicalpolitical economyeconomy forfor thethe problemsproblems ofof development,development, iitt rremainsemains ssurprisingurprising howhow fewfew paperspapers investigateinvestigate kkeyey ppoliticalolitical eeconomyconomy cchannelshannels uusingsing mmicro-icro- ddataata andand carefulcareful empiricalempirical strategies.strategies. I nnowow ddiscussiscuss a fewfew ofof thesethese paperspapers toto givegive a ssenseense ooff wwhathat approachesapproaches areare available.available. LLowow agriculturalagricultural productivityproductivity tthroughouthroughout thethe developingdeveloping worldworld iiss a mmajorajor pproblem,roblem, andand alsoalso a puzzle.puzzle. InIn manymany instances,instances, thethe technologytechnology ofof fallowing,fallowing, wwhichhich mmeanseans plowingplowing tthehe llandand bbutut lleavingeaving iitt uunseedednseeded fforor a periodperiod ofof ttimeime ssoo aass ttoo rreduceeduce wweedeed growthgrowth andand conserveconserve soilsoil moisture,moisture, wouldwould increaseincrease productivityproductivity consider-consider- aably.bly. GoldsteinGoldstein aandnd UUdrydry ((2008)2008) ddocumentocument tthathat inin southernsouthern GhanaGhana thethe amountamount ooff ffallowingallowing isis massivelymassively insuffiinsuffi cient.cient. A nonpoliticalnonpolitical economyeconomy aanswernswer wwouldould bbee ttoo eencouragencourage ffallowing.allowing. BButut inin reality,reality, thisthis recommendationrecommendation (or(or policy)policy) wouldwould bebe iincorrectncorrect oror atat leastleast seriouslyseriously incomplete,incomplete, becausebecause GoldsteinGoldstein andand UdryUdry showshow thatthat ffallowingallowing increasesincreases thethe riskrisk ooff cconfionfi sscationcation ooff llandand bbyy ppowerfulowerful cchiefshiefs aandnd ootherther cconnectedonnected individuals.individuals. InIn fact,fact, tthosehose wwithith ssuffiuffi ccientient ppoliticalolitical ppower,ower, wwhoho ppresum-resum- aablybly faceface a lowerlower riskrisk ooff cconfionfi scation,scation, cchoosehoose signifisignifi cantlycantly hhigherigher llevelsevels ooff ffallowing.allowing. TThishis fi ndingnding illustratesillustrates bbothoth tthehe iimportancemportance ooff ssecureecure ppropertyroperty rrightsights aandnd tthehe rroleole ooff politicalpolitical economyeconomy constraintsconstraints onon productiveproductive investments.investments. IItt aalsolso hhighlightsighlights tthehe rroleole ofof locallocal powerpower structuresstructures inin villagesvillages inin shapingshaping thethe securitysecurity ofof propertyproperty rightsrights aandnd incentivesincentives forfor investment.investment. MMoreore wworkork iiss nneededeeded onon understandingunderstanding howhow thethe politicalpolitical economyeconomy contextcontext iiss sshaped.haped. AAnn eemergingmerging lliteratureiterature iinvestigatesnvestigates tthesehese iissuesssues uusingsing mmicrodata.icrodata. AAss oonene example,example, FFerrazerraz aandnd FFinaninan ((2008)2008) uusese aauditudit rreportseports ffromrom aann aanticorruptionnticorruption pprogramrogram iinn BBrazilrazil ttoo eestimatestimate tthehe eeffectffect ooff eelectorallectoral aaccountabilityccountability oonn ccorruptionorruption aandnd mmisappropriationisappropriation ooff ffundsunds bbyy ppoliticians.oliticians. TTheyhey fi ndnd tthathat mmayorsayors whowho cannotcannot ggetet rreelectedeelected becausebecause ofof termterm limitslimits areare signifisignifi cantlycantly moremore corruptcorrupt andand mmisappro-isappro- ppriateriate 2277 percentpercent mmoreore rresourcesesources tthanhan mmayorsayors wwithith rreelectioneelection incentives.incentives. TheyThey alsoalso sshowhow tthat,hat, cconsistentonsistent wwithith ttheory,heory, tthesehese eeffectsffects areare strongerstronger wwhenhen vvotersoters hhaveave aaccessccess ttoo lessless informationinformation andand whenwhen judicialjudicial punishmentpunishment aagainstgainst ccorruptionorruption iiss wweaker.eaker. IInn a relatedrelated ppaper,aper, FFerrazerraz aandnd FFinaninan ((2009)2009) sstudytudy tthehe eeffectsffects ofof politicianpolitician salariessalaries oonn politicianpolitician behaviorbehavior andand qualityquality ofof publicpublic services.services. TheyThey exploitexploit a ddiscontinuityiscontinuity iinn tthehe ssalariesalaries ooff llocalocal ppoliticiansoliticians aacrosscross BBrazilianrazilian mmunicipalitiesunicipalities rresultingesulting fromfrom a cconstitutionalonstitutional amendmentamendment iimposingmposing ssalaryalary ccapsaps ddependingepending oonn tthehe ssizeize ooff mmunic-unic- iipalpal ppopulation.opulation. UUsingsing rregressionegression discontinuitydiscontinuity techniques,techniques, ttheyhey fi nndd tthathat greatergreater ssalariesalaries areare associatedassociated withwith greatergreater competitioncompetition amongamong potentialpotential candidates,candidates, andand Theory, General Equilibrium, and Political Economy in Development Economics 29
mmoreoveroreover thatthat thethe qualityquality ofof thethe electedelected legislatureslegislatures measuredmeasured byby educationeducation oror eexperiencexperience iimproves.mproves. HHigherigher ssalariesalaries aarere aalsolso aassociatedssociated wwithith iimprovementsmprovements iinn vvariousarious dimensionsdimensions ofof ppoliticianolitician pperformance.erformance.1100 AAnothernother aapproachpproach iiss ttoo aassessssess tthehe eextentxtent ttoo wwhichhich ppastast hhistoricalistorical iinstitutionsnstitutions hhaveave long-runlong-run eeffects.ffects. InIn Acemoglu,Acemoglu, Johnson,Johnson, andand RobinsonRobinson (2005a),(2005a), mymy coauthorscoauthors aandnd I summarizesummarize severalseveral cross-countrycross-country studiesstudies ssuggestinguggesting thatthat certaincertain majormajor eventsevents ssuchuch asas thethe foundationfoundation ofof colonialcolonial institutionsinstitutions oror thethe separationseparation ofof thethe KoreasKoreas cancan hhaveave persistentpersistent eeffects.ffects. However,However, controllingcontrolling forfor confoundingconfounding factorsfactors isis oftenoften ddiffiiffi - ccultult inin cross-countrycross-country studies,studies, aandnd tthehe eexactxact mmechanismechanism lleadingeading ttoo ppersistentersistent eeffectsffects iiss oftenoften ddiffiiffi cultcult oorr iimpossiblempossible ttoo ppinpoint.inpoint. RRecentecent wworkork bbyy DDellell ((2009)2009) ffocusesocuses oonn thethe ppotentialotential eeffectsffects ofof thethe forcedforced laborlabor systemsystem usedused byby thethe SpanishSpanish colonialcolonial ggovernmentovernment iinn PPerueru aandnd BBolivia.olivia. TThishis ssystem,ystem, wwhichhich fforcedorced a largelarge fractionfraction ofof thethe aadultdult malemale populationpopulation ofof villagesvillages nearnear thethe PotosiPotosi silversilver andand HuancavelicaHuancavelica mmercuryercury mminesines ttoo wworkork iinn tthesehese mmines,ines, wwasas uusedsed eextensivelyxtensively iinn tthehe ssixteenthixteenth centurycentury andand waswas aabolishedbolished inin 1812.1812. ThoseThose insideinside andand outsideoutside tthehe bboundaryoundary ooff tthehe ccatchmentatchment aarearea ooff tthehe fforcedorced llaborabor pprogramrogram werewere ssubjectubject toto differentdifferent laborlabor regulations.regulations. InIn a regres-regres- ssionion discontinuitydiscontinuity design,design, DDellell fi nndsds tthathat aareasreas ssubjectedubjected toto forcedforced laborlabor moremore tthanhan 220000 yyearsears aagogo nnowow hhaveave aaboutbout oone-thirdne-third llowerower hhouseholdousehold eequivalentquivalent consumption.consumption. TThehe availableavailable ddataata aalsolso aallowllow aann iinvestigationnvestigation ooff ssomeome ppotentialotential mmechanismsechanisms fforor tthishis vveryery llargearge aandnd ppersistentersistent eeffect,ffect, whichwhich appearsappears toto bebe relatedrelated toto lacklack ofof ppublicublic ggoodsoods iinn aareasreas ssubjectubject toto forcedforced labor.labor. ThisThis lacklack ofof publicpublic goodsgoods inin turnturn maymay bebe rrelatedelated toto thethe policiespolicies ofof thethe SpanishSpanish governmentsgovernments ttoo llimitimit ccompetitionompetition fforor llaborabor inin tthehe catchmentcatchment areasareas fromfrom pprivaterivate llandholdersandholders aandnd bbusinesses.usinesses. FFinally,inally, aagaingain rrelatedelated ttoo tthehe iissuessue ooff ccoercion,oercion, NNaiduaidu aandnd YYuchtmanuchtman ((2009)2009) iinvestigatenvestigate hhowow tthehe aabilitybility ofof employersemployers toto imprisonimprison oror fi nnee aann eemployeemployee forfor bbreachreach ooff ccontractontract uundernder tthehe MMasteraster aandnd SServantervant AActs,cts, wwhichhich rremainedemained iinn eeffectffect iinn BBritainritain uuntilntil 1875,1875, affectedaffected laborlabor marketmarket relations.relations. TheyThey provideprovide evidenceevidence thatthat employersemployers mmadeade extensiveextensive useuse ofof theirtheir ccoerciveoercive aabilitybility underunder thethe law,law, andand asas a consequence,consequence, llaborabor ddemandemand sshockshocks wwereere llargelyargely mmetet bbyy uusingsing iincreasedncreased pprosecutionsrosecutions fforor ccontractontract bbreachreach rratherather tthanhan hhigherigher wwages.ages. TThishis fi ndingnding isis consistentconsistent wwithith ttheoreticalheoretical ppredic-redic- ttionsions ooff rrecentecent modelsmodels ofof llaborabor ccoercionoercion ssuchuch aass iinn AAcemoglucemoglu andand WolitzkyWolitzky (2009).(2009). OOverall,verall, tthehe aabove-mentionedbove-mentioned ppapers,apers, tthoughhough ddistinctistinct iinn mmethodologyethodology andand sscope,cope, sshowhow hhowow mmicrodataicrodata aandnd rregionalegional variationvariation inin institutionsinstitutions andand lawslaws cancan shedshed llightight onon thethe rolerole ofof politicalpolitical economyeconomy factorsfactors inin development.development. EmpiricalEmpirical wworkork iinn ddevelopmentevelopment eeconomicsconomics sshouldhould ppayay mmoreore aattentionttention tto,o, aandnd buildbuild a moremore systematicsystematic uunderstandingnderstanding of,of, ppoliticalolitical eeconomy.conomy. IItt mmustust aalsolso sstudytudy hhowow ddifferentifferent ccounterfac-ounterfac- ttualual aandnd policypolicy experimentsexperiments wwillill iinteractnteract wwithith oorr bbee rresistedesisted byby politicalpolitical factors.factors.
10 Returning to the contrast between different counterfactual exercises, one might question, for example, whether this regression discontinuity estimate would be informative about the effects of a large-scale increase in politician salaries, which might cause different composition effects than cross- municipality variation in salaries induced by salary caps. 30 Journal of Economic Perspectives
Concluding Remarks
A keykey objectiveobjective ofof empiricalempirical workwork inin developmentdevelopment economicseconomics isis toto discriminatediscriminate bbetweenetween theoriestheories aboutabout thethe causescauses ofof economiceconomic growthgrowth andand ttoo cconductonduct ccounterfac-ounterfac- ttualual analysisanalysis ttoo bbuilduild a systematicsystematic uunderstandingnderstanding ooff hhowow aann eeconomyconomy wwillill rrespondespond ttoo llargearge cchangeshanges iinn ffactoractor ssupplies,upplies, technology,technology, oror policy.policy. EconomicEconomic ttheoryheory iiss ccentralentral iinn thisthis endeavor.endeavor. IInn ffact,act, eeconomicconomic ttheoryheory bbecomesecomes mmoreore iimportantmportant iinn tthehe ppresenceresence ooff generalgeneral eequilibriumquilibrium andand politicalpolitical economyeconomy considerations.considerations. GGeneraleneral eequilibriumquilibrium andand ppoliticalolitical eeconomyconomy eeffectsffects areare oftenoften ddiffiiffi cultcult t too eestimatestimate oror toto quantify.quantify. HHowever,owever, ttheyhey aarere ppervasiveervasive aandnd eessentialssential fforor iimportantmportant qquestionsuestions inin developmentdevelopment economics.economics. MostMost researchresearch inin economicseconomics hashas (and(and shouldshould hhave)ave) a narrownarrow focusfocus andand triestries toto investigateinvestigate a pparticulararticular ssetet ooff ffactorsactors iinn a sspecifipecifi c ccontext.ontext. BButut inin developmentdevelopment eeconomicsconomics wwherehere tthehe aagendagenda ooughtught ttoo bbee bbroad,road, wwee sshouldhould aalsolso nnotot lloseose ssightight ooff tthehe bbiggerigger ppictureicture ooff tthehe problemproblem ofof economiceconomic ddevelop-evelop- mment.ent. ThisThis impliesimplies tthathat wewe shouldshould sstrivetrive ttoo iincorporatencorporate ggeneraleneral eequilibriumquilibrium aandnd ppoliticalolitical economyeconomy eeffectsffects whenwhen wewe can,can, andand wewe shouldshould bbee ccognizantognizant ofof ttheirheir iimpor-mpor- ttanceance wwhenhen wewe cannot.cannot. IItt iiss aalsolso uusefulseful toto notenote thatthat generalgeneral equilibriumequilibrium andand politicalpolitical economyeconomy consid-consid- eerationsrations aarere nnotot oonlynly a cconstraintonstraint iinn ppolicyolicy aanalysis,nalysis, eevenven tthoughhough I ffocusedocused oonn ccasesases iinn whichwhich thesethese considerationsconsiderations ttendend ttoo ooffsetffset oorr rreverseeverse ppartialartial eequilibriumquilibrium eeffects.ffects. FForor example,example, tthehe eendogeneityndogeneity ooff ppoliticalolitical eeconomyconomy rresponsesesponses alsoalso impliesimplies tthathat ccertainertain economiceconomic policiespolicies andand shocksshocks mightmight havehave moremore benefibenefi cialcial effectseffects thanthan whatwhat tthehe ppureure eeconomicconomic aanalysisnalysis wwouldould ssuggest,uggest, becausebecause theythey cancan leadlead toto a benefibenefi cialcial cchangehange iinn tthehe ppoliticalolitical equilibrium.equilibrium. OneOne suchsuch example,example, ddiscussediscussed iinn AAcemoglu,cemoglu, JJohnson,ohnson, andand RobinsonRobinson (2005b),(2005b), isis thethe possibilitypossibility thatthat AtlanticAtlantic tradetrade maymay havehave hadhad llong-runong-run benefibenefi cialcial effectseffects inin EuropeEurope mainlymainly byby changingchanging tthehe ppoliticalolitical eequilibriumquilibrium iinn sseveraleveral ccountriesountries ttowardsowards moremore participatoryparticipatory rregimes.egimes.
■ I thank Joshua Angrist, David Autor, Abhijit Banerjee, Timothy Besley, Angus Deaton, Melissa Dell, Esther Dufl o, Fred Finan, Chad Jones, Guido Imbens, Ariel Pakes, James Robinson, Timothy Taylor, Robert Townsend, Chris Udry, and Eric Weese for useful comments and suggestions. Daron Acemoglu 31
References
Acemoglu, Daron. 1998. “Why Do New Tech- ment: Evidence from Hookworm Eradication nologies Complement Skills? Directed Technical in the American South.” Quarterly Journal of Change and Wage Inequality.” Quarterly Journal of Economics, 122(1): 73–117. Economics, 113(4): 1055–90. Boafo-Arthur, Kwame. 1999. “Ghana: Acemoglu, Daron. 2007. “Equilibrium Bias of Structural Adjustment, Democratization and Technology.” Econometrica, 75(5): 1371–1410. the Politics of Continuity.” African Studies Review, Acemoglu, Daron, and Joshua D. Angrist. 42(2): 41–72. 2000. “How Large Are Human Capital Externali- Dell, Melissa. 2009. “The Persistent Effects of ties? Evidence from Compulsory Schooling Laws.” Peru’s Mining Mita.” Available at SSRN: http:// NBER Macroeconomics Annual 2000, vol. 15, pp. papers.ssrn.com/sol3/papers.cfm?abstract 9–59. _id=1596425. Acemoglu, Daron, David H. Autor, and Dufl o, Esther. 2004a. “Scaling up and Evalua- David Lyle. 2004. “Women, War and Wages: The tion.” http://econ-www.mit.edu/fi les/766. Effect of Female Labor Supply on Labor Market Dufl o, Esther. 2004b. “Medium Run Effects of Outcomes.” Journal of Political Economy, 112(3): Educational Expansion: Evidence from a Large 497–551. School Construction Program in Indonesia.” Acemoglu, Daron, and Simon Johnson. 2007. Journal of Development Economics, 74(1): 163–97. “Disease and Development: The Effect of Life Ferraz, Claudio, and Frederico Finan. 2008. Expectancy on Economic Growth.” Journal of “Electoral Accountability and Corruption: Political Economy, 115(6): 925–985. Evidence from the Audits of Local Governments.” Acemoglu, Daron, Simon Johnson, and http://ideas.repec.org/p/nbr/nberwo/14937 James A. Robinson. 2005a. “Institutions as a .html. Fundamental Cause of Long-Run Growth.” In Ferraz, Claudio, and Frederico Finan. 2009. Handbook of Economic Growth, ed. Philippe Aghion “Motivating Politicians: The Impacts of Monetary and Steven Durlauf, 384–473. Amsterdam: North Incentives on Quality and Performance.” NBER Holland. Working Paper 14906. Acemoglu, Daron, Simon Johnson, and James Goldstein, Markus, and Christopher Udry. Robinson. 2005b. “The Rise of Europe: Atlantic 2008. “The Profi ts of Power: Land Rights and Trade, Institutional Change, and Growth.” Agricultural Investment in Ghana.” Journal of American Economic Review, 95(3): 546–79. Political Economy, 116(6): 981–1022. Acemoglu, Daron, Simon Johnson, James A. Haber, Stephen, and Enrico Perotti. 2008. Robinson, and Pablo Querubín. 2008. “When “Political Economy of Finance.” http://fi c.wharton Does Policy Reform Work? The Case of Central .upenn.edu/fi c/sicily/19%20haberperotti.pdf. Bank Independence.” Brookings Papers on Economic Heckman, James J., Lance Lochner, and Activity, no. 1, pp. 351–418. Christopher Taber. 1998. “Explaining Rising Acemoglu, Daron, and Alex Wolitzky. 2009. Wage Inequality: Explorations with a Dynamic “The Economics of Labor Coercion.” NBER General Equilibrium Model of Labor Earnings Working Paper 15581. with Heterogeneous Agents.” Review of Economic Angrist, Joshua D. 1995. “The Economic Dynamics, 1(1): 1–58. Returns to Schooling in the West Bank and Gaza Herbst, Jeffrey. 1993. The Politics of Reform in Strip.” American Economic Review, 85(5): 1065–87. Ghana, 1982–1991. Berkeley and Los Angeles: Angrist, Joshua D., Guido W. Imbens, and University of California Press. Donald B. Rubin. 1996. “Identifi cation of Causal Imbens, Guido. 2009. “Better LATE Than Effects Using Instrumental Variables.” Journal of Nothing: Some Comments on Deaton (2009) and the American Statistical Association, 91(434): 444–55. Heckman and Urzua (2009).” http://ideas.repec Banerjee, Abhijit V., and Esther Dufl o. 2005. .org/p/nbr/nberwo/14896.html. “Growth Theory through the Lens of Develop- Lucas, Robert E. 1976. “Economic Policy ment Economics.” Chap. 7 In Handbook of Economic Evaluation: A Critique.” In The Phillips Curve and Growth, ed. Philippe Aghion and Steven Durlauf. Labor Markets, ed. Karl Brunner and Alan H. Amsterdam: North Holland. Meltzer, 19–46. Amsterdam: North Holland. Behrman, Jere R., and Mark R. Rosenzweig. Marschak, Jacob. 1953. “Economic Measure- 2004. “The Returns to Birthweight.” Review of ment for Policy and Prediction.” In Studies in Economics and Statistics, 86(2): 586–601. Econometric Method, ed. W. Hood and T. J. Koop- Bleakley, Hoyt. 2007. “Disease and Develop- mans, 1–26. New York: Wiley. 32 Journal of Economic Perspectives
Naidu, Suresh, and Noam Yuchtman. 2009. Inference. Boston: Houghton Miffl in. “How Green Was My Valley? Coercive Contract Strauss, John, and Duncan Thomas. 1998. Enforcement in 19th Century Industrial Britain.” “Health, Nutrition and Economic Development.” http://www.irle.ucla.edu/events/2010/naidu Journal of Economic Literature, 36(2): 766–817. .html. Townsend, Robert. Forthcoming. “Financial Schultz, T. Paul. 2002. “Wage Gains Associ- Structure and Economic Welfare: Applied ated with Height as a Form of Health Human General Equilibrium Development Economics.” Capital.” American Economic Review, 92(2): 349–53. Annual Review of Economics. Shadish, William H., Thomas D. Cook, and Weil, David N. 2007. “Accounting for the Donald T. Campbell. 2002. Experimental and Effect of Health on Growth.” Quarterly Journal of Quasi-Experimental Designs for Generalized Causal Economics, 122(3): 1265–1306. This article has been cited by:
1. Sabyasachi Das. 2020. (Don’t) leave politics out of it: Reflections on public policies, experiments, and interventions. World Development 127, 104792. [Crossref] 2. Nimi Hoffmann. 2020. Involuntary experiments in former colonies: The case for a moratorium. World Development 127, 104805. [Crossref] 3. Yana van der Meulen Rodgers, Anthony Bebbington, Catherine Boone, Jampel Dell'Angelo, Jean- Philippe Platteau, Arun Agrawal. 2020. Experimental approaches in development and poverty alleviation. World Development 127, 104807. [Crossref] 4. Judith Favereau, Michiru Nagatsu. 2020. Holding back from theory: limits and methodological alternatives of randomized field experiments in development economics. Journal of Economic Methodology 40, 1-21. [Crossref] 5. Michiru Nagatsu, Judith Favereau. 2020. Two Strands of Field Experiments in Economics: A Historical-Methodological Analysis. Philosophy of the Social Sciences 50:1, 45-77. [Crossref] 6. Eszter Czibor, David Jimenez‐Gomez, John A. List. 2019. The Dozen Things Experimental Economists Should Do (More of ). Southern Economic Journal 86:2, 371-432. [Crossref] 7. Gilles Chemla, Christopher A. Hennessy. 2019. Rational expectations and the Paradox of policy- relevant natural experiments. Journal of Monetary Economics . [Crossref] 8. Florent Bédécarrats, Isabelle Guérin, François Roubaud. 2019. All that Glitters is not Gold. The Political Economy of Randomized Evaluations in Development. Development and Change 50:3, 735-762. [Crossref] 9. Marshall Burke, Lauren Falcao Bergquist, Edward Miguel. 2019. Sell Low and Buy High: Arbitrage and Local Price Effects in Kenyan Markets*. The Quarterly Journal of Economics 134:2, 785-842. [Crossref] 10. Valentina Colcelli. From Cosmopolitan Individual Status (French Civil Code of 1804) and “l’exercice des droits civils” to EU Citizenship 17-24. [Crossref] 11. Eszter Czibor, David Jimenez-Gomez, John A. List. 2019. The Dozen Things Experimental Economists Should Do (More Of ). SSRN Electronic Journal . [Crossref] 12. Tobias Berg, Daniel Streitz. 2019. Handling Spillover Effects in Empirical Research: An Application using Credit Supply Shocks. SSRN Electronic Journal . [Crossref] 13. Evgeny Yakovlev, Ilia Sorvachev. 2019. Short- and Long-Run Effects of Sizable Child Subsidy: Evidence from Russia. SSRN Electronic Journal . [Crossref] 14. Tessa Bold, Mwangi Kimenyi, Germano Mwabu, Alice Ng’ang’a, Justin Sandefur. 2018. Experimental evidence on scaling up education reforms in Kenya. Journal of Public Economics 168, 1-20. [Crossref] 15. Yusuf Neggers. 2018. Enfranchising Your Own? Experimental Evidence on Bureaucrat Diversity and Election Bias in India. American Economic Review 108:6, 1288-1321. [Abstract] [View PDF article] [PDF with links] 16. Kilian Huber. 2018. Disentangling the Effects of a Banking Crisis: Evidence from German Firms and Counties. American Economic Review 108:3, 868-898. [Abstract] [View PDF article] [PDF with links] 17. Joseph Berechman. Interest Groups: Advocacy and Opposition 93-106. [Crossref] 18. Peter Hangoma, Bjarne Robberstad, Arild Aakvik. 2018. Does Free Public Health Care Increase Utilization and Reduce Spending? Heterogeneity and Long-Term Effects. World Development 101, 334-350. [Crossref] 19. Dorothy Kronick. 2018. Vagaries of Police Behavior. SSRN Electronic Journal . [Crossref] 20. Glenn Furton, Adam G. Martin. 2018. What Can Economists Do? Appreciative Theory, Market Design, and Open Systems. SSRN Electronic Journal . [Crossref] 21. Valentin Lang, Marina Mendes Tavares. 2018. The Distribution of Gains from Globalization. IMF Working Papers 18:54, 1. [Crossref] 22. Gloria Colmenares, Andreas Löschel, Reinhard Madlener. 2018. The Rebound Effect and its Representation in Energy and Climate Models. SSRN Electronic Journal . [Crossref] 23. Karthik Muralidharan, Paul Niehaus. 2017. Experimentation at Scale. Journal of Economic Perspectives 31:4, 103-124. [Abstract] [View PDF article] [PDF with links] 24. . Education systems are misaligned with learning 169-188. [Crossref] 25. Maria Sophia Aguirre. Virtue Ethics and Global Development 1297-1305. [Crossref] 26. K. Muralidharan. Field Experiments in Education in Developing Countries 323-385. [Crossref] 27. Gregory P. Casey, Marc Klemp. 2017. Instrumental Variables in the Long Run. SSRN Electronic Journal . [Crossref] 28. Karthik Muralidharan, Paul Niehaus, Sandip Sukhtankar. 2016. Building State Capacity: Evidence from Biometric Smartcards in India. American Economic Review 106:10, 2895-2929. [Abstract] [View PDF article] [PDF with links] 29. AGNÈS LABROUSSE. 2016. Not by technique alone. A methodological comparison of development analysis with Esther Duflo and Elinor Ostrom. Journal of Institutional Economics 12:2, 277-303. [Crossref] 30. Emily Rauscher. 2016. Does Educational Equality Increase Mobility? Exploiting Nineteenth-Century U.S. Compulsory Schooling Laws. American Journal of Sociology 121:6, 1697-1761. [Crossref] 31. Milan Zafirovski. 2016. Toward Economic Sociology/Socio-Economics? Sociological Components in Contemporary Economics and Implications for Sociology. The American Sociologist 47:1, 56-80. [Crossref] 32. Nikolai Boboshko, Stan Veuger. 2016. Business Cycle Shocks to Foreign Aid Flows. Procedia Engineering 159, 326-328. [Crossref] 33. Arno Tausch. 2016. Smart Developmentt. An Essay on a New Political Economy of the Environment. SSRN Electronic Journal . [Crossref] 34. Jean-Noel Barrot. 2016. Can Paying Firms Quicker Impact Aggregate Employment?. SSRN Electronic Journal . [Crossref] 35. Jean-Noel Barrot. 2016. Can Paying Firms Quicker Affect Aggregate Employment?. SSRN Electronic Journal . [Crossref] 36. Miguel Urquiola. 2015. Progress and challenges in achieving an evidence-based education policy in Latin America and the Caribbean. Latin American Economic Review 24:1. . [Crossref] 37. JAMES D. FEARON, MACARTAN HUMPHREYS, JEREMY M. WEINSTEIN. 2015. How Does Development Assistance Affect Collective Action Capacity? Results from a Field Experiment in Post-Conflict Liberia. American Political Science Review 109:3, 450-469. [Crossref] 38. Brendan Nyhan, Jason Reifler. 2015. The Effect of Fact-Checking on Elites: A Field Experiment on U.S. State Legislators. American Journal of Political Science 59:3, 628-640. [Crossref] 39. Jiancai Pi, Yu Zhou. 2015. Rural Property Rights, Migration, and Welfare in Developing Countries. The B.E. Journal of Economic Analysis & Policy 15:3, 997-1029. [Crossref] 40. Erik Berglof. 2015. New structural economics meets European transition. Journal of Economic Policy Reform 18:2, 114-130. [Crossref] 41. Jay Pan, Xuezheng Qin, Qian Li, Joseph P. Messina, Paul L. Delamater. 2015. Does hospital competition improve health care delivery in China?. China Economic Review 33, 179-199. [Crossref] 42. Vlad Tarko. 2015. The challenge of empirically assessing the effects of constitutions. Journal of Economic Methodology 22:1, 46-76. [Crossref] 43. Maria Cheung, Maria Perrotta Berlin. 2015. The Impact of a Food for Education Program on Schooling in Cambodia. Asia & the Pacific Policy Studies 2:1, 44-57. [Crossref] 44. Maria Sophia Aguirre. Virtue Ethics and Global Development 1-9. [Crossref] 45. Valentina Colcelli. The Solidarity Principle in the New EU Member States 247-264. [Crossref] 46. S. Shyam Sundar, Akshaya Sreenivasan. In machines we trust 1-4. [Crossref] 47. Gilles Chemla, Chris Hennessy. 2015. The Paradox of Policy-Relevant Natural Experiments. SSRN Electronic Journal . [Crossref] 48. Andreas Assiotis, Maria Krambia-Kapardis. 2014. Corruption correlates: the case of Cyprus. Journal of Money Laundering Control 17:3, 260-268. [Crossref] 49. Maria Sophia Aguirre. Socioeconomic Impact of the Work of the Home 133-145. [Crossref] 50. T. Bold, J. Svensson. 2013. Policies and Institutions for Effective Service Delivery: The Need of a Microeconomic and Micropolitical Approach. Journal of African Economies 22:suppl 2, ii16-ii38. [Crossref] 51. Cristian Pop-Eleches,, Miguel Urquiola. 2013. Going to a Better School: Effects and Behavioral Responses. American Economic Review 103:4, 1289-1324. [Abstract] [View PDF article] [PDF with links] 52. Ahmed S. Rahman. 2013. THE ROAD NOT TAKEN: WHAT IS THE “APPROPRIATE” PATH TO DEVELOPMENT WHEN GROWTH IS UNBALANCED?. Macroeconomic Dynamics 17:4, 747-778. [Crossref] 53. Samuel Bazzi,, Michael A. Clemens. 2013. Blunt Instruments: Avoiding Common Pitfalls in Identifying the Causes of Economic Growth. American Economic Journal: Macroeconomics 5:2, 152-186. [Abstract] [View PDF article] [PDF with links] 54. Torben K. Mideksa. 2013. The economic impact of natural resources. Journal of Environmental Economics and Management 65:2, 277-289. [Crossref] 55. Quamrul H. Ashraf, David N. Weil, Joshua Wilde. 2013. The Effect of Fertility Reduction on Economic Growth. Population and Development Review 39:1, 97-130. [Crossref] 56. Tessa Bold, Mwangi Kimenyi, Germano Mwabu, Alice Alice Ng'ang'a, Justin Sandefur. 2013. Scaling Up What Works: Experimental Evidence on External Validity in Kenyan Education. SSRN Electronic Journal . [Crossref] 57. Povilas Lastauskas. 2013. Europe's Revolving Doors: Import Competition and Endogenous Firm Entry Institutions. SSRN Electronic Journal . [Crossref] 58. Michael A. Clemens, Gabriel Demombynes. 2013. The New Transparency in Development Economics: Lessons from the Millennium Villages Controversy. SSRN Electronic Journal . [Crossref] 59. Paulo Bastos, Sebastian Miller. 2013. Politics Under the Weather: Droughts, Parties and Electoral Outcomes. SSRN Electronic Journal . [Crossref] 60. Juergen Jung, Chung Tran. 2012. The extension of social security coverage in developing countries. Journal of Development Economics 99:2, 439-458. [Crossref] 61. Pablo Selaya, Eva Rytter Sunesen. 2012. Does Foreign Aid Increase Foreign Direct Investment?. World Development 40:11, 2155-2176. [Crossref] 62. Marco Percoco. 2012. Entrepreneurship, Social Capital and Institutions: Evidence from Italy. Spatial Economic Analysis 7:3, 339-355. [Crossref] 63. Marco Dardi. 2012. The perfect competition paradigm: evolving from its ambiguities. Cahiers d Économie Politique 63:2, 219. [Crossref] 64. Michael A. Clemens, Gabriel Demombynes. 2011. When does rigorous impact evaluation make a difference? The case of the Millennium Villages. Journal of Development Effectiveness 3:3, 305-339. [Crossref] 65. Rohini Pande. 2011. Can Informed Voters Enforce Better Governance? Experiments in Low-Income Democracies. Annual Review of Economics 3:1, 215-237. [Crossref] 66. Alisa Di Caprio. 2011. Backseat Drivers: The Political Economy of Trade Liberalization. International Studies Review 13:2, 322-324. [Crossref] 67. Marc F. Bellemare, Christopher B. Barrett, David R. Just. 2011. The Welfare Impacts of Commodity Price Volatility: Evidence from Rural Ethiopia. SSRN Electronic Journal . [Crossref] 68. Miguel Szekely. 2011. Toward Results-Based Social Policy Design and Implementation. SSRN Electronic Journal . [Crossref] 69. Tessa Bold, Mwangi Kimenyi, Germano Mwabu, Justin Sandefur. 2011. Why Did Abolishing Fees Not Increase Public School Enrollment in Kenya?. SSRN Electronic Journal . [Crossref] 70. Christopher B. Barrett, Michael R. Carter. 2010. The Power and Pitfalls of Experiments in Development Economics: Some Non‐random Reflections. Applied Economic Perspectives and Policy 32:4, 515-548. [Crossref]