RoadblockorAccelerator?TheEffectofElectricVehicleSubsidyEliminationNafisaLohawalaWorkingPaper23-13May2023AbouttheAuthorNafisaLohawalaisafellowatResourcesfortheFuture.SheearnedaPhDineconomicsattheUniversityofMichiganafterreceivingaBS-MSdualdegreeineconomicswithaminorincomputerscienceandengineering(algorithms)fromtheIndianInstituteofTechnology,Kanpur.Lohawala’sresearchliesattheintersectionofindustrialorganization,energyeconomics,andpublicfinance.Shefocusesontheeffectofgovernmentpoliciesonenvironmentalandsafetyexternalitiesgeneratedbythetransportationsector,aswellasothertransportationissues,includingdecarbonizationstrategiesformedium-andheavy-dutyvehicles.AcknowledgementsIthankmyPhDadvisorsYingFan,JoelSlemrod,ZachBrown,andCatherineHausmanfortheirconstantguidanceandsupport.IalsothankBeiaSpiller,GloriaHelfand,JingLi,JoshuaLinn,PaulMcCarthy,NathanSeegert,andseminarandconferenceparticipantsattheUniversityofMichigan,FederalReserveBoard,IndianInstituteofManagementAhmedabad,IndianInstituteofTechnologyKanpur,OklahomaStateUniversity,ResourcesfortheFuture,UniversityofGeneva,theNationalTaxAssociationAnnualConference(2021)andtheAnnualInternationalIndustrialOrganizationConference(2023)forhelpfulcommentsandsuggestions.ThisresearchwasfundedbytheAlfredP.SloanFoundationPre-DoctoralFellowshiponEnergyEconomics(awardedthroughtheNBER),theRackhamGraduateStudentResearchGrant,andtheGraduateResearchAwardbytheDepartmentofEconomics,UniversityofMichigan.Allerrorsaremine.AboutRFFResourcesfortheFuture(RFF)isanindependent,nonprofitresearchinstitutioninWashington,DC.Itsmissionistoimproveenvironmental,energy,andnaturalresourcedecisionsthroughimpartialeconomicresearchandpolicyengagement.RFFiscommittedtobeingthemostwidelytrustedsourceofresearchinsightsandpolicysolutionsleadingtoahealthyenvironmentandathrivingeconomy.Workingpapersareresearchmaterialscirculatedbytheirauthorsforpurposesofinformationanddiscussion.Theyhavenotnecessarilyundergoneformalpeerreview.TheviewsexpressedherearethoseoftheindividualauthorsandmaydifferfromthoseofotherRFFexperts,itsofficers,oritsdirectors.SharingOurWorkOurworkisavailableforsharingandadaptationunderanAttribution-NonCommercial-NoDerivatives4.0International(CCBY-NC-ND4.0)license.Youcancopyandredistributeourmaterialinanymediumorformat;youmustgiveappropriatecredit,providealinktothelicense,andindicateifchangesweremade,andyoumaynotapplyadditionalrestrictions.Youmaydosoinanyreasonablemanner,butnotinanywaythatsuggeststhelicensorendorsesyouoryouruse.Youmaynotusethematerialforcommercialpurposes.Ifyouremix,transform,orbuilduponthematerial,youmaynotdistributethemodifiedmaterial.Formoreinformation,visithttps://creativecommons.org/licenses/by-nc-nd/4.0/.RoadblockorAccelerator?TheEffectofElectricVehicleSubsidyEliminationNafisaLohawala∗May4,2023AbstractFederalandstategovernmentsinmanycountriessubsidizetheearlyadoptersofelectricvehicles(EVs).Theseprogramsoftenusequotasordeadlinestophaseoutthesubsidies,whichcancreatedynamicincentivesforcarmanufacturers.Sincemostoftheliteraturestudiestheeffectofintro-ducingsubsidiesonmarketoutcomesinstaticsettings,littleresearchhasaddressedthedynamiceffectsofsubsidy-cappingdesigns.ThispaperexploresthoseeffectsintheUSvehiclemarket.Idevelopastructuralmodelofconsumers’vehiclechoicesandmanufacturers’pricingdecisionsintheUSautomobileindustry.Ithenestimatethemodelusingcomprehensivedataonnewvehicleregistrations,prices,characteristics,andsubsidiesin30statesbetween2011and2017.Basedontheprimitivesgeneratedfromthemodel,Iconductcounterfactualsimulationstocomparethreedesigns:amarketwidedeadline,aper-manufacturerdeadline,andaper-manufacturerquota.Thesimulationsshowthatforagivengovernmentexpenditure,thequotaleadstoupto18percentlowerEVsalesthanthedeadlines.Moreover,eachdesigninfluencesthesalesofconventionalvehicles,consumersurplus,manufacturerprofits,andliquidfuelconsumptiondifferently.∗.ResourcesfortheFuture,Washington,DC20036,email:nlohawala@rff.org.IthankmyPh.D.advisorsYingFan,JoelSlemrod,ZachBrown,andCatherineHausmanfortheirconstantguidanceandsupport.IalsothankBeiaSpiller,GloriaHelfand,JingLi,JoshuaLinn,PaulMcCarthy,NathanSeegert,andseminarandconferenceparticipantsattheUniversityofMichigan,FederalReserveBoard,IndianInstituteofManagementAhmedabad,IndianInstituteofTechnologyKanpur,OklahomaStateUniversity,ResourcesfortheFuture,UniversityofGeneva,theNationalTaxAssociationAnnualConference(2021)andtheAnnualInternationalIndustrialOrganizationConference(2023)forhelpfulcommentsandsuggestions.ThisresearchwasfundedbytheAlfredP.SloanFoundationPre-DoctoralFellowshiponEnergyEconomics(awardedthroughtheNBER),theRackhamGraduateStudentResearchGrant,andtheGraduateResearchAwardbytheDepartmentofEconomics,UniversityofMichigan.Allerrorsaremine.1IntroductionConsumersubsidiesandrebatesareapopularmeanstopromoteadvancedtechnologyvehiclesinseveralcountries,suchastheUnitedStates,Canada,China,andNorway(BeresteanuandLi2011;Chandra,Gulati,andKandlikar2010;Jenn,Azevedoa,andFerreira2013).Policymakerstypicallyuseprovisionssuchasquotasanddeadlinestocapthesesubsidies.Despitethiswideuse,littleworkhasbeendonetounderstandtheireffectonmarketoutcomes.Thispaperextendstheliteraturebyconsideringthedynamiceffectsofsuchprovisions.Ishowthatdifferentprovisionshavedifferenteffectsthatcanreinforcetheintendedpolicyobjectivesorcreateunintendedconsequencesthatpartlyundothebenefitsofthesubsidy.Policymakerssubsidizeplug-inelectricvehicle(EV)purchasestoaddressvariousexternalities.EVadoptionhasapositiveenvironmentalexternalityduetozerotailpipeemissions.1Itcanenhancetheenergysecurityofoil-importingcountriesbynotrelyingongasoline.IthasinformationspilloverstotheextentthatEVconsumershelpspreadinformationaboutit.Italsomakesentryattractiveforchargingstations,whichiscrucialfordevelopingachargingnetworkandfurtherencouragingdemand.Inaddition,apolicygoalmaybetointegrateEVsintotheautomobileindustrybyovercomingthemostsignificantbarriertotheiradoption—highup-frontcost—bymakingEVspricecompetitivewithconventionalvehicles.ThetraditionalPigouviansolutiontoexternalitiesistosubsidizetheexternality-generatingac-tivityequivalenttothemarginalexternalbenefitattheoptimalquantity.Policymakerstypicallyallowthesesubsidiestoendafteracertainperiod,withseveralpossiblereasonsfordoingso.First,subsidiescanbeprohibitivelyexpensiveifsalessurgeduetohigherbudgetaryrequirements,admin-istrativecosts,orotherpoliticalreasons.Second,marginalgainsfrominformationalspilloversarelikelytofadeasEVsintegrateintotheautomobileindustry.Finally,themarginalcost,andthereforetheprice,islikelytocomedownasmanufacturersfindcheaperwaystoproducethebattery.Policymakersworldwideusedifferentstrategiestocapthesubsidies,suchaslimitingthetotalexpenditure,imposingadeadline,orcombiningboth.Forinstance,in2012,NorwayplannedtoremovefinancialincentivesforEVsafter2018oroncetherewere50,000suchvehiclesontheroad(Steinbacher,Goes,andJörling2018).2In2020,ChinaplannedtocutEVsubsidiesprogressivelybetween2020and2022,withcompleteexpirationin2022.3TheUSfederalEVtax-creditprograminitiatedbytheEnergyImprovementandExtensionActof2008cappedtheincentivesbygivingeachmanufactureraquotaof200,000vehicles,afterwhichitscreditphasedout.TheInflationReductionActofAugust2022(IRA)replacedthiscapwithasingledeadlineforpurchasesubsidies.VehiclesproducedbyallEVmanufacturersnowqualifyforthesubsidyuntil2032(providedtheymeetsomeadditionalrequirements).1.QuestionshavebeenraisedintheliteratureontheenvironmentalbenefitsofdrivingEVsbecausechargingthebatteryincreasespollutionatthepowerplant(seeBabaee,Nagpure,andDeCarolis(2014),Archsmith,Kendall,andRapson(2015),Hollandetal.(2016),andBuekersetal.(2014)).Thecleanerthegrid,thegreatertheenvironmentalbenefitsofreplacingagasolinevehiclewithanEV.2.Althoughthe50,000targetwasreachedearlyin2015,Norwaylaterextendedtheincentives.3.Seehttps://www.globaldata.com/data-insights/automotive/china-will-end-ev-subsidies-after-30-cuts-in-2022/.1Foragivenlevelofexpenditure,arethesedesignsequallyeffectiveinraisingEVpenetration?Whataretheirimplicationsformarketoutcomessuchasconsumersurplus,EVmanufacturers’profits,andoverallgasolineconsumption?Thispapertakesasteptowardansweringthesequestions.IfocusontheUSfederalEVtax-creditprogram,whichprovidesnonrefundableincometaxcreditsofupto$7,500toEVconsumers.4BeforetheIRA,theprogramhadauniquesubsidy-cappingdesign(seeFigure1).Thephaseoutwastriggeredwhenagivenmanufacturerdeliveredthe200,000thsubsidy-qualifyingvehicle.Inthatquarterandthenext,theper-vehiclesubsidyremainedunchanged.Itthenreducedtohalfforthatmanufacturer’svehiclesforthenexttwoquarters,one-fourthforanothertwoquarters,andthenzero.Giventhetimingcomponentinthisdesign,pushingthesaleofthe200,000thvehicletothefollowingquarter(e.g.,inJulyinsteadofJune)coulddelaythephaseoutbyaquarter.Iexaminetheshort-termdynamicincentivescreatedbythisdesign.Todothat,Ibreakthedesignintotwocomponents.Thefirstisaper-manufacturerquota,inwhicheachmanufacturerfacesaseparatequotaonthetotalnumberofqualifyingvehicles.SimilartotheactualUSdesign,ifthemanufacturersellsfewerEVsthanitsquotainanyperiod,thenallEVsthatitsellsinthenextperiodalsoqualifyforthesubsidy.Thesecondcomponentisaper-manufacturerdeadline,whereeachmanufacturerfacesaseparatedeadline.Theactualdesignintheprogramcanbeconsideredasacombinationofaper-manufacturerquotaof200,000vehiclesandthreeper-manufacturerdead-linescorrespondingtothe50%,75%,and100%subsidycuts.IcomparethesecomponentswithamarketwidedeadlineasinstitutedbytheIRA.5Comparingthesedifferentdesignshelpsunderstandthedynamicimplicationsofreplacingtheearlierdesignwithamarketwidedeadline.Ifirstuseastylizedtwo-periodmonopolymodeltoillustratehowthisquotacanpotentiallyin-centivizemanufacturerstoreducetheirEVsalesastheygetclosetothequota,therebyunderminingtheeffectivenessofthesubsidy.Theincentivearisesbecausebystayingbelowthequotainagivenperiod,themanufacturercanensurethatallEVssoldinthefollowingperiodalsoqualify.Becausethemanufacturerhasmarketpower,itcanretainsomebenefitsofthesubsidies.Asaresult,itcanearnhigherprofitsfromalltheEVsalesmadeduringanadditionalperiod.Incontrast,cappingthesubsidyusingadeadlinedoesnotcreatethisincentivebecausethemanufacturercannotcontrolwhenthesubsidyexpires.Next,toquantifytheeffectsofeachdesign,IdevelopandestimateastructuralmodeloftheUSautomobileindustry.Thedemandsidefollowsadiscrete-choiceframework,whereconsumerschooseavehicleamongallavailablefueltypes.BecausetheEVmarketisstillnascent,adoptionmaydependoninformationgainsfromearlyadoptersandmobilitygainsfromthedevelopmentofachargingnetwork(KalishandLilien1983;HeutelandMuehlegger2015;Springel2021;Lietal.2017).IcapturethisnetworkeffectinthemodelbyallowingconsumerstocareaboutthenumberofEVspreviouslysoldintheirlocalgeographicarea.Incontrast,thesupplysideisanoligopoly4.Startingin2024,thesubsidywillbeavailableasa"point-of-salerebate"ratherthanataxcredit.5.TheIRAalsomadeotherchangestotheprogram.Forinstance,itaddedafinalassemblyrequirement,wherebythesubsidyisonlyavailableforEVsthatwereassembledinNorthAmerica,andapricecap,anincomecap,andrestrictionsonsourcingcriticalbatteryminerals.Becausemyfocusisonunderstandingtheeffectofthesubsidy-cappingdesigns,Idonotconsidertheseotherchanges.2withproductdifferentiationwherecarmanufacturerscompeteinprices.Themodel’skeyfeatureisthat,inadditiontocurrentprofits,itallowsmanufacturerstocareaboutthefollowingyear’sprofitswhenchoosingvehicleprices.Suchtwo-periodpricingcapturesmanufacturers’responsestothedynamicincentivesinducedbythesubsidycaps,whichastaticmodelwouldmiss.Thetwo-periodmodelalsoallowsmanufacturerstointernalizethedemand-sidenetworkeffect.Givenaper-manufacturerquota,thenetwork-effect-inducedincentivesworkintheoppositedirectiontoquota-inducedincentives.Ontheonehand,exhaustingthequotashrinksthefutureEVdemandbyeliminatingthesubsidy;ontheotherhand,attractingearlyadoptersincreasesthefutureEVdemandduetothenetworkeffect.Thus,themanufacturer’spricingresponseisaprioriambiguousanddependsonmarketparameterssuchasown-andcross-priceelasticitiesandthenetworkeffect.Next,Iestimatethesedemandparametersusingproduct-leveldataonvehicleregistrations,characteristics,andfederalandstate-levelsubsidiesin30statesbetween2011and2017.Basedontheestimateddemandparametersandthefirst-orderconditionsofthemanufacturers’profitfunctions,Ithenrecoverthevehiclemarkupsandmarginalcostsin2017.Finally,basedontheseprimitives,Irecomputepricingequilibriaunderthreecounterfactualdesigns:amarketwidedeadline,aper-manufacturerdeadline,andaper-manufacturerquota.Icomparethesedesignswithacounterfactualwithnosubsidy.Thesimulationsshowthateventhoughallsubsidy-cappingdesignsboosttheEVmarketpen-etration,therearenotequallyeffective.Foragivengovernmentexpenditure,aper-manufacturerquotacanleadtomuchlowerEVsalescomparedtothedeadlines.Inmyexperiments,theper-manufacturerquotareducesEVsalesbyupto18percentcomparedtothedeadlines.Twofactorsdrivethisresult.First,stayingbelowthequotainanyperiodallowsamanufacturertoqualifyforthesubsidyonallEVsitsellsinthefollowingperiod,whichallowsittoearnhigherprofits.Second,becausethesubsidyisonlyeliminatedformanufacturersthatexhaustthequota,notdoingsopro-tectsmanufacturersfromcompetitionfrommanufacturersbelowthequota.Incontrast,asdeadlinesdonotallowmanufacturerstocontrolwhenthesubsidyexpires,theycanbemorecost-effectiveinincreasingEVmarketpenetration.Theseresultssuggestthat,allelseequal,replacingtheearlierdesignwithamarketwidedeadlinewilllikelyboosttheEVmarketpenetrationclosertosubsidyexpiration.Inadditiontotheeffectonmarketpenetration,thesubsidy-cappingdesignscanhavespillovereffectsonconventionalvehicles’salesandcanaffectconsumersurplus,manufacturers’profits,andliquidfuelconsumption.Theyalsoaffectprofitdistributionacrossmanufacturers.Comparedtoamarketwidedeadline,aper-manufacturerdeadlineshiftsprofitsawayfromthemanufacturersthatfacethelimit.Aper-manufacturerquotadoesnotnecessarilydoso,becauseitallowsthemtocontrolwhenthesubsidyexpires.ThisfindingshedslightontheargumentmadebythedominantEVmanufacturers,suchasTesla,GeneralMotors(GM),andNissan,whichhaveclaimedthattheper-manufacturercapputthematacompetitivedisadvantagecomparedtonewlyenteringrivals.EVsubsidiesbecameatopicofvigorousdebateduringthetaxreformof2017partlybecauseofthis3cap;thedominantmanufacturersandothersupportersformedanEV-drivecoalitionandargued(amongotherreforms)toremovethecap.Aftertheoriginaldesignsurvivedthattaxreform,thetopEVmanufacturersthatinitiallylobbiedtopreservetheseincentivesstartedfavoringtheirremovalaltogether(Lambert2018).Theresultssuggestthat,allelseequal,replacingtheearlierdesignwithamarketwidedeadlineislikelytoimpactthedistributionofprofitsandothermarketoutcomesmentionedabove.Thepaperaddstomultiplestrandsoftheliterature.First,itcontributestothegrowingeco-nomicliteratureontheroleofgovernmentpoliciesindecarbonizingtransportation.AccordingtotheParisAgreement,manygovernmentsfromdifferentcountrieshavesetatargettoachievenet-zeroemissionsby2050.TransitioningfromconventionalvehiclestoEVsisessentialtothisgoal(Williamsetal.2012).Somepapers,suchasDeShazo,Sheldon,andCarson(2017),Jenn,Springel,andGopal(2018),andClintonandSteinberg(2019),analyzetheeffectivenessofincentivesinencouragingcon-sumeradoptionofEVsandgenerallyfindthatconsumersrespondtosubsidiesandotherincentives.Otherpapers,suchasLietal.(2017),Li(2018),andSpringel(2021),explorethepositivefeedbackloopbetweenEVpurchasesandcharginginfrastructure.Thisbodyofliteraturesuggestsanindirectnetworkeffectthatisimportantforpolicydesign.Yetotherpapers,suchasAghionetal.(2016),Jacobsen(2013),andGillingham(2022),modelvehiclemanufacturers’responsestoenvironmentalregulation.IcontributebyanalyzinghowthesubsidydesigncanhelpimproveEVmarketpenetra-tion,accountingformanufacturers’responsestothedynamicsofsubsidyelimination.Comparingthemarketoutcomesunderdifferentdesignsallowsforsystematicpolicymaking—basedonarray-ingalternativedesignsandcomparingtheadvantagesanddisadvantagesofeach.Recognizingtheimportanceofthenetworkeffectonmanufacturers’pricingdecisions,Iinternalizeitbyallowingconsumers’utilitytodependonpreviousEVpurchasesintheirgeographicarea.Morebroadly,thepapercontributestotheliteratureinvestigatingtheroleofgovernmentin-centivesinpromotinggreentechnology.ExamplesincludeBeresteanuandLi(2011),GallagherandMuehlegger(2011),andJenn,Azevedoa,andFerreira(2013)onhybridvehicles,VanBenthem,Gillingham,andSweeney(2008),CragoandChernyakhovskiy(2017),andLangerandLemoine(2022)onsolarpower,andHitaj(2013)onwindpowerdevelopment.Mostpapersstudytheeffectofintroducingsubsidiesonpricesandwelfareinastaticequilibriumbutignorethedynamicsofsubsidyelimination.Forinstance,BeresteanuandLi(2011)buildanequilibriummodelofthenewcarmarketandestimatethatfederalincome-taxcreditsforhybridvehiclesaccountedforabout20percentofsuchsalesin2006.Mypaperaddstotheliteraturebyexplicitlymodelingtheresponsesofforward-lookingvehiclemanufacturerstothesubsidy-cappingdesignsinamicrofoundedmodel.Theanalysisisrelevantforothercountries,aswellasotherenvironmentallyfriendlyproducts,suchasfuel-cellvehicles,solarpanels,smallwindturbines,andgeothermalheatpumps,wherepolicymakersusesimilarsubsidy-cappingdesigns.Finally,mypaperaddstotheliteratureontheincidenceeffectsofsubsidyprograms.SomepapersonUScleanenergysubsidiesincludeSallee(2011),BorensteinandDavis(2016),Gulati,McAusland,andSallee(2017),andPlessandVanBenthem(2019).Examplesfromothersubsidy4contextsincludeCabral,Geruso,andMahoney(2018)onhealthinsurance,PolyakovaandRyan(2019)ontheAffordableCareAct,andFanandZhang(2022)oncellphones.Thispaperaddstotheincidenceliteraturebyhighlightingthat,foragivenvalueofthesubsidy,theincidencedependsonprogramdesign.Itakeastructuralapproachthatallowsforadetailedanalysisofmediatingfactorsandasimulationofmarketoutcomesunderthecounterfactualsubsidy-cappingdesigns.Therestofthepaperisorganizedasfollows.Section2providesabriefbackgroundoftheUSplug-inEVindustry.Section3describesanillustrativeexampletoprovideeconomicintuitionandidentifiesthekeyparametersgoverningmanufacturers’responsestothesubsidy-cappingdesigns.Section4outlinestheutilityspecificationandthesupply-sideproblem.Section5reportsdataandsummarystatistics.Section6discussesidentification,estimation,andresults.Section7describesthecounterfactualexperimentsanddiscussesthefindings.Section8concludes.2IndustrialBackgroundThissectionbeginswithabriefdescriptionoftheUSplug-inEVmarketandthefederaltax-creditprogramthatisthefocusofthispaper.Itthendescribesthekeymechanismsofinterestinthefederalprogram.Finally,itdescribesotherregulationsthathaveinfluencedEVdevelopment.2.1Plug-InEVMarketandFederalTaxCreditsPlug-inEVsareroadvehiclespoweredbybatteriesthatcanberechargedbypluggingintotheelectricgrid.Theycomeintwovarieties:(i)batteryEVs(BEVs),whicharepoweredexclusivelythroughelectricity,and(ii)plug-inhybridEVs(PHEVs),whichuseanelectricmotorastheprimarypowersourceandtheinternalcombustionengineasabackup.Bothdifferfromfuel-cellEVs(FCEVs),suchastheHondaClarity,andconventionalhybrids(HEVs),suchastheToyotaPrius,neitherofwhichcanbepluggedintoanelectricgrid.TheUSplug-inEVmarketmostlydevelopedafterNissanintroducedtheLeafinlate2010.Withfuelefficiencyandenvironmentalregulationsbecomingincreasinglystringent,mostUSvehi-clemanufacturershaveaddedplug-instotheirportfolios.Asof2023,Teslaisthehighest-sellingmanufacturer,followedbyGMandNissan.TheUSfederalgovernmentstartedatax-creditprogramforPHEVsandBEVsundertheEn-ergyImprovementandExtensionActof2008.TheprogramofferednonrefundabletaxcreditsforpurchasesmadeafterDecember31,2009(IRS2009).Thecreditvariedbycarmodelandwasworth$2,500plus$417foreachkilowatt-hourofbatterycapacityover4kWh,cappedat$7,500.6BEVsqualifyforahighercreditthanPHEVsduetotheirlargerbatterycapacity.PopularBEVs,suchasallTeslamodelsandChevroletBolt,qualifiedforthefull$7,500subsidy.Until2022,theUSprogramusedauniquephaseoutprovision.AssummarizedinFigure1,thephaseouttriggeredforamanufactureronceitsold200,000subsidy-qualifyingcarsforUSuseafter6.Aconsumer’spurchaseneededtomeetspecificrequirementstobeeligible.SeeInternalRevenueCodeSection30Dfordetails.5December31,2009.Thecreditwasunchangedinthequarterwhenthemanufacturerdeliveredthe200,000thvehicleandthenextquarter.Itreducedto50percentforthenexttwoquarters,25percentforthenexttwoquarters,andthenzero.Alleligibleplug-invehiclessoldduringthephaseoutperiodqualifiedforthecredit.Thisdesignfollowedthetax-creditprogramforconventionalhybridvehicles(EnergyPolicyAct,2005),allegedlydesignedtopreventdominantforeignmanufacturers,suchasToyotaandHonda,frombenefitingmorethandomesticmanufacturersovertheprogram’slife(Lazzari2006;Leonhardt2006).Thefirsttwomanufacturersthathitthethreshold(TeslaandGM)areAmerican.Tesladeliveredthe200,000thqualifyingvehicleinJuly2018;Teslacarsqualifiedfora$7,500creditJuly–December2018,$3,750January–June2019,and$1,875July–December2019(IRS2018).GMdeliveredthe200,000thqualifyingvehicleinNovember2018andfacedthesubsidyexpirationinApril2020(IRS2019).TheIRAreplacedthisphaseoutdesignwithamarketwidedeadlineof2032.Allmanufacturers,includingtheoneswiththe200,000thvehiclebefore2022,areeligibleuntil2032,providedtheirvehiclesmeetadditionalrequirements,suchasdomesticassembly.7BecauseonlytwoEVmanufacturershavefacedtheeliminationofpurchasesubsidies,Irelyonstructuralmethodstounderstandtheimplicationsofdifferentsubsidy-cappingdesigns.Specifically,IdevelopandestimateastructuralmodeloftheUSautomobileindustry,explicitlyaccountingforconsumers’andmanufacturers’decisions.Iusetheestimatedmarketparameterstosimulatepricingequilibriumunderthecounterfactualdesignsandcomparemarketoutcomesacrossdifferentdesigns.AppendixAshowstime-seriesevidencethatEVsalesrespondeddifferentlytotheper-manufacturerquotaandtheper-manufacturerdeadline,basedonTeslaandGM’sexperiences.IncontrasttotheEVtax-creditprogramstudiedhere,theconventionalhybridtax-creditpro-graminitiatedbytheEnergyPolicyAct(2005)wouldallowbetterdataavailabilityduringandafterthesubsidyeliminationbecausetheprogramexpiredin2010.Nonetheless,Ifocusontheplug-inEVtax-creditprogramfortworeasons.First,incontrasttoplug-inEVs,conventionalhybridscomparedbetterthanthedominantalternativesfromaconsumer’sperspective,astheycombinethebenefitsofgasolineenginesandelectricmotors.Asaresult,hybridswerealreadyinhighdemandbeforethetaxcreditsstarted.Second,thehybridprogramofferedtaxcreditsonlyupto$3,150withamuchlowerper-manufacturercapof60,000.Toyotaexhaustedthequotawithinafewmonths(IRS2006).Duetothesereasons,vehiclemanufacturersaremorelikelytocareabouttheconsumersubsidiesintheEVmarketand,hence,morelikelytorespondtotheirelimination.7.SeeIRS(2023)fordetails.6Figure1:Subsidy-CappingDesignAdoptedintheUnitedStatesQ1Q2Q3Q4Q5Q6Q7Q8QuarterSubsidy(λ)N=200kλ=λ0λ=λ0/2λ=λ0/4λ=0Q1Q2Q3Q4Q5Q6Q7Q8QuarterSubsidy(λ)N=200kλ=λ0λ=λ0/2λ=λ0/4λ=0Notes:Panel(a)explainshowthesubsidywouldevolveifthemanufacturerexhaustedthe200,000thresholdattheendofquarterQ1.Thephaseoutistriggeredinthesecondquarteraftertheelectricvehicle(EV)manufacturerdeliversthe200,000thsubsidy-qualifyingvehicle(Q3).Inthefirstsixmonthsofthephaseout,aqualifyingvehiclefromthatmanufacturerreceives50percentoftheoriginalsubsidy.Inthesecondsixmonths,thesubsidyreducesto25percent.Itiseliminatedthereafter.Thenumberofvehiclesthatcanreceivesubsidiesduringthephaseoutperiodisunlimited.Panel(b)showsthesubsidyevolutionifthemanufacturerhitthethresholdatthebeginningofQ2instead,indicatingasubstantialincentivetoreduceEVsalesattheendofQ1becausedoingsoprolongsthesubsidyforanotherquarter.2.2KeyFeaturesoftheFederalSubsidyDesignThedesigninFigure1isacombinationofaper-manufacturerquotaof200,000vehiclesandthreeper-manufacturerdeadlines.Thefirsttworeducethevalueofthecredit,andthefinaldeadlineeliminatesit.Comparedtotheper-manufacturerdeadline,thequotaincentivizesmanufacturerstoreducetheirEVsalesfortworeasons.First,thequotaholdsupthefirstdeadline.Thesubsidyreducestohalfinthesecondquarterafterthe200,000thsubsidy-qualifyingEVisdelivered.Thus,pushingthesaleofthatvehicletothenextquartercandelaythephaseoutbythreemonths.Second,thesubsidyiseliminatedonlyforthosemanufacturerswhoexhaustthequota;doingsobeforeothersexposesanEVmanufacturertoincreasedcompetitionbecauseothermanufacturerscontinuetoqualify.Byreducingthesalesofsubsidy-qualifyingvehicles,EVmanufacturerscanavoidthissituation.Ontheotherhand,thedeadlinesthatfollowedthequotacreatenosuchincentivebecausemanufacturerscannotcontrolwhenthesubsidyexpires.7Figure2:NetworkEffectChargingInfrastructureEVAdoptionConsumerlearningNotes:Thefiguredepictsthepositivefeedbackeffect(orthenetworkeffect)ofelectricvehicle(EV)adoptiononfuturedemandthroughtwoindependentchannels.Adoptionallowspotentialconsumerstoexperientiallyinferthequalityofplug-inEVs,whichincreasesfutureadoption.Similarly,itmakesentrymoreappealingforchargingstations,andmorechargingstationsallowmoreconsumerstopurchaseanEV.AnEVmanufacturer’sresponsetotheper-manufacturerquotacanbemorecomplexifitan-ticipatesgainsfromsellingearly.Suchgainsmayariseduetomultiplereasons.Onthedemandside,earlysalesmaycreateanetworkeffectthatencourageslatersalesthroughtwomechanismsdescribedinFigure2.Thefirstmechanismisconsumerlearning,whereby“wordofmouth”effectsmitigatetheuncertaintyinproductquality(KalishandLilien1983;HeutelandMuehlegger2015)forfuturecarbuyers.WhenbuyersseemoreEVadoption,theirexposuretothisnewtechnologyincreases,whichmayincreasetheirwillingnesstopurchaseEVsinsteadofconventionalvehicles.Asaresult,earlysalescanshiftthefutureEVdemandtotheright.This,inturn,canleadtomoreconsumerlearningandevenhigheradoption.Thesecondmechanismisdevelopingcharginginfrastructure,whichcreatesasimilarfeedbackeffect:moreEVsontheroadmakeentrymoreappealingforchargingstations,andmorechargingstationsallowmoreconsumerstoadoptEVs.Thus,earlysalescanshiftthefutureEVdemandtotherightbyfacilitatingtheentryofchargingstations.Theroleofcharginginfrastructuremaynotseemobvious,consideringthatconsumerscanplugEVsintoanordinaryelectricoutlet.However,thatprocessisveryslowandnotviablefortravelinglongdistancesthatwouldexceedthe’sbatterycapacity.Fast-charginginfrastructureiscrucialtoensuremobility,especiallyforBEVs,becausetheydonothaveagasolinebackup.Inadditiontothesedemand-sidegains,earlysalesmayoffersupply-sidegainsbyhelpingmanufacturersreducecoststhroughinnovationandself-perfection(learningbydoing).Givensuchgains,aper-manufacturerquotacreatestwoconflictingforces.Ontheonehand,surpassingthequotameansforgoingfuturesubsidies.Ontheotherhand,stayingbelowitmeansforgoingthenetworkgainsfromadditionalsales.Asaresult,theresponsewoulddependontherelativestrengthsofthetwochannels.IdiscussthismechanismfurtherinSection3andaccountforthenetworkeffectinmymodelbyallowingconsumerutilitytodependonthepreviousEVsalesbythemanufacturer.Forsimplicity,Idonotmodelthesupply-sidegainsseparately.82.3State-LevelSubsidiesandZEVMandatesSomestateandlocalgovernmentsalsooffermonetaryornonmonetaryincentives.Monetaryincen-tivesareupto$5,000perconsumer(ontopofthefederaltaxcredits)instatessuchasCalifornia.Nonmonetaryincentivesincludeaccesstocarpoollanesandfreemeterparking.California’sZEVprogramhasalsosignificantlyinfluencedthedevelopmentoftheplug-inEVmarket.DesignedbytheCaliforniaAirResourcesBoardinthe1990stoachievethestate’slong-termemissionreductiongoals,theprogramrequiresagrowingpercentageofmanufacturers’overallsalestohavelowemissions.Nineotherstates(collectivelycalled“ZEVstates“)alsoadopttheZEVregulationsand,togetherwithCalifornia,representnearly30percentoftheUScarmarket.AlthoughZEVmandatesdonotaffectconsumerdecisions,theyaffectmanufacturers’profitfunc-tion.Theprogramworksthroughacreditsystem,whereeachmanufacturermustshowZEVcreditsasapercentageofvehiclesalesintheZEVstatesineachmodelyear.Manufacturerswithashortfallcanusecreditsaccumulatedinotheryearsorbuycreditsfromothermanufacturers.Conversely,manufacturersthatexceedtheircreditrequirementscanbankcreditsforlateryearsorsellthem.Forinstance,TeslaandNissansoldrelativelyhigherBEVvolumesthanothermanufacturersstartingin2012andsoldcreditstoothers.IdiscusstheZEVprogramfurtherinSection4andincorporateitintomymodelbyincludingthevalueofZEVcreditsinthefirms’profitfunctions.3AnIllustrativeModelThissectiondemonstratestheeffectofsubsidy-cappingdesignsonEVsalesusingamonopolyexam-ple.Althoughthefullmodelinvolvesanoligopolywithstrategicinteractions,thissimpleexampleprovideseconomicintuitionandidentifiesthekeyparametersgoverningthedesigns’effect.Section4generalizestothefulloligopolymodel,whichIestimateanduseforcounterfactualexperiments.Consideramonopolistthatmaximizesthesumofprofitsacrosstwoperiods.Themarketdemandinthefirstperiodislinearintheconsumerprice:Q1(P1)=A−BP1,whereAandBarepositivescalars.Themarketdemandinthesecondperiodissimilarbutdependsonthefirst-periodadoptiontoaccountforthenetworkeffect:Q(P1,P2)=(A−BP2)+ηQ1(P1).Here,ηrepresentsthenetworkeffect.Thehigheritsvalueis,themorevaluabletheearlyadoptersare.AsdescribedinSection2,suchanetworkeffectmayberelevantfornewtechnologiesduetoconsumerlearningorchargingnetworkdevelopment.Letλtdenotethepurchasesubsidyinperiodt.Theconsumerpriceisthedifferencebetweenthemanufacturer-setpriceptandthesubsidyλt.Thefirmproducesataconstantmarginalcostc9inbothperiodsandchoosesthepricesp∗1andp∗2tomaximizethesumofprofitsinbothperiods:(p∗1,p∗2)=argmaxp1,p2(p1−c)Q1(p1−λ1)+(p2−c)Q2(p1−λ1,p2−λ2).Considertwosubsidy-cappingdesignsinspiredbythecurrentUSphaseout.Thefirstdesignintroducesadeadlinesothatonlythefirst-periodbuyersqualifyforthesubsidy:λt=s,ift=10,ift=2.Incontrast,theseconddesignintroducesacapΓonthenumberofqualifyingsales.Allfirst-periodbuyersareeligible.Second-periodbuyersqualifyonlyiffirst-periodsalesfailtoexceedthequotaΓ:λt=s,ift=1sI[Q1(p1−s)<Γ],ift=2.Thecrucialdistinctionbetweenthetwodesignsisthatthelattergrantsthefirmcontroloverthesecond-periodsubsidy.Correspondingly,theprivatelyoptimalresponseswoulddiffer.Fortheexposition,Figure3plotstheoptimalfirst-periodsalesunderthedeadlineandquotadesignsasafunctionofquotaΓunderdifferenthypotheticalparametervalues.Panel(1)usesparametervaluesA=300,B=12,c=20,andη=0.Thefirst-periodsalescanvarysubstantiallyunderthequotaordeadline.Thefirmsells90vehiclesinthefirstperiodwhenfacingadeadline.WhenfacingaquotaΓ<90,itsellsonlyΓcarsinthefirstperiodtosecurethesubsidyinthesecondperiod.Thedifferenceintheprivatelyoptimalpricesandsalesacrossthetwodesignsdependsontheunderlyingparameters,suchaspricesensitivityBandnetworkeffectη.Panels(2)and(3)demon-stratethisbyvaryingBandη.InPanel(2),IreducethepricesensitivitytoB=10whilekeepingthenetworkeffectas0,asinPanel(1).TheeffectofthesubsidyonconsumerdemandislowercomparedtoPanel(1)duetothelowerpriceelasticityofdemand.Asaresult,thefirmfacingaquotastaysbelowitonlyifΓ>Γ∗.WhenΓ<Γ∗,thefirmbehavesasiffacingadeadline,andthetwodesignsproduceequivalentmarketoutcomes.InPanel(3),Iincreasethenetworkeffectηto0.2whilekeepingthepricesensitivityasinPanel(2).Becauseofthenetworkeffect,thefirmfacingadeadlinesellsmoreEVsinthefirstperiodthaninPanels(1)and(2).Thefirmwithaquotanowfacesanontrivialdilemma:exhaustingthequotashrinksthesecond-perioddemandduetoreducedsubsidy,butattractingadoptersinthefirstperiodincreasessecond-perioddemandduetothenetworkeffect.AlthoughΓ∗isthesameasinPanel(2),thedifferenceinsalesbetweenthedeadlineandquotaislarger.Severallessonsemergefromthissimpleanalysis.First,aper-manufacturerquotacanincentivizemanufacturerstosellfewerEVs.Inthemonopolyexample,thisincentivearisesbecausebystayingbelowthequotainagivenperiod,themanufacturercanensurethatallEVssoldinthefollowingperiodalsoqualify.Thisincentivewillbereinforcedinanoligopolybecausestayingbelowthequota10alsoprotectsthefirmfromcompetitionfromothermanufacturers.Second,theeffectofthesubsidy-cappingdesignsonmarketoutcomesdependsonparameterssuchasown-priceelasticityandthenetworkeffect.Inanoligopoly,marketoutcomes,suchasprofitdistribution,willalsodependonthecross-priceelasticities.Byrecoveringthekeyparameters,Icananswerhowthedesignsaffectmarketpenetration,gasolineconsumption,consumersurplus,andfirms’profits.4FullModelInowdescribethecompletemodelwiththeconsumerandmanufacturerdecisionproblemsintheUSautomobileindustry.IobservenewvehiclesalesinMgeographicmarkets(indexedbym=1,2,...,M)overTyears(indexedbyt=1,2,...,T).Eachyearhasafixedsetoffirmsthatproduceanexogenoussetofproducts.Thedemandspecificationfollowsthediscrete-choiceframeworkofBerry,Levinsohn,andPakes(1995),whereconsumerschooseasinglevehiclefromallavailablefueltypes.Includingallfueltypesallowsforsimulatingwhathappenstotheentiremarketofautomobilesunderthecounterfactualscenarios.Toeasecomputation,Iassumethatconsumersaremyopicinthattheydonotconsiderthefutureevolutionofpricesorinfrastructureandonlypurchaseifthevehicleservestheirpresentdrivingneeds.Incontrast,thesupplysideisanoligopolywithproductdifferentiationwheremanufacturerschoosepricesforallvehiclesintheirportfolio.Iusetheestimateddemandelasticitiestorecoverthemarginalcostsin2017andinvestigatewhatwouldhavehappenedifthesubsidyeliminationbeganin2017underdifferentsubsidy-cappingdesigns.Inpractice,thesubsidyeliminationbeganin2018.However,asdiscussedlater,Iavoidthisyearintheestimationtoensurethatthedemandelasticitiesarenotinfluencedbyintertemporalsubstitution.Althoughthe2017dataareimperfecttodirectlyinformtheeffectoftheactualsubsidy-cappingdesign,itallowsexaminingthedynamictrade-offshighlightedinSection3andpredictingtheeffectthatdifferentdesignswouldhavehadduring2017.Ielaborateonconsumerdemandandmanufacturers’decisionproblemsnext.4.1DemandEachperiod,consumersarriveatthemarket.Theproductsavailableinmarketminmodelyeartareindexedbyj∈Jmt.Consumeri’sindirectutilityfromchoosingvehiclejisafunctionofvehicleandindividualcharacteristics:Uijmt=−αipjmt+xjtβi+Njmtη+ξjmt+ϵijmt,(1)wherepjmtrepresentstheconsumerpriceandequalsthemanufacturer’ssuggestedretailprice(MSRP)minusallpurchaseincentives:pjmt=MSRPjt−RDjtRetaildiscount−λ0jtFederalsubsidy−λjmtlocalsubsidy.Inequation1,xjtisaK×1vectorofvehicleattributes,includingsize,performance,costof11Figure3:Deadlinevs.QuotainaMonopolyNotes:Thisfigureshowsthefirst-periodsalesasafunctionofquotaΓinthreedifferentsituations.Eachpanelfixesc=20andA=300butchangeseitherthepricecoefficientBorthenetworkeffectη.Panel(1):ThepricecoefficientB=12,andthenetworkeffectη=0.Whenfacingaquota,thefirmalwaysstaysbelowthequotatosecurethefuturesubsidy,whichmayresultinfewersalescomparedtoadeadline.Panel(2):Thenetworkeffectisasincase(1),butthepricecoefficientisloweredtoB=10.Thefirmreducesthefirst-periodsalesonlywhenthequotaishigherthanΓ∗.Panel(3):Thepricecoefficientremainsasincase(2),butthenetworkeffectisraisedtoη=0.2.Asincase(2),thefirmreducesthefirst-periodsalesonlywhenthequotaishigherthanΓ∗.However,thenetworkeffectleadstoalargerdifferenceinfirst-periodsalesbetweendeadlineandquotacomparedtocase(2).12driving,batteryrange,fueltypeindicators,and14vehiclesegmentindicatorsbasedonmarketorientations.αidenotesthemarginalutilityfromprice,assumedtofollowalog-normaldistributionwithparametersαandσα.Inotherwords,log(αi)=α+σαviα,whereviαfollowsastandardnormaldistribution.βiisaK×1vectoroftastecoefficients,assumedtofollowanormaldistributionwithparametersβandσβlforthelthdimensionofβi.NjmtindicatesavectorofthenetworkeffectvariablesandincludestheinteractionofBEVandPHEVindicatorswiththelogcumulativeEVsalesbythemanufacturerofvehiclejinmarketmuptoyeart−1,alongwiththeinteractionofBEVandPHEVindicatorswiththelogcumulativeEVsalesbyallmanufacturerswhoseEVsusethesametypeofLevel3chargerinmarketmuptoyeart−1.ξjmtrepresentstheaverage,orcommon,utilityfromtheattributesofvehiclejinmarketmandyeartthatisunobservabletotheresearcherbutknowntoconsumersandproducers;thesemayincludequality,promotionalactivity,orsystematicdemandshocks.Imodelξjmt=ξm+ξt+∆ξjmt.Econometrically,ξmiscapturedbymarket-specificdummiesthatcontrolfortime-invariantmarket-levelvariations,suchasthequalityofpublictransitorlocalinclinationstobegreen.ξtiscapturedbytimedummiesthatcontrolfornationalfactorsthatdonotvaryacrossmarkets,suchasnationalmacroeconomic,climate,andglobalfuelpriceshocks.∆ξjmtisleftasaneconometricerrorterm.Finally,ϵijmtrepresentsidiosyncratictastesassumedtofollowi.i.d.type-Iextremevaluedistribution.Thespecificationincorporatesthenetworkeffectinareduced-formfashionbyallowingconsumerutilityfromanEVtodependonthecumulativeEVsalesfromthesamemanufactureruntilthepreviousperiodandthecumulativeEVsalesfromallmanufacturerswithcompatibleLevel3chargersuntilthepreviousperiod.8TherichdatasetincludescarregistrationssincetherecentdevelopmentoftheUSplug-inEVmarket.ItallowsmetocalculatethecumulativeEVsalesineachgeographicmarketprecisely.TherationaleforincludingpreviousadoptionofEVsfromthesamemanufactureristhatadoptershelpspreadinformationaboutthatmanufacturer’sEVsamongthenewconsumerpool.WhencarbuyersobservemoreEVsproducedbyagivenmanufacturer,theymaybemorewillingtobuyitsEVs.Therefore,thesignsofthesenetwork-effectvariablesareexpectedtobepositive.TherationaleforincludingthepreviousadoptionofallEVsthatuseacompatibleLevel3chargeristhatitisassociatedwiththeavailablecharginginfrastructurenetworkvitaltoguaranteedrivers’mobility.However,previousadoptionofEVsproducedbycompetingmanufacturersalsoincreasesconsumers’exposuretocompetingEVs,whichmayinducethemtopurchasethosevehiclesinstead.Therefore,thesignsofthesenetwork-effectvariablesaretheoreticallyambiguous.Modelingthevehiclepurchasedecisionsasstaticisreasonableforbuyersofconventionalgasoline-poweredvehicles,asthesedonotevolvesubstantiallyovertime.However,EVbuyersmayalsotimetheirpurchasestotakeadvantageofbetterprices.Forinstance,iftheybelievethatsubsidieswillexpire,theymaybuyearlier.Toensurethatthedemandparametersreflectactualpurchasechoicesandnotanintertemporalsubstitution,Iestimatethedemandmodelusingdataunaffected8.AlthoughLevel1andLevel2chargingstandardsareuniformacrossallvehiclebrands,Level3chargingstationswereofferedthroughthreeincompatiblestandardsduringthesampleperiod.TeslauseditsownSuperchargernetwork.Nissan,Mitsubishi,Kia,andToyotausedtheJapanese-developedCHAdeMOstandard.FCA,GM,Ford,Volkswagen,andBMWusedtheSAEInternationalCombostandard.13bythesubsidychanges(2011–2017).ThestaticchoiceframeworkisagoodapproximationforEVpurchasingdecisionsduringtheseyearsbecausethefederalpriceincentiveswereputinplacein2009—wellbeforethestartoftheEVmarket—anddidnotphaseoutuntil2018.Asaresult,thesubsidy-inducedtimingeffectsareunlikelytoberelevant,andtheelasticitieswilllikelyreflecttheactualchangesinvehiclechoice.Althougheliminatingsubsidiesmayinducechangesinpurchasetiming,itisunlikelytoaffectthetrueparametersgoverningvehiclechoicebehavior.Asaresult,Icanapplythedemandparametersestimatedfromthe2011–2017datainthecounterfactualexperiments.EVbuyersmayalsocareaboutthetimingoftheirpurchasesiftheybelievethatthechargingnetworkorqualitywillimprove.Thestaticassumptionimposesthatcarpurchasebehaviorisgovernedbypresentdrivingneeds,whichisreasonablebecauseconsumersmaybelimitedinchangingtheirresidenceorworkplacelocationintheshortrun.Moreover,asdiscussedinSection5.2,theimprovementsinbatteryrangehavebeenslow,suggestingthatconsumerswouldhavetowaitforalongtimeforsignificantupgrades.Consumersmaketheirpurchasebymaximizingtheirutilitiesacrossallvehicleoptionswiththeoutsideoptionofpurchasingausedvehicleornothing.Astheyaremyopic,theoutsidegooddoesnotincludepurchasingthevehicleinthefuture.TheutilityfromchoosingtheoutsidechoiceisUi0mt=ξ0mt+ϵi0mt.Themeanutilityoftheoutsidegoodisnotidentified,soInormalizeξ0mt=0.ConsumerichoosesamodeljifandonlyifUijmt≥Uij′mt,∀j′̸=j.ThechoiceprobabilityisPrijmt=I(ϵUijmt≥Uij′mt∀j′)dF(ϵ)=exp(−αipjmt+xjtβi+Njmtη+ξjmt)1+j′∈Jmtexp(−αipj′mt+xj′tβi+Nj′mtη+ξj′mt).Correspondingly,theshareofvehiclejinthemarketmarketmandperiodtissjmt=exp(−αipjmt+xjtβi+Njmtη+ξjmt)1+j′∈Jmtexp(−αipj′mt+xj′tβi+Nj′mtη+ξj′mt)dF(αi,βi).LetHmtdenotethenumberofhouseholdsinmarketmandperiodt.ThedemandforvehiclejinmarketmandperiodtisQjmt=Hmtsjmt.4.2SupplyImodelthemarketasservedbyamultifirm,differentiated-productindustrywherefirmsengageinBertrandpricecompetition.Giventhemarketprimitives,eachfirmfchoosespricesforall14vehiclesinitsportfoliotomaximizethesumofprofitsinthecurrentandfollowingyearacrossallitsproductsinallthegeographicmarkets,assumingthatproductofferings,marginalcosts,state-levelsubsidies,anddemandshocksstayasinperiodt.Vehiclecharacteristicsotherthanthepriceevolveexogenously,whichisreasonablebecausemanufacturerstypicallymakeproductdecisionsoveralongerhorizonthanpricingdecisions.TheprofitforfirmfisΠft=mj∈Jft[(pjt−cjt+hjmt)Qjmt(p.t)+(pj,t+1−cjt+hjmt)Qjm,t+1(p.t,p.t+1)].(2)ThepricepjtisuniformacrossallmarketsandequalstheMSRPminusretaildiscounts:pjt=MSRPjt−RDjtRetaildiscount.MSRPsandretaildiscountsareconstantacrossmarketswithinamodelyear.Idonotobserveandthereforedonotconsidermarket-specificdiscounts.cjtisthemarginalcost,andhjmtisthevalueofZEVcreditsformodeljinmarketm.Fornon-ZEVstates,hjmttakesthevalue0.ForZEVstates,hjmtistheproductofthevalueofthecreditformodeljandthepriceofZEVcredit.In2017,batteryEVs,plug-inhybrids,andhydrogenfuel-cellvehiclesearnedcreditsdependingontheirbatterychargetimeandrange.Vehicleswitharangeoffewerthan50milesearnedonecredit;thosewitharangeofmorethan300milesandarechargetimeoffewerthan15minutesearnedninecredits.Inaddition,conventionalhybrids,suchastheHondaCivicandToyotaPrius(AT-PZEV),earnedupto0.8ofacredit,andgasolinevehicleswithloweremissions(PZEV)thanfederalstandardsearnedupto0.2ofacredit.AlthoughtheZEVcreditmarketdoesnothavepricetransparency,literaturehasbackedoutpricesfromtherevenuesreportedbythemanufacturers.FollowingMcConnellandLeard(2021),IassumethatthevalueofZEVcreditsin2017wasUSD2,218.Atwo-periodpricingmodelisvitaltoexaminehowfirmsrespondwhenthedynamicsofsubsidyeliminationbecomerelevant.Forinstance,firmsfacingaper-manufacturerquotamayincreaseEVpricestodelayexhaustingit.Astaticmodelwillfailtocapturesuchadjustments.Thetwo-periodmodelisalsocrucialforthefirmstointernalizethenetworkeffectandreactstrategicallytothesubsidyeliminationbasedontheirexpectationsofhowcurrentpricesaffectfutureEVdemand.Forinstance,themodelallowsthefirmsfacingaper-manufacturerquotatokeeppriceslowandsurpassthequotaiftheybelievethatraisingpriceswoulddiminishtheirfutureprofits.Althoughthetwo-periodassumptionisguidedbycomputationalsimplicity,itisnotrestrictive,becauseiffirmscareaboutalongerhorizon,thatissimilartosolvingthesameproblemwithalowerdiscountfactor.Thepriceschoseninperiodtaffecttheprofitsinperiodt+1byinfluencingonlyasetofcommonlyobservedstatevariables:thesubsidyvalueandnetworkeffect.Giventhesestatevariables,allfirmssimultaneouslychoosepricesforalltheproductsinperiodt+1.Iderivetheoptimalityconditionsusingbackwardinduction.Givenp.t,theoptimalpricevectorp∗.t+1(p.t)inperiodt+1isthesolutiontothesystemofJfirst-orderconditions:15mQjm,t+1+k∈Jft(pk,t+1−ckt+hkmt)∂Qkm,t+1∂pj,t+1=0.Theoptimalpricevectorp∗.t+1(p.t)canbeusedtosimulatetheoptimalprofitvectorΠ∗f,t+1(p.t)asafunctionofp.t.Inperiodt,thevectorofpricesp.tmaximizesΠft=mj∈Jft[(pjt−cjt+hjmt)Qjmt(p.t)]+Π∗f,t+1(p.t).In2017,allEVmanufacturerswerefarbehindthequota.Asaresult,theirprofitmaximizationproblemsin2017arenotconstrainedbyit,ensuringthatΠ∗f,t+t(p.t)isdifferentiableinp.t.Therefore,thenecessaryoptimalityconditioninperiodtwithrespecttothepriceofproductjismQjmt+k∈Jft(pkt−ckt+hkmt)∂Qkmt∂pjt+∂Π∗f,t+1(p.t)∂pjt=0.(3)Thesefirst-orderconditionsinvolveownandcross-pricederivativesofthedemandforeachproduct,calculatedastheweightedsumsofindividualderivatives:∂Qkmt∂pjt=Hmt∂skmt∂pjt=−Hmtsijmt(1−sijmt)αidF(αi,βi),ifj=kHmtsijmtsikmtαidF(αi,βi),otherwise.(4)Thederivatives∂Π∗f,t+1(p.t)∂pjtcanbecomputednumerically.Iusethesefirst-orderconditionstorecoverthemarginalcostsforallproductsin2017.Themodelhassomecaveats.First,firmscontrolsalesonlythroughshort-runpricechanges.Inpractice,firmscanusemorewaystoreduceEVsales.Forexample,givenaper-manufacturerquota,theycanlowertheproductionofsubsidy-qualifyingvehicles;thiswouldcreateanartificialshortageandkeepthesalesbelowthequota.Intheabsenceofdataonvehicleinventories,Idonotmodelthismechanism.Suchsimplificationaffectshowfirmsrespondtothecounterfactualsubsidy-cappingdesigns,whichIdiscussfurtherinSection7.Second,throughouttheanalysis,Iabstractfromentryandexitdecisions.Inpractice,thedesignsmayalsoaffectfirms’entryintotheEVmarket.Althoughthisconcernwasimportantwhenthesubsidywasenactedin2009,itislessrelevanttodaybecausemostmajorvehiclemanufacturersalreadyhavesomeEVsintheirportfolio.Finally,otheroverlappingregulationsimposedonvehiclemanufacturers,suchasfederalcorpo-rateaveragefueleconomy(CAFE)andgreenhousegas(GHG)standards,alsoincentivizemanufac-turerstoincreaseEVsales.CAFEandGHGstandardsinfluencethemarketonthesupplysidebyimposinglimitsontheaveragefueleconomyandGHGemissionsofthevehiclesthatamanufac-turersellseachyear.BothregulationsgrantextracreditstoEVs,incentivizingmanufacturerstosellmoreEVs.Forsimplicity,Idonotaccountfortheseregulationsinthemanufacturers’profit16function.Oneconcernisthattheseincentivesmayinteractwiththedynamicincentivescreatedbyaper-manufacturerquota.Forinstance,amanufacturernearingitsquotamaywanttosellmoreEVstooffsetCAFEliabilities,evenifitmeansexceedingthequotaandforgoingfuturesubsidies.Suchconcernisirrelevanttotheestimationbecauseitreliesontheyearsbeforesubsidyelimination.ItisalsounlikelytoberestrictiveforcounterfactualanalysisbecausebothCAFEandGHGprogramsallowadditionalflexibilities,suchasbankingcreditsfromovercompliancetouseforcomplianceinanothermodelyear.Asaresult,amanufacturerfacingaquotaonEVsubsidiescanusesuchflexi-bilitiestomeettheirCAFEandGHGrequirements.Thus,ignoringtheseregulationsstillprovidesagoodapproximationofthemarketoutcomesunderdifferentsubsidy-cappingdesigns.5Data5.1DataSourcesThedataforthispapercomefromvarioussources.Thevehiclesalesdatawerepurchasedfromamarketresearchcompanyandcontainnewlight-dutyvehicleregistrationsin30statesduringthecalendaryears2011–2017.TheselectedstatescapturedthehighestEVmarketsharein2016.Iusethesestatestodefineageographicmarket.AstheEVmarketprimarilydevelopedafter2010,thedatacaptureitfromtheoutset.Avehicleisauniquemodelyear,make,model,andfueltype.Iuseregistrationsforallfueltypes(exceptfuelcell)toaccountforsubstitutionbetweenfueltypes.9Iexcludeexoticvehicles,suchasFerrariandLamborghini.Foreachmarket,IestimateitssizeusingtheUSCensusBureau’sstate-levelannualestimatesofhousingunitsandcalculatethemarketsharesbydividingthestate-levelsalesvolumebythenumberofhouseholdsinthatyear.Themarketshareoftheoutsidegoodisthedifferencebetween1andthesumofinsidegoodsmarketshares.Thedistinctionbetweenacalendaryearandamodelyearpresentsatechnicalissueindefiningthechoicesets.Amodelyearisamanufacturer’sannualproductionperiod,includingJanuary1ofthecalendaryear.IttypicallyrunsfromOctobertoSeptemberofthenextyear(e.g.,2016modelswerereleasedaroundOctober2015.)Idefinethechoicesetsbasedonmodelyear,thusassumingthatallvehiclesreleasedinanymodelyearsellinthesamemodelyearandthatmodelyearsperfectlyalignforeachmanufacturer.Idonotobservethe2011modelssoldinthe2010calendaryear,soIuse2011datatocalculatecumulativeEVsalesineachmarketbutnotthedemandestimation.Thesamplecomprises62,186observationsofvehiclesharesover30statesduring2012–2017.Iobtainvehicle-levelcharacteristicsfromtheWARDSIntelligenceDataCenterandEnviron-mentalProtectionAgencyandfillinmissingvaluesbasedontheinformationfromEdmunds.Ad-ditionally,IobtainmarketsegmentationdatafromAutomotiveNews.AlthoughIobservevehiclecharacteristicsatthetrimlevel,registrationsareatthemake-model-fuellevel.Hence,Iaveragethecharacteristicsacrossdifferenttrimstomatchtheregistrations.VehiclecharacteristicsincludeMSRP,size-relatedmeasures(wheelbaseandwidth),horsepower,curbweight,fueltype,fueleffi-9.Idropfuel-cellvehiclesfromthesample,asIdonotobservestate-levelhydrogenprices.Theseobservationscompriselessthan0.01percentofthetotalUSsalesduring2011–2017.17ciency,andbatteryrange(forEVs).Thedemandmodelallowsconsumerutilitytodependonthesize,performance,costofdriving,andbatteryrange.Imeasuresizebytheproductofwheelbaseandwidthandperformancebytheratioofhorsepowerandcurbweight.Forgasolineanddieselvehicles,Icalculatethecostofdrivingasthestate-levelfuelpricepergallondividedbythevehicle’sfueleconomy;thefuelpricescomefromtheUSEnergyInformationAdministration(EIA)andareexpressedindollarspermillionBritishthermalunits(Btu).IconverttheseintodollarspergallonusingtheBtucontentofeachfuel.10ForBEVs,Icalculatethecostofdrivingbydividingtheelec-tricitypriceperkW-hrsbythevehicleelectricityconsumptioninkW-hrs/mile.11Finally,IcalculatethecostofdrivingforPHEVsbyassumingthattheyrun50percentofthetimeonelectricityand50percentongasordiesel.Itvarieswithtwosources:fueleconomyandmarket-levelfuelprices.Thus,ahighgaspriceinastateraisesthecostofdrivingallgasolinevehiclesinthatstate.Althoughaveragetransactionpricesarepreferredindemandestimation,suchdataarenotreadilyavailable.Instead,IcombineMSRPdatawiththemanufacturers’retaildiscountsfromAutomotiveNewsandfederalandstate-levelsubsidiesfromtheUSDepartmentofEnergytoapproximatethepricesthatconsumersandfirmsface.Federalandstate-levelsubsidiesvaryacrossmodelsandtime.Givenpurchasesubsidies,thepricethatentersthefirms’profitfunctiondiffersfromtheconsumerprice—itisthedifferencebetweenMSRPandtheaveragediscountsprovidedbythemanufacturerinthatyear.TheconsumerpriceisthedifferencebetweenMSRPandallpurchaseincentives,includingmanufacturerdiscountsandfederalandstatesubsidies.Federalsubsidiesarenonrefundabletaxcredits,soinpractice,theamountreceivedbyaconsumerdependsontheirincome-taxliability.Forsimplicity,Iassumethatallconsumerscanclaimthefulltaxcredit.Thejustificationisthatthenewvehiclemarketistypicallyusedbywealthyhouseholdswithhighincome-taxliabilities.IdeflateallvehicleandfuelpricesusingtheBureauofLaborStatisticsConsumerPriceIndexandadjustthemto2015$.Finally,IobtainthelistofmakesproducedbyeachmanufacturerfromtheannualEPAAutotrendreports.Consistentwiththeregulatorydefinitions,Iassumethatdifferentmakesofthesameparentmanufacturerbelongtoasinglefirm.Forexample,Buick,Cadillac,Chevrolet,andGMCareallpartofGM.5.2SummaryStatisticsTable1summarizesthesalesandsales-weightedaveragecharacteristicsofvehiclesinthesample.Thefirstcolumnreportsthemodelyear,andthesubsequentcolumnsshowthetotalmodels,therealprice(inthousandsofdollars),totalsales(inthousandsofdollars),size(inthousandsofin2),performance(inHPper10lb),costofdriving(indollarspermile),andbatteryrange(inmiles)separatelyforplug-inEVsandotherfuel-typevehicles.TheavailableEVmodelsrosefrom9in10.Seehttps://www.eia.gov/energyexplained/units-and-calculators/fordetailsonthiscalculation.11.ForBEVs,EPAreportsfueleconomyusinganMPGemetric,calculatedas33.705kWhrs/gallondividedbythevehicleelectricityconsumptioninkW-hrs/mile.Thismeasuresthemilesthevehiclecantravelonanamountofenergyequaltothatstoredinagallonofgasoline.Iusethesevaluestocalculatethevehicle’selectricityconsumption(kW-hrs/mile)as33.705kWhrs/gallondividedbytheEPA-reportedMPGevalues.182012to34in2017;theirtotalsalesinthesamplestatesrosefrom43,000in2012to207,000in2017.Modelsofotherfueltypesrosefrom315in2012to348in2017;theirtotalsalesinthesamplestatesrosefrom9.4millionin2012to12.1millionin2017.TheaveragesizeofEVsremainedfairlystableacrossbothtypes.TheaverageperformanceforEVsincreasedfrom0.32Hp/10lbsin2012to0.42Hp/10lbsin2017,butthatofotherfueltypesremainedstableataround0.61Hp/10lbs.TheaveragecostofdrivingforEVsremainedstableataround$0.05permile,butthatofotherfueltypesreducedfrom$0.16in2011to$0.10in2017.TheaveragecostofdrivingismuchlowerforEVsduetohighfueleconomyandlowelectricityprices.Finally,theaveragebatteryrangeforEVsrosefrom53.98milesin2012to115.43milesin2017.Table1:NewVehicleSalesandCharacteristicsintheSampleStatesYearModelsMSRPSalesSizePerformanceDrivingCostBatteryRange($’000)(’000)(’0000in2)(Hp/10lb)($/mile)(miles)EVOtherEVOtherEVOtherEVOtherEVOtherEVOtherEVOther2012931543.1130.47439,4070.750.810.320.610.040.1653.980.0020131033643.8131.169510,7390.760.820.380.610.050.1574.760.0020142034945.6331.889810,9750.750.830.440.610.050.1478.940.0020152135748.9632.7811512,0870.780.830.420.610.040.1099.280.0020162734360.2632.8512411,6850.840.830.520.610.050.09123.240.0020173434848.0533.4620712,1960.800.830.420.610.040.10115.430.00Notes:Thistableshowstheevolutionofkeyvariablesinthesamplestatesbetween2012and2017.Columns(6)–(9)showthesales-weightedaveragevehiclecharacteristics.Sizeiswheelbase×width(inthousandsofin2),performanceishorsepowerbycurbweight(inHp/10lb),drivingcostisfuelcost(indollarspermile),andbatteryrangeistheall-electricrange(in10miles)forelectricvehicles.Figure4showsthattheannualsalesandshareofbothBEVsandPHEVswentupinthesamplestatesbetween2012and2017;theyrepresented1.6percentofdomesticautomobilesalesin2017.19Figure4:SalesandShareofNewLight-DutyPlug-InEVsintheSampleStatesNotes:ThefigureshowstheevolutionoftheUSplug-inelectricvehicle(EV)industrybetween2012and2017.Thehorizontalaxisshowsthemodelyears.Theleftandtherightaxesshowthetotalplug-inEVsalesandtheshareofplug-inEVsintotalnew-vehiclesalesinthesamplestates,respectively.Figure5showsallEVmanufacturerswiththeiryearofentryonthehorizontalaxisandthetotalvehiclessoldinthesamplestatesbetween2011and2017ontheverticalaxis.TeslaandGMsoldaround150,000EVs;Nissansoldaround110,000.Table2summarizestheplug-inmodelsandthenominalvalueoffederalsubsidiesforeachmanufacturerin2017.Thesubsidyrangedfrom$3,793forPHEVssuchasBMWI8to$7,500forBEVssuchasTeslaModel3.Itremainedunchangedforallmodelsduringthesampleperiod.20Figure5:MajorPlayersinthePlug-InEVIndustryNotes:ThisfigureshowsthemajorUSplug-inelectricvehicle(EV)manufacturers.Foreachmanufacturer,thex-coordinateshowstheyearinwhichitsfirstplug-inEVsaleappearsinthesample.They-coordinateshowsthetotalplug-insalesbetween2011and2017.Table2:FederalSubsidiesforPlug-InEVsManufacturerPlug-inEVModelsSubsidyRange(USD)BmwBMW330,BMW740,BMWI3,BMWI8,BMWX53,793–7,500DaimlerMercedes-BenzB-Class,Mercedes-BenzGLE,SmartFortwo4,460–7,500FiatChryslerChryslerPacifica,Fiat5007,500FordFordC-Max,FordFocus,FordFusion4,007–7,500GeneralMotorsCadillacCT6,ChevroletBolt,ChevroletVolt7,500HyundaiHyundaiIoniq,HyundaiSonata4,919–7,500KiaKiaOptima,KiaSoulEV4,919–7,500MitsubishiMitsubishii-MiEV7,500NissanNissanLeaf7,500TeslaTeslaModel3,TeslaModelS,TeslaModelX7,500ToyotaToyotaPriusPrime4,502VolkswagenAudiA3,PorscheCayenne,VolkswagenGolf4,502–7,500VolvoVolvoXC904,585Notes:Thistablesummarizestheplug-inelectricvehiclemodelsandfederalconsumersubsidiesin2017foreachmanufacturer.Table3summarizestheplug-inEVsalesandregulationsinall30statesin2017.Columns(1)21and(2)listthetotalsalesineachstateandtheirshareasapercentageofoverallnew-vehiclesales.SalesarehighestinCalifornia(3.67percent)andlowestinOklahoma(0.15percent).Columns(3)and(4)showpurchaseincentivesandZEVmandates.Sixstatesprovidesomesubsidiesforplug-inEVs,andeighthavetheZEVrequirement.StateswithsubsidiesorZEVmandateshavehigherEVsales.Table3:State-LevelEVSalesandIncentivesMarketPlug-InPercentofPlug-InZEVSalesTotalSalesIncentivesStateArizona10,3740.68-California350,5633.67YesYesColorado12,5271.10Yes-Connecticut7,6070.80YesYesFlorida25,9830.43-Georgia27,4481.29-Hawaii7,1582.30-Illinois14,5530.49-Indiana3,7600.30-Maryland11,5750.76YesYesMassachusetts14,0600.77YesYesMichigan14,0640.46-Minnesota4,6910.39-Missouri3,9150.33-Nevada3,7660.57-NewHampshire2,3540.50-NewJersey16,6940.57YesNewYork30,2380.57YesYesNorthCarolina8,0410.37-Ohio8,1660.27-Oklahoma1,1890.15-Oregon14,8692.00YesPennsylvania11,8830.35-Tennessee4,2660.34-Texas21,6190.30-Utah4,3680.87-Vermont2,5921.17YesVirginia9,8490.51-Washington27,8042.05-Wisconsin6,2050.48-Notes:Columns(1)and(2)showthetotalplug-inelectricvehicle(EV)salesandtheirshareasapercentageofoverallnewcarsalesduring2011–2017basedondataforthe30statesinthesample.Col-umn(3)and(4)showstheavailabilityofstate-levelplug-inEVin-centivesandZEVrequirement.226EstimationandResultsThenextstepistoestimatethestructuralmodeldescribedinSection4.Iestimatethedemandsys-temandrecovermarginalcost,assumingthatthedataaregeneratedbyNash-Bertrandequilibriumbehavior.Thebenefitofsequentialestimationisthatdemandestimationdoesnotrelyonsupply-sideconduct.Section6.1describestheestimationandidentificationofthedemandparameters,andSection6.2reportstheresultsfromestimatingthestructuralmodel.6.1EstimationandIdentificationThefundamentalissuethatmotivatesdemandestimationisthepriceendogeneityarisingfromtwosources.First,themodelimpliesthatpriceandquantityaredeterminedinequilibrium,sothepricepartlydependsontheunobservableproductcharacteristics∆ξjmt.Forinstance,vehiclecharacteristics,suchascomfort,ridesmoothness,andexpectedresalevalue,cannotbemeasureddirectly.However,thepricewilllikelyreflectthemiftheyarecostlyforthemanufactureroraffectdemand.Similarly,advertisementeffortsareunobservedbutmaybecorrelatedwiththepricingdiscounts.Second,IdonotobservetheaveragevehicletransactionpriceandinsteadapproximateitusingMSRPminuspurchaseincentives.Asaresult,variationsinthevehicletransactionpriceacrossmarketsenter∆ξjmtinequation1.Bothcasesresultinpriceendogeneity.Identificationrequiresasetofexogenousinstruments.Vehiclecharacteristicsotherthanpricearevalidinstrumentsforthemselves,astheyarepartofanexogenousdevelopmentprocess.Appropriateinstrumentsforpriceincludeanyfactorsthatarecorrelatedwiththepricebutnotwith∆ξjmt.IfollowBerry,Levinsohn,andPakes(1995)andusethesumoverthecharacteristicsoffirm’sothervehiclesandthesumoverthecharacteristicsofallthecompetingvehiclesasinstrumentsforprice.Specifically,foreachvehiclecharacteristick(constant,size,performance,drivingcost,andbatteryrange),Iincludethefollowingtermsasinstrumentsforprice:zkjmt=(xkjt,r̸=j,r∈Jfmtxkrt,r̸=j,r/∈Jfmtxkrt).(5)Overall,thereare10excludedinstruments.Theseinstrumentsvaryovervehiclemodelsineachmarketandacrosstime.Theyarerelevantbecausetheyproxyforthedegreeandclosenessofcompetitionthatabrandfaces,thusaffectingthefirm’smarkups.Therationaleforseparatelyincludingfirms’ownvehiclesandotherfirms’vehiclesisthatwhenafirmpricesitsvehicles,itwouldtreatthesubstitutionwithitsownandotherfirms’vehiclesdifferently,asconsumerswhowillswitchawaytoanotherofitsvehiclesfollowingapriceincreasedonotrepresentasmuchofaloss.Theidentifyingassumptionisthatthedemandunobservables∆ξjmtaremeanindependentoftheobservedcharacteristics.Theunderlyingtimingassumptionisthatcarmanufacturersdonotobserve∆ξjmtwhenchoosingvehiclecharacteristics.Theidentificationissuesassociatedwithincludingcumulativesalesasaproductcharacteristicaresimilartothoseinvolvedinusingalaggeddependentvariableasaregressor.Specifically,consistent23estimationofthenetwork-effectparametersrequiresthatdemandunobservablesarenotseriallycorrelated.Iassumethat,conditionalonthemarketandyearfixed-effects,thisisthecase.FollowingBerry,Levinsohn,andPakes(1995)andNevo(2000),Iestimatethedemandparam-etersusingthesimulatedGeneralizedMethodofMomentsusingthepopulationmomentconditionthatisaproductofthedescribedinstrumentalvariablesandunobservabledemandshocks∆ξjmt.6.2ResultsTable4showstheresultsfromtheestimatingdemandderivedfromtheindirectutilityspecificationinEquation1.12Thefirsteightrowsshowthecoefficientsmeasuringthemeanparametersαandβ.Mostcoefficientsarepreciselyestimatedandhaveexpectedsigns.Vehiclesizeandthehorsepower-to-weightratiohavepositivecoefficients,indicatingthatconsumersvaluesizeandperformance.Thenegativecoefficientonthecostofdrivingpermileimpliesthatconsumerspreferhighfuelefficiency,whichreducesthecostpermile.ThesignontheBEVindicatorisnegative,indicatingthatintheabsenceofanetwork(i.e.,zerocumulativeEVsales)andceterisparibus,BEVsarelesspreferredtoconventionalmodels,possiblybecausetheydonothaveagasolinebackup.Moreover,thecoefficientonthePHEVindicatorispositive,indicatingthatconsumersvaluevehicleswithagasolinebackup.ThenexttworowsshowtheinteractionsbetweenEVindicatorsandcumulativeEVsalesinthestatebymanufacturer.Thesetermshavepositivesigns,indicatingthatconsumersgainmoreutilityfromBEVsandPHEVsasthemanufacturer’snetworkdevelops.ThenexttworowsshowtheinteractionsbetweenEVindicatorsandcumulativesalesofallEVsthatuseacompatibleLevel3chargerinthestate.Bothcoefficientsarenegative,possiblybecausethevehiclesalesfromcompetingmanufacturersinduceconsumerstobuythosevehiclesinstead,thusshiftingthevehicle’sdemandleftward.Thesubsequentrowsshowtheestimatesofsixrandomcoefficientsthatmeasurethedispersioninhouseholds’tastes.Table5presentsasampleofownandcross-priceelasticities,markups(i.e.,pjt−cjt),andmarginalcostsimpliedbythedemandestimatesandthefirms’first-orderconditions.Eachrowcorrespondstoadifferentvehicle;thefirstfourrowsarethetopfourplug-inEVcarsbysalein2016,thefollowingthreerowsarethetopconventionalcars,andthebottomthreerowsarethetoppickuptrucks.Eachentrygivesthepercentagechangeindemandoftherowvehicleassociatedwitha1percentincreaseinthepriceofthecolumnvehicle.Priceelasticitiesdifferacrossmarketsforeachproduct,butmarginalcostsareidentical.Ratherthanpresentingelasticitiesforaparticularmarket,Ipresenttheaverageacrossallmarketsin2017.Thecross-priceelasticitiesarelargeramongsimilarproducts.Forinstance,anincreaseinthepriceoftheChevroletSilverado(pickup)shiftsconsumersdisproportionatelytotheFordF-150(pickup)comparedtotheChevroletVolt(car).Moreover,thelesselasticthedemandistothevehicle’sownprice,thelargertheratioofthemarkuptoprice.12.AppendixBprovidesthedetailsofthefirst-stageestimation.Thefirst-stageF-statisticfortheexcludedinstru-mentsis494.99.24Table4:DemandEstimatesVariableCoefSEMeanParametersαandβPrice(0000USD)0.2820.235Constant-12.663∗∗∗0.728Size(0000in2)6.580∗∗∗0.754Performance(Hp/10lb)5.976∗∗∗0.528FuelCost($/mile)-33.734∗∗∗8.090BatteryRange(10miles)0.0090.022BEV-1.826∗∗∗0.665PHEV2.823∗∗∗0.744NetworkEffectsBEV×log(1+ManufacturerEVSales)0.500∗∗∗0.062PHEV×log(1+ManufacturerEVSales)0.529∗∗∗0.079BEV×log(1+Same-ChargerEVSales)-0.339∗∗∗0.067PHEV×log(1+Same-ChargerEVSales)-0.580∗∗∗0.117Std.Dev.ParametersσαandσβPrice(0000USD)0.414∗∗∗0.141FuelCost($/mile)4.56011.550Car1.6081.420Van0.01326.081Pickup0.15915.037SUV7.523∗∗∗1.057FixedEffectsStateFEYesTimeFEYesSegmentFEYesObs62186Notes:Thistableshowstheestimatesfromtheflexiblelogitmodel.Aunitofobservationisavehicle-state-year.Sizeiswheelbase×width(inthousandsofin2),performanceishorsepowerbycurbweight(inHp/10lb),drivingcostisfuelcost(indollarspermile),andbatteryrangeistheall-electricrange(in10miles)forelectricvehicles(EVs).Thevariable“ManufacturerEVSales”showscumulativeEVssoldbythevehicle’smanufacturerinthegeographicmarketuntilthepreviousyear.Thevariable“Same-ChargerEVSales”showscumulativeEVsalesbyallmanufacturerswiththesameLevel3chargingstandardinthegeographicmarketuntilpreviousyear.25Table5:ASampleofOwnandCross-PriceElasticities(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)PriceMarkupMarginal(USD)(USD)Cost(USD)(1)FordFusion(PHEV)-2.9510.0030.0060.0030.0330.0280.0260.0130.0160.01134,7979,44126,155(2)ChevroletVolt(PHEV)0.001-2.6930.0050.0030.0330.0290.0270.0120.0150.01034,3428,19927,303(3)TeslaModelS(BEV)0.0010.003-5.3210.0020.0290.0220.0220.0190.0230.01485,13310,68275,118(4)ToyotaPriusPrime(PHEV)0.0010.0030.005-2.4620.0320.0290.0260.0110.0130.01028,2048,84220,539(5)HondaAccord(Gas)0.0010.0030.0060.003-2.7520.0300.0280.0140.0160.01227,9529,82518,127(6)HondaCivic(Gas)0.0010.0030.0050.0030.033-2.3590.0280.0120.0140.01022,4659,18413,282(7)ToyotaCamry(Gas)0.0010.0030.0050.0030.0350.030-2.6020.0130.0160.01125,7259,75315,973(8)FordF(Gas)0.0000.0010.0030.0010.0100.0080.008-3.1110.0280.01935,78611,41524,372(9)ChevroletSilverado(Gas)0.0000.0010.0030.0010.0100.0080.0080.024-3.2710.02038,65311,80826,846(10)ToyotaTacoma(Gas)0.0000.0010.0030.0010.0100.0080.0080.0220.026-2.87131,70410,78620,918Notes:Columns(1)–(10)reportaveragecross-priceelasticitiesfor10vehiclesacrossallsamplestatesin2017,calculatedfromthedemandestimatesinTable4.Eachentry(i,j),whereiistherowandjisthecolumn,referstotheaveragepercentagechangeindemandformodeljwhenthepriceofmodelichangesby1percentinthemarketswherebothproductsareavailable.Columns(11),(12),and(13)reporttheprices,markupsandmarginalcosts,respectively.26Table6summarizestheestimatedelasticities,markups,andmarginalcostsforall382modelsobservedin2017.Panel(a)summarizestheseforallvehicles.Theaverageown-priceelasticityis-3.97,whichiswithintherangeoftheestimatesintheliterature(Berry,Levinsohn,andPakes1995;Li2018).13Themarginalcostsrangefrom$19,372atthe25thpercentileto$42,246atthe75thpercentile.Panels(b)and(c)breakthesampleintoplug-inEVsandconventionalvehicles.Theaveragepricesofplug-inEVsarehigher,buttheestimatedaverageelasticitiesandmarkupsaresimilar.Table6:MarginalCostEstimatesVariableMean25%Median75%StdDevObsPanel(a):AllvehiclesPricebeforesubsidy(USD)45,98428,20438,06154,086270,679382Own-priceelasticity-3.97-4.66-3.64-2.941.46382Markup(USD)10,5849,00410,47011,7982,521382Marginalcost(USD)35,55219,37228,13342,24625,078382Panel(b):Plug-inEVsPricebeforesubsidy(USD)51,99031,52738,88673,496294,07434Own-priceelasticity-3.86-5.32-3.08-2.541.7634Markup(USD)9,0617,0888,94010,8383,73234Marginalcost(USD)44,62426,50634,07060,81526,66434Panel(c):ConventionalvehiclesPricebeforesubsidy(USD)45,39727,47337,80053,307268,019348Own-priceelasticity-3.98-4.64-3.71-3.051.43348Markup(USD)10,7329,18910,55411,8052,325348Marginalcost(USD)34,66618,64627,71541,69524,781348Notes:Thistablesummarizesthepriceelasticities,markups,andvehiclemarginalcostscal-culatedfromthedemandestimatesinTable4andthefirst-orderconditionsoffirms’profitmaximization.7CounterfactualExperimentsThenextstepistocomparemarketoutcomesunderthedifferentsubsidy-cappingdesigns.Iexaminethreedesigns:(1)amarket-widedeadlinewhereallmanufacturersfacethesamedeadline,(2)aper-manufacturerdeadlinewhereeachmanufacturerfacesaseparatedeadline,and(3)aper-manufacturerquotawhereeachmanufacturerfacesaseparatequotaonthenumberofsubsidy-eligiblevehicles.Inthelastdesign,asintheactualUSdesign,ifthemanufacturersellsfewerEVsthanitsquotainanyperiod,thenallEVsitsellsinthenextperiodalsoqualifyforthesubsidy.Ineachcase,IusetheparameterestimatesfromSection6torecomputethepricingequilibriaunderthetwo-stagegamefromSection7.2andcalculatethemarketoutcomesofinterest,assumingthatproductofferings,marginalcosts,state-levelsubsidies,anddemandshocksstayatthe201713.Berry,Levinsohn,andPakes1995estimateelasticitiesfortheconventionalvehiclesintherange−3to−6.Li2018findsaveragepriceelasticityof-2.7forEVs.27levels.7.1CounterfactualSubsidy-CappingDesigns1.Marketwidedeadline:Thisdesigninstitutesasingledeadlineforallfirms.ConsumerswhopurchaseEVsuptotheendof2017qualifyforthesubsidy:λ(1)jt=λ0j1(t≤2017),whereλ0jistheinitialfederalsubsidyforvehiclejasobservedinthedata.2.Per-manufacturerdeadline:ThisdesigninstitutesadeadlineforTeslaandGM.Consumerswhopurchaseuptotheendof2017qualify.Consumerswhobuyotherbrandsqualifyinboth2017and2018.Idiscusstherationaleforthisdesignbelow.Thesubsidyevolvesasfollows:λ(2)jt=λ0j1(t≤2017)iff∈{Tesla,GM}λ0jotherwise.3.Per-manufacturerquota:Thisdesigngiveseachmanufactureraquotaκ.Allconsumersqualifyin2017.Consumerswhopurchaseavehiclein2018qualifyifthemanufacturersellsfewerthanκsubsidy-eligiblevehiclesbetween2011and2017.Thesubsidyevolvesasfollows:λ(3)jt=λ0jt−1τ=2011j∈Jfτ∩JEVQ(3)jτ≤κ,whereJEVisthesetofallEVsandt−1τ=2011j∈Jfτ∩JEVQ(3)jτisthemanufacturer’snationwidecumulativeEVsalesbetween2011andt−1.BecauseIonlyobserveannualvehiclesales,Iallowcounterfactualdesignstoaffecttheper-vehiclesubsidiesyearly.Inpractice,theUSphaseoutdesignaffectedthemquarterly.Theper-manufacturerdeadlineandquotaareinspiredbytheUSdesign,whichcombinesbothofthese.ThechoiceofTeslaandGMforaper-manufacturerdeadlineisbecausethesemanufacturershadthehighestcumulativeEVsalesupto2017,whichallowsforconvenientcomparisonwiththequota—adesignthatonlyaffectsTeslaandGMinthesimulations.Inpractice,policymakerscouldbasetheper-manufacturerdeadlinesonyearofentry,whichcanhelpraisemarketpenetrationbyreducingthebarrierstoentryandensuringmultiplesuppliersinthenewindustry.Myanalysisfocusesontheshort-termdynamicimplicationsofthesubsidy-cappingdesignsonthemarketoutcomes.Inthelongterm,thesedesignscouldaffectmarketpenetrationdifferentlyiftheyaffectmanufacturers’entry.IdonotmodelentryintotheEVmarketbecausemostmajormanufacturersalreadyhadEVsintheirportfoliobythetimethesubsidybegantophaseout.Table7summarizesthefeaturesofeachdesign.Underamarketwidedeadline,thesubsidyiselim-inatedforallmanufacturerssimultaneously;noincentivetoreduceEVsalesexistsin2017because28firms’actionsdonotaffectwhethertheyqualifyforasubsidyin2018.Underaper-manufacturerdeadline,thesubsidyiseliminatedformanufacturersaccordingtotheirseparatedeadlines.Asbe-fore,noincentiveexiststoreduceEVsales.Finally,underthequota,thesubsidyiseliminatedformanufacturersbasedonwhentheyexhaustit,creatinganincentivetoreducesalesin2017.Table7:FeaturesofCounterfactualSubsidy-EliminationDesignsFeatureMarketwidePer-ManufacturerPer-ManufacturerDeadlineDeadlineQuotaDifferentialelimination✕✓✓IncentivetoreduceEVsales✕✕✓7.2ComputingEquilibriumUnderaPer-ManufacturerQuotaSolvingfortheoptimalpricevectorp∗.tunderthequotaintroducesimportantcomputationalchal-lengesbecauseΠ∗f,t+1(p.t)isnotdifferentiableinp.t.Thisisbecausethesubsidyinperiodt+1dependsonfirms’actionsinperiodt.Iaddressthischallengeusingthefollowingstrategytocomputetheequilibrium.14First,basedontheobservationthatmanufacturersotherthanTeslaandGMsoldveryfewEVsupto2016,Iconjecturethatthesemanufacturersstaywithinthequotain2017intheequilibrium.Ithenconsiderfourscenarios,dependingonTeslaandGM’schoiceofcrossingthequotain2017(discussednext).Finally,Iconfirmmyconjecturebyverifyingthatcumulativesalesbyothermanufacturersstaybelowthequota.Theconjectureholdsinthecounterfactualanalysis.Whentheconjectureholds,theoriginalgamecanbereformulatedasatwo-playergamerepre-sentedinthenormalformbythefollowingpayoffmatrix:Tesla/GMCrossDon’tCrossCross(ΠTeslaCC,ΠGMCC)(ΠTeslaCD,ΠGMCD)Don’tCross(ΠTeslaDC,ΠGMDC)(ΠTeslaDD,ΠGMDD)Thepayoffvectorineachcellrepresentsthesumofprofitsinperiodstandt+1forTeslaandGM.Ineachcase,TeslaandGMsolveaconstrainedmaximizationproblemin2017.Forinstance,whenTeslaplays“Don’tCross,”itchoosespricestomaximizethetwo-periodprofitssubjecttotheconstraintthatitscumulativeEVsalesstaybelowthequotain2017.maxpjt,j∈JTesla,tmj∈JTesla,t[pjt−cjt+hjmt]Qjmt+Π∗Tesla,t+1(p.t)s.t.2017τ=2011j∈JTesla,tQjτ≤κ.14.Analternativesolutionistousegrid-searchalgorithms.However,thesealgorithmstendtobeveryslow.29Incontrast,allothermanufacturerssolveforpricesusingEquation3.Thefinalequilibriumundertheper-manufacturerquotaistheNashequilibriuminthis2×2game.Unlikeper-manufacturerquota,Π∗f,t+1(p.t)isstilldifferentiableinp.tunderthemarketwideandper-manufacturerdeadlinesbecausefirms’actionsinperiodtdonotaffectthesubsidiesinperiodt+1.Inthesecases,IuseafixedpointofEquation3tocomputethenewpricingequilibrium,calculatingthepartialderivative∂Qkt∂pjtusingEquation4and∂Π∗f,t+1(p.t)∂pjtnumerically.7.3OutcomesofInterestTherelevantmarketoutcomesincludegovernmentexpenditure,consumersurplus,firmprofits,salesofelectricandconventionalvehicles,andtotalgasolineconsumption.15Governmentexpenditureinperiodtunderthecounterfactualcisjλ(c)jtQ(c)jt.Asitchangesacrosstheexperiments,Ireportittofacilitatedirectcomparisonbetweendifferenteliminationdesigns.16Consumersurplusrepresentscompensatingvariation(McFaddenetal.1973;SmallandRosen1981).Forhouseholdi,thecompensatingvariationinanycounterfactualscenario(c)fromacomparisonscenarioisgivenby∆CSimt=1−αilnj∈Jmtexp(δ(c)jmt+µ(c)ijmt)−lnj∈Jmtexp(δ0jmt+µ0ijmt),(6)whereαiishousehold’smarginalutilityofincome.Giventhecompensatingvariationforaspecifichousehold,thechangeinaveragesurplusinmarketmisi∆CSimtdF(αi,βi).Thetotalchangeinconsumersurplusisthesumofchangesinallmarketsmi∆CSimtddF(αi,βi).ProfitsarecalculatedusingEquation2.Finally,thetotalgasolineconsumptionfromthevehiclessoldinperiodtunderthecounterfactualcisj1mpgj×Q(c)jt×VMTjwhereQ(c)jtisthetotalsalesofvehiclejandVMTjisthemilestraveledduringitslifetime.Iassumethatvehiclestravel12,000milesperyearandhavealifeof15years.7.4CounterfactualResultsThissectionreportsthemarketoutcomesfromsimulatingthealternativedesigns.Overall,theresultsshowthateachdesignhasdifferentimplicationsformarketpenetration,environmentalimpact,anddistributionofgainsacrossconsumersandmanufacturers.Ielaborateontheresultsnext.15.Thesumofconsumersurplusandmanufacturerprofitsdoesnotreflectwelfarefortworeasons.First,economicagentsdonotinternalizetheenvironmentaleffectsofEVs.Second,eliminationdesignsarelikelytohavedifferentlong-termimpactsduetothenetworkeffect,whicharenotcapturedbyaggregatingthetwo-periodoutcomes.16.Alternativestrategiestocomparethedesignsincludefixingthegovernmentexpenditureacrosstheexperimentsbychangingtheamountordurationofthesubsidy.Bothapproacheshavelimitations.Changingtheamountaffectsconsumerpricesand,hence,purchasebehavior,makingitdifficulttodisentangletheeffectofsubsidyelimination.Changingthedurationrequiresmodelingmorethantwoperiods,whichcomplicatescomputation.307.4.1EVPricesandSalesFigure6showsthecumulativeEVsalesforTesla,GM,andNissanunderamarketwidedeadline,aper-manufacturerdeadline,andaper-manufacturerquotaof120,000.17,18Thebluebarsindicatetotalsalesbetween2011and2016,asobservedinthedata,andtheorangeandyellowbarsindicatesalesin2017and2018,respectively,undertherecomputedequilibria.Figure6:EffectofSubsidy-CappingDesignsonEVSalesNotes:Thisfigureshowsthecumulativeplug-inelectricvehiclesales(inthousands)underdifferentsubsidy-cappingdesignsforthethreedominantmanufacturers.Thebluebarsindicatetotalsales2011–2016,asobservedinthedata,andtheorangeandyellowbarsindicatesalesin2017and2018,respectively,undertherecomputedequilibria.ForTeslaandGM,salesin2017remainthesameunderthemarketwideandtheper-manufacturerdeadlinesbecausetheycannotcontrolthestatusofsubsidiesin2018.Incontrast,thequotaactsasabindingconstraintforbothmanufacturers,andtheyremainbelowittoensurethesubsidyin2018.Table8illustratesthemechanismbehindthisfinding,showingthatwhenfacingthequota,TeslaandGMlowerEVsalesin2017byraisingthepricesofEVsin2017.Forinstance,comparedtothecounterfactualwithnosubsidy(Column(1)),GMraisesthepriceoftheChevroletVoltby$4,924(Column(4)).Notably,forTeslaModelS,theincreaseinpriceis$13,500,whichisevenhigherthantheactualsubsidyof$7,500,implyinganegativepass-throughtoconsumersin2017.Suchastrikingeffectofthequotaonpricesispartlyaconsequenceofthemodelingassumptionthatfirmscontrolsalesthroughpricealone.Inpractice,manufacturerscanalsocreateanartificialshortageofqualifyingvehiclesbyloweringproductionandmaynotdrasticallyincreaseprices.How-ever,datalimitationsdonotallowexaminingthismechanism.EventhoughthemodelspecificationleadstoanunconvincinglylargeincreaseinEVpricesunderthequota,itisvaluablebecauseithigh-17.AppendixTable13reportsEVsalesseparatelyforallmanufacturers.18.Iexploreothervaluesoftheper-manufacturerquotainAppendixD.31lightsthestrongincentivetoreduceEVsalesunderthisdesignthatisrobusttothespecification.Twofactorsdrivethisincentive:(1)stayingbelowthequotainanyperiodallowsmanufacturerstoqualifyforthesubsidyonallEVsinthenextperiod,and(2)asthesubsidyiseliminatedonlyformanufacturersthatexhaustthequota,stayingbelowitpreventsexposuretoincreasedcompetitionfrommanufacturersbelowit.UnlikeTeslaandGM,Nissan’sEVsalesin2017remainunaffectedbecauseitisfarbelowtheper-manufacturerquotain2017.AlthoughNissanhadcomparablesalesduring2011–2016,itsellsmuchlessthanTeslaandGMin2017undereachcounterfactual.ThisisbecauseTeslaandGMintroducednewmodels,suchastheChevroletBoltandTeslaModel3,in2017,whichgainedpopularitysincetheirintroduction.Incontrast,Nissanonlysoldasinglemodel(theLeaf)in2017.Table8:EffectofSubsidy-CappingDesignsonVehiclePricesandSalesin2017VehicleOutcomeNoMarketPer-MfrPer-MfrSubsidyDeadlineDeadlineQuota(120,000)ChevroletBolt(BEV)Price(USD)40,15739,04139,02644,905Sales8,20821,59421,63410,826ChevroletVolt(PHEV)Price(USD)35,77435,00634,99240,698Sales15,30032,86432,89718,929TeslaModelS(BEV)Price(USD)85,86985,41085,41299,369Sales14,80124,65124,6399,854ToyotaPriusPrime(PHEV)Price(USD)29,05028,65628,20228,215Sales18,77530,20531,66131,714HondaAccord(Gas)Price(USD)27,95927,95227,95227,958Sales330,695329,571329,491329,994HondaCivic(Gas)Price(USD)22,47422,46622,46622,471Sales341,661340,506340,423340,911ToyotaCamry(Gas)Price(USD)25,72625,72525,72525,730Sales270,928270,025269,961270,301FordF(Gas)Price(USD)35,78835,78735,78635,788Sales277,994277,768277,755277,888ChevroletSilverado(Gas)Price(USD)38,65338,65338,65338,653Sales305,132304,825304,809305,018ToyotaTacoma(Gas)Price(USD)31,70731,70531,70531,708Sales220,559220,273220,253220,391Notes:Thistableshowstheequilibriumprices(beforesubsidy)andsalesacrossthe30samplestatesin2017forasampleofvehiclesusingcounterfactualsimulationsdescribedinSection7.32Next,considertheeffectofsubsidy-cappingdesignsonEVsalesin2018.AsTeslaandGMdonotqualifyunderthemarketwideorper-manufacturerdeadlines,theysellalmostthesamenumberofEVsunderboth.Incontrast,theyqualifyin2018underthequota.Asaresult,theysellmoreEVsin2018thanundereitherofthedeadlines,whichpartlyoffsetstheirlowsalesin2017.Forinstance,Tesla’sEVsalesriseby65percentunderthequotacomparedtotheper-manufacturerdeadline.NissansellsmoreEVsin2018underaper-manufacturerdeadlineandthequotathanunderamarketwidedeadlinebecauseitqualifiesinthefirsttwocasesbutnotinthelast.7.4.2AggregateMarketOutcomesAsmarketoutcomesdifferbymanufacturerandovertime,Ireporttheaggregatemarket-leveloutcomesinTable9usingnosubsidyasthebenchmarkcounterfactual.Asthetotalgovernmentexpenditurechangesacrossthesedesigns,Ialsoreportitforeachdesign.Panel(a)showstheaggregatemarketoutcomesin2017.Boththemarketwideandper-manufacturerdeadlinesleadtoasimilarboostinEVsales,asexpected.Moreover,allothermarketoutcomeslooksimilarunderthetwodesigns.Incontrast,theoutcomesunderthequotaaregovernedbyTeslaandGMtryingtostaybelowitbycharginghigherprices.Becauseoftheseefforts,thesubsidy-inducedEVsalesarearound54percentlowerthantheper-manufacturerdeadline.Moreover,thereductioninconventionalvehiclesalesislowerasconsumerssubstitutefromthehigh-pricedEVstowardlower-costconventionalalternativesin2017.Thesubsidy-inducedconsumersurplusreducesby62percent,theaggregatemanufacturerprofitsincreaseby23.5percent,andthegovernmentexpenditurereducesby29.4percent.Panel(b)showstheaggregatemarketoutcomesin2018.Thoseunderthemarketwidedeadlineareclosetothecounterfactualwithnosubsidybecausethesubsidyexpiresforeveryonein2018.However,EVsalesarehigherthaninthecounterfactualwithnosubsidybecauseofthenetwork-effectgainsfromthe2017subsidies.Thisoutcomeshowsthatinthepresenceofnetworkeffect,EVsubsidiescanhavealong-termimpactonmarketpenetration.Specifically,byreducingtheup-frontcostofEVs,purchasesubsidiesraiseEVsalesin2017.And,becauseEVconsumerscareaboutpreviousEVadoption,thedemandforEVsshiftsrightin2018.Comparedtothemarketwidedeadline,thesubsidy-inducedEVsalesaresubstantiallyhigherunderaper-manufacturerdeadlinebecauseallmanufacturersotherthanTeslaandGMqualifyin2018.Moreover,subsidy-inducedEVsalesareevenhigherunderthequotabecauseallmanufacturers,includingTeslaandGM,qualifyin2018.Panel(c)showsthemarketoutcomessummedoverthetwoyears.Byconstruction,themar-ketwidedeadlinedesignonlydisbursessubsidiesin2017.Asaresult,itrequirestheleastgovernmentexpenditureandresultsinthelowestEVsales,consumersurplus,andmanufacturerprofits.Theper-manufacturerdeadlinerequiresadditionalgovernmentexpenditure,asitalsopaysoutin2018.ItresultsinthegreatestboostinEVsalesandhighestreductioninconventionalvehiclesalesandgasolineconsumption.Finally,thequotarequiresthehighestgovernmentexpenditurebutresultsintheleastreductioninconventionalvehiclesales,lowestconsumersurplus,andhighestmanufacturer33profits.Table9:EffectofSubsidy-CappingDesignsonAggregateOutcomesOutcomeMarketPer-MfrPer-MfrDeadlineDeadlineQuota(120,000)Panel(a):2017Outcomes∆EVSales89,00794,55143,856∆ConvSales-34,478-35,974-12,427∆GasConsumption(MillionGallons)-2,546.11-2,651.85-853.45∆ConsumerSurplus(MillionUSD)1,066.841,135.19428.01∆TotalProfits(MillionUSD)461.24432.16534.07GovtExpenditure(MillionUSD)1,219.411,250.22881.56Panel(b):2018Outcomes∆EVSales1,90443,16189,807∆ConvSales-833-12,464-34,283∆GasConsumption(MillionGallons)-64.4-878.96-2,535.68∆ConsumerSurplus(MillionUSD)35.93560.971,109.05∆TotalProfits(MillionUSD)12.54280.7525.41GovtExpenditure(MillionUSD)0511.751,200.88Panel(c):Total∆EVSales90,911137,712133,663∆ConvSales-35,311-48,438-46,710∆GasConsumption(MillionGallons)-2,610.51-3,530.8-3,389.13∆ConsumerSurplus(MillionUSD)1,102.781,696.161,537.06∆TotalProfits(MillionUSD)473.78712.861,059.48GovtExpenditure(MillionUSD)1,219.411,761.972,082.43Panel(d):Total(Normalized)∆EVSales757864∆ConvSales-29-27-22∆GasConsumption(MillionGallons)-2.14-2-1.63∆ConsumerSurplus(MillionUSD)0.90.960.74∆TotalProfits(MillionUSD)0.390.40.51Notes:Thistableshowsthechangeinaggregatemarketoutcomesunderthecounterfactualsimulationsdiscussedinsection7comparedtothecounterfactualwithnosubsidies.Panel(a)showsthemarketoutcomesin2017,Panel(b)showstheoutcomesfor2018,Panel(c)showstheaggregateoutcomesoverthetwoyears,andPanel(d)showsaggregatetwo-periodoutcomesnormalizedbygovernmentexpenditure.Allsimulationsassumethatthesubsidyeliminationbeganafter2017.Panel(d)showstheaggregatetwo-periodoutcomes,normalizingthegovernmentexpenditureat$1million.Becausegovernmentexpenditureisnotheldconstantacrossdifferentexperiments,suchnormalizationisrequiredtocomparethecost-effectivenessofdifferentdesigns.Panel(d)showsthatsubsidieswithamarketwidedeadlinesellaround75moreEVsand29fewerconventionalvehiclesthanthecounterfactualwithnosubsidy.Theprogramresultsin2.14milliongallonsoflowerfuelconsumption,$0.9millionhigherconsumersurplus,and$0.39millionhighermanufacturer34profits.Similarly,subsidieswithaper-manufacturerdeadlinesellaround78moreEVsand27fewerconventionalvehiclesthanthecounterfactualwithnosubsidy.Theprogramresultsin2milliongallonslowerfuelconsumption,$0.96millionhigherconsumersurplus,and$0.4millionhighermanufacturerprofits.Finally,subsidieswithaquotasell64moreEVsand22fewerconventionalvehiclesthanthecounterfactualwithnosubsidy.Theprogramresultsin1.63milliongallonshigherfuelconsumption,$0.74millionhigherconsumersurplus,and$0.51millionhighermanufacturerprofits.Ineachcase,consumersandmanufacturerssharethebenefitsunequally.However,thebenefitsaccruedtoconsumersarethelowestunderthequota,indicatingasubstantialleakageofbenefitstomanufacturers.ThisoutcomeisaresultofTeslaandGMraisingthepricesoftheirEVsin2017tostaybelowthequota.Asnoted,theeffectofthequotaonpricesispartlyaconsequenceofthemodelspecification,wherebyfirmscontrolEVsalesonlythroughprices.Inpractice,manufacturerscanalsostaybelowthequotabyloweringEVproduction.However,doingsowouldalsolowertheconsumersurpluscomparedtothedeadlinedesigns.Overall,theresultsshowthatforagivengovernmentexpenditure,amarketwideorper-manufacturerdeadlinearealmostequallyeffectiveatraisingEVpenetrationintheshorttermandresultinsimilarbenefitsforconsumers.Moreover,thequotaislesseffectiveatraisingEVpenetrationcomparedtothedeadlinesandresultsinthelowestbenefitsforconsumers.7.4.3ProfitDistributionAcrossManufacturersTable10decomposestheprofitsacrossmanufacturersunderdifferentcounterfactuals.Teslaearnsatleast$237million(or32.6percent),GMearnsatleast$208million(0.5percent),andNissanearnsatleast$17million(0.08percent)moreprofitsfromEVsubsidiesthanthecounterfactualwithnosubsidy.NotethatcomparedtoGMandNissan,EVsubsidieshaveamuchhigherimpactonTesla(asapercentoftotalprofits)becauseitfocusesexclusivelyonEVs.TeslaandGMearnthelowestprofitsundertheper-manufacturerdeadlinebecausetheydonotqualifyforsubsidiesin2018,whereasothersstilldo.Theyearnslightlyhigherprofitsunderamarketwidedeadlinewhennomanufacturerqualifiesin2018.Theyearnthehighestprofitsunderthequota,astheyqualifyin2018.Forinstance,Tesla’sprofitsriseby9.6percentcomparedtotheper-manufacturerdeadline.Incontrast,otherEVmanufacturers,suchasNissanandToyota,earntheleastprofitsunderthemarketwidedeadlinewhentheydonotqualifyforsubsidiesin2018.Theyearnslightlyhigherprofitsunderaper-manufacturerdeadlinewhentheyqualifyin2018,whichallowsthemtohaveacompetitiveadvantageoverTeslaandGMin2018.LikeTeslaandGM,theyalsoearnthehighestprofitsunderthequota.ThisislikelybecausetheyhaveacompetitiveadvantageoverTeslaandGMin2017,bothofwhichmustlimittheirEVsalesin2017toqualifyin2018.357.4.4DiscussionOverall,theresultsshowasubstantialincentiveformanufacturerstoreduceEVsalesunderaper-manufacturerquotawhenitactsasabindingconstraint,whichcandecreasetheeffectivenessofthesubsidy.ThisobservationconfirmstheintuitionfromthemonopolyexampleinSection3.AlthoughtheriseinEVsalesin2018partiallyoffsetsthesalesreductionin2017,thecombinednormalizedsalesoverthetwoyearsare18and14percentlowercomparedtotheper-manufacturerandmarketwidedeadlines,respectively.Thisobservationshowsthatdeadlinesarelikelytobemorecost-effectiveinaidingEVmarketpenetrationcomparedtoabindingper-manufacturerquota.Moreover,becauseEVmanufacturersaremultiproductoligopolists,thedesignsmayalsoaffectthepricesandsalesofconventionalvehicles.Onaggregate,thesedesignsaffectconsumersurplus,manufacturerprofits,andliquidfuelconsumption.Finally,eachdesignaffectstheprofitdistributionacrossmanufacturersdifferently.Comparedtoamarketwidedeadline,aper-manufacturerdeadlineshiftsprofitsawayfromthemanufacturersfacingit.Thus,ifthedeadlineisbasedontheyearofentry,thiswouldimplymoresupportfornewermanufacturers.Incontrast,aper-manufacturerquotadoesnotnecessarilyshiftprofitsawayfrommanufacturersfacingitbecauseitallowsthemtocontrolwhenthesubsidyexpires.BeforeIRA,theEVsubsidieswereeliminatedusingacombinationofaper-manufacturerquotaandper-manufacturerdeadlines(seeFigure1).IRAreplacedthatdesignwithamarketwidedead-line.Theresultsfromtheanalysisindicatethatamarketwidedeadlineisalmostascost-effectiveasaper-manufacturerdeadline.Itisalsoatleastascost-effectiveasaper-manufacturerquotawhenthequotaactsasabindingconstraint.Thus,allelseequal,replacingtheearlierdesignislikelytopositivelyaffectEVmarketpenetrationandcreateahigherreductionintheliquidfuelconsumptioncloserto2032,whenthedynamicsofsubsidyeliminationbecomerelevant.Moreover,thepass-throughtoconsumersisalsolikelytobehighercomparedtotheearlierdesign.However,astheper-manufacturerquotaanddeadlinehavedifferentimplicationsforprofitdistribution,theeffectofreplacingthesubsidy-cappingdesignontheprofitdistributionacrossEVmanufacturersisambiguous.36Table10:EffectofSubsidy-CappingDesignsonProfitDistributionManufacturerNoMarketPer-MfrPer-MfrSubsidyDeadlineDeadlineQuota(120,000)BMW6,1856,2326,2926,300Daimler6,0736,0646,0666,069FiatChrysler24,81424,82124,86124,867Ford29,02829,04629,08829,099GeneralMotors41,36841,59441,57641,775Honda27,78727,73027,70527,709Hyundai11,01611,00511,00411,006JaguarLandRover2,4172,4102,4082,410Kia7,3987,3987,4097,410Mazda3,7253,7183,7153,715Mitsubishi1,3571,3561,3551,355Nissan20,05220,06920,10520,108Subaru9,9509,9309,9249,925Tesla7269649631,056Toyota41,40541,43741,52241,528Volkswagen11,78411,78111,79411,800Volvo1,2221,2261,2321,234Notes:Thistableshowsmanufacturer-levelprofits(inmillionUSD)during2017–2018fromsalesinthe30samplestatesunderthecounterfactualsimula-tions.8ConclusionThispaperdemonstratestheimplicationsofthesubsidy-cappingprovisionsinpurchase-subsidyprogramsdesignedtopromoteinfantgreentechnologies.IfocusontheUSplug-inEVmarket,whichisimportanttounderstandgivenitspotentiallyenormousenvironmentalbenefits.Usingamonopolyexample,IfirstshowthattheprovisionsmayaidorhindertheEVmarketpenetration,andthemagnitudeoftheeffectdependsonstructuralprimitivessuchasown-andcross-pricedemandelasticitiesandthenetworkeffect.Next,tocomparealternativeprovisions,Idevelopastructuralmodeloftheautomobileindustry,whereconsumerschoosevehiclestopurchaseamongallfueltypesbymaximizingutility,andfirmschoosepricesforvehiclestomaximizetheirprofits.Then,Iestimatethedemand-sideparametersusingproduct-leveldataonthenewlyregisteredvehicles,prices,characteristics,andsubsidiesacross30statesintheinitialyearsoftheEVmarketthatwereunaffectedbyeliminatingsubsidies.Usingthedemandparameters,IrecovervehiclemarkupsundertheassumptionofstaticNash-Bertrandequilibrium.Finally,Iusethemarketprimitivesandatwo-stagepricingmodeltopredictfirms’responseswhentheyfacethreecounterfactualsubsidy-cappingdesigns:amarketwidedeadline,aper-manufacturerdeadline,andaper-manufacturerquota.Overall,theresultsshowthat,allelseequal,thequota,whenbinding,incentivizesmanufacturerstoreduceEVsalescomparedtothedeadlinedesigns.Twofactorsdrivethisincentive:(1)stayingbelowthequotainanyperiodallowsmanufacturerstoqualifyforthesubsidyonallEVsinthenext37period,and(2)asthesubsidyiseliminatedonlyforthosethatexhaustthequota,stayingbelowitprotectsthemfromcompetitionfromothermanufacturersbelowit.Asaresult,givengovernmentexpenditure,deadlinescanbemorecost-effectiveinincreasingEVmarketpenetrationthanthequota.Becausemanufacturershavemarketpower,thiscanalsotranslateintolowerpass-throughofsubsidiestoconsumerscomparedtothedeadlinedesigns.Inaddition,becausemanufacturersaremultiproductoligopolists,thedesignsaffectthesalesofconventionalvehiclesand,hence,theconsumersurplus,manufacturerprofits,andliquidfuelconsumption;theyalsoaffectthedistributionofprofitsacrossmanufacturers.Thesefindingsfacilitateadeeperunderstandingoftheroleofpolicyininfluencingtechnologychangeinthreeways.First,theyelucidatetheeffectofthesubsidydesignonmarketpenetrationinatheoreticallymotivatedanalysis.BecauseEVsofferaviablesolutiontofuelefficiencyandenergysecurity,policymakersareeagertoincreaseadoption.TheUSmarketshareofEVshasremainedlimiteddespiteseveralincentiveprograms.Carefuldesignisthereforecrucial,especiallyconsideringthatEVtaxincentivescostbillionsofdollarsandreceivemuchscrutiny.Second,thepapershedslightontheimpactofthesubsidydesignonconsumersurplusandthedistributionofprofitsacrossmanufacturers,whichishelpfulfortargeting.Forinstance,comparedtoamarketwidedeadline,aper-manufacturerdeadlineshiftsprofitsawayfromthemanufacturersfacingit.Despitethepenaltyondominantmanufacturers,aper-manufacturerdeadlinemaybejustifiedifsignificantbarrierstomanufacturers’entryexist,becausepositiveexternalitiesfromtheentryappearintheformofenvironmentalbenefits,innovationspillovers,andhighernationalenergysecurity.Finally,theimplicationsfromtheplug-inEVmarketmayalsoholdforothercountriesandsustainabletechnologies,suchassolarpanelsandwindenergy.ReferencesAghion,Philippe,AntoineDechezlepretre,DavidHemous,RalfMartin,andJohnVanReenen.2016.“CarbonTaxes,PathDependency,andDirectedTechnicalchange:EvidencefromtheAutoIndustry.”JournalofPoliticalEconomy124(1):1–51.Archsmith,James,AlissaKendall,andDavidRapson.2015.“FromCradletoJunkyard:AssessingtheLifeCycleGreenhouseGasBenefitsofElectricVehicles.”ResearchinTransportationEconomics52:72–90.Babaee,Samaneh,AjaySNagpure,andJosephFDeCarolis.2014.“HowMuchDoElectricDriveVehiclesMattertoFutureUSEmissions?”EnvironmentalScience&Technology48(3):1382–1390.Beresteanu,Arie,andShanjunLi.2011.“GasolinePrices,GovernmentSupport,andtheDemandforHybridVehiclesintheUnitedStates.”InternationalEconomicReview52(1):161–182.Berry,Steven,JamesLevinsohn,andArielPakes.1995.“AutomobilePricesinMarketEquilibrium.”Econometrica63(4):841–890.38Borenstein,Severin,andLucasWDavis.2016.“TheDistributionalEffectsofUSCleanEnergyTaxCredits.”TaxPolicyandtheEconomy30(1):191–234.Buekers,Jurgen,MirjaVanHolderbeke,JohanBierkens,andLucIntPanis.2014.“HealthandEn-vironmentalBenefitsRelatedtoElectricVehicleIntroductioninEUCountries.”TransportationResearchPartD:TransportandEnvironment33:26–38.Cabral,Marika,MichaelGeruso,andNealeMahoney.2018.“DoLargerHealthInsuranceSubsi-diesBenefitPatientsorProducers?EvidencefromMedicareAdvantage.”AmericanEconomicReview108(8):2048–87.Chandra,Ambarish,SumeetGulati,andMilindKandlikar.2010.“GreenDriversorFreeRiders?AnAnalysisofTaxRebatesforHybridVehicles.”JournalofEnvironmentalEconomicsandManagement60:78–93.Clinton,BentleyC,andDanielCSteinberg.2019.“ProvidingtheSpark:ImpactofFinancialIncen-tivesonBatteryElectricVehicleAdoption.”JournalofEnvironmentalEconomicsandManage-ment98:102255.Crago,ChristineLasco,andIlyaChernyakhovskiy.2017.“ArePolicyIncentivesforSolarPowerEffective?EvidencefromResidentialInstallationsintheNortheast.”JournalofEnvironmentalEconomicsandManagement81:132–151.DeShazo,JR,TamaraLSheldon,andRichardTCarson.2017.“DesigningPolicyIncentivesforCleanerTechnologies:LessonsfromCaliforniaâsPlug-InElectricVehicleRebateProgram.”JournalofEnvironmentalEconomicsandManagement84:18–43.Fan,Ying,andGeZhang.2022.“TheWelfareEffectofaConsumerSubsidywithPriceCeilings:TheCaseofChineseCellPhones.”TheRANDJournalofEconomics53(2):429–449.Gallagher,KellySims,andErichMuehlegger.2011.“GivingGreentoGetGreen?IncentivesandConsumerAdoptionofHybridVehicleTechnology.”JournalofEnvironmentalEconomicsandManagement61(1):1–15.Gillingham,KennethT.2022.“DesigningFuel-EconomyStandardsinLightofElectricVehicles.”EnvironmentalandEnergyPolicyandtheEconomy3(1):111–154.Gulati,Sumeet,CarolMcAusland,andJamesMSallee.2017.“TaxIncidencewithEndogenousQualityandCostlyBargaining:TheoryandEvidencefromHybridVehicleSubsidies.”JournalofPublicEconomics155:93–107.Heutel,Garth,andErichMuehlegger.2015.“ConsumerLearningandHybridVehicleAdoption.”EnvironmentalandResourceEconomics62(1):125–161.Hitaj,Claudia.2013.“WindPowerDevelopmentintheUnitedStates.”JournalofEnvironmentalEconomicsandManagement65(3):394–410.39Holland,StephenP,ErinTMansur,NicholasZMuller,andAndrewJYates.2016.“AreThereEnvironmentalBenefitsfromDrivingElectricVehicles?TheImportanceofLocalFactors.”AmericanEconomicReview106(12):3700–3729.IRS.2006.“Phase-outofcreditfornewqualifiedhybridmotorvehiclesandnewadvancedleanburntechnologymotorvehicles.”Notice2006-78..2009.“EnergyprovisionsoftheAmericanRecoveryandReinvestmentActof2009(ARRA).”FS-2009-10..2018.“Phase-outofcreditfornewqualifiedplug-inelectricdrivemotorvehicles.”Notice2018-96..2019.“Phase-outofcreditfornewqualifiedplug-inelectricdrivemotorvehicles.”Notice2019-22..2023.“CertaindefinitionsoftermsinSection30DCleanVehicleCredit.”Notice2023-1.Jacobsen,MarkR.2013.“EvaluatingUSFuelEconomyStandardsinaModelwithProducerandHouseholdHeterogeneity.”AmericanEconomicJournal:EconomicPolicy5(2):148–87.Jenn,Alan,InesL.Azevedoa,andPedroFerreira.2013.“TheImpactofFederalIncentivesontheAdoptionofHybridElectricVehiclesintheUnitedStates.”EnergyEconomics40:936–942.Jenn,Alan,KatalinSpringel,andAnandRGopal.2018.“EffectivenessofElectricVehicleIncentivesintheUnitedStates.”Energypolicy119:349–356.Kalish,Shlomo,andGaryL.Lilien.1983.“OptimalPriceSubsidyPolicyforAcceleratingtheDif-fusionofInnovation.”MarketingScience2(4):407–420.Lambert,Fred.2018.“There’saNewEfforttoRemovetheEVTaxCreditCapJustasTeslaandGMAreAbouttoHitIt.”Electrek.Langer,Ashley,andDerekLemoine.2022.“DesigningDynamicSubsidiestoSpurAdoptionofNewTechnologies.”JournaloftheAssociationofEnvironmentalandResourceEconomists9(6):1197–1234.Lazzari,Salvatore.2006.“TaxCreditsforHybridVehicles.”CongressionalResearchService,LibraryofCongress.Leonhardt,David.2006.“USHybridsGetMoreMilesperCongress.”TheNewYorkTimes.Li,Jing.2018.“CompatibilityandInvestmentintheUSElectricVehicleMarket.”MITWorkingPaper.Li,Shanjun,LangTong,JianweiXing,andYiyiZhou.2017.“TheMarketforElectricVehicles:IndirectNetworkEffectsandPolicyDesign.”JournaloftheAssociationofEnvironmentalandResourceEconomists4(1):89–133.40McConnell,Virginia,andBenjaminLeard.2021.“PushingNewTechnologyintotheMarket:Cali-fornia’sZeroEmissionsVehicleMandate.”ReviewofEnvironmentalEconomicsandPolicy15(1):169–179.McFadden,Daniel,etal.1973.“ConditionalLogitAnalysisofQualitativeChoiceBehavior.”Nevo,Aviv.2000.“APractitioner’sGuidetoEstimationofRandom-CoefficientsLogitModelsofDemand.”JournalofEconomics&ManagementStrategy9(4):513–548.Pless,Jacquelyn,andArthurVanBenthem.2019.“Pass-ThroughasaTestforMarketPower:AnApplicationtoSolarSubsidies.”AmericanEconomicJournal:AppliedEconomics11(4):367–401.Polyakova,Maria,andStephenPRyan.2019.“SubsidyTargetingwithMarketPower.”NBERWorkingPaper,No.26367.Sallee,JamesM.2011.“TheSurprisingIncidenceofTaxCreditsfortheToyotaPrius.”AmericanEconomicJournal:EconomicPolicy3(2):189–219.Small,KennethA,andHarveySRosen.1981.“AppliedWelfareEconomicswithDiscreteChoiceModels.”Econometrica49(1):105–130.Springel,Katalin.2021.“NetworkExternalityandSubsidyStructureinTwo-SidedMarkets:Evi-dencefromElectricVehicleIncentives.”AmericanEconomicJournal:EconomicPolicy13(4):393–432.Steinbacher,K,MGoes,andKJörling.2018.“IncentivesforElectricVehiclesinNorway:FactSheet.”EcofysforFederalMinistryfortheEnvironment,NatureConservationandNuclearSafetyofGermany3.VanBenthem,Arthur,KennethGillingham,andJamesSweeney.2008.“Learning-by-DoingandtheOptimalSolarPolicyinCalifornia.”TheEnergyJournal29(3):131–151.Williams,JamesH,AndrewDeBenedictis,RebeccaGhanadan,AmberMahone,JackMoore,WilliamRMorrowIII,SnullerPrice,andMargaretSTorn.2012.“TheTechnologyPathtoDeepGreen-houseGasEmissionsCutsby2050:ThePivotalRoleofElectricity.”Science335(6064):53–59.41AppendixAEVManufacturersandBuyersRespondtoSubsidy-CappingDe-signsThissectionoffersevidencethatEVmanufacturersandbuyersrespondtothesubsidy-cappingdesignsbasedonTeslaandGM’sexperiences.TeslaandGMsurpassed200,000plug-insalesinJulyandNovember2018,respectively.Figure7comparesthemonthlynationwidesalesofTesla’smostaffordablemodel(Model3)withGM’smostpopularplug-inhybrid(ChevroletVolt)duringMarch2018–June2019.Threeobser-vationsareworthnoting.First,thesubsidyaffectsconsumers’purchasedecisions,asevidentfromtheintertemporalbunchingsurroundingchangesinthefederaltaxcredit.ThecreditavailableforModel3inJanuarywashalfasgenerouscomparedtoDecember,andJanuarysalescorrespondinglyplummetedto50percentoftheDecembersalesvolume.SuchbunchingdoesnotappearforChevro-letVolt,whichdidnotfaceareductionintaxcreditsinJanuary2019,pointingtothepossibilitythatthechangesinfederalsubsidyaffectedconsumers’buyingdecisions.Thisbunchingconflatesthechangesinpurchasechoiceandpurchasetimingbecausesomeconsumersawareoftheloomingchangemayhaveboughtearliertotakeadvantageofthemoregeneroussubsidy.Suchtimingeffectsposeachallengeinestimatingdemandelasticities,whichIdiscussinSection4.Second,deadlinesareeffectiveininducingcarsales.OnceTeslaexhaustedthethreshold,itfacedasix-monthdeadline:allvehiclesdeliveredbyDecember2018qualifiedfora$7,500subsidy.BetweenJulyandDecember,Teslasetnewproductionanddeliveryrecords.Third,manufacturerslikelyrespondtoaquotabyreducingEVsales,asevidentbythe130percentspikeinModel3salesinJuly2018.IfTeslareachedthethresholdinJuneinsteadofJuly,itssubsidywouldhavereducedtohalfinOctober2018insteadofJanuary2019.ThelowsalesvolumeinJuneisconsistentwiththeincentivetopushthedeliveryofthe200,000thvehicletoJuly.ComparingthesalesoftheChevroletVoltshowsthatseasonalityindemandisnotenoughtoexplainthedifferencebetweentheJuneandJulysalesvolumes.Toquantifytheeffectofquotaanddeadlinesonthequarterlysalesvolume,Ialsoestimatethefollowingregression:Salesjft=β0+β1.t+β2.IDft+β3.IQft+δj+δqtr+ϵjftwhereIDfttakesvalue1iffirmffacesadeadlinewithinnexttwoquarters;IQftis1iffirmfhassoldlessthan200,000vehiclesuptoquartertandisrestrictedbythequota.δjandδqtrindicateproductandquarterfixedeffects.Thecoefficientβ2indicatestheeffectofdeadline,andβ3indicatestheeffectofquotaonEVsales.ThevariationcomesonlyfromTeslaandGMbecausenootherfirmhasexhaustedthequotayet.Controllingforproductandquarterfixedeffects,thesefirms,onaverage,sell3,000moreEVsunderthetwo-quarterdeadlineand3,000fewerEVsunderthequotaof200,000.Althoughthesefindingsinformusoftheimportanceofthesubsidy-cappingdesigns,theydonotexplainwhathappenswithamarketwidedeadline.Moreover,thevariationisonlydrivenbyTesla42Figure7:TeslaModel3andGMChevroletVoltSales,2017–2018Notes:ThisfigureplotsthetotalnationwidesalesoftheModel3andVoltbetweenMarch2018andJune2019basedonthemonthlysalesestimatesbyAutomotiveNewsDataCenter.TheblackverticallineindicatesJuly2018—whenTesladeliveredthe200,000thqualifyingvehicle.GMreachedthisthresholdinNov2018.TheshadesofgreyindicatethevalueofTesla’sfederalsubsidy;darkershadesindicatehighervalues.Tesla’sphaseoutbeganinJanuary2019,andthesubsidyforallitsmodelswasreducedfrom$7,500to$3,750.GM’ssubsidywasreducedinApril2019.43Table11:Evidence:EVSalesDependontheSubsidy-CappingDesign(1)(2)(3)(4)VARIABLESSales(1,000s)Sales(1,000s)log(Sales)log(Sales)ApproachingDeadline2.7975.1710.4320.593(0.926)(1.011)(0.477)(0.498)QuotaConstrained-3.0760.649-0.240-0.580(0.876)(1.478)(0.451)(0.728)trend0.004-0.007(0.020)(0.010)Constant4.3230.8136.2136.430(0.973)(1.410)(0.501)(0.694)Observations836836836836R-squared0.3760.4250.2660.380ProductFEYesYesYesYesQuarterFEYesYesYesYesFirm-leveltrendYesYesSource:WardsAutoUSLightVehicleQuarterlySalesNotes:Standarderrorsareinparentheses.indicates99percentlevelofsignificance.indicates95percentlevelofsignificance.indicates90percentlevelofsignificance.andGM,asthesubsidywasnoteliminatedforanyothermanufacturer.Therefore,tocomparethedifferentsubsidy-cappingdesigns,Irelyonstructuralmethodsinthepaper.BFirst-StageRegressionResultsTable12reportstheresultsofthefirst-stageregression.IVs1–5arethesumoverthecharacter-isticsoffirm’sothervehicles.IVs6–10aresumsoverthecharacteristicsofcompetingvehicles.Vehiclecharacteristicsusedtoconstructtheexcludedinstrumentsincludeaconstant,vehiclesize,performance,drivingcost,andbatteryrange.44Table12:First-StageRegressionResultsDependentVariable:Price($’0000)VariableCoefSEIncludedIVConstant-8.447∗∗∗1.447Size(0000in2)6.439∗∗∗0.143Performance(Hp/10lb)9.026∗∗∗0.065FuelCost($/mile)6.797∗∗∗0.484BatteryRange(10miles)-0.038∗∗∗0.009BEV2.778∗∗∗0.250PHEV2.927∗∗∗0.252BEV×log(1+ManufacturerEVSales)-0.0000.028PHEV×log(1+ManufacturerEVSales)-0.187∗∗∗0.022BEV×log(1+Same-ChargerEVSales)-0.274∗∗∗0.036PHEV×log(1+Same-ChargerEVSales)0.0360.037ExcludedIVIV10.149∗∗∗0.034IV2-0.323∗∗∗0.037IV30.165∗∗∗0.014IV40.168∗∗∗0.020IV50.009∗∗∗0.002IV6-0.0350.033IV70.0420.036IV80.0030.013IV9-0.032∗∗∗0.010IV10-0.0000.002Obs62186R-squared0.583FtestF(66,62119)1,317.700FtestofexcludedIVF(10,62119)494.992Notes:Sizeiswheelbase×width(inthousandsofin2),performanceishorsepowerbycurbweight(in10lb),drivingcostisfuelcost(indollarspermile),andbatteryrangeistheall-electricrange(inmiles)forelectricvehicles(EVs).Thevariable“ManufacturerEVSales”showsthetotalEVssoldbythemanufacturerinthegeographicmarketuntilthepreviousyear.Thevariable“Same-ChargerEVSales”showscumulativeEVsalesbyallmanufacturerswiththesameLevel3chargingstandardinthegeographicmarketuntilthepreviousyear.IVs1–5arethesumoverthecharac-teristicsoffirm’sothervehicles.IVs6–10aresumsoverthecharacteristicsofcompetingvehicles.Vehiclecharacteristicsusedtoconstructtheexcludedinstrumentsincludeaconstant,vehiclesize,performance,drivingcost,andbatteryrange.indicates99percentlevelofsignificance.indicates95percentlevelofsignificance.indicates90percentlevelofsignificance.45CCounterfactualsTable13:EffectofSubsidy-CappingDesignsonEVSalesManufacturerYearNoMarketPer-MfrPer-MfrSubsidyDeadlineDeadlineQuota(120,000)BMW2011–201631,26131,26131,26131,26120179,59216,30417,29817,43620188,7798,68815,07015,630DAIMLER2011–20168,4858,4858,4858,48520171,5462,3872,4672,48920181,4011,3482,1192,179FIATCHRYSLER2011–201618,03118,03118,03118,03120173,6927,6038,6268,73920184,0094,2649,5189,856FORD2011–201668,41468,41468,41468,414201712,43119,17819,70619,806201811,30910,75616,68117,323GENERALMOTORS2011–201690,09090,09090,09090,090201723,64254,68754,76129,910201822,25323,42723,25451,116HONDA2011–20168578578578572017222220182222HYUNDAI2011–20161,3201,3201,3201,32020171,0121,6891,9131,94220189581,0041,8101,882KIA2011–20163,5713,5713,5713,57120171,6402,8963,3713,37620181,6791,8503,6033,595MITSUBISHI2011–20161,6521,6521,6521,652201741111112018441010NISSAN2011–201696,16596,16596,16596,16520174,4309,96610,21210,23720184,2254,2309,5149,490TESLA2011–2016103,550103,550103,550103,550201725,15842,90042,88016,450201825,19725,72125,64742,567TOYOTA2011–201642,14442,14442,14442,144201718,77530,20531,66131,714201818,71519,26431,63631,572VOLKSWAGEN2011–201618,08418,08418,08418,08420173,9136,4376,7436,80620183,5373,4305,7405,944VOLVO2011–20161,9791,9791,9791,97920171,4802,0602,2202,25520181,4001,3822,0252,109Notes:Thistableshowstheelectricvehiclesalesforeachmanufacturer.Salesin2011–2016arereportedasobservedinthedata.Salesin2017and2018arecomputedunderthecounterfactualpolicysimulationsdiscussedinSection7.46Table14:EffectofSubsidy-CappingDesignsonVehiclePricesandSales,2018VehicleOutcomeNoMarketPer-MfrPer-MfrSubsidyDeadlineDeadlineQuota(120,000)ChevroletBolt(BEV)Price(USD)41,07441,04541,04339,740Sales7,8568,4748,44120,778ChevroletVolt(PHEV)Price(USD)36,60436,58836,59035,644Sales14,27014,82214,68230,124TeslaModelS(BEV)Price(USD)86,36286,36386,37285,850Sales14,85315,16415,11124,510ToyotaPriusPrime(PHEV)Price(USD)30,24330,26729,66729,660Sales18,71519,26431,63631,572HondaAccord(Gas)Price(USD)27,96027,96027,95727,952Sales330,769330,755330,124329,657HondaCivic(Gas)Price(USD)22,47422,47422,47122,466Sales341,737341,724341,066340,602ToyotaCamry(Gas)Price(USD)25,72625,72625,73025,726Sales270,967270,949270,359270,043FordF(Gas)Price(USD)35,78835,78735,78735,786Sales277,998277,994277,872277,758ChevroletSilverado(Gas)Price(USD)38,65338,65338,65238,653Sales305,136305,127304,998304,816ToyotaTacoma(Gas)Price(USD)31,70731,70731,70831,705Sales220,568220,560220,387220,268Notes:Thistableshowstheequilibriumprices(beforesubsidy)andsalesacrossthe30samplestatesin2018forasampleofvehiclesusingcounterfactualsimulationsdescribedinSection7.47DOtherValuesofthePer-ManufacturerQuotaThissectionexplorestheeffectofEVsubsidiesonthemarketoutcomesunderdifferentvaluesoftheper-manufacturerquota.UsingtheparameterestimatesfromSection6,IrecomputetheequilibriumforfivedifferentvaluesofthequotaandplottheresultingoutcomesinFigure8.Ineachcase,IfixthetotalEVsalesbetween2011and2016asobservedinthedata.Panel(a)showsthetotalEVssoldbyTeslaandGMbetween2011and2017.Whenthequotais110,000,Teslachoosestoexceeditin2017,evenifthatsacrificeseligibilityin2018.Whenthequotaisbetween120,000and140,000,bothTeslaandGMchoosetostaybelowthequotatoensuresubsidiesin2018.Whenthequotais150,000,itisnotbindingforeithermanufacturer.Inthatcase,bothmanufacturersrespondsimilarlycomparedtowhentheyfaceadeadline.Panel(b)showstheaggregateboostinEVsalesacrossallmanufacturersduring2017and2018comparedtothecounterfactualwithnosubsidy,fixingthegovernmentexpenditureat$1million.AlthougheachpolicyincreasesEVsales,theboostdependsonthevalueofthequota.Thefartherthequotaisfrommanufacturers’privatelyoptimalsales,themoretheyareincentivizedtoreduceEVsalesandthefewerEVsmaybesoldforthesamelevelofexpenditure.However,whenthequotaistoosmall,somemanufacturersmayfinditmorebeneficialtoexceedit,evenifitmakesthemineligibleforfuturesubsidies.Finally,whenthequotaisnotbinding,theoutcomesaresimilartowhenmanufacturersfaceadeadline.48Figure8:EffectofEVSubsidiesonEVSalesUnderDifferentValuesofthePer-ManufacturerQuotaNotes:Panel(a)showsthecumulativeelectricvehicle(EV)sales(inthousands)byTeslaandGMbetween2011and2017underthedifferentvaluesofaper-manufacturerquota.Ineachcase,Ifixthesalesbetween2011and2016asinthedataandrecomputetheequilibriumin2017.Panel(b)showstheaggregateboostinEVsalesduetosubsidybetween2017and2018comparedtothecounterfactualwithnosubsidy,fixingthegovernmentexpenditureat$1million.49