ScienceDirect
EconomiA18(2017)244–259
Impact
of
infrastructure
expenses
in
strategic
sectors
for
Brazilian
poverty
Emerson
Marinho
a,∗,
Guaracyane
Campelo
b,
João
Franc¸a
a,
Jair
Araujo
caPost-GraduateCourseinEconomy,UniversityFederalofCeara—(CAEN/UFC),UniversityAvenue2700,2ndFloor,Benfica,Fortaleza,Ceará 60020-181,Brazil
bUniversityFederalofCeará—(UFC/Sobral),DepartmentofEconomicsandFinance,CoronelStanislausFleets/n,Centro,Sobral-Ce,BlockI, CampusSobral,Mucambinho,Fortaleza,Ceará62010-560,Brazil
cUniversityFederalofCeará,AgriculturalScienceCenter,ACCampusdoPici,AvMisterHull.Pici,Fortaleza,Ceará60440-970,Brazil
Received1April2015;receivedinrevisedform2May2016;accepted18January2017 Availableonline6March2017
Abstract
Thispaperanalyzesthe impactof infrastructureinvestmentsinthe reductionof povertyin Brazil,controlledthroughother determinantssuchaseconomicgrowth,incomeinequality,averageschoolingyears,unemploymentrateandstatebudgetsfrom1995 to2011.Amodelforadynamicpaneldata,estimatedbythegeneralizedmethodofmoments(GMM)intwostepsasdeveloped byArellano-Bond(1991)andBlundell-Bond(1998)foundamongotherconclusions,asignificantinverserelationbetweenpublic investmentininfrastructureandpoverty.TheGrangercausalitytestforpaneldataproposedbyHurlinandVenet(2001,2004)and Hurlin(2004,2005)reinforcedresultsvalidation.
© 2017 The Authors. Production and hosting by Elsevier B.V. on behalf of National Association of Postgraduate Cen-ters in Economics, ANPEC. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/ licenses/by-nc-nd/4.0/).
JELclassifications: H54;I30
Keywords:Poverty;Infrastructure;Dynamicpanel
Resumo
Otrabalho temcomoobjetivo principalanalisaroimpacto dosinvestimentosem infraestrutura nossetores estratégicosda economia(transporte,energia,comunicac¸ão,saúdeesaneamento)nareduc¸ãodapobrezacontrolandoporoutrosdeterminantes taiscomocrescimentoeconômico,desigualdadederenda,anosmédiodeestudo,taxadedesempregoereceitasgovernamentais orc¸amentáriasparaosestadosbrasileiros,noperíodode1995a2011.Ummodeloparadadosempaineldinâmico,estimadopelo métododemomentosgeneralizados-sistema(MMG-S)emdoispassos,desenvolvidoporArellano-Bond(1991)eBlundel-Bond (1998),detectou,entreoutrasconclusões,umarelac¸ãosignificanteentreosinvestimentospúblicoseminfraestruturaepobreza, sendoestesumaferramentaeficientenocombatedesta.Osoutrosdeterminantesinvestigadosdesempenhamumpapelimportante
∗Correspondingauthor.
E-mailaddresses:emarinho@ufc.br(E.Marinho),guaracyane@ufc.br(G.Campelo),joaomario@caen.ufc.br(J.Franc¸a),jaraujoce@gmail.com
(J.Araujo).
http://dx.doi.org/10.1016/j.econ.2017.01.002
nadinâmicadapobrezanoBrasil.OtestedecausalidadedeGrangerparadadosempainel,propostoporHurlineVenet(2001, 2004)eHurlin(2004,2005)validaosresultados.
© 2017 The Authors. Production and hosting by Elsevier B.V. on behalf of National Association of Postgraduate Cen-ters in Economics, ANPEC. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/ licenses/by-nc-nd/4.0/).
Palavras-chave: Pobreza;Infraestrutura;Paineldinâmico
1. Introduction
Historicallyspeaking,allexpensesinBrazilianinfrastructurehavealwaysbeen theresponsibility of thepublic sector.Inthenineties,however,partnershipsbetweenthepublicandprivatesectorsenabledthebeginningofsignificant participationofnationalandinternationalprivatecompaniesthroughprivatizationsinthetelecommunicationssector andpart of theenergy business,roadandrailroad concessionaries,etc. However,despite thesechanges,the State continuestobethemainresponsibleagentforthesupplyofinfrastructure.
Seekingtodiscusstheroleofinfrastructureinpovertyreduction,itmustbesaidthatmoreaccesstoinfrastructure servicesalsoaffectsthematerializationoftheso-called“MillenniumDevelopmentGoals”(MDG)Brazilisinvolved in.ThecontributionofinfrastructuretotheMillenniumDevelopmentGoals(MDG)isreflectedintheincreasein pro-ductivityandwellbeingamongpoorpeople,thusimprovingtheiraccesstolocalmarketsandotherregions,optimizing thecoverageandthequalityofservicesofferedthroughtheimprovementofeducation,health,transportationservices, energy,informationtechnologyandbasicsanitation.
Theinfrastructuresupplyisavitalcomponentoftheincentivetonationaleconomicgrowth,bothforitspotential togenerateemploymentandforitsinfluenceinalleconomysectors.Inthissense,itimproveseconomicactivityand helpsreducepersistent poverty.Additionally,wide accesstoinfrastructurecontributes toreduceinequality(Sílvia andTriches,2014;BertussiandEllery, 2012;MussoliniandTeles,2010;CalderonandServen,2004;Ferreiraand Malliagros,1998).
Adequateinfrastructureisanecessaryconditionforeconomicdevelopment.Therefore,anygrowthstrategyplanned tohelpthepoormustnecessarilyincludethepromotionofinvestmentininfrastructureinordertoallowwiderpopulation accesstothepositiveexternalitiescreatedbysuchinvestments(Hirschman,1958;DattandRavallion,2002).
Anadequateinfrastructureisanecessaryconditionforeconomicdevelopment.Therefore,anygrowthstrategythat involveshelpingthepoormustnecessarilycontemplatethepromotionofinfrastructureinvestments,seekingtoallow thispopulationsegmentabetteraccesstothepositiveexternalitiesgeneratedbyadequateinfrastructure.
AccordingtotheInter-AmericanDevelopmentBank(IDB,2000)itispossibletodefineinfrastructureasasetof engineeringstructuresandfacilitiesthatarethenecessarybasisforthedevelopmentofproductiveactivitiessuchas services,policies,socialandpersonalactivities.Theregionsthatdirectlybenefitfrominfrastructureservicesachieve positiveexternalities,attractingindustriesandhumancapital,thusincreasingproductivityandstimulatingeconomic growth.
Amonginternationalworksthatempiricallytestedtheroleofinfrastructureinthefightagainstpoverty,weshould mentionthoseofJacoby(2000),Runsinarith(2008),Roy(2009),Ogun(2010),Seetanahetal.(2009),Escobaland Ponce(2001)andAparicioetal.(2011)amongothers.
Inthelocalenvironment,economicsliteratureontheimpactofdirectpublicinvestmentsininfrastructureforpoverty reductionpurposesismainlycoveredbytheworksofCruzetal.(2010)andindirectly,bytheKageyamaandHoffmann study(2006).
Withthisperspective,consideringthetemporaleffectofpovertyandusingstate-provideddata,thisarticleanalyzes iftheresultsofinfrastructureinvestmentpolicieshaveaffectedthedynamicsofpovertyinBrazilfrom1995to2011. Forthispurpose,we appliedadynamic panel datamodel that usestheGeneralized Method ofMoments (GMM) developedbyArellanoandBond(1991),ArellanoandBover(1995)andBlundell andBond (1998).Additionally, theGrangerCausalityTestwasappliedforHurlinandVenetpaneldata(2001,2004)andHurlin(2004,2005)which validatedresultsbyrevealingthatinfrastructureisanefficienttooltofightpoverty.Thistestpointsatboththeexistence ofthistoolandthecausalitylinkbetweenpovertyandinfrastructure.
betweenpovertyandinfrastructure,whichismeasuredthroughtheState’spublicexpensespercapitainstrategicareas oftheeconomy(energyandmineralresources,transportation,communications,healthandsanitation).Thisrelation iscontrolledbyotherpovertydeterminants,suchasthegrossdomesticproduct(GDP)percapita,averageschooling years,incomeinequality,unemploymentrateandtheStatebudgets.
Themostcommonwaytomeasurepoverty,becauseofitssimplicity,isthesettingofapovertyline,inotherwords, anincome level belowwhichpeopleare classifiedas poor. Thepoverty line used ismade available bythe IPEA (InstituteofAppliedEconomicsResearch)anditsvalueisequaltohalfminimummonthlysalaryaccordingtoprices ofSeptember2009.ThecalculationofthislinefollowstheCorseuilandFoguelmethod(2002).Theindicatorusedto measureabsolutepovertywastheproportionofpoorpeople.
ThepovertylineappliedinthisarticlewasmadeavailablebytheInstituteofStudiesonLaborandSocietyIETS, forseveralBrazilianstates.Thislinedoublestheindigencelineanditisdefinedasthefinancialvaluenecessaryforan individualtopurchaseminimumcalorieconsumptionfood.Theindicatorusedtomeasureabsolutepovertywasthe proportionofpoorindividuals.
Amongother important findings,the main onewas that public investment inpublic infrastructureprovokeda significantimpactonpovertyreduction.Asfortheotherdeterminants,suchastheincreaseinyearsofschooling,State budgetsandthereductionofunemployment,theyalsocontributedtodecreasepovertyincidence.Thesamehappened withregardstotheStateGDPpercapita,howeverthisimpacthasbeenlowerthanthatofincomeinequality.Thisfact maybeduetothehighincomeconcentrationfoundinBrazil,whichinacertainwayamortizestheeffectsofeconomic growth.
Theremainingofthearticleisorganizedinsevensections.Section2reviewstherelationbetweenpublicinvestment ininfrastructureandpoverty.Section3analyzespovertydeterminants.Thefourthsectionintroducesadiscussionon thedatabaseandtheconstructionofmodelvariables.Inthefifthsectionwespecifytheeconometricmodelanddiscuss theGrangerCausalityTestforpaneldata.Inthesixthsection,estimatedeconometricdataandtheGrangerCausality Testdataareanalyzed.Thelastsectiondrawsthefinalconclusions.
2. Theoreticalandempiricalaspectsoftherelationbetweenpovertyandinfrastructure
Theeffectsthatinfrastructureexertsonpovertyhavebeentheobjectofaseriesofstudiesineconomicsliterature specializedonthesubject.Ingeneral,itissupposedthattheprovisionofadequateinfrastructureisakeyelementto reducepoverty,asittriggersadirecteffectintheimprovementinemploymentratesandwageswhentheeconomy growthsandbecomesmorecompetitive.
Initstheoreticalstudies,Hirschman(1958)statesthatpublicinvestmentininfrastructureisvitalforthesocialand economicdevelopmentofacountry,onceitprovidesanattractiveenvironmentforprivateinvestments,thusmaking servicescheaperandmorecompetitiveandthereforesupportingallothereconomicactivities.
Theinfrastructurecomponentsthatexertthehighestinfluenceonthesystemiccompetitivenessofcompaniesare relatedtotheofferofenergy,transportationandtelecommunications.Theofferofthesecomponentsplaysakeyrolein thosestatesthatofferitatlowcostinanefficient,regularandreliableway.AccordingtoHirschman(1958),infrastructure iscomposedofbasicservicessuchasthejudiciarypower,education,publichealth,transports,communications,water andelectricitysupplyandtheagriculturalsupportinservicessuchasirrigationanddrainage.
Withregardstointernationalempiricalevidence,thereareseveralworksthatestimatetheimpactofinfrastructure inpovertyreductioninmanycountries.AcasestudywithdataonNepal’spopulationlivingstandardswasdeveloped byJacoby(2000)fortheyears1995and1996.Thestudyfoundthattheconstructionofmarketaccessroadsoffered substantialbenefitstopoorfamilies.However,suchimprovementwasnotconsistentenoughtosignificantlyreduce incomeinequality.
TheroleofroadsasoneofthefactorsthatcontributetoaffectpovertyincidencewasstudiedbyKwon(2001)in25 Indonesianprovincesbetween1976and1996.Withtheuseofinstrumentalvariabletechniques,resultsshowedthat thesignificanteffectthatroadsexertedonpovertyreductionwashigherinprovinceswithgoodroadaccessthanthose thatdidnothavesuchinfrastructure.
roadaccesspresented lowerpovertylevels.InPeru,Toreroetal.(2001)analyzedthesignificanceofinfrastructure (drinkingwater,sewagesystem,electricityandtelephone)onpovertyintheyears1985,1991,1994and1996.Obtained resultsshowedthathavingatelephonelinecontributedtoreduceurbanpovertymorethanotherinfrastructureservices, althoughthistypeofinfrastructurehasnosignificanteffectsinruralareas.InanotherPerustudyontheevolutionof povertythroughtimeanditsdeterminantsfrom1997to1999,HerreraandRoubaud(2002)demonstratedthataccess topublicinfrastructureservicessignificantlyreducestheprobabilitiesoffallinginto“permanent”poverty.Besides,in thecaseoffamiliesthathavealwaysbeenpoor,theaccesstotheseservicesincreasestheirchancesofleavingpoverty.
Fanetal.(2002)analyzedtheeffectsofdifferenttypesofpublicexpensesoneconomicgrowthandruralpovertyin ChineseandVietnameseprovinces,findingthatpublicexpensesonruralroadshaveabigimpactonpovertyincidence. Theresearchshowedthat poorfamilieslivinginruralmunicipalitieswithpavedroadshave67%morechancesof overcomingpovertythanthoselivingincommunitieswithoutthiskindofinfrastructure.
Confirmingtheseempiricalevidencesontherelationbetweenpovertyandinfrastructure,Warr(2005)provedthat intheyears1997/1998and2002/2003,thereductionofruralpovertyinLaoswasattributedtotheimprovementsin roadaccesses.
ThemethodologyofstaticanddynamicpaneldataisusedinastudycarriedoutbySeetanahetal.(2009)tomeasure therelevanceofinfrastructureonurbanpovertyina20-countrysamplecollectedbetween1980and2005.Government expensesonroadsandcommunicationsareappliedasaninfrastructureproxy.In bothmodels,theydiscoveredthat transports andcommunicationsare efficient instrumentstocombat poverty inurban areas. Applying the Granger CausalityTest,theyidentifiedareversecausalitybetweenpovertyandinfrastructure.
Inastudycarriedoutin91countriesincludingBrazil,RajkumarandSwaroop(2008)usedacross-sectiondataon governmentexpendituresonhealthandeducationin1990,1997and2003,verifyingtheexistenceofareversecausality betweenpovertyandhealthinfrastructure.Theycorroboratedthatwhenagovernmentfacesasituationofpovertyand healthdeteriorationamongitscitizens,orevenincriticaleducationsituations,theStatetendstoincreaseexpensesin theseareas.
Anotherstudythatusedpublic expensesasinfrastructureproxyfor25Indonesianprovincesfrom1976to1996 wasdevelopedbytheAsianDevelopmentBankandtheResourcesCenterforEconomicDevelopment(1999).The studyprovedthatthereductioninpovertyrateswastheresultofinvestmentsinroads,health,agriculture,education, sciencesandtechnology.Byapplyingpaneldatamodels(fixedandrandomeffects)toreflecttheimpactofdifferent infrastructuretypes(access towater, sewage system, electricity andtelephone)on domestic expensesinPeru for the2007–2010period,Aparicioetal.(2011)verifiedadifferentiatedandsignificantimpactofinfrastructureonthe reductionoftransientandchronicpoverty,dependingonthegeographicalarea(urbanorrural),thesexandthefamily head.
Astudy of 73Philippines provincesdeveloped byBalisacan (1999)withdatafrom 1988to1997showed that changes in access toelectricity were strong andpositively correlatedto the reduction in poverty levels. Another researchdevelopedinthePhilippinesbyBalisacanandPernia(2002)intheeighties andninetiescorroboratedthat electricitypositivelyaffectedtheincomeofthepoor.
Runsinarith (2008)found significant impacts of mobiletelephones, electricity,irrigationand roads onpoverty incidencethroughtheapplicationofquantileregressionsinCambodiafortheyear2006.Heconcludedthatmobile technology wasthe infrastructurewiththehighest impactonpovertyreduction, followedbyelectricity,roads and irrigation.
Withthedevelopmentoftwoinfrastructureindexes(physicalandsocial)createdthroughthemethodofmain compo-nents,Roy(2009)detectedastrongnegativecorrelationbetweenthehumanpovertyindexandphysicalinfrastructure (roads,electricity,irrigationprojects,etc.)andsocialinfrastructure(hospitals,schools,etc.)inIndiafortheperiod 1981–2001.Ontheotherhand,Ogun(2010)basedondatarelatedtothe1970–2005periodfoundthatthedevelopment insocialandphysicalinfrastructuresignificantlyreducedpovertyinurbanareasofNigeria.
Ghoshand De (2000),evaluatedphysical infrastructures insouth Asian countriesinthe eighties andnineties, demonstratingthatdifferentiatedamountsofphysicalinfrastructurewereresponsibleforthegrowingregionaldisparity insouthernAsia.
Byusinganon-balancedpanelof121countriesfrom1960to2000,CalderonandServen(2004)appliedquantitative infrastructureindexesandqualityindicatorsthatdemonstratedpositiveandsignificantinfrastructurestockeffectson theincomelevelandeconomicgrowthofthesecountries.Theseauthorssustainthatthedevelopmentofinfrastructure favorsabetterincomedistributionandconsequently,areductioninpovertylevels.
Thenationalliteratureontheimportanceofinfrastructureinthereductionofpovertyisveryrich.Wecanhighlight theworksofCruzetal.(2010),whobasedondataobtainedfrom1980to2007concludedthatfederalandstatepublic expensesoneducation,healthandphysicalcapital(roadsandenergy)areextremelyrelevantfortheincomegeneration andproductivitygrowth,whichsomehowallowtoreducepovertylevels.
InamultidimensionalanalysisofpovertyinBrazilfortheperiodfrom1990to2004,usingdatafromtheNational ResearchperSampleDomiciles(PNADs)KageyamaandHoffmann(2006)verifiedthattherewasanimprovement tendencyininfrastructureconditions,beingthistrendlargelyresponsibleforthepovertyreduction.
BertussiandEllery(2012)investigatedtheimpactofpublicexpensesintransportbasedontheeconomicgrowth ofBrazilian statesbetween1986 and2007,using panel data.Theseauthorsverified that the publicinvestmentin thetransportsectorprovokedapositiveandstatisticallysignificanteffectonthelong-termeconomicperformanceof Brazilianstates.Besides,thisinvestmentpotentiallycontributedtothereductionofincomeinequalityamongdifferent states.
WhenanalyzingtheeffectsofinfrastructureonBrazilianproductivityfrom1950to1995fromanempirical perspec-tive,FerreiraandMalliagros(1998)estimatedproductandproductivityelasticitywithregardstothecapitalandthe infrastructureinvestment.Theyestimatedtheunbundledimpactofinfrastructureexpensesinfivesectors(electricity, telecommunications,railroads,highwaysandports)ontheGDPandtheproductivityofprivatefactors.Amongthemain resultsobtained,theyverifiedastrongrelationbetweeninfrastructureandthelongtermproduct,alsocorroborating thatthetotalproductivityoffactorsisnotGranger-causedbyproductivity,butrather,theotherwayround.
Likewise,MussoliniandTeles(2010)analyzedtherelationbetweeninfrastructureandtotalproductivityoffactors, whichwasconsideredthemainlong-termgrowthtriggerfrom1950to2000.ResultswereoppositetothoseofFerreira andMalliagros(1998).Thismeansthatinfrastructureimprovementforacertainprivatecapitalstockmaycauseinthe Grangersense,thetotalproductivityoffactors.However,SílviaandTriches(2014)analyzedeffectsofgovernment expensesintheinfrastructure,communications,transport,healthandbasicsanitationsectorsintheBrazilianeconomy productfrom1980to2005.Amongthemainresultsobtained,publicexpensesininfrastructureintheanalyzedsectors showedpositiveandstatisticallysignificantimpactsonthegrowthofBrazilianproductduringthestudiedperiodof time.
3. Povertyandotherdeterminants
3.1. Relationbetweenpoverty,economicgrowthandincomeinequality
Severalnationalandinternationalempirical worksconfirm thecommonsense ideathateconomicgrowthhelps povertyreliefeffortsintwoways:by expandingthenumberof jobsandbyincreasing the realwages paidtothe workers.Povertyreductiondependsbothongrowthratesaswellasondecreasinginequalitylevels.
Forexample,Anderson(1964),Thorntonetal.(1978)andHirsch(1980)analyzedthisrelationthrougha trickle-downeconomicgrowthmodel for theUnitedStates. Thebasicconcept isthat althoughgrowth primarilybenefits peopleinthehighestincomedistributionpyramid,arobustgrowthtendstohelpthosewhoareinthelowestincome distributionlevel as well.However,anumberof recentstudiessuggestthat theeconomicexpansion that America experiencedintheeightieshadnostatisticallysignificanteffectonaggregatedpoverty.Blank(1993)andFormbyetal. (2001)affirmthataggregatedpovertywaslesssensitivetotheAmericaneconomicexpansionof1980thantotheone enjoyedin1960.
UnlikeotherauthorssuchasRavallionandHuppi(1991),DattandRavallion(1992)andKakwani(1993)dotake intoconsiderationpovertyanditscauses.Theyareparticularlycarefultodistinguishtheeffectsofgrowthonpoverty reductionaswellasdistributionvariations.
AccordingtoRanisandStewart(2002)whenanalyzingdatafromseveralLatinAmericancountries,itisclearthat economicgrowthisnotalwaysenoughtoeliminatepoverty.Inthesixties,seventiesandeightiesinBrazil,forexample, therewasastrongbiasaboutbeingpro-economicgrowthbutwithalowhumandevelopment.
Inthenineties,KageyamaandHoffmann(2006)affirmedthatBrazilhadentereda“viciouscircle”inwhichlow humandevelopmentstandardsstartedtolimiteconomicgrowth.Contrarytowhathappenedintheeighties, inthe ninetiestherewasageneralincreaseinsocialexpensesinLatinAmericaincluding Brazil,whichcouldpotentially preparethepathfornewgrowthstandardsinthecurrentdecade.
EconomicgrowthisfundamentaltoreducepovertybutBarreto(2005),Hoffmann(2005),deLimaetal.(2003)and
MenezesandPinto(2005)highlightthatitseffectsaremultipliedonthepoorestwhenaccompaniedbyredistribution policies.
ForGafar(1998),growthisanecessaryconditiontoreducepovertyasitincreasesemploymentopportunities,the standardoflivingandrealwages;howeverthisisoftennotenough,aswhengrowthisurban-biasedwithintensive capitalandqualifiedemploymentconcentration,povertycanincreaseevenwithagrowthinthepercapitaGDP.
Economicgrowthopportunities,accordingtoRocha(2006)tendtohaveconcentratingeffects,astheyimplythe applicationofmoderntechnologiesassociatedtotheuseofqualifiedmanpower,whichrequirescompensatorymeasures toavoidanincreaseininequality,aswellasinitiativestopromotethereductionofabsolutepoverty.
Thepersistenceofabsolutepovertyinthecountryis,accordingtoRocha(2006)originatedbytheinequalityexistent whenyieldisconsidered.Heremarksthatabsolutepovertymaybereducedboththroughtheincreaseinincomeas wellasthroughtheimprovementinitsdistribution,howeverthereisaconsensusthatthefocusmustbeonreducing inequality.Thisisbecauseincomegrowthwithoutinequalityreductionmeanstransferringtheeliminationofabsolute povertyinthecountrytoadistantfuture.
BarrosandMendonc¸a(1997)andBarrosetal.(2007),usingPNADdatafrom1993,verifiedthat areductionin povertylevelswithinasocietyrequireseconomicgrowthoradeclineininequalitylevels.Thisfactiscertainlyoneof themainreasonswhythegoalsofpublicpoliciesarefocusedonthesearchforbothgrowthandequality.Theynotice analmostlinearrelationbetweeneconomicgrowthandpovertyreduction.
Inanother,thereferredauthors,whenanalyzingtheperiodbetween2001and2005,verifiedthattheincomegrowth ofthepoorest10%reached8%peryear,whichresultedinaremarkablepovertyreductionthatincludedadeclinein inequalitylevels.TheyalsoobservedtwodesirabletransformationsintheBrazilianincomedistribution:therewas growth(howevermodest)andinequalitydecreasedsignificantly(theGinicoefficientfell4.6%).Thenoveltyduring thisperiodwasthatcontrarytootherhistoricepisodesofdramaticpovertyreduction,thistime,thedrivingforcefor thisimprovementwasnotgrowthbutastrongdecreaseininequalitylevels.
NederandSilva(2004)developedmethodologicalapplicationstoestimatepovertyratesandincomedistribution inBrazil’sruralareasbasedonPNADdatafrom1995to2001.Theirresultsalsoconfirmedasignificantreductionin incomeconcentrationinruralareasinmostoftheanalyzedstates.
Moreiraetal.(2010)evaluatedtheeffectofgrowthandinequalitycomponentsinpovertyvariationsforBrazilian statesbetween1996and2007.Theyanalyzedtheexistenceofbarriersoriginatedintheinequalityandpovertyconditions thatmayimpedeapercapitaincomegrowthinBrazilianstates.Theyverifiedthatvariationsinstatepovertydynamics continue tobe more sensitive to income distributionthan to growth,reflecting the persistent regional disparities observed.
Ontheotherhand,MarinhoandAraújo (2012)basedonpaneldataforBrazilianstatesinthe1995–2009term, analyzedtheimpactofeconomicgrowthvariationsandincomeinequalityonpovertyalterationsinBrazil.Estimated resultsconcludedthattheincomegrowthrelatedtopovertyreductionislowerwhentheinitialdevelopmentlevelis lowandthatthesameresultisseenwhentheinitialinequalitylevelishigh.
3.2. Relationbetweenpovertyandaverageyearsofschooling
Shultz(1973),oneofthepioneersofthehumancapitaltheory,affirmsthatnowadayspeopleareinvestingmoreand moreinthemselvesashumanassetsandthatsuchhumaninvestmentsareconstitutingagrowinginfluenceoneconomic growth,asthebasicinvestmentinhumancapitalstartswithformaleducationandtraining.Therefore,educationhas as itsfundamental role,the developmentof skillsandknowledge seekingtoincrease productivity, promoting the obtainmentofcognitiveskills.Finally,thehighertheproductivitygain,thebiggertheincomeapersonwillget,thus improvingitssocialposition.
ForEnrenbergandSmith(2000),theimprovementineducationlevelsresultinproductivitygrowths,whichalso raise real wagesin agreement withthe humancapital theory.Thisway, regionswithahigh humancapitalstock boastahigheraveragesalarythantheothers.Besidesincreasingwages,knowledgeconcentrationgeneratespositive externalitiesfortheregionasgrowthstandardsbecomemoredynamic,inducingthearrivalofnewinvestmentsand thedisseminationofnewknowledgeandskills.
ReisanddeBarros(1990)andQueiroz(1999)statethattheeducationvariablemeasured byyearsofschooling canbetterexplaindifferencesinindividualperformanceamongdifferentregionsthroughacertainperiodoftime.A concentrationofhumancapitalstocktendstobenefitthemostdevelopedcities(withmoreformaleducation)tothe detrimentofthelessdevelopedmunicipalities(lesseducated)generatinganever-growinggapinregionalwagelevels. AhumancapitalanalysisthrougheducationinBrazil,ascarriedoutbyVilela(2005)demonstratedthatduringthe 1991–1996quinquennium,regionaleducationparticipationintheHumanDevelopmentIndex(HDI)wasveryhigh: 62%inthe Midwest,60% intheSoutheast,59%intheNortheast,54%intheSouthand42%intheNorth.These figuresevidencedtheimportantcontributionofthisvariableinpovertyreductionamongBrazilianregions.
UsingPNADdatafrom1999,Rocha(2006)noticedthateducationindicatorsinBrazilprovide evidenceof the correlationbetween loweducational level andpoverty. Foradult individuals(25 or over) the disadvantagethat a lowschoolinglevelrepresentsintermsofpovertyincidencewasevident,asthepercentageofpoorpeopledeclines monotonicallyaccordingtotheschoolinglevel.Only2.1%ofindividualswithsomesortofhighereducationarepoor.
Marinhoetal.(2011)analyzedtheimpactofincometransferprogramsonpovertyinBrazil,controlledbyother determinantssuchaseconomicgrowth,incomeinequalityandmeanyearsofschooling,percentageoffamiliesledby womenandmaleunemploymentratefrom2000to2008.Amongthemostrelevantresults,theyfoundthatanincrease inthemean yearsof schooling contributedtoreduce povertyduring theanalyzed period.They highlight that the effectofeducationpoliciesonpovertylevelsisrelativelyhigherontheproportionofpovertythanonthoseconsidered extremelypoor.Theyalsoverifiedthatincometransferprogramsdonotsignificantlyaffectpovertyreduction.
3.3. Relationbetweenpoverty,unemploymentratesandgovernmentrevenues
Thereisageneralconsensusintraditionaleconomicsliteraturethatunemploymentispositivelyrelatedtopoverty. WhenanalyzingpovertydeterminantsintheUnitedStates,Formbyetal.(2001),verifiedthatthevariationinthe maleunemploymentratehasasignificantimpactonpovertywhenthelinearregressionmodelisapplied.However,
EndersandHoover(2003)foundthatusingthesamedatabase,thiseffectisonlysignificantifnon-linearregression modelsareused.
Forthissameeconomy,Hirsch(1980)analyzedthereasonswhypovertyonlyexperiencedaslightdeclineeven withthestrongeconomicgrowthinAmericain1980.Oneoftheexplanationswasthatevenwhenareductioninthe unemploymentrate benefitedthe poorest,adeclineinreal wagesmorethancompensatedfor theprioreffect. The inclusionoftheunemploymentrateinhismodelcontrolledtheeffectofthebusinesscycle.
Analyzingdatafrom the2000Census, Barbosa(2004)suggestedthat thereare unemploymentrate differences betweenthepoorandthenon-poorinthedifferentgeographicalareasthroughoutBrazil;howevertheyappeartobe higherinmetropolitanareas.He noticedthat povertyandunemploymentarehighlyrelated, thatis,povertycanbe explainedthroughunemploymentorthestructureofthelabormarket.
Seekingtomeasuretherelevanceofinfrastructureonurbanpoverty,asampleof20countriesfortheperiodbetween 1980and2005waselaboratedbySeetanahetal.(2009),whoverifiedthateducation,inflation,governmentrevenues, pavedroads,communicationsandtheunemploymentratesignificantlyimpactedonthereductionofurbanpoverty.
Anotherrelevantresultwasthatwhenthegovernmentrevenuesareincreasedin1%,urbanpovertydeclines23%. Itisworthhighlightingthatthisisoneofthefewstudiesthatrelatepovertytogovernmentexpenses.
4. Databasedescriptionsandanalysis
Thedatabase usedwasobtained fromPNAD,IPEADATAandFINBRAforthe differentstatesandtheFederal DistrictofBrazil, comprisingthe1995–20111 periodoftime.It isworthhighlightingthat theArellanoandBover (1995)andBlundellandBond(1998)models,describedindetailinthenextsectionarevalidforasmalltemporal dimensioninwhichobservationalunitsmustbeconsiderablyhigher.
Theabsolutepovertyindicatorappliedwastheproportionofpoorpeople(P0)whobelongtotheclassproposed
byFosteretal.(1984).Forthecalculationofsuchindex,thepovertylinedefinedbyIPEAwasconsideredaccording topricesofSeptember2011,whichconsidersthevalueofthelineequaltohalfminimummonthlysalary.Toupdate familyincome,2 theINPC(National ConsumerPricesIndex—Restricted) wasapplied,as publishedbyIBGEand correctedthroughthemethodologysuggestedbyCorseuilandFoguel(2002).Thispovertyindexiscalculatedbased onthefollowingequation:P0=q/n,whereqisthenumberofpoorindividuals(numberofpersonswithapercapita
incomebelowthepovertyline)andnisthepopulationsize.
Despitebeingimportantandeasytocalculate,thisindicatoronlydescribesthepovertyextension,beinginsensitive tothepovertyintensity.Thismeasurementisnotalteredbyreducingtheincomeofanindividualbelowthepoverty line,or ifsuchincomegrowsbutdoesnotreach thepovertyline.Theproportion isalsoinsensitive tothe income distributionamongthepoor,notbeingalteredwhenincomeistransferredfromaverypoorindividualtoanotherone whoislesspoor.
Theunemploymentratewascalculatedbasedontherelationbetweentheunemployedpopulationandthe econom-icallyactivesectorasobtainedfromPNADsdata.
ThedataobtainedfromIPEADATAwas:stateGDPpercapitaatconstantpricesinBrazilianreaisfor2011deflated bytheimplicitnationalGDPdeflator,theaverageschoolingyearsforpeopleagedtwenty-fiveormoreandtheGini indexasameasurementofincomeinequality.
StategovernmentbudgetfundsandpublicStateinfrastructureexpensespercapitainthestrategiceconomysectors (energyandmineralresources,transportation,communication,healthandsanitation)werecollectedfromFINBRA anddeflated bythe GeneralPrice Index—InternalAvailability(IGP-DI) basedonthe year2011.Aproxy forthe infrastructureindexwascreatedwithapercapitaaverageofStateexpensesinthesestrategicsectors.
5. Econometricmodel
Thissectionintroducestheempiricalmodelthroughwhichtherelationbetweenpovertyandinfrastructurecontrolled byotherdeterminantsisanalyzed.Theeconometricspecificationisbasedontheassumptionthatcommonpoverty tendstobeperpetuatedand/oraffectitsfuturedevelopment.TheempiricalevidenceofthisphenomenonforBrazil canbeobservedinthestudyperformedbyRibasetal.(2006),wheretheauthorsdemonstratethatpovertyinBrazilis essentiallychronic.
Therefore,takingintoconsiderationthisevidencetogetherwithotherpovertydeterminants,theadequate econo-metricmodeltoanalyzetheseinteractionsmustbeadynamicpaneldatamodel.Consequently,forBrazilianstatesand comprisingtheperiodfrom2000to2009,themodelisdefinedasfollows:
Pit =β0+β1Pit−1+β2infit+β3pibit+β4aemit+β5giniit+β6regovit+β7desit+ηi+εit (1)
1 Datafortheyear2000wasgeneratedthroughinterpolation(arithmeticalaverage)usingPNADsfrom1999and2001.Datafor2000and2010
wasgeneratedbyinterpolation(arithmeticmean)usingPNADsfrom1999to2001and2009to2011respectively.
2 Asmonthlyfamilyyield,thesumofthemonthlyincomeofalljobsfromallfamilymemberswasconsidered,excludingthosefamilymembers
wherethe variablePit is the povertyindex proportion, infit is theinfrastructure index,pibit isthe state GDP per
capita,aemitistheaverageyearsofschoolingforpeopleaged25orover,giniitistheGiniindex,regovitarethestate
governmentsbudgetrevenues,desitistheunemploymentrate,ηirepresentsnon-observableuniteffects,εitarerandom
disturbancesandiandtrepresenttransversalandtemporalobservationindexes(states).
Thehypothesisadoptedinthismodelis:E[ηi]=E[εit]=E[ηiεit]=0eE[εitεis]=0fori=1,2,...,Ne∀t =/ s.
Addi-tionally,thereisastandardhypothesisrelatedtotheinitialconditions:E[Pk,i0εit]=0fori=1,2,...,Net=1,2,...,T(Ahn andSchimdt,1995).
TraditionalestimationtechniquesareinadequateforEq.(1)duetotwomaineconometricproblems.Thefirstoneis thepresenceofnon-observationaleffectsintheηiunitsandthesecondistheendogeneityoftheexplanatoryvariable Pit−1(one-periodlaggeddependentvariable).3Inthiscase,omittingfixedindividualeffectsinthedynamicmodelpanel
makesordinaryleastsquare(OLS)estimatorstendentiousandinconsistent.Forexample,duetoapossiblepositive correlationbetweenthelaggeddependantvariableandthefixedeffects,thecoefficientestimationβ1isupwardbiased
(Hsiao,2004).
Ontheotherhand,thefixedeffectestimator,whichcorrectsforthepresenceofheterogeneityintransversalunits, generatesandestimationofdownward-biasedβ1inpanelswithasmalltemporaldimension.Studiesdevelopedby
MonteCarlo,JudsonandOwen(1999)showedthatthisbiascanreach20%,eveninpanelswhereT=30.Thesecond problemisduetotheprovableendogeneityofexplanatoryvariables.Inthiscase,anendogeneityontherightsideof Eq.(1)mustbetreatedtoavoidapossiblebiasgeneratedbyasimultaneityproblem.
Seekingtocorrecttheseproblems,ArellanoandBond(1991)proposedtheestimatorfortheGeneralized Differen-tiatedmethodofMoments(GMM).Suchmethodconsistsofeliminatingfixedeffectsthroughthefirstdifferencein Eq.(1),thatis:
Pit=β1Pit−1+β2infit+β3pibit+β4aemit+β5giniit+β6regovit+β7txdesit+εit (2)
wherefor avariableZit any,Zit=Zit−Zit−1.Bythe constructionofEq. (2),Pit−1 andεit are correlatedand
therefore,OLSestimatorsfortheircoefficientsshallalsobetendentiousandinconsistent.Inthiscase,itisnecessary toemployinstrumentalvariablesforPit−1.Thesetof hypothesisadoptedinEq. (1)implythattheconditionsin
momentsE[Pit−sεit]=0,fort=3.4,...Tands≥2,arevalid.Basedonthesemoments,ArellanoandBond(1991)
suggesttousePk,it−s,fort=3.4,...Tes≥2,asequationinstruments(2).
Withregardstootherexplanatoryvariables,therearethreepossiblesituations.AnexplanatoryZitmaybequalified
as(i)strictlyexogenous,ifnotcorrelatedtothetermsofpast,presentandfutureerrors;(ii)franklyexogenous,ifitis onlycorrelatedtopasterrortermvaluesand;(iii)endogenous,ifcorrelatedwithpast,presentandfutureerrorterms. Inthesecondcase,Zitlaggedvaluesinoneormoreperiodsarevalidinstrumentstoestimateequationparameters(2).
Asforthelastcase,ZitlaggedvaluesfortwoormoreperiodsarevalidinstrumentsforEq.(2).
Meanwhile,ArellanoandBover(1995)andBlundellandBond(1998)arguethattheseinstrumentsareweakwhen thedependantandexplanatoryvariableshaveastrongpersistenceand/ortherelativevarianceoffixedeffectsincreases. Thisproducesanon-consistentbiasedGMMestimatorforpanelswithasmalltemporaldimension.ArellanoandBover (1995)andBlundellandBond(1998)suggestasystemthatcombinesasetofequationsindifferenceasawaytoreduce thebiasandimprecisionproblems(Eq.(2))withasetofleveledEq.(1).Thatiswherethegeneralizedmomentssystem comesfrom.Fordifferenceequations,thesetofinstrumentsisthesamedescribedabove.Forthelevelregression,the mostadequateinstrumentsarethelaggeddifferencesoftherespectivevariables.
Forexample,assumingthatexplanatoryvariabledifferencesarenotcorrelatedtothefixedindividualeffects(for
t=3.4,...T)andE[Pi2ηi]=0,fori=1,2,3,...,N,thenthedifferentexplanatoryvariables,eitherexogenousorfrankly
exogenous,andPit−1,arevalidinstrumentsforlevelequations.ThesamehappenstothePit−1explanatoryvariables
inlaggeddifferencesforagivenperiodiftheyareendogenous.
TheconsistenceoftheGMMsystemestimatordependsonthesuppositionoftheabsenceofserialcorrelationin theerrortermandthevalidityofadditionalinstruments.Consequently,atfirst,hypothesiswithnocorrelationabsence
3 ThevariableP
it−1isendogenoustothefixedeffectηiinEq.(1),originatingabiasinthedynamicpanel.Considerastatethatexperiencesa
of firstandsecond residualorderaretested.Seekingtocheckthatparameterestimatorsare consistent,thelackof firstorderautocorrelationhypothesismustberejectedandthesecondorderautocorrelationisaccepted.Lateron,the HansentestisperformedtoverifythevalidityofinstrumentsusedandtheSargantestisalsocarriedouttocheckthe validityoftheadditionalinstrumentsrequiredbytheGMMsystemmethod.
Resultsareintroducedinthefollowingsectionandvariancesestimatorsforparametersarerobustwithregardsto theheteroscedasticityandtheautocorrelationobtainedthroughtheGMMsystem.Theestimatorobtainediscorrected throughtheWindmeijer’smethod(2005)toavoidthattherespectivevarianceestimatorsunderestimatetherealvariances inafinitesampling.
5.1. Grangercausalitytestforpaneldata
Theconceptofcausalityreferstothecapacityofavariabletohelpinthebehaviorforecastofanothervariableof interest.Itisabouttheexistenceoftemporalanteriorityintheexplanationofadeterminedvariable.Oneadvantageof non-causalitytestsisrelatedtothefactthatintheory,theyareimmunetotheproblemofendogeneity(orsimultaneity bias)asonlylaggedvaluesofendogenousvariablesappearontherightsideoftheequations.
AccordingtoGranger(1969),inabi-variablestructure,thefirstvariableissaidtocausethesecondvariableinthe Grangersenseiftheforecastforthesecondvariableimproveswhenlaggedvaluesfromthefirstoneareconsidered.In thisarticleweusetheGrangercausalityprocedureforpaneldataassupportedbyHurlinandVenet(2004)andHurlin (2004,2005).Thistechniqueisthereforeutilizedtoperformaspecificcausalityexistencetest, aswellas todetect thedirectionofanycausalityamongvariables,beingconsistentwiththestandardGrangercausalitywherevariables withinthesystemneedtobestableovertime.
ThecausalitytestproposedbyHurlinandVenet(2004)andHurlin(2004,2005)appliestoheterogeneouspanel datawithfixedeffects.Initially,considerthefollowingautoregressivemodelwithTperiodsandNcross-sectionunits:
Yit=αi+ K
k=1
γi(k)Yi,t−k+ K
k=1
β(ik)Xi,t−k+εi,t
whereKisthenumberoflagsandγi(k)andβ(ik)arethecoefficientsofYi,t−kandXi,t−ktobeestimated,i=1.2,...N
andt=1.2,....T.
ItisassumedthatindividualeffectsαiarefixedandKlagsareequalforallunits.Asfortheautoregressiveparameters γi(k)andinclinationregressioncoefficientsβi(k),theydifferamongindividualunits,however,thesefixedparametersin timeareconsidered.
ThistesthasanullhypothesisH0,thehomogeneousnon-causality(HNC)fromthevariableXtothevariableY,
whichmeansthatthereisnocausalrelationforallstatepanels.ThealternativehypothesisH1considerstheexistence
ofacausalrelationoraheterogeneousnon-causality(HENC),fromXtoY,foratleastonecross-sectionunit. Therefore,thetestfortheHomogeneousNon-Causality(HNC)shallbegivenbyH0:βi=0,∀i=1,...Nagainstthe
alternativehypothesisofHeterogeneousNon-Causality(HENC)H1:βi =/ 0∀i=N1+1,N1+2,...,N,whereN1isan
unknownvalue,howeveritsatisfiestheconditionthat0≤N1/N<1.
ThistestconsidersthemeanoftheindividualWaldtestofnon-causalitybetweenNcross-sections.Theindividual WaldstatisticassociatedtothenullhypothesisofHomogeneousNon-Causalityisgivenby:
WN,THNC = 1 N
N
i=1
Wi,T
whereWi,TrepresentstheindividualWaldtestfortheuniti.BasedonthehypothesisofNon-Causality,wehavethat
eachoftheindividualWaldstatisticsasymptoticallyconvergestoachi-squaredistributionwithKdegreesoffreedom. Ontheotherhand,themeancrosssectionWN,THNCconvergestoanormaldistributionwhenTandNtendtobeinfinite.
ThestandardizedstatisticofWN,THNCmaybeobtainedasfollows:
ZN,THNC=
N
2K
WN,THNC−K →d
whereT,N→∞representsthefactthatT→∞firstandthenN→∞.
Forsometemporaldimension, when Tis fixed,the convergence of Wi,T maynot bereached. Thismeans that
althoughtheWaldstatisticsmaypossessthesamesecondordermoments,theymaynothavethesamedistribution. Takingthisintoaccount,HurlinandVenet(2004)andHurlin(2004,2005)proposedandapproximationofthetwo firstunknowndistributionmomentsofWi,T withthetwoFisherdistributionmoments.Withthisapproximationand
consideringT>5+2K,asemi-asymptoticstatisticmaybedefinedthroughthefollowingexpression:
˜
ZHNCN,T =
N
2K×
(T−2K−5) (T −K−3) ×
(T −2K−3) (T −2K−1)W
HNC
N,T −K
d
→
N→∞N(0,1)
6. Results
Theestimatedresultsoftheparametersobtained fromEq.(1)withtheaidofEq.(2)wereobtained throughthe econometrictechniquesintroducedinSection5andarenowenteredinTableA1.
Inthe modelestimated through theGMM-S, explanatoryvariablesconsideredendogenous werethe dependent one-periodlaggedvariable(Pit−1)andthepercapitaGDP(pibit).Thevariablesinfitandginiitweretreatedasfrankly
exogenousandtheotherswereconsideredasbeingstrictlyexogenous.
Initially,itwasverifiedthatthevalueofthecoefficientestimationofβ1forPit−1throughtheGMMmethod(column
[c])wasbiggerthantheoneobtainedthroughtheEF(column[B]andsmallerthantheoneobtainedthroughMQO (Column[a]).AsdiscussedinSection5,MQOandEFestimationsforβ1areupwardanddownwardbiasedrespectively,
providingapproximatesuperiorandinferiorlimitstoguidetheβ1estimationthroughtheGMM-S.4Inthissense,the
estimationbiasofβ1seemstohavebeenminimized.
ThetestsperformedthroughtheGMMmodelrevealthatthemodelstatisticalpropertieshavebeenrespected.The HansenandSargantests,whichcheckifinstrumentsusedandadditionalinstrumentsrequiredbytheGMMsystem arevalid,aresatisfied.Finally,wealsoincludetheArellanoandBond(1991)statisticalteststoevaluatetheexistence offirstandsecondorderautocorrelation.Notethat theabsenceofsecond orderautocorrelation isessentialfor the consistenceoftheGMM-systemestimator.Thetestconfirmsthenon-rejectionoffirstorderautocorrelationalthough thesecondorderautocorrelationhypothesisisindeedrejected.
ThepositiveandsignificantpositiveestimatedcoefficientachievedthroughtheGMMrelatedtothelaggedpoverty index(P0,it−1,column[c])suggeststhatpoverty isadynamic andpersistence process,once theresponsecapacity
ofpovertyinthecurrentperiodwithregardstopastvaluesishigh,thusconfirmingthehypothesisofitspersistence (viciouscircle).
Theremarkableresultisthestatisticalsignificanceoftheeffectofinfrastructureonthemeasurementoftheanalyzed poverty.Weobservethateventhroughinadequateestimationmethods(MQOandEF),thecoefficientofthisvariable isstatisticallysignificant andpresentsthe expectedsigns. Thissuggeststhat investmentsininfrastructuredirectly affectthetemporaltrajectoryofpovertyinBrazil,implyingthatthereisapovertyreversion.Thisempiricalevidence validatestheideathatinfrastructurehasbeenfundamentalforpovertyreduction,beingconsistentwithresultsfound inthespecializedliteratureasdiscussedinSection3.
Somehow,theseresultscorroboratethoseofBertussiandEllery(2012)astheystudiedtherelationbetweenpublic spendingandeconomicgrowthinBrazilfrom1986to2007.Theseauthorsconcludedthatpublicinvestmentinthe transportsectorprovokesapositiveeffectontheeconomicperformanceofBrazilianstatesandcontributestoanincome inequalityreductionamongthem.Inagreementwiththesefindings,FerreiraandMalliagros(1998)andMussoliniand Teles(2010)highlightthatimportantlong-termrelationbetweeninfrastructureandeconomicgrowthinBrazil.
Thispro-poorimpact inthe Braziliancontextmaybe assigned toseveralfactors, as the proxyapplied for the infrastructurevariableisanindexcomposedoffourtypesofpublicexpensespercapita:transportation,energyand mineralresources,communicationsandhealthandsanitation.Electricityreflectstheaccesstotechnologyanddirectly contributestotheimprovementofemploymentlevelsandincomeforthepoorestthrougheconomicgrowth.Investors tendtolocatetheirbusinessesinareasthathaveservicesbasedonefficienttechnology,informationandcommunications.
Theexpansionoftheseservicesinacountryasawholemaystrengthentheinvestmentandcorporateatmosphere,thus improvingthegeneralstateoftheeconomyandcreatingapositiveenvironmentforthelow-incomepopulation.
Theprovisionofdrinkingwaterandadequatesanitationservicescanhelpincreasetheaggregatedeconomicgrowth, whichtranslatesintomorejobsandbettersalariesforthepoor.Thisgrantsthatthepoorcanhaveaccesstodrinking water sourcesandadequatesanitation servicesthat improve healthconditions,thusreducing job absenteeismand increasingincome.
Amongtheotherdeterminants,thepercapitaGDPandtheaverageyearsofschoolingalsoprovidedtheirrespective estimatedcoefficientswithinexpectedresultsandwerestatisticallysignificant.Allthesefactorshavecontributedto povertyreduction.Suchempiricalevidencesareconfirmedbynationalandinternationaleconomyliterature,asdetailed inSection3.
The unemploymentrate alsopresented asignificantpositivecorrelationwiththepoverty index.Inthe end,the highertheunemploymentrate,thehigherthepercentageofpoorpeople.Thisshowsthatincludingtheunemployment rateinthemodelforbusinesscyclesandmacroeconomicpoliciescontrolpurposeswasagoodidea.
ThepositiveandsignificantcoefficientoftheGiniindexsuggeststhatincomeinequalityinBrazilstronglycontributes tothegrowthofpovertylevels.ThisresultcorroboratesthosefoundininternationalarticlessuchasKalwijandVerschoor (2004)andBourguignon(2004)aswellasstudiesbyMarinhoandSoares(2003),andHoffmann(2006)andSantos (2008)forBrazil.Besides,theestimatedvalueof thecoefficientforthisvariableismuchhigherthantheGDPper capita.Therefore,policiesaimedatinequalityreductionaremoreeffectivetofightpovertythanthoseaimedsolelyat boostingeconomicgrowth.
Asforthestatebudgetrevenues,theyofferedresultsaccordingtoourexpectations.Resultsrevealanegativerelation betweenthisvariableandpoverty,whichratifiesconclusionsreachedbySeetanahetal.(2009)forasampleof 20 developingcountriesincludingBrazil.Infact,inrecentyears,moregovernmentfundshasbeenusedtoreassignincome tothepoorestthroughincometransferprograms.
Theseresultsprovethatpoliciesthatencourageinfrastructureinvestmentandfostergrowth,incomedistribution andeducationareimportanttofightpovertyintensity.However,ifinfrastructure,GDPgrowthandeducationpolicies increaseincomeconcentration,theymayresultinonlymoderateresultsorevenworsenpoverty.
Thecausalitytestresultsamongmodelvariables,accordingtoHurlinandVenet(2004)andHurlin(2004,2005)
areintroducedinTableA2.
Itisobservedthatwiththreelags,p-valuesofstatisticsZN,THNCand ˜ZHNCN,T allowustoconcludethattheinfrastructure indexcausesthepovertyvariableintheGrangersense.Likewise,sincethecorrelationbetweenpovertyand infrastruc-tureisnegative,wecanconfirmthattheexpensesininfrastructurehelpreducepoverty.Inthecaseofoneortwolags only,thestatisticZHNCN,T issignificant.
Anotherinterestingresultverifiedisthereversecauseofpovertyinthedirectionoftheinfrastructureindex,which mightbeexplainedbythe factthat morepovertymayimplylessinfrastructureexpenses. Thisempirical evidence confirmstheresultsobtainedbyRajkumarandSwaroop(2008)andSeetanahetal.(2009)asdescribedinSection2.
Withregardstootherpovertydeterminants,wecouldverifythroughthep-valuesofstatisticsZHNCN,T and ˜ZN,THNCthat thereisalsoareversecausalityofpovertyagainstallofthem,thusvalidatingsuchdeterminants.
Finally,sincepovertycausesanegativeimpactonthepercapitaGDPintheGrangersense,itsuggeststheexistence ofaviciouscircle.
7. Conclusions
Resultsobtainedfromeconometricmodelssuggestthatpovertyisadynamicandpersistentprocess,asitsresponse capacityinthecurrentdaysishighincomparisontothepast,thusconfirmingthehypothesisofaviciouscircle.
Withregardstotheotheranalyzeddeterminants,thepercapitaGDPandtheaverageschoolingyearshavealso contributedtodiminishedpovertylevels.Itisworthhighlightingthatsuchempiricalevidencesconfirmresultsfound inthenationalandinternationalliterature.
Theunemploymentratealsoshowedasignificantpositivecorrelationincomparisontothepovertyindex.Inthe end,thehighertheunemploymentrate,the highertheproportionofpoorpeople.Sincethisvariableisaffectedby businesscyclesandmacroeconomicpolicies,thegovernmentshouldbeconcernedaboutimplementingmeasuresthat canstabilizetheeconomy.Althoughincomeinequalityhasbeenreducedinrecentyears,itstillcontributesstrongly tothepovertygrowth.Thisresultconfirmsfindingsfromnationalandinternationalpapers.Besides,theimpactofthis variableonpovertyismuchhigherthanthepercapitaGDP.Therefore,policiesaimedatthereductionofinequalities aremoreefficienttofightpovertythanthosesolelyfocusedoneconomicgrowth.
Asforthestatebudgets,wenoticedanegativecorrelationofpovertywithregardstothepovertyindex,whichratifies resultsobtainedbySeetanahetal.(2009)forasampleof20developingcountriesincludingBrazil.Infact,inrecent years,moregovernmentfundshavebeenusedtoreassignincometothepoorestthroughincometransferprograms.
Anotherinterestingresultisthereversecausationofpovertyinthedirectionoftheinfrastructureindex,whichmay beexplainedbythefactthathigherpovertylevelsmayimplyless infrastructureexpenses.Thisempiricalevidence corroboratesresultsobtainedbyRajkumarandSwaroop(2008)andSeetanahetal.(2009).Additionally,sincepoverty causesanegativepercapitaGDPimpactintheGrangersense,wecanconcludethatthereisindeedapovertyvicious cycle.Thisviciouscycleisaggravatedduetotheintensepersistenceofpoverty.
Summarizing,resultsobtainedshowthatpoliciesaimedatencouraginginfrastructureinvestment,aswellas sus-tainablegrowth,incomedistributionandeducationareimportanttofightpovertyintensity.However,ifinfrastructure investment,GDPgrowthandeducationpoliciesincreaseincomeinequality,theymayresultinonlymodestresultsor evenworsenpoverty.
AppendixA.
TableA1
Estimationsandstatisticsofmodel2parameters.
MQO[a] Fixedeffects[b] GMM—system[c]
Coefic. Value-p Coefic. Value-p Coefic. Value-p
P0,it−1 0.9393(0.0139) 0.000 0.5567(0.0339) 0.000 0.5952*(0.0772) 0.000
infit −3.13E−05(8.74E−06) 0.000 −4.70E−05(9.06E−06) 0.000 −2.5E−05*(1.25E−05) 0.056
pibit −3.45E−07(8.57E−07) 0.687 8.67E−07(1.39E−06) 0.533 −7.8E−06*(2.85E−06) 0.011
aemit −0.0046(0.0028) 0.103 −0,0260(0.0030) 0.000 −0.0078**(0.0043) 0.082
giniit 0.1335(0.0584) 0.023 0.3723(0.0728) 0.000 0.7202*(0.2204) 0.003
regovit 2.23E−09(1.94E−08) 0.909 −7.78E−08(1.73E−08) 0.000 −4.64E−08*(1.96E−08) 0,026
desit 0.0026(0.0007) 0.000 0.0006(0.0011) 0.575 0.0042**(0.0024) 0.090
Const. −0.0386(0.0399) 0.334 0.1455(0.0521) 0.000 −0.1540(0.1130) 0.185
F(7.370)=1331.31 F(7,344)=296.31 F(7.26)=99.18 Prob>F=0.0000 Prob>F=0.000 Prob>F=0.000 R2=0.96
N◦ofobs:378 N◦ofobs:378 N◦ofobs:378
N◦ofgroups:27 N◦ofgroups:27
N◦ofinstruments:20
H0:absenceoffirstorderresidualautocorrelation Value-p 0.01
H0:absenceofsecondorderresidualautocorrelation Value-p 0.98
Hansentest Prob>chi2 0.29
Sargantest Prob>chi2 0.32
Source:ResultsobtainedbytheauthorthroughtheSoftwareStata11.0.
Notes:(i)ValuesbetweenparenthesesarestandarddeviationscorrectedthroughtheWindmeijer(2005)method.
(ii)ValuesfortheHansentestarethep-valuesforthenullhypothesisthatinstrumentsarevalid.Thistestisnotrobustbutitsperformanceisnot affectedbythepresenceofseveralinstruments.
(iii)ValuesfortheSargantestarethep-valuesforthevalidityofadditionalinstrumentsrequiredbythesystemmethod.Thisisarobusttestbutits performanceisaffectedbythepresenceofseveralinstruments.
TableA2
Causalitytestforthemodel.
Lags K=1 K=2 K=3
Testsstatistics ZHNC
N,T Z˜HNCN,T ZHNCN,T Z˜HNCN,T ZHNCN,T Z˜HNCN,T
Pit→infit 13.0(0,000) 8.7(0.000) 27.8(0.000) 7.8(0.000) 35.2(0.000) 4.8(0.000)
infit→Pit 2.1(0,033) 1.0(0.319) 6.0(0.000) 1.0(0.296) 28,5(0,000) 3.7(0,000) Pit→Pibit 3.8(0.000) 2.1(0.037) 12.1(0.000) 2.9(0.003) 20.3(0,000) 2.3(0.020) Pibit→Pit 9.0(0.000) 5.9(0.000) 16.0(0.000) 4.2(0.000) 88.6(0.000) 13.7(0.000) Pit→aemit 6.0(0.000) 3.8(0.000) 23.9(0.000) 6.4(0.000) 119.4(0.000) 18.84(0.000) aemit→Pit 6.5(0.000) 4.1(0.000) 13.3(0.000) 3.3(0.001) 49(0,000) 7.3(0.000) Pit→giniit 7.4(0.000) 4.77(0.000) 16.3(0.000) 4.2(0.000) 28.3(0.000) 3.7(0.000) giniit→Pit 8.2(0.000) 5.3(0.000) 34.4(0.000) 9.9(0.000) 92.7(0,000) 14.4(0.000) Pit→regovit 0.9(0.379) 0.11(0.915) 65.0(0,000) 19.4(0.000) 16.4(0,000) 1.7(0.093) regovit→Pit 1.3(0.21) 0.4(0.707) 1.8(0.074) −0.3(0.80) 37.2(0,000) 5.1(0.000) Pit→desit 1.9(0.062) 0.8(0.421) 35.6(0.000) 10.2(0.000) 103.4(0.000) 16.2(0,000) desit→Pit −0.9(0.393) −1.13(0.259) 6.5(0.000) 1.2(0.227) 84.5(0.000) 13.03(0.000)
Source:ResultsobtainedbytheauthorsthroughSoftwareMATLAB7.9.
Notes:Valuesinparenthesisarepvalues.Thesymbol→pointsattheGrangercausalitydirection.
References
Ahn,S.C.,Schimdt,P.,1995.Efficientestimationofmodelsfordynamicpaneldata.J.Econom.68,5–28.
Anderson,W.,1964.Tricklingdown:therelationshipbetweeneconomicgrowthandtheextentofpovertyamongAmericanfamilies.Q.J.Econ. 78,511–524.
Aparicio,C.,Jaramillo,M.,Román,M.C.,2011.Desarrollodelainfraestructurayreduccióndelapobreza:elcasoperuano.CentrodeInvestigación (UniversidaddelPacífico),setembro.
Arellano,M.,Bond,S.,1991.Sometestsofspecificationforpaneldata:MonteCarloevidenceandanapplicationtoemploymentequations.Rev. Econ.Stud.58(April(2)),277–297.
Arellano,M.,Bover,O.,1995.Anotherlookattheinstrumental-variableestimationoferror-componentsmodel.J.Econom.68,29–52.
AsianDevelopmentBANK,1999.FightingPovertyinAsiaandPacific:ThePovertyReductionStrategy.ADB,Manila.
Balisacan,A.M.,1999.PovertyProfileinthePhilippines:AnUpdateandReexaminationofEvidenceintheWakeoftheAsianCrisis.Report PreparedfortheWorldBank.U.P.SchoolofEconomics,QuezonCity.
Balisacan,A.M.,Pernia,E.M.,2002.ProbingBeneathCross-NationalAverages:Poverty,Inequality,andGrowthinthePhilippines,ERDWorking PaperSeries,7.EconomicsandResearchDepartment,AsianDevelopmentBank,Manila.
Barbosa,A.F.O.,2004.Mercadodetrabalhobrasileiropós-1990:mudanc¸asestruturaiseodesafiodainclusãosocial.In:SEMINÁRIO INTER-NACIONALSOBREEMPLEO,DESEMPLEOYPOLÍTICASDEEMPLEOEMELMERCOSURYLAUNIONEUROPEA,2004,Buenos Aires,BuenosAires:WordBank.
Barreto,F.A.,2005.CrescimentoEconômico,PobrezaeDesigualdade:oquesabemossobreeles?UFC/CAEN/LaboratóriodeEstudosdeEstudos daPobreza,Fortaleza(SérieEnsaiossobreaPobreza).
Barros,R.P.,Mendonc¸a,R.,1997.OImpactodoCrescimentoEconômicoedeReduc¸õesnoGraudeDesigualdadesobreaPobreza.IPEA,Riode Janeiro(TextoparaDiscussão,528).
Barros,R.P.,deCarvalho,M.,Franco,S.,Mendonc¸a,R.,2007.DeterminantesImediatosdaQuedadaDesigualdadeBrasileira.IPEA,RiodeJaneiro (TextoparaDiscussão,1253).
Bertussi,G.L.,ElleryJunior,R.,2012.InfraestruturadetransporteecrescimentoeconômiconoBrasil.J.Transp.Lit.6(October(4)),101–132.
Blank,R.M.,1993.WhyWerePovertyRatesSoHighintheLateTwentiethCentury.WolfMacmillian,London,pp.21–55.
Blundell,R.,Bond,S.,1998.Initialconditionsandmomentrestrictionsindynamicpaneldatamodels.J.Econom.87,115–143.
Bond,S.R.,Hoeffler,A.,Temple,J.,2001.GMMEstimationofEmpiricalGrowthModels.CentreforEconomicPolicyResearch,London,United Kingdom(CEPRDiscussionPaper3048).
Bourguignon,F.,2004.ThePoverty–Growth–InequalityTriangle.TheWorldBank,WashingtonD.C.
Calderon,C.,Serven,L.,2004.TrendsinInfrastructureinLatinAmerica,1980–2001.WorkingPaper,269.BancoCentraldeChile.
Corseuil,C.H.,Foguel,M.N.,2002.UmasugestãodedeflatorespararendasobtidasapartirdealgumaspesquisasdomiciliaresdoIBGE.IPEA,Rio deJaneiro(TextoparaDiscussão,897).
Cruz,A.C.,Teixeira,E.C.,Braga,M.J.,2010.Osefeitosdosgastospúblicoseminfraestruturaeemcapitalhumanonocrescimentoeconômicoena reduc¸ãodapobrezanobrasil.Rev.Econ.11(4),163–185.
Datt,G.,Ravallion,M.,1998.WhyhavesomeIndianstatesdonebetterthanothersatreducingruralpoverty?Economica65(February(257)), 17–38.
Datt,G.,Ravallion,M.,2002.IsIndia’seconomicgrowthleavingthepoorbehind?J.Econ.Perspect.16(3),89–108.
deLima,F.S.,Barreto,F.A.,Marinho,E.,2003.Impactodocrescimentoeconômicoedaconcentrac¸ãoderendasobreoníveldepobrezadosestados brasileiros.ENCONTROREGIONALDEECONOMIA,7.,2003,Fortaleza.Anais...Fortaleza.
Dercon,S.,Krishnan,P.,1998.ChangesinPovertyinRuralEthiopia1989–1995:Measurement,RobustnessTestsandDecomposition,WPS/98-7. CenterfortheStudyofAfricanEconomics,Oxford.
Enders,W.,Hoover,G.A.,2003.Theeffectofrobustgrowthonpoverty:anonlinearanalysis.Appl.Econ.35,1063–1071.
Enrenberg,R.G.,Smith,R.S.,2000.Amodernaeconomiadotrabalho:teoriaepolíticapública.MakronBooks,SãoPaulo,pp.319–409.
Escobal,J.,Ponce,C.,2001.Elbeneficiodeloscaminosrurales:ampliandooportunidadesdeingresosparalospobres.DocumentodeTrabajo. Grade,Lima,pp.40.
Fan,S.,Zhang,L.,Zhang,X.,2002.Growth,InequalityandPovertyinChina;TheRoleofPublicInvestments.ResearchReport125.International FoodPolicyResearchInstitute,WashingtonD.C.
Ferreira,P.C.,Malliagros,T.G.,1998.Impactosprodutivosdainfra-estruturanoBrasil—1950/95.PesquisaePlanejamentoEconômico28(August (2)),315–338.
Formby,J.P.,Hoover,G.A.,Kim,H.,2001.EconomicgrowthintheUnitedStates:comparisonsofestimatesbaseduponofficialpovertystatistics
andSen’sindexofpoverty.J.IncomeDistrib.10,6–22.
Foster,J.,Greer,J.,Thorbecke,E.,1984.Aclassofdecomposablepovertymeasures.Econometrica52(May(3)),761–766.
Gafar,J.,1998.Growth,inequalityandpovertyinselectedCaribbeanandLatinAmericacountries:withemphasisonGuyana.J.LatinAm.Stud. 30,591–617.
Ghosh,B.,De,P.,2000.Infrastructure,economicgrowthandtradeinSAARC.BIISSJ.21(2).
Granger,C.W.J.,1969.Investigatingcausalrelationsbyeconometricmodelsandcross-spectralmodels.Econometrica34,541–551.
Herrera,J.,Roubaud,F.,2002.DinámicadelaPobrezaUrbanaenPerúyemMadagascar1997–1999:Unaanálisisdedatosdepanel.Bulletinde l’InstitutFrancaisd’EtudesAndines,Tome31,3.
Hirsch,B.T.,1980.Povertyandeconomicgrowth:hastrickledownpeteredout?Econ.Inq.18,151–157.
Hirschman,A.O.,1958.TheStrategyofEconomicDevelopment.YaleUniversityPress,NewHaven,CT.
Hoffmann,R.,2005.ElasticidadedaPobrezaemRelac¸ãoàRendaMédiaeàsDesigualdadesnoBrasilenasUnidadesdaFederac¸ão.Economia6 (July(2)),255–289.
Hoffmann,R.,2006.Transferênciaderendaeareduc¸ãodadesigualdadenoBrasilecincoregiõesentre1997e2004.Econômica8(June(1)), 55–81.
Hsiao,C.,2004.AnalysisofPanelData.CambridgeUniversityPress.
Hurlin,C.,2004.TestingGrangercausalityinheterogeneouspaneldatamodelswithfixedcoefficients.In:53èmeCongrèsAnnueldeL’Association Franc¸aisedeScienceEconomique,Paris.
Hurlin,C.,2005.Testingforgrangercausalityinheterogeneouspaneldatamodels.RevueEconomique56,1–11.
Hurlin,C.,Venet,B.,2001.GrangerCausalityTestsinPanelDataModelswithFixedCoefficients.WorkingPaperEurisco.UniversityofParis Dauphine,pp.9.
Hurlin,C.,Venet,B.,2004.FinancialDevelopmentandGrowth:ARe-ExaminationUsingaPanelGrangerTest.WorkingPaper.Universityof Orléans,UniversityofParisDauphine.
InstitutoDEPesquisaEconômicaAplicada—IPEA,2006.SobreaRecenteQuedadaDesigualdadedeRendanoBrasil.NotaTécnica.Ago.
Inter-AmericanDevelopmentBank—BID,2000.UnnuevoimpulsoparalaintegracióndelainfraestructuraregionalenAméricadelSur.Mimeo, Agosto.
Jacoby,H.,2000.Accesstomarketsandthebenefitsofruralroads.Econ.J.110,713–737.
Judson,R.A.,Owen,A.L.,1999.Estimatingdynamicpaneldatamodels:aguideformacroeconomists.Econ.Lett.65,9–15.
Kageyama,A.,Hoffmann,R.,2006.PobrezanoBrasil:UmaPerspectivaMultidimensional.EconomiaeSociedade,Campinas15(January/July(1)), 79–112.
Kakwani,N.,1993.PovertyandeconomicgrowthwithapplicationtoCoted’Ivoire.Rev.IncomeWealth39,121–139.
Kalwij,A.,Verschoor,A.,2004.HowGoodisGrowthforthePoor?TheRoleoftheInitialIncomeDistributioninRegionalDiversityinPoverty Trends.WorkingPaperSeries.CENTER-UniversityTilburgno,pp.115.
Kwon,E.,2001.Infrastructure,Growth,andPovertyReductioninIndonesia:ACross-SectionalAnalysis.Mimeo,Manila:AsianDevelopment Bank.
Machado,A.F.,Hermeto,A.M.,Viegas,M.,Totino,B.,2003.EconomiaSocial—MercadodeTrabalho,pobrezaedesigualdadeecriminalidade. BeloHorizonte,Mimeo.
Marinho,E.,Soares,F.,2003.Impactodocrescimentoeconômicoedaconcentrac¸ãoderendasobreareduc¸ãodapobrezanosestadosbrasileiros. ENCONTRONACIONALDAECONOMIA,XXXI,2003,PortoSeguro.Anais...,PortoSeguro:ANPEC.
Marinho,E.,Linhares,F.,Campelo,G.L.,2011.OsprogramasdetransferênciaderendadogovernoimpactamapobrezanoBrasil?RevistaBrasileira deEconomia65(n.3),267–288.
Marinho,E., Araújo,J.A., 2012.Crescimento econômico econcentrac¸ão derenda:seusefeitos napobrezano Brasil.SérieWorkingPaper BNDES/ANPEC,n.24,marc¸o.
Menezes,T.A.,Pinto,R.F.,2005.ÉPrecisoEsperaroBoloCrescer,paraDepoisRepartir?ENCONTROREGIONALDEECONOMIA,8.,2005, Fortaleza.Anais...Fortaleza.
Mussolini, C.C., Teles, V.K., 2010. Infraestrutura e produtividade no Brasil. Revista de Economia Política 30 (October–December (4)), 645–662.
Neder,H.D.,Silva,J.L.M.,2004.Pobrezaedistribuic¸ãoderendaemáreasrurais:umaabordagemdeinferência.RevistadeEconomiaeSociologia Rural42(Setembro(3)),Brasília.
Ogun,T.P.,2010.InfrastructureandPovertyReduction—ImplicationsforUrbanDevelopmentinNigeria.UNU-WINDERWorkingPaper,43. Marc¸o.
Queiroz,B.L.,1999.EfeitosdocapitalhumanolocalsobreodiferencialregionaldesaláriosemMinasGerais.Dissertac¸ão(Mestradoemdemografia). UFMG,BeloHorizonte.
Rajkumar,A.S.,Swaroop,V.,2008.Publicspendingandoutcomes:doesgovernancematter?WorldBank,Washington,DC.J.Dev.Econ.86, 96–101.
Ranis,G.,Stewart,F.,2002.CrecimientoeconómicoydesarrollohumanoenAméricaLatina.RevistadelaCEPAL,SantiagodeChile78,7–24, dic.
Ravallion,M.,Huppi,M.,1991.Measuringchangesinpoverty:amethodologicalcasestudyofIndonesiaduringanadjustmentperiod.WorldBank Econ.Rev.5,57–82.
Reis,J.G.A.,deBarros,R.P.,1990.Desigualdadesalarialedistribuic¸ãodaeducac¸ão:aevoluc¸ãodasdiferenc¸asregionaisnoBrasil.IPEA,Riode Janeiro.
Ribas,R.P.,Machado,A.F.,Golgher,A.B.,2006.FluctuationsandPersistenceinPoverty:ATransient-ChronicDecompositionModelfor Pseudo-PanelData.UFMG/CEDEPLAR,BeloHorizonte(TextoparaDiscussão,289).
Rocha,S.,2006.PobrezanoBrasil.Afinaldequesetrata?,3ed.FGV,Riodejaneiro.
Roy,K.,2009.EffectonpublicinfrastructureonpovertyreductioninIndia:astatelevelstudyfortheperiod1981–2001.IndianJ.Millenn.Dev. Stud.4(March(1)),99–111.
Runsinarith,P.,2008.InfrastructureDevelopmentandPovertyReduction:EvidencefromCambodia’sBorderProvinces.GraduateSchoolof InternationalStudies.NagoyaUniversity.
Santos,B.A.F.,2008.Aumentodarendaoureduc¸ãodadesigualdade?Oqueémaiseficientenocombateàpobreza:novasevidênciasparaaszonas rurais,urbanasemetropolitanasnoBrasil.2008.51f.Dissertac¸ão(MestradoemEconomia).CAEN,UniversidadeFederaldoCeará,Fortaleza.
Seetanah,B.,Ramessur,S.,Rojid,S.,2009.Doesinfrastructurealleviatepovertyindevelopingcountries?Int.J.Appl.Econom.Quant.Stud.6,2.
Shultz,T.W.,1973.OValorEconômicodaEducac¸ão,2ed.Zahar,RiodeJaneiro.
Thornton,J.R.,Agnello,R.J.,Link,C.R.,1978.Povertyandeconomicgrowth:trickledownpetersout.Econ.Inq.16,385–394.
Torero,M.,Escobal,J.,Saavedra,J.,2001.Distribution,AccessandComplementarity:CapitalofthePoorinPeruenPortraitofthePoor:An Assets-BasedApproach.In:AttanasioySzékely(Ed.).Inter-AmericanDevelopmentBank,Washington.
Vilela,A.I.,2005.CapitalHumanoeCrescimentoEconômiconosmunicípiosdoestadodoCeará—1991a2000.46f.Dissertac¸ão(Mestradoem Economia).CAEN,UniversidadeFederaldoCeará,Fortaleza.
Warr,P.,2005.RoadsandPovertyinRuralLaos.DepartmentalWorkingPapers.AustralianNationalUniversity,EconomicsRSPASseries.