《chinese roads in india-the effect of transport infrastructure on economic development》由會員上傳分享，免費在線閱讀，更多相關內(nèi)容在工程資料-天天文庫。
ChineseRoadsinIndia:TheEffectofTransportInfrastructureonEconomicDevelopmentSimonAlderyUniversityofZurichJobMarketPaperzJanuary28,2014AbstractThispaperusesageneralequilibriumframeworkasinEatonandKortum(2002)toestimatethecontributionoftransportinfrastructuretoregionaldevelopment.IapplytheanalysistoIndia,acountrywithanotoriouslyweakandcongestedtransportationinfrastructure.I?rstanalyzethedevelopmenteffectsofarecentIndianhighwayprojectthatimprovedconnec-tionsbetweenthefourlargesteconomiccenters.Iestimatetheeffectofthisnewinfrastruc-tureonincomeacrossdistrictsusingsatellitedataonnightlights.TheresultsshowaggregatenetgainsfromtheIndianhighwayproject,butveryunequaleffectsacrossregions.Chinahasfollowedadifferenthighwayconstructionstrategyandhasexperiencedmoresigni?-cantconvergenceacrossregionsthanIndia.IthereforeusethemodeltogaugetheeffectsofacounterfactualhighwaynetworkforIndiathatreplicatestheChinesestrategyofconnect-ingintermediate-sizedcities.I?ndthatthiscounterfactualnetworkwouldhavesigni?cantlybene?tedthelaggingregionsofIndia,inducingmoreequalregionaldevelopment.However,theeffectsonaveragenationalgrowthwouldhavebeenmodest.Ialsoconstructadditionalcounterfactualsanddiscusstheireffectsoneconomicdevelopment.JELCodes:F11,F14,F15,O11,O18,R12,R13Keywords:TransportInfrastructure,EconomicGrowth,RegionalDevelopment,Trade,India,China,GeographicInformationSystem,SatelliteData,Luminosity.IwouldliketothankFabrizioZilibottiforhissupportduringthisproject.IalsothankSimeonD.Alder,George-MariosAngeletos,MarcoBassetto,TimoBoppart,FilippoBrutti,MariusBrulhart,AdrienBussy,GuidoCozzi,GregoryCrawford,MariacristinaDeNardi,DavidDorn,¨PeterKondor,MichaelKonig,RafaelLalive,SergiJim¨enez-Mart′′?n,OmarLicandro,AndreasMuller,AlessandroPavan,MichelleRendall,Jos¨e-V′′?ctorR′?os-Rull,DominicRohner,EstebanRossi-Hansberg,LinShao,KjetilStoresletten,ViktorTsyrennikov,RainerWinkelmann,ChristophWinter,GabrielZucman,JosefZweimuller,andseminarparticipantsattheUniversityofZurich,¨theUniversityofLausanne,andtheZurichWorkshoponEconomicsforhelpfulcomments.Se-bastianOttingerprovidedoutstandingresearchassistance.RonaldSchmidtandLarryCrissmanprovidedvaluablesupportwithGISsoftwareanddata.yUniversityofZurich,Muhlebachstrasse86,¨8008Zurich,Switzerland.E-Mail:simon.alder@econ.uzh.ch.Personalwebpage:https://sites.google.com/site/simonalderch.zThenewestversioncanbefoundhereoronmywebpage.1 1IntroductionChinaandIndia,thetwomostpopulouscountriesintheworld,aredevelopingatunprecedentedrates.Yet,theirspatial,orregional,developmentpatternsaresur-prisinglydifferent.ThroughoutChina,newclustersofeconomicactivityemergeandthereisastrongpatternofconvergenceacrossChinesecounties.Incontrast,asubstantialnumberofIndiandistrictsofintermediatedensityexperiencelowgrowthandthereisgenerallylessconvergence.WhilesuchdifferencesinthespatialdevelopmentofChinaandIndiahavebeendocumentedintheliterature(Desmetetal.,forthcoming;ChaudhuriandRavallion,2006),westilllackpreciseexplanationsandpossiblepolicymeasures.Thispaperlinksthedifferencesinthespatialdevelopmentofthetwocoun-triestotheirtransportnetworks.TheIndiangovernmentlaunchedanationalhighwayprojectin2001thatimprovedconnectionsbetweenthefourlargesteco-nomiccentersDelhi,Mumbai,Chennai,andCalcuttawiththe“GoldenQuadri-lateral”(GQ).Incontrast,ChinabuiltaNationalExpresswayNetwork(NEN)thathadtheexplicitgoalofconnectingallintermediate-sizedcitieswithapopu-lationabove500,000andallprovincialcapitalswithmodernhighways.Overall,ChinainvestedabouttentimesmoreinitshighwaynetworkthanIndia,whichisseenasbeingseverelyconstrainedbyitsinsuf?cientinfrastructure(Harraletal.,2006).Iftransportinfrastructureisadeterminantofdevelopment,thenonemayaskhowanetworkshouldbedesignedinordertofostergrowthandregionalde-velopment.Inthispaper,I?rstidentifytheeffectofamajorhighwayprojectinIndia,theconstructionoftheGQ.Then,inlightofthestarkdifferenceinthetransportinfrastructurestrategiesofIndiaandChina,IaskhowIndiawouldhavedevelopedifithadbuiltanetworkliketheChineseNEN.Tothisaim,IconstructacounterfactualIndianhighwaynetworkthatmimicstheChineseap-proachofconnectingintermediate-sizedcities.ThecounterfactualisbuiltbasedonthepreciselocationofcitiesandonthetopographicfeaturesofIndia,whicharemodeledusingageographicinformationsystem(GIS).Thepathsofthecoun-terfactualhighwayconnectionsarechosentominimizetheconstructioncostsbasedonslopeandlandcover.Theresultingroadnetworkthenallowsmeto2 computebilateraltransportcostsbetweenall590mainlanddistrictsinIndiaus-ingashortestpathalgorithm.Intheempiricalanalysis,thebilateraltransportcostsarerelatedtoincome,whichismeasuredusingdataonluminosityatnight.Thisdataisavailablefromsatelliteimagesandcaptureshumaneconomicactiv-ityatahighspatialresolutionandovertime.Luminosityhasbeenshowntobeagoodproxyforincomegrowth(Hendersonetal.,2012)andcanbeaggregatedwithdigitizedmapstothelevelofdistricts,forwhichof?cialGDPdataisnotavailable.Theempiricalanalysisbuildsongeneralequilibriumtradetheory.IfollowDonaldsonandHornbeck(2013)whoderivefromaRicardiantrademodelareduced-formmeasurefortheaggregateimpactoftransportinfrastructureonincome.1Thisreducedformcapturesthemarketaccessofalocationbysummingovertheincomeoftradingpartners,discountedbythebilateraltradecostsandbythedestination’smarketaccess.2Transportinfrastructuredeterminesbilateraltradecostssuchthatchangesintheinfrastructureovertimegeneratevariationinmarketaccess.Moreprecisely,thebilateraltransportcostscanbecomputedforthetransportnetworkin2000(beforethestartoftherecentIndianhighwayproject),in2009(aftercompletionofthe?rstphases),andforthecounterfactual(replicatingtheChinesenetwork).Theresultingbilateraltransportcostsarethenusedtoderivedistricts’marketaccessforeachversionofthetransportnetwork.Themodelpredictsalog-linearrelationshipbetweenincomeandmarketac-cess.Thetimevariationinmarketaccessbetweentheactualnetworksin2000and2009allowsmetoestimatethisrelationship.Giventheresultingestimatefortheelasticityofincomewithrespecttomarketaccess,Ipredicteachdistrict’sincomebasedonthemarketaccessimpliedbythecounterfactualnetwork.Importantly,marketaccesscapturesthegeneralequilibriumconsequencesoftransportinfras-tructureandtheresultingpredictionsthereforerepresentaggregateeffects.31DonaldsonandHornbeck(2013)estimatetheeffectofAmericanrailwaysonlandvalue,whileIestimatetheeffectofIndianhighwaysonrealincome.Sections4and5willdiscussthediffer-encesinmoredetail.2Arelatedmeasureismarketpotential,whichhasbeenderivedfrommodelsintheneweco-nomicgeographyliterature.SeeforexampleReddingandVenables(2004)andHanson(2005).3Anincreaseinmarketaccessofthetradingpartner(e.g.becauseitisbetterconnectedtoathirddistrict)canreducethemarketaccessofanorigin.Marketaccessthereforecapturesgeneralequilibriumconsequencessuchastradediversion(seeSection5).3 Theempiricalanalysismakesthreecontributions.First,Iquantifytheaggre-gateeffectoftherealizedGQ,India’smajorhighwayinvestmentprojectbetween2001and2009.Theresultssuggestthataggregateincomewas2.4-3.5percenthigherin2009thanitwouldhavebeeniftheGQhadnotbeenbuilt.Thisimpliesagainafteronedecadethatismorethanthreetimestheconstructioncosts.Sec-ond,Ipredicttheaggregateeffectsofthecounterfactualtransportinfrastructure,whichreplicatesthesalientaspectsoftheChinesenetworkinIndiainawaythatminimizesroadconstructioncosts.Takingintoaccounttheconstructioncostsofcounterfactualroads,theresultsimplyintheaggregateamodestdifferencerelativetotheexistinginfrastructure.Thethirdcontributionistoevaluatethedistributionalconsequencesoftheactualandcounterfactualnetworks.There-sultsshowthatinitiallylessdevelopedregionswouldgainsubstantiallyfromthecounterfactual.Thereasonisthatthisnetwork,byconnectingallintermediate-sizedcities,alsoreachesintoregionsthatpreviouslyhadlowgrowthandwereneglectedbytheGQ.Thus,atransportnetworkthatfollowstheChinesestrategywouldincreasegrowthparticularlyinIndia’slaggingregions.ThisprovidesanexplanationfortheweakerconvergenceinIndiacomparedtoChina.Twoalter-nativewaystoreplicatetheChinesenetworkinIndialeadtoqualitativelysimilarresults.Thedistributionalconsequencesareparticularlyrelevantinlightoftheun-equalregionaldevelopmentofIndia.Policymakersareawareofthisandthenationalhighwaydevelopmentstrategydidincludeplansforotherhighwaycon-nectionsbesidestheGQ.Inparticular,thegovernmentplannedtheNorth-SouthandEast-WestCorridorswhichcrossthroughregionsthatwerenotreachedbytheGQ.However,theseotherprojectsweredelayedandby2009onlyasmallparthasbeen?nished.Inanadditionalcounterfactualexercise,I?ndthatthecompletionofthesecorridorswouldindeedincreaseincomeinsomeofthelag-gingstates.However,theexplicitstrategyofconnectingallintermediate-sizedcitieswouldhavelargeraggregateeffectsandbene?tmorelaggingdistricts.Theremainderofthepaperisstructuredasfollows.Section2reviewstherelatedliterature.Section3discussesthespatialdevelopmentandtransportin-frastructureinIndiaandChina.Section4showshowthecounterfactualnetworkisconstructedandwhatdataisused.Section5presentstheconceptualframe-4 workandSection6discussestheresults.Section7showsalternativewaystoreplicatetheChinesenetworkanddiscussestherobustnessoftheresults.Section8concludes.2RelatedLiteratureTheroleoftransportinfrastructurefordevelopmenthasbeenthesubjectofalargeliterature.4Arecentincreaseinthisliteraturewastriggeredbyacombina-tionofeconomictheorywithgeographicinformationsuchastheexactlocationoftransportinfrastructure.Mymethodologyforevaluatingtheimpactofinfras-tructurebuildsonDonaldsonandHornbeck(2013)whoestimatetheaggregateeffectoftheexpansionoftheAmericanrailwaynetworkinthe19thcentury.Theyderivemarketaccessasareducedformmeasurefortheimpactoftransportin-frastructureinageneralequilibriumtrademodelasinEatonandKortum(2002).DonaldsonandHornbeck(2013)alsocomparetheeffectoftheactuallybuiltnet-worktocounterfactualscenariosinwhichrailwaysarereplacedbyroadsandcanals.5Iadapttheirframeworktotheuseoflightdataasameasureofrealin-comeinIndia.Todeterminetheprecisepathsofthecounterfactualroads,Iusetheleast-costnetworkthatconnectsagivensetofcities.Suchanetworkhaspre-viouslybeenusedbyFaber(2013)withinChinainordertoconstructaninstru-mentfortheactuallybuilthighways.Ifollowtheapproachofconnectingcitieswhichful?llthecriteriaoftheNENinawaythatminimizesroadconstructioncosts,butIapplyittoIndiancitiesandtothelocalterraininordertoconstructacounterfactualnetwork.Theempiricalexerciseofthispaperisalsorelatedtorecentstudiesonthelo-caleffectsoftransportinfrastructure.Datta(2012)andGhanietal.(2012)studytheeffectsoftheGQand?ndapositiveimpacton?rmslocatedintheproximityofthenewhighways.6Animportantaspectofthesestudiesistheidenti?cationofexogenoussourcesofvariationintransportinfrastructure.Theyrelyonan4SeeforexampleRedding(2010)andarecentsurveybyBreinlichetal.(2013).5SuchacounterfactualexercisewasalsoproposedintheseminalworkbyFogel(1964).6Bothstudiesuse?rmsurveystoevaluatetheeffectoftheGQ.Ghanietal.(2012)pointoutthatitwouldbevaluabletoestimatetheeffectwithluminositydata.5 identi?cationstrategysimilartotheoneproposedbyChandraandThompson(2000)andMichaels(2008)whoestimatetheeffectofUShighwaysoncountiesthatliebetweentwoimportantnodalcities.Thisisbasedontheobservationthatthehighwayswerebuilttoconnectlargercitiesandtherebypassedthroughothercountieswhichconsequentlyobtainedaccesstothenewtransportinfrastructurewithoutbeingtargeted.TheresultsofDatta(2012)andGhanietal.(2012)sug-gestpositiveeffectson?rmslocatedclosetotheGQ.IusethisstrategyinordertoestimatetheeffectoftheGQonnon-nodaldistricts,excludingthefourcitiesthatweretargetedbytheGQ.SeveralrelatedstudieshaveanalyzedtheChinesetransportnetwork(seeBanerjeeetal.,2012;Baum-Snowetal.,2013;andFaber,2013).7Theabovestudiesfocusonidentifyingthelocaleffectsintheproximityofnewroads.Tworecentcontributionsthatestimatethegeneralequilibriumcon-sequencesofthenationaltransportinfrastructureinIndiaandChinaareDonald-son(forthcoming)andRobertsetal.(2012).Donaldson(forthcoming)estimatestheeffectofrailwaysincolonialIndiaand?ndsthathistoricalincomelevelsofIndiandistrictshadincreasedby16percentwhentheywereconnectedtotherailwaynetwork.Healsoshowsthatthereisasuf?cientstatisticforthegeneralequilibriumeffectoftransportinfrastructureonincome,whichexplainsmostofthevariationduetotransportinfrastructure.Robertsetal.(2012)useastructuralneweconomicgeographymodeltomeasuretheaggregateeffectoftheexpan-sionoftheNENinChinaand?ndthataggregateincomewas6percenthigherin2007duetotheNEN.Myanalysisdiffersfromtheabovestudiesbyestimat-ingtheaggregateeffectoftransportinfrastructurethroughmarketaccess,aswasproposedbyDonaldsonandHornbeck(2013),andusingthisestimatetopredictincomeundervariouscounterfactualinfrastructures.Toidentifyexogenousvari-ationinmarketaccess,Iapplytheidenti?cationstrategyproposedbyChandraandThompson(2000)andMichaels(2008).Themarketaccessapproachusedinthispaperiscloselyrelatedtomodelsintheneweconomicgeographyliterature.Severalauthorsanalyzetheroleofmar-ketaccess(ormarketpotential),whichcanbeaffectedbytransportcosts(Puga,7TransportinfrastructureinothercountrieshasrecentlybeenstudiedbyAtack(2008),Baum-Snow(2007),GollinandRogerson(2010),Herrendorfetal.(2012),andStoreygard(2013).6 2002;ReddingandVenables,2004;Hanson,2005;ReddingandSturm,2007;HeadandMayer,2011,2013).They?ndthatmarketaccessisassociatedwithtrade,in-come,andpopulationwithinandbetweencountries.Thispaperalsorelatesmorebroadlytoalargeliteratureontrade,inparticularonthegravitystructure(An-dersonandvanWincoop,2003;AllenandArkolakis,2013,AtkinandDonaldson,2013;Cos?arandFajgelbaum,2013;Redding,2012).HeadandMayer(2011)pointoutthatthegravitystructureandmarketaccesscanbederivedfromvarioustrademodelswithdifferentmarketstructuresandsourcesofgainsfromtrade.Acon-tributionofthispaperisthatthedigitaltransportnetworkcanmodelexplicitlyhowtradecostsandthusproximitychangeduetotransportinfrastructure.Thus,changesintransportcostsgeneratevariationinmarketaccesswhichallowstostudytherelationshipbetweenincomeandmarketaccessovertime.Whilethesemodelsarestatic,DesmetandRossi-Hansberg(forthcoming)proposeamodelofspatialdevelopmentbasedontechnologyspilloverswheregrowthdependsonthedensityofeconomicactivity.Theassessmentofthedevelopmenteffectsoftransportinfrastructurealsore-latestocost-bene?tanalysesofindividualinfrastructureinvestments.Forexam-ple,asamajorinvestorintransportinfrastructureindevelopingcountries,theWorldBankhasdevelopedprocedurestoevaluatetheeffectivenessofinfrastruc-tureprojects(seeWorldBank,2007aforanoverview).Whilethoseconceptshaveadvantagesincapturingproject-speci?caspectssuchassafetyandroaddeterio-ration,themethodologyappliedinthispaperisabletocapturethegeneralequi-libriumeffectsatalargescale,whichallowsevaluatingandcomparingnationalinfrastructurestrategies.3SpatialDevelopmentandTransportInfrastructureThispapermakesalinkbetweenregionalgrowthandtransportinfrastructure.IndiaandChinaprovideaninterestingcontexttostudythisrelationship.Whilebothcountriesaregrowingfast,theyalsoshowsubstantialdifferencesintheirregionaldevelopmentpatternsandintheirtransportinfrastructure.Thissection?rstreviewstheevidenceonthespatialdistributionofincomeandgrowthinthe7 twocountriesandthendiscussestheirtransportinfrastructure.3.1TheSpatialDevelopmentofIndiaandChinaDuringthepasttwodecades,realGDPpercapitainIndiahasbeengrowingatanaveragerateof4.8percent(WorldBank,2013).China’sgrowth,averagingat9.2percent,hasbeenevenmorespectacularanditsincomepercapitaovertookIndia’sintheearly1990s.Althoughthereissubstantialvariationintheregionalgrowthrateswithinbothcountries,previousstudiesfoundthatChinahasover-allseenmoreconvergence(Desmetetal.,forthcoming;ChaudhuriandRavallion,2006).Thesame?ndingemergeswhenusinglightasameasureofincome.8Fig-ure1showsthespatialdistributionoflightintheyear2000.Notsurprisingly,thereisastrongclusteringofincomeinbothcountries.Asimilarpictureariseswhenaggregatingthelightpixelstothesub-nationalunitsofChineseprefecturesandIndiandistricts.Interestingly,therearesubstantialdifferencesinthespatialdevelopmentovertime.Toillustratethis,Figures2and3showthespatialdistributionofinitialden-sityandgrowthinthetwocountries.WhileChinahasseenthehighestgrowthratesinprefectureswithinitiallylowdensity(measuredasaveragelightintensityperpixelin2000),thishasnotbeenthecaseforIndiandistricts.Inparticular,therightpanelofFigure3suggestthatthedistrictswithinitiallyloworintermediatelightdensityhadexperiencedsurprisinglylowgrowth.ThisisconsistentwiththeevidencepresentedinDesmetetal.(forthcoming)andChaudhuriandRaval-lion(2006)whofoundstrongerconvergencepatternsinChinathaninIndia.9Whilethe?ndingthatregionalgrowthpatternsdifferbetweenthetwocoun-triesiswellknown,westilllackpreciseanswersforwhatisdrivingthesediffer-ences.Transportinfrastructureisapotentialcandidatesinceitisanimportantde-terminantofthespatialdistributionofeconomicactivity.IndiaandChinaindeedhavefolloweddifferentstrategiesforhowtoinvestintheirtransportnetworks8Section4discussesthelightdatainmoredetail.Hendersonetal.(2012)showthatlightcorrelatesstronglywithGDPinapanelof188countries.9Theobservationiscon?rmedwhenregressinglightgrowthofeachprefectureordistrictonitsinitialdensity.Theslopecoef?cientissigni?cantlysmallerinChinathaninIndia,suggestingstrongerconvergence.8 andthesedifferenceswillbeoutlinednext.3.2TransportInfrastructureinIndiaandChinaInfrastructureisakeydeterminantoftransportcostsandtrade(LimaoandVen-ables,2001)andinvestmentsintransportinfrastructurehavebeenusedexten-sivelytopromotedevelopment(WorldBank,2007a).IndiaandChinahavebothinvestedintheirtransportinfrastructureduringthepastdecades,butwithdif-ferentintensitiesandstrategies(Harraletal.,2006).Inthissection,I?rstreviewthekeyelementsoftheinfrastructureinvestmentsinthetwocountriesandthendiscusstheconstructionofacounterfactualnetworkforIndiawhichmimicstheChinesestrategy.Intheearly1990s,theIndianroadinfrastructurewassuperiortotheChineseintermsoftotalkmlengthandkmperperson,butbothcountrieshadaboutthesamelowqualityofroads.Travelspeedsonroadswerefurtherreducedbythesimultaneoususebypedestriansandslowvehicles.10Overthe1990s,China’shighwayandrailwaynetworkdevelopedsigni?cantlyfasterthantheIn-diancounterpart.Inparticular,itbuilttheNationalExpresswayNetwork(NEN)withtheexplicitobjectiveofconnectingallcitieswithmorethan500,000peopleandallprovincialcapitalsinamodernhighwaysystem.11Atthattime,China’stransportinfrastructurewasatriskofbecomingaconstraintforeconomicde-velopmentwhichwasgainingspeedsincethereformsstartedinthelate1970s(AsianDevelopmentBank,2007).Thenewnetwork,showninredinFigure4,hadreachedalengthof40,000kmby2007anditcontinuedtobeexpanded.Itconsistsoffour-lanelimitedaccesshighwaysthatallowedsigni?cantlyhigherdrivingspeedthantheexistingroads.12Indiaalsoinvestedinitsroadinfrastructure,butabouttentimeslessthan10TherailwayinfrastructureinthetwocountrieswassimilarintermsofpassengersbuttheChineserailwaystransportedfourtimesmorefreightthantheIndianrailways.Thenumbersinthissectionaretaken(ifnototherwisestated)fromHarraletal.(2006).11ThisisalsoreferredtoastheNationalTrunkHighwaySystem.Theprogramwaslaterex-pandedtoincludeallcitieswithmorethan200,000people.SeeChineseMinistryofTransporta-tion(2004),WorldBank(2007b),Robertsetal.(2012),andFaber(2013)foradiscussion.12AdescriptionofthehistoryoftheChinesehighwaynetworkanditsdifferentcomponentsisprovidedbyACASIAN.Seewww.acasian.comforfurtherdetails.9 Chinaandwithafocusonthemaineconomiccenters.Inparticular,itlaunchedaNationalHighwaysDevelopmentProject(NHDP)in2001andthe?rstachieve-mentofthatprojectwastheGQ,whichconnectsthefourmajoreconomiccenterswithfour-lanehighways(showninFigure4).Construction,mostlyupgradesofexistinghighwaystohigherquality,beganin2001andwascompletedby2012withatotalnetworklengthof5,846kmandatacostofUSD6billion(1999prices).13TheNHDPinIndiawasnotrestrictedtotheGQandalsoincludedtheso-calledNorth-SouthandEast-West(NS-EW)Corridors.However,theseprojectsweredelayedandnotfullycompletedby2010.Figure5showsthepartswhichwerecompletedby2010.TheGQinIndia,liketheNENinChina,hassigni?cantlyreducedthetrans-porttimesbetweenplaceswithaccesstothesenewhighways.Theaveragedriv-ingspeedonaconventionalnationalhighway(i.e.ahighwaywhichwasnotupgradedorbuiltaspartoftheNHDP)wasbelow40km/h(WorldBank,2002),whilethedrivingspeedontheGQisaround75km/h.14However,thereisampleevidencethat,eventoday,insuf?cienttransportinfrastructureisaseverecon-straintfortheIndianeconomy.RaghuramRajan,thecurrentGovernoroftheReserveBankofIndia,recentlystatedthatIndianeedstoimproveitsinfrastruc-turewiththesamedisciplineinordertocatchupwithChina(FAZ,2013).ThesameviewisheldbytheWorldBankandseveralconsultanciesandlogistic?rms,statingthatalackofadequateinfrastructurehamperstheregionaldevelopmentinIndia(WorldBank,2008;DHL,2011;ErnstandYoung,2013;KPMG,2013).Theroadinvestmentprojectsdescribedabovewereamongthelargestinter-citytransportinfrastructureinvestmentsinthetwocountriesanddominatedin-vestmentsinothermeansoftransportation.ThespendingontheNENinChinawasaroundUSD30billionperyear,roughlythreetimesasmuchasitsinvest-mentsinthenationalrailwaysystemduringtheperiod1992-2002.Theimpor-tanceofhighwaysrelativetorailwaysalsoincreasedinIndiaandtheshareof13Mostpartswerealreadycompletedby2007.SeethewebpageoftheNationalHighwayAu-thorityofIndia(http://www.nhai.org/index.asp)fordetails.ThecostestimatesarebasedonGhanietal.(2012).14Theof?cialspeedlimitwasincreasedto100km/hin2007,buttheactualdrivingspeedissigni?cantlylower.Thiswasderivedbyselectingarandomsampleoflocationsandexportingbilateraltransporttimeswitharoutinefromgooglemaps.10 expendituresonrailwaysintotaltransportinfrastructuredeclinedfrom50%inthe1990sto30%bytheendofthe2000s(MinistryofRailways,2012).Today,roadsarebyalargemarginthemostimportanttransportmodeinIndia,carrying60%ofthefreightturnovercomparedto31%forrailways.15Thehighwayprojectsundertakeninthetwocountriesarethereforethecrucialpartsoftheirtransportstrategiesandofhighimportanceforthedevelopmentofthetwocountries.AlthoughtheanalysisundertakenherecapturesakeyaspectofthemoderntransportinfrastructureinIndiaandChina,somecaveatsmustbepointedoutthatconcernpossiblechangesinothertypesofinfrastructure.ButIwillshowthattheseconcernsaremitigatedbymyempiricalstrategywhichexploitsexoge-nousvariationintransportinfrastructureandcontrolsforlocationandtime?xedeffects.The?rstconcernistheomissionofothertypesofdomestictransportin-frastructuresuchasrailwaysorurbansystemssuchassubways.Second,accesstointernationalmarketsviaseaportsorairportsisnotmodeledaspartofthetransportnetworkhere.Third,villages’accesstothetransportinfrastructureviaruralroadsisnotconsideredduetoalackofprecisedata.16Finally,non-transportinfrastructuresuchaselectricityandwateralsoaffecteconomicdevelopment.However,thesecaveatswouldlimitthevalidityoftheexercisehereonlyiftheomittedfactorsweretime-varyingatthedistrictlevelandcorrelatedwiththeex-planatoryvariablemarketaccess.Section5discussesindetailhowIaddressthiswithasuitableempiricalstrategy.Theabovediscussion,andtheillustrationinFigure4,makesclearthatIndiaandChinahavefolloweddifferentstrategiestoimprovetheirroadinfrastructure.WhileIndia’shighwayinvestmentsfocusedonconnectingitslargesteconomiccenterswithbetterhighways,Chinahasbuiltanetworkthatconnectsallcitiesthathavearegisteredpopulationofmorethan500,000andallprovincialcapitals.Furthermore,itisclearthathighwaysplayakeyroleintheoveralltransportin-15TheshareofhighwaysinthetotalfreightturnoverisevenhigherinIndiathaninChina(KPMG,2013).16AccordingtoHarraletal.(2006),IndiahaspriortothestartoftheNHDPin2001focuseditsinfrastructureinvestmentsontheimprovementofroadswhichprovideaccesstohighways,whileChinahasfromthestartofitsprogramin1992puttheemphasisoninvestmentsinarterialhighwaystoconnectcities.WhileIcannotobservetheupgradesoflocalroadspriortomysampleperiod,theNHDPwasalargeinfrastructureprogramtowhichtheChineseNENcanbedirectlycompared.11 frastructure.Indiacurrentlyfacessevereconstraintsduetoinsuf?cienttransportinfrastructure,whichislessthecaseforChina.AnaturalquestionthereforeishowIndiawoulddevelopifithadatransportinfrastructurelikeChina.Toan-swerthisquestion,IproposeacounterfactualroadnetworkforIndiabyapplyingthepolicyobjectiveoftheChinesegovernmenttoidentifytheIndiancitieswhichwouldbeconnectedwiththeChinesenetwork.Theexactroutesarechosensuchthatthecostsofbuildingtheroads,whichdependoncharacteristicsoftheIndianterrain,areminimized.Thenextsectionwillpresentthedatathatisrequiredtobuildsuchacounterfactualnetworkandtoevaluateitseffectoneconomicdevel-opment.4CounterfactualRoadsInthissectionIbuildacounterfactualnetworkofhighwaysandcomputebilat-eraltransportcosts.First,IusetopographicfeaturesofIndiainordertodeter-minetheprecisepathsofthecounterfactualroadssuchthatroadconstructioncostsareminimized.Basedonthegeographicallyexplicitroadnetwork,IthencalculatethebilateraltransportcostsbetweenallIndiandistrictsusingashortestpathalgorithm.Thedataandmethodsrequiredforthisprocedurearedescribednext.4.1DataInordertoanalyzetheeffectoftransportinfrastructureoneconomicdevelop-ment,Iconstructadatasetthatincorporatestheprecisegeographiclocationofdifferenttypesofroadsandoflocalincome.IaddtothisdatasettheboundariesofIndianstatesanddistrictsandmultiplemapsoftheIndianterrain.TransportInfrastructureandTerrainIusegeographicinformationsystem(GIS)methodstoprocessthespatialdata.17DigitalmapswiththelocationoftheactualIndiantransportinfrastructureare17TheapplicationsusedhereareArcGISDesktop10.1andthespmatfunctionsinStata13.12 takenfromthreesources:CIESIN(2013)providesadigitizedroadnetworkthatincludesbothhighwaysandlocalroads.Esri(2013)alsohasdigitizedroadsbutislimitedtothenationalhighwaynetworks.These?rsttwosourcesallowtolocalizethecurrenttransportinfrastructureinspace,buttheydonotallowtoac-curatelytrackchangesovertimeandcannotdistinguishthehigherqualityofto-day’sGQ.Therefore,IuseasathirdsourcemapsoftheNHDPissuedbytheNa-tionalHighwayAuthorityofIndia(NHAI,2010andNHAI,2013).Thesemaps,whichweredigitizedmanually,showthelocationofseveralnewhighways,in-cludingtheGoldenQuadrilateralandthecompletedpartsoftheNorth-SouthandEast-WestCorridors.Theaveragedrivingspeedonexistingroadsaretakenfromseveraltransportef?ciencystudies.WorldBank(2005)reportsthatthetyp-icaldrivingspeedontheexistingIndiannationalandstatehighwaysisbetween30and40km/handIthereforeassumeaspeedof35km/hforallhighwaysbuiltbeforethestartoftheNHDP.18Roadsoflowerorunknownqualityarealsocontrolledforandtheassumedtravelspeedis25km/h,whichissuggestedbysurveysofruralIndiantransportinfrastructure(Liu,2000).Forareaswheretherearenoroadsreportedinthedigitizedmaps,Iassumeatravelspeedof10km/h,whichcorrespondstothespeedonunpavedroads(Robertsetal.,2012).ThetravelspeedonthecounterfactualnetworkistakentobethesameasforthetheChineseexpresswaysandtheGQ,whichaccordingtogooglemapsis75km/h.Foracomparisonofthehighwaynetworks,thedigitalmapsoftheChineseex-presswaynetworkwereobtainedfromACASIAN(2013).Inordertodeterminetheconstructioncostsforthecounterfactualroads,Ineeddigitizedinformationontheterrain.IusedigitalelevationdataproducedbyJarvisetal.(2008)forameasureofslope.Forlandcover,Iusetheclassi-?cationbytheGlobalLandCoverFacility(2013)attheUniversityofMarylandDepartmentofGeography.PoliticalBoundariesandLuminosityTheunitsofanalysisinthispaperareIndiandistricts.Ifocusonmainlanddis-tricts,ofwhichthereare590.Luminositymeasuredbyweathersatelliteshas18TheseestimatesareinlinewithmorerecentnumbersbyKPMG(2013).13 beenshowntobeagoodproxyforincome(Hendersonetal.,2012).Twoimpor-tantadvantagesofthelightdataarethatishasahighspatialresolutionandisindependentofcountries’statisticalcapacity.Itisparticularlyusefulwhenof-?cialGDP?guresarenotavailable,forexampleforsubnationaladministrativeunitssuchasIndiandistricts.ThedigitizeddistrictboundariesareprovidedbyGlobalAdministrativeAr-eas(2012)andthelightdataisavailablefromtheEarthObservationGroup(2013)oftheNationalGeophysicalDataCenteroftheUnitedStates.ThesatelliteimagesoriginatefromtheDefenseMeteorologicalSatelliteProgram(DMSP)OperationalLinescanSystem(OLS)todetectcloudcover.Thedataisavailablefrom1992to2012ascompositesovercloud-freeevenings.19Therasterare30arcsecondgrids,spanning-180to180degreeslongitudeand-65to75degreeslatitude.ToderiveameasureofrealincomeIforeachdistrict,Iaggregatelightwithindistrictbound-ariesusinganequalareaprojection.ThelightsummarystatisticsofthesampleofmainlandIndiandistrictsarepresentedinTable1.InordertointerpretthelightdataintermsofGDP,one?rsthastoanalysetherelationshipbetweenGDPandlight.GDPdataisnotavailableforIndiandistrictssuchthattherelationshiphastobeinferredfromothersamples.GDPatthelevelofsub-nationalunitssuchasdistrictsisalsoinothercountriesdif?culttoobtain.AnexceptionisChinawhereonecanrelyonGDPdataattheprefecturelevelbasedonof?cialstatisticalyearbooks.20Thisdataallowsestimatingtherelation-shipwithlightatasimilarlevelofaggregationasIndiandistricts.However,therearesomecaveatswhenusingthisapproach.First,lightintensityismeasuredbyvarioussatellitesovertheyears.Thesesatellitescanhavesomewhatdifferentcalibrationssuchthatobservationsfromdifferentsatelliteyearscannotdirectlybecompared.Second,GDPdataismeasuredwithinadministrativeboundariesandthesemayalsochangeovertime.21Ithereforerelyonlyontheyear2010forwhichIhaveGDP?guresforChineseprefecturesandthedigitalmapsofadmin-19Thelasttwoyearshaverecentlybeenmadeavailableandtheyhavenotbeenincludedinthepresentanalysis.20SeeAlderetal.(2013)foramoredetaileddescriptionoftheGDPdataintheChinesestatisticalyearbooks.21TheChinesestatisticalyearbooksalsoreportthelandarea.These?guresshowthatsubna-tionalbordershaveindeedbeenchanging.14 istrativeboundariesthatmatchtheunitsinthestatisticalyearbooks.ThisallowsmetomeasurelightandGDPonthesamelandareaandwiththesamesatellite,therebyaddressingthetwoconcernsabove.IthenregressthelogarithmofGDPonthelogarithmoflightwithinChineseprefecturesinordertoestimatethere-lationship.Theestimatedelasticityis1.05andsigni?cant,butindistinguishablefrom1.Intheempiricalanalysis,IthereforeinterpretthemagnitudeofaneffectonlightasanequallylargeeffectsintermsofGDP.22WhiletheelasticitythatisusedtotranslatetheestimatefromlighttoGDPclearlyaffectsthemagnitudeoftheaggregateeffects,thedistributionalimplicationsarenotaffectedbythischoice.4.2BuildingaCounterfactualHighwayNetworkforIndiaToreplicatetheChinesenetworkinIndia,I?rstidentifythecitiesinIndiawhichwouldhavebeenchosenbytheChinesepolicyandthenbuildacounterfactualnetworktoconnectthemthroughtheIndianterraininawaythatminimizescon-structioncosts.2368Indiancitiesful?lloneofthetwocriteria,i.e.havingapopu-lationabove500,000orbeingastatecapital.ThelocationofthesecitiesisshowninFigure6.Inordertodeterminethenetworkwhichconnectsalltargetedcitiesinaleast-costmanner,one?rstneedstoobtainameasureforroadconstructioncostsontheIndianterrain.IfollowFaber(2013)andassumethattheconstructioncostsonagiven1x1kmcelloflanddependsontheslopeandtheshareofwaterandbuiltupareainthefollowingway:ConstructionCostsc=1+Slope+25Builtup+25Water(1)SlopeismeasuredinpercentandBuiltupandWaterarebinaryindicatorswhich22NotethattoestimatetherelationshipbetweenGDPandlightIdonotexploittimevariation.Analternative,whichhasalsobeenusedinHendersonetal.(2012),istoestimatetherelationshipinapanelorinlongdifferences.Intheperiodfrom2000to2009,theresultforthelatersuggestsanelasticityofroughly0.5.However,itissubjecttothetwoconcernedraisedaboveandinthiscontextIprefertorelyontheestimatebasedonthe2010crosssection.23ThisapproachhaspreviouslybeenappliedbyFaber(2013)inChinatoconstructaninstru-mentfortheactuallybuiltexpressways.Inthestepsbelow,IadaptthisapproachtoreplicatetheChinesenetworkinIndia.15 taketheunitvalueifthemajorityofthecellisbuiltuporwater,respectively.24Applyingthisformulausingdetailedterraindataproducesa?ne11kmgridofconstructioncostsfortheentireIndianlandscape.Giventhisgridofconstructioncosts,onecaninasecondstepapplytheDijkstraalgorithmto?ndthecheapestconnectionbetweenanytwogivenpointsthroughthecostgrid.25TheprocedureisillustratedinFigure7,wherethecellsrepresentdifferentconstructioncosts(basedonEquation1)andthelinesaretheleast-costpathstoconnectthecities(shownascircles).Thethirdstepinordertoobtainthecounterfactualnetworkisto?ndthecheapestpossiblewaytoconnectalltargetedcitiestothenetwork.26ThisisachievedbytheKruskalalgorithm(Kruskal,1956),whichusesasinputsallbilateralconstructioncostsand?ndstheminimallinksneededtoconnectallcitiesatleastoncetothecommonnetwork.Thisproducestheleast-costnetwork.Onceitisdeterminedwhichbilateralconnectionsmustbemade,thecounter-factualhighwayscanbedrawnwithGISsoftwarefollowingtheleast-costpathcomputedabove(illustratedbylinesinFigure7).TheresultingcounterfactualnetworkisshowninFigure9.Itrepresentsthecheapestwaytoformallyful?lltheChinesepolicyobjective(connectingcitieswhichhaveapopulationabove500’000orarestatecapitals)inIndia.27Despiteitsimmediatelinktotheof?cialpolicyobjective,thisminimalisticnetworkisnotonethatwouldtypicallybeimplementedbygovernments.Onereasonisthatplannerswouldmostlikelycomplementitwithadditionalconnec-tionsbetween”looseends”createdbythealgorithm.Forexample,iftwocitiesareindirectlyconnectedthroughothercities,thenconnectingthemisredundantfromtheperspectiveofthepolicyobjective,evenifthecitiesareclosetoeach24Theimplicationofthisformulationisthata25percentagepointsincreaseinsloperaisestheroadconstructioncostsinthesamewayaswhentheroadhastobebuiltthroughanareawithexistinghouses,otherinfrastructure,orwater.DifferentfromFaber(2013),myformulationdoesnotincludewetlands.25ThisalgorithmisimplementedintheArcGISNetworkAnalystextension.Thesamealgo-rithmcanbeusedtocomputetheleast-costtransportroute(insteadofleastcostconstructionpath).Thealgorithmhasalreadybeenwidelyusedintheeconomicsliterature,forexampleinDell(2012),Faber(2013),DonaldsonandHornbeck(2013),andDonaldson(forthcoming).26Notethatthepreviousstepcomputedtheleast-costconstructionpathbetweenallbilateralpairsofcities.Mostofthesepathsareredundantbecauseagivencitymayalreadybeconnectedtothenetworkthroughanothercity.27Moreprecisely,itisthecheapestwaytoconnectagivensetofnodeswithbilaterallinks.Theproceduredoesnotallowendogenousnodesorhubs.16 other.However,inreality,theadditionallinkmaypossiblybeeffectiveforreduc-ingtransporttimes.Thisillustratesthattheleast-costnetworkisnotminimizingtransportcostsnormaximizingaggregateincome.However,theleast-costnet-workisausefulbenchmarkforthecounterfactualanalysisbecauseitimprovestransportinfrastructureforaparticularsetofcitieswhichwouldhavebeentar-getedbytheNEN.Importantly,itisanobjectivewaytoreplicatetheChinesenetworkbecausetheleast-costnetworkisunique.Thecounterfactualnetworkthereforeallowsaninterestingcomparisonbetweenanetworkthatfocusedonthefourlargesteconomiccenters(theIndianGQ)andonethatconnectsallinter-mediatesizedcities(theChineseNEN).Intherobustnesssection,IdiscusstheresultsforalternativecounterfactualnetworksinIndia.Inparticular,IproposeanalternativenetworkexploitingthattheChinesegovernmentalsospeci?edthatthetargetedcitiesshouldbeconnectedwithraysoutofthecapitalcityandwithhorizontalandverticalcorridors.ThesealternativenetworksresemblemorethestructureofthenetworkwhichwasactuallybuiltinChina.Thedisadvantageisthattheyarenotuniqueastherewouldbeseveralwaystomaketheconnections.Inthemainpartofthispaper,Iwillthereforefocusontheleast-costnetworkanddiscussthealternativesintherobustnesssection.4.3ComputingTransportCostsThroughaRoadNetworkTransportinfrastructureaffectseconomicactivityinseveraldimensions,suchasthetimeittakestomovegoodsandpeople,pecuniarycostsfromtolls,orrisksassociatedwiththeuseofinadequateoroverusedinfrastructure.Iwillfocusonthetransporttimesasadeterminantoftransportcosts.Higherroadquality,limitedaccess,andmorecapacityareallre?ectedinthetimeittakestomovegoodsbetweentwolocations.Thecounterfactualanalysisrequiresinformationonthetransporttimesbe-tweenallpairsofIndiandistrictsfordifferentversionsof(actualandcounter-factual)transportnetworks.Whilethetransporttimesonthecurrentnetworkcanbederivedfromautomatedsearchesonapplicationslikegooglemaps,thisisnotthecaseforpastorcounterfactualnetworks.MyapproachistousetheDi-jkstraalgorithmthat?ndstheshortestpath(intermsoftransporttime)between17 anytwolocationsonadigitizedroadnetwork.Theadvantageofthisapproachisthatthesamealgorithmcancomputeallbilateraltransporttimesfordifferentroadnetworks.Therequiredinputs(describedinSection4.1)arethegeographi-callyreferencedroadsandthetransportspeedondifferenttypesofroads.Withtheseinputs,itispossibletoconstructagridofIndiawherethevalueofeach1010kmcellrepresentsthecostsoftravelingthroughthiscell.Thesetravelcostsdependonthequalityoftheroadinsideofeachcell,i.e.thetravelcostsarehighifthereareonlyroadsofpoorqualitywithlowtravelspeeds.SuchagridoftransportcostsisshowninFigure8.TheDijkstraalgorithmthencalculatesthecheapestwaytotravelfromonelocation(districtcentroids,representedbydotsinFigure8)toanotherlocation.Dependingontheroadinfrastructureandthusonthetransportcostsineachcell,thecheapestpathmaynotbetheshort-estintermsofdistance.Moreimportantly,thetransporttimesassociatedwiththecheapestpathchangewhentheinfrastructureisimproved,thusgeneratingtimevariationinthetransportcosts.FollowingRobertsetal.(2012),Iassumethatthereareeconomiesofscaleintransport,suchthattransportcostsincreaselessthanproportionallyintransporttimes.28Moreprecisely,Icalculatetransportcostsbetweenanoriginoandadestinationdas0:6TransportCostsod=1+TransportTimeod;(2)wheretheexponentof0:6isanaveragevaluethatRobertsetal.(2012)derivedfortheChinesenetwork.Aparticularcaseiswheno=d,i.e.transportcostswithinadistrict.Althoughadistrictisrepresentedherebyitscentroidandtheicebergassumptioninsuchacaseimpliestransportcostsof1,thiswouldnotbeaccurateforactualIndiandistrictwhichdiffersubstantiallyinsize.Onesolutionwhichhasbeenusedintheliteratureistonormalizeittotheobservationwiththesmallestlandarea(AuandHenderson,2006).Iusethedistancebetweenthedistrictcentroidandthenearestdistrictborderasameasureforwithin-districtcostsandnormalizealltravelcoststothesmallestdistanceinthesample.ByapplyingtheDijkstraalgorithmtoallversionsofthetransportnetworks(past,current,andcounterfactualnetwork),onecanderivethebilateraltrade28Thisisacommonassumption,seeforexamplealsoAuandHenderson(2006)whoassumethattransportcostsincreaselessthanproportionallyindistance.18 costsofanypairofdistrictsforeachversion.Intheempiricalanalysis,thisvaria-tionintradecostsovertimeisrelatedtogrowthinincome.Thechannelthroughwhichthisrelationshipworksisshownbythetheconceptualframeworkinthenextsection,whichservesasguidefortheempiricalanalysis.5ConceptualFrameworkThissection?rstpresentsaconceptualframeworkthatillustrateshowtransportinfrastructureaffectseconomicdevelopment.29ThenIdiscusshowthelightdataandthetransportnetworkspresentedinSections3and4areusedtoestimatethiseffect.ThesetupisageneralequilibriumtrademodelasinEatonandKortum(2002).DonaldsonandHornbeck(2013)derivefromavariationofthatmodelareducedformexpressionfortheimpactoftransportinfrastructureonincome.Thatexpressioncapturesthe”marketaccess”ofalocation,whichisthesumovertradingpartners’income,discountedbythebilateraltradecostsandbythemar-ketaccessofthetradingpartners.TheyusethisframeworktoestimatetheeffectoftheexpansionoftheAmericanrailwaynetworkonlandprices.IfollowthisapproachtoestimatetheeffectoftheIndiantransportnetworkonincomebyadaptingtheirframeworktoaversionwhichcanbeestimatedwithlightdataasameasureforrealincome.5.1ARicardianModelofTradeThebasicsetupisaRicardiantrademodelwiththeimmobileproductionfactorslandandlaborandthemobilefactorcapital.30Theeconomyconsistsofmanytradingregions(i.e.Indiandistricts),wheretheoriginofatradeisdenotedbyoandthedestinationbyd.Eachdistrictproducesvarietiesindexedbyjwitha29Thepresentationinthissectionfocusesonthekeyaspectsofthemodel.Thedetailsarediscussedintheappendix.30DonaldsonandHornbeck(2013)assumethatlaborismobile.Themotivationforandim-plicationoftheassumptionthatlaborisimmobilewillbediscussedbelow.Thetwoimmobileproductionfactorscannotactuallybeseparated,butIdistinguishtheminordertoallowacom-parisontothemodelwithmobilelaborasinDonaldsonandHornbeck(2013).19 Cobb-Douglastechnologyusingland(L),labor(H),andcapital(K),