WegratefullyacknowledgetheNationalBankofBelgiumfortheuseofitsresearchfacilitiesanddataandinparticularValereBogaertsforhelpwiththedatacollectionandEmmanuelDhyneandCatherineFussforcommentsanddataclar ID: 294638
Download Pdf The PPT/PDF document "Importers,Exporters,andExchangeRateDisco..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Importers,Exporters,andExchangeRateDisconnectMaryAmitiMary.Amiti@NY.FRB.orgFederalReserveBankofNewYorkOlegItskhokiitskhoki@princeton.eduPrincetonUniversityJozefKoningsJoep.Konings@KULeuven.beUniversityofLeuvenandNationalBankofBelgiumThisdraft:September23,2013Firstdraft:July2012AbstractLargeexportersaresimultaneouslylargeimporters.Inthispaper,weshowthatthispatterniskeytounderstandinglowaggregateexchangeratepass-throughaswellasthevariationinpass-throughacrossexporters.First,wedevelopatheoreticalframeworkthatcombinesvariablemarkupsduetostrategiccomplementaritiesandendogenouschoicetoimportintermediateinputs.Themodelpredictsthat rmswithhighimportsharesandhighmarketshareshavelowexchangeratepass-through.Second,wetestandquantifythetheoreticalmechanismsusingBelgian rm-product-leveldatawithinformationonexportsbydestinationandimportsbysourcecountry.Wecon rmthatimportintensityandmarketsharearekeydeterminantsofpass-throughinthecross-sectionof rms.Asmallexporterwithnoimportedinputshasanearlycompletepass-through,whilea rmatthe95thpercentileofbothimportintensityandmarketsharedistributionshasapass-throughofjustabove50%,withthemarginalcostandmarkupchannelsplayingroughlyequalroles.Thelargestexportersaresimultaneouslyhigh-market-shareandhigh-import-intensity rms,whichhelpsexplainthelowaggregatepass-throughandexchangeratedisconnectobservedinthedata.Keywords:F14,F31,F41JELclassi cation:exchangeratepass-through,pricing-to-market,importintensity WegratefullyacknowledgetheNationalBankofBelgiumfortheuseofitsresearchfacilitiesanddata,andinparticularValereBogaertsforhelpwiththedatacollection,andEmmanuelDhyneandCatherineFussforcommentsanddataclari cations.WethankGeorgeAlessandria,ArielBurstein,ElhananHelpman,DoireannFitzgerald,LindaGoldberg,PennyGoldberg,LoganLewis,NickLi,BenMandel,UlrichMüller,SteveRedding,EstebanRossi-Hansberg,DavidWeinstein,HylkeVandenbusscheandseminarparticipantsatmultiplevenuesforinsightfulcomments.WealsothankSydneeCaldwell,StefaanDecramer,CecileGaubert,DiegoGilsanz,PrestonMui,andMarkRazhevforexcellentresearchassistance.TheviewsexpressedinthispaperarethoseoftheauthorsanddonotnecessarilyrepresentthoseoftheFederalReserveBankofNewYorkortheNationalBankofBelgium. 1IntroductionOneofthecentralpuzzlesininternationalmacroeconomicsiswhylargemovementsinex-changerateshavesmalleectsonthepricesofinternationallytradedgoods.Thisexchangeratedisconnecthasgeneratedavastliterature,yetnoempiricalpass-throughstudyhastakenintoaccountoneofthemostsalientfeaturesofinternationaltrade,thatisthatthelargestexportersaresimultaneouslythelargestimporters.Inthispaper,weshowthatthispatterniskeytounderstandingthelowaggregatepass-through,aswellasthevariationinpass-throughacross rms.UsingdetailedBelgianmicrodata,we ndthatmoreimport-intensiveexportershavesig-ni cantlylowerexchangeratepass-throughintotheirexportprices,astheyfaceosettingexchangerateeectsontheirmarginalcosts.Thesedatarevealthatthedistributionofim-portintensityamongexportersishighlyskewed,withtheimport-intensive rmsbeingamongthelargestexporters,accountingforamajorshareofinternationaltrade.Consequently,theimport-intensive rmsalsohavehighexportmarketsharesandhencesethighmarkupsandactivelymovetheminresponsetochangesinmarginalcost,providingasecondchannelthatlimitstheeectofexchangerateshocksonexportprices.Thesetwomechanismsreinforceeachotherandacttointroduceabuerbetweenlocalcostsandinternationalpricesofthemajorexporters,thusplayingacentralroleinlimitingthetransmissionofexchangerateshocksacrosscountries.Theavailabilityof rm-leveldatawithimportsbysourcecountryandexportsbydestination,combinedwithdomesticcostdata,enablesustoestimatethemagnitudeofthesetwochannels.Toguideourempiricalstrategy,wedevelopatheoreticalframeworktostudytheforcesthatjointlydeterminea rm'sdecisionstosourceitsintermediateinputsinternationallyandtosetmarkupsineachdestinationofitsexports.ThetwobuildingblocksofourtheoreticalframeworkareanoligopolisticcompetitionmodelofvariablemarkupsfollowingAtkesonandBurstein(2008)andamodelofthe rm'schoicetoimportintermediateinputsata xedcostfollowingHalpern,Koren,andSzeidl(2011).Thesetwoingredientsallowustocapturethekeypatternsinthedatathatwefocuson,andtheirinteractiongeneratesnewinsightsonthedeterminantsofexchangeratepass-through.Inequilibrium,themoreproductive rmsenduphavinggreatermarketsharesandchoosetosourcealargershareoftheirinputsinternationally,whichinturnfurtherampli estheproductivityadvantageofthese rms.Thetheoryfurtherpredictsthata rm'simportintensityandexportmarketshareformasu cientstatisticforitsexchangeratepass-throughwithinindustry-destination,withimportintensityproxyingformarginalcostsensitivitytotheexchangerateandmarket1 sharesproxyingformarkupelasticity.1WetestthepredictionsofthetheorywitharichdatasetofBelgianexportersfortheperiod2000to2008.Adistinctivefeatureofthesedataisthattheycomprise rm-levelimportsbysourcecountryandexportsbydestinationattheCN8-digitproductcodes(closeto10,000distinctproductcodes),whichwematchwith rm-levelcharacteristics,suchaswagesandexpenditureoninputs.Thisallowsustoconstructanimportintensitymeasureforeach rmastheshareofimportsintotalvariablecostsandameasureof rm'smarketshareforeachexportdestination,whicharethetwokey rmcharacteristicsinouranalysis.Further,withtheinformationonimportsbysourcecountry,wecanseparateinputsfromEuroandnon-Eurocountries,whichisanimportantdistinctionsinceimportedinputsfromwithintheEuroareaareintheBelgian rms'currency.Westartourempiricalanalysisbydocumentingsomenewstylizedfactsrelatedtothedistributionofimportintensityacross rms,lendingsupporttotheassumptionsandpredic-tionsofourtheoreticalframework.Weshowthatinthealreadyveryselectgroupofexportersrelativetotheoverallpopulationofmanufacturing rms,therestillexistsasubstantialhet-erogeneityintheshareofimportedinputssourcedinternationally,inparticularfromthemoredistantsourcecountriesoutsidetheEuroZone.Theimportintensityisstronglycor-relatedwith rmsizeandother rmcharacteristicsandisheavilyskewedtowardthelargestexporters.Ourmainempiricalspeci cation,assuggestedbythetheory,relatesexchangeratepass-throughwiththe rm'simportintensitycapturingthemarginalcostchannelandthedestination-speci cmarketsharescapturingthemarkupchannel.Weestimatethecross-sectionalrelationshipbetweenpass-throughanditsdeterminantswithinindustriesanddes-tinations,holdingconstantthegeneralequilibriumforcescommontoall rms.2Ourmethod-ologicalcontributionistoshowthatsucharelationshipholdsindependentlyofthegeneralequilibriumenvironment,thuswedonotneedtomakespeci cassumptionsaboutthesourcesofvariationintheexchangerate.Theexchange-ratepass-throughcoe cientsinthisrela- 1Notethattherelationshipbetweenimportintensityandmarginalcostisverygeneralanddoesnotrelyonaparticularstructuralmodel.Inturn,therelationshipbetweenmarketshareandmarkupisnotuniversal,yetitemergesinaclassofmodelscommonlyusedininternationalmacro(seeBursteinandGopinath,2012).Thestructuralmicroliterature,however,adoptsmoresophisticateddemandsystemswheremarkupvariabilitydependsnotonlyonmarketshare,butalsoonprices,productcharacteristicsandthedistributionofconsumercharacteristics.Ourapproachprovidesasimpleapproximationofthemarkup,linkingitexclusivelytothemarketshareofthe rm,andwhilebeinglessgeneralatthelevelofindividualindustries,thisapproachallowsustoproceedwithestimationacrossbroadsectorsandmultipleexportmarkets.We ndthatthevariationinthemarketsharealoneexplainssubstantialvariationinthemarkupvariabilityacross rms.2Suchcommonforcesincludethecorrelationsofsector-destination-speci cpriceindex,sector-speci cproductivityandcostindexwiththeexchangerate.2 tionshipcanbedirectlyestimatedwithoutimposingstrongpartialequilibriumorexogeneityassumptions.Theoryfurtherprovidesclosed-formexpressionsforthesecoe cients,whichallowsustodirectlytestforthestructuralmechanismemphasizedinthemodel.3Theresultsprovidestrongsupportforthetheory.First,weshowthatimportintensityisanimportantcorrelateofa rm'sexchangeratepass-through,witheachadditional10percentagepointsofimportsintotalcostsreducingpass-throughbyover6percentagepoints.Second,weshowthatthiseectisduetoboththemarginalcostchannel,whichimportintensityaectsdirectly,andthemarkupchannelthroughtheselectioneect.Finally,whenweincludemarketshare,whichproxiesforthemarkupchannel,togetherwithimportintensity,we ndthesetwovariablesjointlytoberobustpredictorsofexchangeratepass-throughacrossdierentsub-samplesandspeci cations,evenaftercontrollingforother rmcharacteristicssuchasproductivityandemploymentsize.Quantitatively,theseresultsarelarge.A rmatthe5thpercentileofbothimportintensityandmarketshare(bothapproximatelyequaltozero)hasanearlycompletepass-through(94%andstatisticallyindistinguishablefrom100%).Incontrast,a rmatthe95thpercentileofbothimportintensityandmarketsharedistributionshasapass-throughslightlyabove50%,withimportintensityandmarketsharecontributingnearlyequallytothisvariationacross rms.Inotherwords,whilesmallexportersbarelyadjusttheirproducerpricesandfullypassontheexchangeratemovementstoforeignconsumers,thelargestexportersosetalmosthalfoftheexchangeratemovementbyadjustingtheirpricesalreadyatthefactorygates.Activemarkupadjustmentbythelarge rmsexplainsthisonlyinpart,andanequallyimportantroleisplayedbythemarginalcostchannelbymeansofimportedinputs.Theseresultshaveimportantimplicationsforaggregatepassthroughsincethelowpass-through rmsaccountforadisproportionatelylargeshareofexports.RelatedliteratureOurpaperisrelatedtothreestrandsofrecentliterature.First,itrelatestotherecentandgrowingliteratureontheinteractionofimportingandexportingdecisionsof rms.Earlierwork,forexample,Bernard,Jensen,andSchott(2009),hasdocumentedalargeoverlapintheimportandexportactivityof rms.4Indeed,majorexportersarealmostalwaysmajorimporters,andthisisalsotrueinourdataset.Wefocusexclusivelyonthealreadyselectgroupofexporters,mostofwhomarealsoimportersfrom 3Inparticular,weshowinthedatathatimport-intensiveexportershavelowerpass-throughduetogreatersensitivityoftheirmarginalcoststoexchangerates.Furthermore,theeectofimportintensityonpass-throughislargerwhentheimportandexportexchangeratesarecloselycorrelatedandwhenthepass-throughintotheimportpricesishigh,con rmingthetheoreticalpredictions.4OtherrelatedpapersincludeKuglerandVerhoogen(2009),ManovaandZhang(2009),Feng,Li,andSwenson(2012),andDamijan,Konings,andPolanec(2012).3 multiplesourcecountries.Weemphasizethestrongselectionthatstilloperateswithinthegroupofexporters,inparticulartheheterogeneityintheintensitywithwhich rmsimporttheirintermediateinputsanditsconsequencesforexport-pricepass-through.Second,ourpaperisrelatedtotherecentempiricalandstructuralworkontherelation-shipbetween rmimportintensityand rmproductivity.AlthoughwebaseourmodelonHalpern,Koren,andSzeidl(2011),whoestimatetheeectsofimportuseontotalfactorproductivityforHungarian rms,similarmodelsweredevelopedinAmitiandDavis(2012)tostudytheeectsofimporttarison rmwagesandinGopinathandNeiman(2012)tostudytheeectsoftheArgentinetradecollapsefollowingthecurrencydevaluationof2001ontheeconomy-wideproductivity.5Inourstudy,wefocusinsteadonhowtheinterplaybetweenimportintensityandmarkupvariabilitycontributestoincompleteexchangeratepass-through.Third,ourpapercontributestothevastliteratureontheexchangeratedisconnect(seeObstfeldandRogo,2001;Engel,2001)andinparticularontheincompletepass-throughofexchangerateshocksintointernationalprices.Inthepastdecade,substantialprogresshasbeenmadeinthestudyofthisphenomenon,boththeoreticallyandempirically.6Thisliteraturehasexploredthreechannelsleadingtoincompletepass-through.The rstchannel,assurveyedinEngel(2003),isshort-runnominalrigiditieswithpricesstickyinthelocalcurrencyofthedestinationmarket,labeledintheliteratureaslocalcurrencypricing(LCP).UnderLCP,the rmsthatdonotadjustpriceshavezeroshort-runpass-through.GopinathandRigobon(2008)providedirectevidenceontheextentofLCPinUSimportandexportprices.Thesecondchannelpricing-to-market(PTM)arisesinmodelsofvariablemarkupsinwhich rmsoptimallychoosedierentpricesfordierentdestinationsdependingonlocalmarketconditions.AtkesonandBurstein(2008)provideanexampleofarecentquantitativeinvestigationofthePTMchannelanditsimplicationforinternationalaggregateprices.7Finally,thethirdchannelofincompletepass-throughintoconsumerpricesoftenconsidered 5Blaum,Lelarge,andPeters(2013)documentstylizedfactsaboutimportbehaviorofFrench rmsandprovideanotherrelatedmodel.AmitiandKonings(2007)provideanempiricalanalysisofthemicro-leveleectsofimportson rmproductivity.6Forasurveyofearlierwork,seeGoldbergandKnetter(1997),whoinparticularemphasizethat[l]essisknownabouttherelationshipbetweencostsandexchangerates...(seep.1244).ThehandbookchapterbyBursteinandGopinath(2012)providesasummaryofrecentdevelopmentsinthisarea.7GopinathandItskhoki(2011)showtheimportanceofPTMinmatchingpatternsintheinternationalaggregateandmicropricedata.FitzgeraldandHaller(2012)providethemostdirectevidenceonPTMbycomparingtheexchangerateresponseofpricesofthesameitemsoldtoboththedomesticandtheinternationalmarket.Gopinath,Itskhoki,andRigobon(2010)andGopinathandItskhoki(2010)showthatthePTMandLCPchannelsofincompletepass-throughinteractandreinforceeachother,withhighlyvariable-markup rmsendogenouslychoosingtopriceinlocalcurrencyaswellasadoptinglongerpricedurations.4 intheliteratureislocaldistributioncosts,asforexampleinBurstein,Neves,andRebelo(2003)andGoldbergandCampa(2010).Ourimportedinputschannelissimilarinspirittothelocaldistributioncostsinthattheymakethecostsofthe rmmorestableinthelocalcurrencyofexportdestination.Thedierencewiththedistributioncostchannelisthattheuseofimportedinputsresultsinincompletepass-throughnotonlyintoconsumerprices,butalsointoproducerfactorygateprices.Arelatedlineofliterature,surveyedinGoldbergandHellerstein(2008),hasidenti edthePTMchannelbystructurallyestimatingindustrydemandtobackoutmodel-impliedmarkupsofthe rms.GoldbergandVerboven(2001)applythismethodologyinthecontextoftheEuropeancarmarket,whileNakamuraandZerom(2010)andGoldbergandHellerstein(2011)studythecoeeandthebeermarkets,respectively,incorporatingstickypricesintotheanalysis.Basedonthe ndingthatmarkupsdonotvaryenoughtocapturethevariationinexporterpricesarisingfromexchangeratevolatility,thesestudiesconcludetheremustbearesidualroleforthemarginalcostchannel,duetoeitherthelocaldistributionmarginorimportedinputs.Ourpaperisthe rsttodirectlyestimatetheimportanceofthemarginalcostchannelforincompletepass-throughintoexporterpricesarisingfromtheuseofimportedintermediateinputs.OurworkiscloselyrelatedtoBerman,Martin,andMayer(2012)inthatwealsostudythevariationinpass-throughacrossheterogeneous rms.Whiletheyfocusontheroleof rmproductivityandsize,weemphasizetheroleofimportedinputsanddestination-speci cmarketshares.8Ourapproachalsoenablesustoprovideaquantitativedecompositionofthecontributionofthemarginalcostandvariablemarkupchannelstoincompleteexchangeratepass-through.Therestofthepaperisstructuredasfollows.Section2laysoutthetheoreticalframe-workandprovidesthetheoreticalresultsthatmotivatetheempiricalanalysisthatfollows.Section3introducesthedatasetanddescribesthestylizedpatternsofcross-sectionalvaria-tioninthedata.Section4describesourmainempirical ndings.Section5concludes.Thetechnicalderivationsandadditionalresultsareprovidedintheonlineappendix.2TheoreticalFrameworkInthissection,wedevelopatheoreticalframeworklinkinga rm'sexchangeratepass-throughtoitsimportintensityandexportmarketshares,allofwhichareendogenously 8Anumberofearlierpapershavelinkedpass-throughwithmarketshareofexporters(seeFeenstra,Gagnon,andKnetter,1996;Alessandria,2004;Garetto,2012;AuerandSchoenle,2012).5 determined.Weusethisframeworktoformulatetestableimplicationsandtoderiveanequilibriumrelationship,whichwelaterestimateinthedata.WestartbylayingoutthetwomainingredientsofourframeworktheAtkesonandBurstein(2008)modelofstrategiccomplementaritiesandvariablemarkupsandtheHalpern,Koren,andSzeidl(2011)modelofthe rm'schoicetoimportintermediateinputs.Wethenshowhowtheinteractionofthesetwomechanismsgeneratesnewtheoreticalinsightsonthedeterminantsofexchangeratepass-through.Thekeypredictionsofthistheoryarethata rm'simportintensityandmarketsharesarepositivelycorrelatedinthecross-sectionandtogetherconstituteasu cientstatisticfortheexchangeratepass-throughofthe rmwithinsector-exportmarket,withimportintensitycapturingthemarginalcostsensitivitytotheexchangerateandmarketsharecapturingthemarkupelasticity.Wecharacterizetheequilibriumrelationshipbetweenmarketshare,importintensityandpass-through,whichholdsparametricallyindependentlyoftheparticulargeneralequilibriumenvironment.Theparametervaluesinthisrelationshipdependonthespeci csofthegeneralequilibrium,inparticulartheequilibriumco-movementbetweenaggregatevariablessuchasexchangerates,pricelevelsandcostindexes.Nonetheless,weneednottakeastandonthespeci cgeneralequilibriumassumptionsaswedirectlyestimatetheseparametersusingcross-sectionalvariationbetween rmswithinindustriesandexportdestinations,withoutrelyingonstrongpartialequilibriumorexchangerateexogeneityassumptions.Tofocusouranalysisontherelationshipbetweenimportintensityandpass-throughofthe rms,wemakeanumberofsimplifyingassumptions.First,weconditionouranalysisonthesubsetofexporting rms,andhencewedonotmodelentry,exit,orselectionintoexporting(as,forexample,inMelitz,2003),butratherfocusontheimportdecisionsofthe rms.Similarly,wedonotmodelthedecisiontoexporttomultipledestinations,butsimplytakethisinformationasexogenouslygiven.Furthermore,weassumeall rmsaresingle-product.Intheempiricalsectionwediscusstheimplicationsofrelaxingtheseassumptions.Second,weassume exiblepricesettingasinAtkesonandBurstein(2008)andhencedonotneedtocharacterizethecurrencychoice(i.e.,localversusproducercurrencypricing).Thismodelingchoiceismotivatedbythenatureofourdatasetinwhichweuseunitvaluesasproxiesforprices.Empirically,incompletepass-throughisatleastinpartduetopricestickinessinlocalcurrency,andinlightofthisweprovideacarefulinterpretationofourresultsinSection4.5.9 9Itisusefultokeepinmindthat,asshowninGopinath,Itskhoki,andRigobon(2010),the exible-pricepass-throughforcesshapethecurrencychoiceofthe rms,i.e. rmswithalowpass-throughconditionalonapricechangechoosetopriceinlocalcurrency,whichfurtherreducestheshort-runpass-throughofthese rms.Inthispaper,wefocusontheendogenousdeterminantsof exible-price(orlong-run)pass-through6 Last,whilethemarginalcostchannelemphasizedinthepaperisinherentlyamechanismofrealhedging,inmodeling rms'importdecisionsweabstractfromchoosingorswitchingimportsourcecountriestobetterhedgetheirexportexchangeraterisk.Empirically,we ndthatthepositivecorrelationbetweena rm'sdestination-speci cexchangerateanditsimport-weightedexchangeratedoesnotvarywiththemain rmvariablesthatarethefocusofouranalysis(seeSection4.4).102.1DemandandmarkupsConsidera rmproducingadierentiatedgoodiinsectorsandsupplyingittodestinationmarketkinperiodt.ConsumersineachmarkethaveanestedCESdemandoverthevarietiesofgoods,asinAtkesonandBurstein(2008).Theelasticityofsubstitutionacrossthevarietieswithinsectorsis,whiletheelasticityofsubstitutionacrosssectoralaggregatesis,andweassume�1.Underthesecircumstances,a rmifacesthefollowingdemandforitsproduct:Qk;i=k;iP�k;iP�kDk;(1)whereQk;iisquantitydemanded,k;iisarelativepreference(quality)parameterofthe rm,Pk;iisthe rm'sprice,Pkisthesectoralpriceindex,andDkisthesectoraldemandshifter,whichthe rmtakesasgiven.Indexkemphasizesthatallthesevariablesaredestinationspeci c.Forbrevity,wedroptheadditionalsubscriptssandtforsectorandtime,sinceallofouranalysisfocusesonvariationwithinagivensector.ThesectoralpriceindexisgivenbyPkPik;iP1�k;i1=(1�),wherethesummationisacrossall rmsinsectorsservingmarketkintimeperiodt,andwenormalizePik;i=1.Asaconvention,wequoteallpricesinthelocalcurrencyofthedestinationmarket.Animportantcharacteristicofthe rm'scompetitivepositioninamarketisitsmarketsharegivenby:Sk;iPk;iQk;i Pi0Pk;i0Qk;i0=k;iPk;i Pk1�2[0;1];(2) inthecross-sectionof rms,whichinthestickypriceenvironmentwouldalsocontributetotheprevalenceoflocalcurrencypricing,withthetwoforcesworkinginthesamedirection.10Notethatundertheassumptionofriskneutralityofthe rmandintheabsenceofliquidityconstraints(forexample,ofthetypemodeledinFroot,Scharfstein,andStein,1993), nancialhedgingconstitutesonlyasidebettothe rmanddoesnotaectitsimportandpricingdecisions.Fauceglia,Shingal,andWermelinger(2012)provideevidenceontheroleofimportedinputsinnaturalhedgingofexportexchangerateriskbySwiss rmsandMartinandMéjean(2012)providesurveyevidenceontheroleofcurrencyhedgingininternationaltransactionsof rmsintheEuroZone.7 wheremarketshareissector-destination-timespeci c.Theeectivedemandelasticityforthe rmisthenk;i�dlogQk;i dlogPk;i=(1�Sk;i)+Sk;i;(3)since@logPk=@logPk;i=Sk;i.Inwords,the rmfacesademandelasticitythatisaweightedaverageofthewithin-sectorandtheacross-sectorelasticitiesofsubstitutionwiththeweightonthelatterequaltothemarketshareofthe rm.High-market-share rmsexertastrongerimpactonthesectoralpriceindex,makingtheirdemandlesssensitivetotheirownprice.When rmscompeteinprices,theysetamultiplicativemarkupMk;ik;i=(k;i�1)overtheircosts.Firmsfaceademandwithelasticitydecreasinginthemarketshare,andhencehigh-market-share rmschargehighmarkups.Wenowde neameasureofthemarkupelasticitywithrespecttothepriceofthe rm,holdingconstantthesectorpriceindex:11�k;i�@logMk;i @logPk;i=Sk;i ��Sk;i1�� �1Sk;i�0:(4)Alowerpricesetbythe rmleadstoanincreaseinthe rm'smarketshare,makingoptimalalargermarkup.Furthermore,themarkupelasticityisalsoincreasinginthemarketshareofthe rm: rmswithlargermarkupschoosetoadjustthembymoreinresponsetoshocksandtokeeptheirpricesandquantitiesmorestable.Wesummarizethisdiscussionin:Proposition1Marketshareofthe rmSk;iisasu cientstatisticforitsmarkup;bothmarkupMk;iandmarkupelasticity�k;iareincreasinginthemarketshareofthe rm.Thistheoreticalframeworkhastwosharppredictionsaboutthemarkup.First,thevariationinthemarketsharefullycharacterizesthevariationinthemarkupelasticityacross rms.Aswediscussintheintroduction,thisislessthangeneral,andalternativedemandstructuresemphasizeotherdeterminantsofmarkupvariability.Nonetheless,ourempiricalanalysisshowsthatthispredictionprovidesausefulapproximationacrossbroadsectorsandmultipleexportdestinations.Second,markupvariabilityismonotonicallyincreasinginthemarketshare.Althoughthispredictionisalsomodel-speci c,ourempiricalanalysisprovidessupportforthismonotonicrelationship(seeSection4.3). 11Wefocusonthispartialmeasureofmarkupelasticitybecausethisistherelevantmeasureinourempiricalanalysis,wheretheidenti cationstrategyexploitsvariationwithinindustriesacross rmsthatallfacethesamepriceindex.Itshouldbenotedthatanalternativemeasureofmarkupelasticity,whichdoesnotholdthesectorpriceindex xed,ingeneralresultsinaU-shapedrelationshipbetweenpass-throughandmarketshare(as,forexample,inGaretto,2012,andAuerandSchoenle,2012).8 2.2ProductionandimportedinputsWebuildonHalpern,Koren,andSzeidl(2011)tomodelthecoststructureofthe rmanditschoicetoimportintermediateinputs.Considera rmi,whichuseslaborLiandintermediateinputsXitoproduceitsoutputYiaccordingtotheproductionfunction:Yi= iX iL1� i;(5)where iis rmproductivity.Parameter 2[0;1]measurestheshareofintermediateinputsin rmexpenditureandissectorspeci cbutcommontoall rmsinthesector.Intermediateinputsconsistofabundleofintermediategoodsindexedbyj2[0;1]andaggregatedaccordingtoaCobb-Douglastechnology:Xi=exp10 jlogXi;jdj:(6)Thetypesofintermediateinputsvaryintheirimportanceintheproductionprocessasmeasuredby j,whichsatisfy10 jdj=1.Eachtypejofintermediategoodcomesintwovarietiesdomesticandforeignwhichareimperfectsubstitutes:Xi;j=Z 1+i;j+a1 1+jM 1+i;j1+ ;(7)whereZi;jandMi;jare,respectively,thequantitiesofdomesticandimportedvarietiesoftheintermediategoodjusedinproduction.Theelasticityofsubstitutionbetweenthedomesticandtheforeignvarietiesis(1+)�1,andajmeasurestheproductivityadvantage(whenaj�1,anddisadvantageotherwise)oftheforeignvariety.Notethatsincehomeandforeignvarietiesareimperfectsubstitutes,productionispossiblewithouttheuseofimportedinputs.Atthesametime,importedinputsareusefulduebothtotheirpotentialproductivityadvantageajandtothelove-of-varietyfeatureoftheproductiontechnology(7).A rmneedstopaya rm-speci csunkcostfiintermsoflaborinordertoimporteachtypeoftheintermediategood.ThecostoflaborisgivenbythewagerateW,andthepricesofdomesticintermediatesarefVjg,bothdenominatedinunitsofproducercurrency(hencestarred).ThepricesofforeignintermediatesarefEmUjg,whereUjisthepriceinforeigncurrencyandEmisthenominalexchangeratemeasuredasaunitofproducercurrencyforoneunitofforeigncurrency.12Thetotalcostofthe rmisthereforegivenby 12Wedenotebymagenericsourceofimportedintermediates,andhenceEmcanbethoughtofasanimport-weightedexchangeratefacedbythe rms.Thegeneralizationofthemodeltomultipleimport9 WLi+10VjZi;jdj+J0;i�EmUjMi;j+Wfidj,whereJ0;idenotesthesetofintermediatesimportedbythe rm.Withthisproductionstructure,wecanderivethecostfunctionofthe rm.Inparticular,givenoutputYiandthesetofimportedintermediatesJ0;i,the rmchoosesinputstominimizeitstotalcostssubjecttotheproductiontechnologyinequations(5)(7).Thisresultsinthefollowingtotalvariablecostfunctionnetofthe xedcostsofimporting:TVCi(YijJ0;i)=C B i iYi;(8)whereCisthecostindexforanon-importing rm.13Theuseofimportedinputsleadstoacost-reductionfactorBiB(J0;i)=expnJ0;i jlogbjdjo,wherebj1+aj(EmUj=Vj)�1=istheproductivity-enhancingeectfromimportingtype-jintermediategood,adjustedfortherelativecostoftheimportvariety.Wenowdescribetheoptimalchoiceofthesetofimportedintermediategoods,J0;i.Forsimplicity,wediscussherethecasewithoutuncertainty,andtheappendixgeneralizestheresults.First,wesortallintermediategoodsjby jlogbj,fromhighesttolowest.Then,theoptimalsetofimportedintermediateinputsisanintervalJ0;i=[0;j0;i],withj0;i2[0;1]denotingthecutointermediategood.Theoptimalchoiceofj0;itradesothe xedcostofimportingWfiforthereductionintotalvariablecostsfromtheaccesstoanadditionalimportedinput,whichisproportionaltothetotalmaterialcostofthe rm.14Thisre ectsthestandardtrade-othatthe xedcostactivityisundertakenprovidedthatthescaleofoperation(heretotalspendingonintermediateinputs)issu cientlylarge.Withthiscoststructure,thefractionoftotalvariablecostspentonimportedintermediateinputsequals:'i= j0;i0 j�1�b�jdj;(9)where istheshareofmaterialcostintotalvariablecostand j(1�b�j)istheshareofmaterialcostspentonimportsoftype-jintermediategoodforj2J0;i.Wereferto'iastheimportintensityofthe rm,anditisoneofthecharacteristicsofthe rmwemeasuredirectlyinthedata. sourcecountriesisstraightforward;inthedata,wemeasureEmasanimport-weightedexchangerateatthe rm-level,aswellassplitimportsbysourcecountry(seeSection4.4).13ThiscostindexisgivenbyC=�V= �W=(1� )1� withV=exp10 jlog�Vj= jdj .14Themarginalimportedinputsatis es j0;ilogbj0;iTMCi=Wfi,wheretheleft-handsideistheincrementalbene tproportionaltothetotalmaterialcostofthe rmTMCi CYi=B i iandthecost-savingimpactofadditionalimports j0;ilogbj0;i.10 Finally,holdingthesetofimportedvarietiesJ0;iconstant,thiscoststructureresultsinthefollowingmarginalcost:MCi=C=B i i:(10)ThepartialelasticityofthismarginalcostwithrespecttotheexchangerateEmequalstheexpenditureshareofthe rmonimportedintermediateinputs,'i=@logMCi=@logEm,whichemphasizestheroleofimportintensityintheanalysisthatfollows.Wesummarizetheseresultsin:Proposition2(i)Withinsectors, rmswithlargertotalmaterialcostorsmaller xedcostofimportinghavealargerimportintensity,'i.(ii)Thepartialelasticityofthemarginalcostofthe rmwithrespecttothe(import-weighted)exchangerateequals'i.2.3EquilibriumrelationshipsWenowcombinetheingredientsintroducedabovetoderivetheoptimalpricesettingofthe rm,aswellastheequilibriumdeterminantsofthemarketshareandimportintensityofthe rm.Consider rmisupplyinganexogenouslygivensetKiofdestinationmarketsk.The rmsetsdestination-speci cpricesbysolvingmaxYi;fPk;i;Qk;igk(Xk2KiEkPk;iQk;i�C B i iYi);subjecttoYi=Pk2KiQk;ianddemandequation(1)ineachdestinationk.Wequotethedestination-kpricePk;iintheunitsofdestination-klocalcurrencyandusethebilateralnominalexchangerateEktoconvertthepricetotheproducercurrency,denotingwithPk;iEkPk;itheproducer-currencypriceofthe rmfordestinationk.AnincreaseinEkcorrespondstothedepreciationoftheproducercurrency.Thetotalcostofthe rmisquotedinunitsofproducercurrencyandhenceisstarred.15NotethatwetreatthechoiceofthesetofimportedgoodsJ0;iandtheassociated xedcostsassunkbythepricesettingstagewhentheuncertaintyaboutexchangeratesisrealized.TheproblemofchoosingJ0;iunderthiscircumstanceisde nedandcharacterizedintheappendixandSection3.2providesempiricalevidencesupportingthisassumption.Takingthe rst-orderconditionswithrespecttoPk;i,weobtaintheoptimalpricesetting 15Wedonotexplicitlymodelvariabletradecosts,butiftheytakeanicebergform,theyarewithoutlossofgeneralityabsorbedintothek;iDkterminthe rm-idemand(1)indestinationk.11 conditions:Pk;i=k;i k;i�1MCi=Mk;iC B i i;k2Ki;(11)whereMCiisthemarginalcostasde nedin(10)andMk;i=k;i=(k;i�1)isthemultiplica-tivemarkupwiththeeectivedemandelasticityk;ide nedin(3).Thissetof rst-orderconditionstogetherwiththeconstraintsfullycharacterizestheallocationofthe rm,givenindustry-levelvariables.Intheappendix,weexploittheseequilibriumconditionstoderivehowrelativemarketsharesandimportintensitiesaredeterminedinequilibriumacross rms,andsincetheseresultsareveryintuitive,hereweprovideonlyabriefsummary.Weshowthatotherthingsequalandundermildregularityconditions,a rmwithhigherproductivity i,higherquality/demandk;i,lower xedcostofimportingfi,andservingalargersetofdestinationsKihasalargermarketshareSk;iandahigherimportintensity'i.Intuitively,amoreproductiveorhigh-demand rmhasalargermarketshareandhenceoperatesonalargerscalewhichjusti espayingthe xedcostforalargersetofimportedintermediateinputs,J0;i.Thismakesthe rmmoreimportintensive,enhancingitsproduc-tivityadvantagethroughthecost-reductioneectofimports(largerBiin(8)).Thisimpliesthat,independentlyofthespeci csofthegeneralequilibriumenvironment,weshouldexpectmarketsharesandimportintensitiestobepositivelycorrelatedinthecross-sectionof rms,apatternthatwedocumentinthedatainSection3.2.4Importedinputs,marketshare,andpass-throughWearenowinapositiontorelatethe rm'sexchangeratepass-throughintoitsexportpriceswithitsmarketshareandimportintensity.Thestartingpointforthisanalysisistheoptimalpricesettingequation(11),whichwerewriteasafulllogdierential:dlogPk;i=dlogMk;i+dlogMCi:(12)Consider rstthemarkupterm.Using(2)(4),wehave:dlogMk;i=��k;i�dlogPk;i�dlogPs;k+�k;i �1dlogk;i;(13)whereconvertingtheexportpricetolocalcurrencyyieldsdlogPk;i=dlogPk;i�dlogEk,andwenowmakeexplicitthesubscriptsindicatingthatPs;kistheindustry-destination-speci cpriceindex.Themarkupdeclinesintherelativepriceofthe rmandincreasesinthe rm'sdemandshock.FromProposition1,�k;iisincreasinginthe rm'smarketshare,12 andhencepriceincreasesforlargermarket-share rmsareassociatedwithlargerdeclinesinthemarkup.Next,thechangeinthemarginalcostinequation(10)canbedecomposedasfollows:dlogMCi='idlogEmUs Vs+dlogCs s+MCi:(14)ThisexpressiongeneralizestheresultofProposition2ontheroleofimportintensity'ibyprovidingthefulldecompositionofthechangeinthelogmarginalcost.HereUsandVsarethepriceindexesfortheimportedintermediates(inforeigncurrency)anddomesticinterme-diates(inproducercurrency),respectively.Thesubscriptsemphasizesthattheseindexescanbespeci ctosectorsinwhich rmioperates.Finally,dlogCs= sisthelogchangeintheindustry-averagemarginalcostfora rmthatdoesnotimportanyintermediates,andMCiisa rm-idiosyncraticresidualtermde nedexplicitlyintheappendixandassumedorthogonalwiththeexchangerate.16Combiningandmanipulatingequations(12)(14),weproveourkeytheoreticalresult:Proposition3Inanygeneralequilibrium,the rst-orderapproximationtotheexchangeratepass-throughelasticityintoproducer-currencyexportpricesofthe rmisgivenby k;iEdlogPk;i dlogEk= s;k+s;k'i+ s;kSk;i;(15)where( s;k;s;k; s;k)aresector-destinationspeci canddependonlyonaveragemomentsofequilibriumco-movementbetweenaggregatevariablescommontoall rms.Thepass-throughelasticity k;imeasurestheequilibriumlogchangeofthedestination-kproducer-currencypriceof rmirelativetothelogchangeinthebilateralexchangerate,averagedacrossallpossiblestatesoftheworldandshocksthathittheeconomy.17Underthisde nition,thepass-throughelasticityisameasureofequilibriumco-movementbetweenthepriceofthe rmandtheexchangerate,andnotapartialequilibriumresponsetoanexogenousmovementintheexchangerate.Thus,wedonotneedtoassumethatmovements 16ThisorthogonalityassumptionisnecessaryforProposition3,anditholdsprovidedthat rmidiosyn-craticshocksarenotsystematicallycorrelatedwithexchangeratemovements.Moreprecisely,weallow rmdemand,productivityandimportpricestomovewithexchangeratesinarbitraryways,butwerequirethattherelativedemand,productivityandimportpricesofanytwo rmswithinsector-destinationdonotsystematicallychangewithexchangeratemovements.17Formally,theexpectationin(15)isoverallpossibleexogenousshocksthataecttheexchangerateingeneralequilibrium,aswellasoverallstatevariables(e.g.,thedistributionof rmproductivities).Further,theexpectationin(15)isconditionalonthepersistentcharacteristicsofagiven rm( i;k;i;fi),which,asProposition3emphasizes,aectthe rm'spass-throughonlythroughthesu cientstatistic('i;Sk;i).13 inexchangeratesareexogenous,nordowerelyonanysourceofexogenousvariationinourestimation.Infact,weallowforarbitraryequilibriumco-movementbetweenexchangeratesandotheraggregatevariables,includingpriceandcostindexes.Proposition3showsthatwecanrelate rm-levelpass-throughtomarketshareandimportintensityofthe rm,whichformasu cientstatisticforcross-sectionvariationinpass-throughwithinsector-destination,independentlyofthespeci csofthegeneralequilibriumenvironmentand,inparticular,independentlyofwhatshockshittheeconomyandshapethedynamicsoftheexchangerate.Thisrelationshipisthefocusofourempiricalinvestigationinthenextsection.Thevaluesofthecoe cientsinthisrelationship( s;k;s;k; s;k)aresector-destinationspeci c,and,althoughtheydependonthegeneralequilibriumenvironmentinwhich rmsoperate,weshowinSection4.1thatwecandirectlyestimatetheminthedatawithoutimposinganygeneralequilibriumassumptions.18Furthermore,thetheoryprovidesclosedformexpressionsforthecoe cientsin(15).Thecoe cients s;kand s;kdependontheunconditionalmomentsofequilibriumco-movementbetweentheexchangerateandaggregatevariablessuchasthepriceindexinthedestina-tionmarketandthedomesticcostindex(seeappendix).Forexample,pass-throughintodestinationpricesislower(i.e., s;kishigher)inanequilibriumenvironmentwherethedo-mesticcostindexosetssomeoftheeectsoftheexchangerateonmarginalcosts.Andpass-throughisrelativelylower(i.e., s;kishigher)forlargemarket-share rmswhenthedes-tinationpriceindexrespondsweaklytotheexchangerate,becauseofthestrongerstrategiccomplementaritiesintheirpricesetting.Theclosedformexpressionfors;khighlightsthestructuraldeterminantsoftherelation-shipbetweenpass-throughandimportintensity:s;k=1 1+�s;kEdlogEm dlogEkdlog(EmUs=Vs) dlogEm;(16)where�s;kisthemarkupelasticityevaluatedatsomeaveragemeasureofmarketshareSs;k.Intuitively,s;kdependsontheco-movementbetweenexportandimportexchangeratesandthepass-throughofimportexchangerateintotherelativepriceofimportedintermediates,asre ectedbythetwotermsinsidetheexpectationin(16).Indeed,importintensityproxiesforthemarginalcostcorrelationwiththeexportexchangerateonlytotheextentthatimport 18Weestimatethecoe cientsbyexploitingthecross-sectionalvariationinthepaneldatawithinindustriesanddestinationsintheresponsestoshocksacross rms.Theseestimates,however,arenotsuitableforundertakingcounterfactualsacrossgeneralequilibriumenvironments,aswhenthesourceofvariationinexchangeratechanges,sodothecoe cients.Suchcounterfactualsneedtobecarriedoutinthecontextofaspeci ccalibratedgeneralequilibriummodel.14 andexportexchangeratesarecorrelatedandmovementsintheimportexchangerateareassociatedwiththechangesintheimportpricesoftheintermediateinputs.Importantly,wedonotrestricttheexchangeratepass-throughintoimportpricestobecomplete.Inourempiricalwork,wedirectlyexamineourtheoreticalmechanismbyusingthevariationinexchangeratecorrelationandimportpricepass-throughacrossimportsourcecountries.Proposition3predictsthatpass-throughintodestination-currencyprices,equalto(1� k;i),islowerformoreimport-intensive rmsandfor rmswithhigherdestinationmarketshare,provideds;kand s;karepositive.19Althoughtheoreticallythesecoe cientscanhaveei-thersign(orequaltozero),weexpectthemtobepositive,andthisiswhatwe ndinthedata.Intuitively,amoreimport-intensive rmiseectivelyhedgedfromadomesticcurrencyappreciationviadecreasingimportpricesandhencekeepsitsdestination-currencypricemorestable.Further,ahigh-market-share rmhasalowerpass-throughasitchoosestoaccommodatethemarginalcostshockswithalargermarkupadjustmentasopposedtoalargermovementinitsdestination-currencyprice.3DataandStylizedFactsInthissectionwedescribethedatasetthatweuseforourempiricalanalysisandthebasicstylizedfactsonexportersandimporters.3.1DatadescriptionandconstructionofvariablesOurmaindatasourceistheNationalBankofBelgium,whichprovidedacomprehensivepanelofBelgiantrade owsby rm,product(CN8-digitlevel),exportsbydestination,andimportsbysourcecountry.Wemergethesedata,usingaunique rmidenti er,with rm-levelcharacteristicsfromtheBelgianBusinessRegistry,comprisinginformationon rms'inputs,whichweusetoconstructtotalcostmeasuresandtotalfactorproductivityestimates.Oursampleincludesannualdatafortheperiod2000to2008,beginningtheyearaftertheeurowasintroduced.WefocusonmanufacturingexportstotheOECDcountriesoutsidetheEuroZone:Australia,Canada,Iceland,Israel,Japan,theRepublicofKorea,NewZealand,Norway,Sweden,Switzerland,theUnitedKingdomandtheUnitedStates,accountingfor 19NotethatProposition3providesalinearapproximation(15)tothegenerallynonlinearequilibriumrelationship.Inthedata,wetestdirectlyforthenonlinearityinthisrelationshipand ndnostatisticallysigni cantevidence.Ourinterpretationisthatthisisnotbecausenonlinearitiesareunimportantingeneral,butbecausethese nerfeaturesoftheequilibriumrelationshiparelessrobustacrossbroadindustriesthatweconsiderinouranalysis.Thisiswhywechoosetofocusonthe rst-orderqualitativepredictionofthetheorycapturedbytheapproximationin(15).15 58percentoftotalnon-Euroexports.20Wealsoincludearobustnesstestwiththefullsetofnon-Eurodestinations.Weprovideadetaileddescriptionofallthedatasourcesinthedataappendix.Thedependentvariableinouranalysisisthelogchangeina rmf'sexportpriceofgooditodestinationcountrykattimet,proxiedbythechangeina rm'sexportunitvalue,de nedastheratioofexportvaluestoexportquantities:pf;i;k;tlogExportvaluef;i;k;t Exportquantityf;i;k;t;(17)wherequantitiesaremeasuredasweightsorunits.Weusethechangeintheratioofvaluetoweights,whereavailable,andthechangeintheratioofvaluetounitsotherwise.WenotethatunitvaluesareanimpreciseproxyforpricesbecausetheremaybemorethanonedistinctproductwithinaCN8-digitcodedespitethehighdegreeofdisaggregationconstitutingcloseto10,000distinctmanufacturingproductcategoriesoverthesampleperiod.Somepricechangesmaybeduetocompositionalchangeswithinaproductcodeorduetoerrorsinmeasuringquantities.21Totrytominimizethisproblem,wedropallyear-to-yearunitvaluechangesofplus200percentorminus67percent(aroundsevenpercentoftheobservations.)Adistinctivefeatureofthesedatathatiscriticalforouranalysisisthattheyalsocontain rm-levelimportvaluesandquantitiesforeachCN8-digitproductcodebysourcecountry.Weincludeall234sourcecountriesandall13,000productcodesinthesample.Studiesthatdrawonpricedatahavenotbeenabletomatchimportandexportpricesatthe rmlevel.Ingeneral,many rmsengagedinexportingalsoimporttheirintermediateinputs.InBelgium,around80percentofmanufacturingexportersimportsomeoftheirinputs.WeusethesedatatoconstructthreekeyvariablestheimportintensityfromoutsidetheEuroZone'f;t,thelogchangeinthemarginalcostmcf;tandthe rm'smarketshareSf;s;k;t. 20TheEuroZonewasformedonJanuary1,1999,inAustria,Belgium,Finland,France,Germany,Ireland,Italy,Luxembourg,theNetherlands,Portugal,andSpain.GreecejoinedonJanuary1,2001,Sloveniajoinedin2007,CyprusandMaltajoinedin2008.WealsoexcludeDenmarkfromthesetofexportdestinationsbecauseitsexchangeratehardlymovesrelativetotheEuro.21Thisisthetypicaldrawbackofcustomsdata(as,forexmaple,isalsothecasewiththeFrenchdatasetusedinBerman,Martin,andMayer,2012),wheredespitetherichnessof rm-levelvariables,wedonotobservetradepricesofindividualitems.Asaresult,therearetwopotentialconcerns:one,aggregationacrossheterogeneousgoodsevenatthevery nelevelofdisaggregation( rm-destination-CN8-digitproductcodelevel);and,two,aggregationovertimeofstickyprices.Thismeanswecannotconditionouranalysisonapricechangeofagood,aswasdoneinGopinath,Itskhoki,andRigobon(2010)usingBLSIPPitem-leveldata.TheBLSdata,however,islimitedintheavailable rmcharacteristicsandhenceisnotsuitableforouranalysis.Weaddressthesetwocaveatsbyconductinganumberofrobustnesstestsandprovidingacautiousinterpretationofour ndingsinSection4.5.16 Speci cally,'f;tTotalnon-euroimportvaluef;t Totalvariablecostsf;t;(18)wheretotalvariablecostscomprisea rm'stotalwagebillandtotalmaterialcost.Weoftenaveragethismeasureovertimetoobtaina rm-levelaverageimportintensitydenotedby'f.Thechangeinmarginalcostisde nedasthelogchangeinunitvaluesof rmimportsfromallsourcecountriesweightedbyrespectiveexpenditureshares:mcf;tXj2Jf;tXm2Mf;t!f;j;m;tlogUf;j;m;t;(19)whereUf;j;m;tistheeuroprice(unitvalue)of rmf'simportsofintermediategoodjfromcountrymattimet,theweights!f;j;m;taretheaverageofperiodtandt�1sharesofrespectiveimportvaluesinthe rm'stotalvariablecosts,andJf;tandMf;tdenotethesetofallimportedgoodsandimportsourcecountries(includinginsidetheEuroZone)forthe rmatagiventime.Notethatthismeasureofthemarginalcostisstillaproxysinceitdoesnotre ectthecostsofdomesticinputsand rmproductivity.Wecontrolseparatelyforestimated rmproductivityandaverage rmwagerate,however,detaileddataonthepricesandvaluesofdomesticinputsarenotavailable.Nonetheless,controllingforourmeasureofthe rm-levelmarginalcostisasubstantialimprovementoverpreviouspass-throughstudiesthattypicallycontrolonlyfortheaggregatemanufacturingwagerateorproducerpricelevel.Furthermore,ourmeasureofmarginalcostarguablycapturesthecomponentofthemarginalcostmostsensitivetoexchangeratemovements.Ideally,wewouldliketoconstruct'f;tandmcf;tforeachoftheproductsia rmproduces;however,thismeasureisavailableonlyatthe rm-flevel,whichmaynotbethesameforalloftheproductsproducedbymultiproduct rms.Toaddressthismultiproductissue,wekeeponlythe rm'smainexportproducts,whichweidentifyusingBelgium'sinput-outputtablefortheyear2005,comprising56IOmanufacturingcodes.Foreach rm,weidentifyanIOcodethataccountsforitslargestexportvalueoverthewholesampleperiodandkeeponlytheCN8-digitproductswithinthismajor-IOcode.Theobjectiveistokeeponlythesetofproductsforeach rmthathavesimilarproductiontechnologies.Thisleavesuswith60percentoftheobservationsbut90percentofthevalueofexports.Wealsopresentresultswiththefullsetofexportproductsandexperimentwithde ningthemajorproductusingmoredisaggregatedproductlines,suchasHS4-digit.22Further,itispossible 22Byonlykeepingthe rm'smajorproducts,wealsodealwiththepotentialproblemofincludingproductsthat rmsexportbutdonotproduceaphenomenonreferredtoascarry-alongtrade(Bernard,Blanchard,VanBeveren,andVandenbussche,2012).Asafurtherrobustnesscheck,inSection4.5weuseIOtablesto17 thatsomeofthe rm'simportsmightbe nalgoodsratherthanintermediateinputs.Weattempttoidentifyimportedintermediateinputsusinganumberofdierentapproaches.First,weomitanyimportfromtheconstructionof'f;tthatisde nedasa nalproductusingBroadEconomicCodes(BEC).Second,weconstruct'fusingonlytheintermediateinputsforagivenindustryaccordingtotheIOtables.Thelastkeyvariableinouranalysisisa rm'smarketshare,whichweconstructasfollows:Sf;s;k;tExportvaluef;s;k;t Pf02Fs;k;tExportvaluef0;s;k;t;(20)wheresisthesectorinwhich rmfsellsproductiandFs;k;tisthesetofBelgianexporterstodestinationk,insectorsattimet.Therefore,Sf;s;k;tmeasuresaBelgium rm'smarketshareinsectors,exportdestinationkattimetrelativetoallotherBelgiumexporters.Notethat,followingthetheory,thismeasureisdestinationspeci c.Thetheoryalsosuggeststhattherelevantmeasureisthe rm'smarketsharerelativetoall rmssupplyingthedestinationmarketinagivensector,includingexportersfromothercountriesaswellasdomesticcompetitorsinmarketk.But,sinceouranalysisisacrossBelgianexporterswithinsector-destinations,thecompetitivestanceinaparticularsector-destinationiscommonforallBelgianexporters,andhenceourmeasureofSf;s;k;tcapturesallrelevantvariationforouranalysis(seeSection4.1).23Wede nesectorsattheHS4-digitlevel,atwhichwebothobtainanontrivialdistributionofmarketsharesandavoidhavingtoomanysector-destinationsservedbyasingle rm.243.2StylizedfactsaboutexportersandimportersAsalientpatterninourdatasetisthatmostexportersarealsoimporters,apatternalsopresentinmanyearlierstudiescitedintheintroduction.AsreportedinTable1,inthefullsampleofBelgianmanufacturing rms,thefractionof rmsthatareeitherexportersor isolatetheinputsthatareusedintheproductionof rms'majorproducts,andonlyusetheseinputstoconstructthe rm'simportintensityvariable.SeeDeLoecker,Goldberg,Khandelwal,andPavcnik(2012)foranalternativestructuraltreatmentofmultiproduct rms.23Inanextensionofthetheory,amultiproduct rmfacinganested-CESdemand(1)setsthesamemarkupforallitsvarietieswithinasector,asin(11),whereitsmarkupdependsonthecumulativemarketshareofallthesevarieties.Therefore,Sf;s;k;tisindeedtheappropriatemeasureofmarketpowerforallvarietiesiexportedby rmftodestinationkinsectorsattimet.SeeChatterjee,Dix-Carneiro,andVichyanond(2012)forastudyofpass-throughofmultiproduct rmsunderanalternativedemandstructure.24ThemedianofSf;s;k;tis7.8%,yetthe75thpercentileisover40%andtheexport-value-weightedmedianis55%.24%ofSf;s;k;tobservationsarelessthan1%,yettheseobservationsaccountforonly1.4%ofexportsales.3%ofSf;s;k;tobservationsareunity,yettheyaccountforlessthan2.5%.Ourresultsarerobust(and,infact,becomemarginallystronger)totheexclusionofobservationswithverysmallandverylargemarketshares.WedepictthecumulativedistributionfunctionofSf;s;k;tinFigureA1intheappendix.18 Table1:Exporterandimporterincidence ExportersAlland/orimportersexporters Fractionofall rms32.6%23.7%ofthem:exportersandimporters57.0%78.4%onlyexporters15.8%21.6%onlyimporters27.2% Note:Thedataincludesallmanufacturing rms.Thefrequenciesareaveragedovertheyears20002008.importersis33%.Outofthese rms,57%bothimportandexport,28%onlyimportand16%onlyexport.Thatis,22%ofmanufacturing rmsinBelgiumexportand78%ofexportersalsoimport.25Thishighcorrelationbetweenexportingandimportingsuggeststhattheselectionintobothactivitiesisdrivenbythesame rmcharacteristics,suchasproductivity,whichdeterminesthescaleofoperations.Weshowhowthisoverlapinimportingandexportingactivitiesturnsouttobeimportantforunderstandingtheincompleteexchangeratepass-through.Interestingly,thedatarevealalotofheterogeneitywithinexporting rms,whichareanalreadyveryselectsubsampleof rms.Thelargedierencesbetweenexportersandnonexportersarealreadywell-knownandarealsoprevalentinourdata.Thenewstylizedfactswehighlightherearethelargedierenceswithinexportersbetweenhighandlowimport-intensityexporting rms.WeshowinTable2thatthesetwogroupsofexporting rmsdierinfundamentalways.Wereportvarious rm-levelcharacteristicsforhighandlowimport-intensityexporters,splittingexportersintotwogroupsbasedonthemedianimportintensityoutsidetheEuroZone('f)equalto4.2%.26Forcomparison,wealsoreporttheavailableanalogousstatisticsfornon-exporting rmswithatleast5employees.FromTable2,weseethatimport-intensiveexportersoperateonalargerscaleandaremoreproductive.Theshareofimportedinputsintotalcostsforimport-intensiveexportersis37%comparedto17%fornonimport-intensiveexporters,andsimilarlyforimportssourcedoutsidetheEuroZoneitis17%comparedto1.2%.Andofcourse,thesenumbersaremuchlowerfornon-exportersat1.6%forimportsoutsideBelgiumand0.3%forimportsoutsidetheEuroZone.Import-intensiveexportersare2.5timeslargerinemploymentthannon-import- 25Thesestatisticsareaveragedoverthesamplelength,buttheyareverystableyear-to-year.InthesubsampleofexportersweuseforourregressionanalysisinSection4.2,thefractionofimporting rmsissomewhathigherat85.5%,re ectingthefactthatdataavailabilityisslightlybiasedtowardlarger rms.26Theunitofobservationhereisa rm-year.Ifwesplitoursamplebasedon rm-product-destination-year(whichistheunitofobservationinourregressionanalysis),themedianimportintensityishigherat8.2%,however,thishasnomaterialconsequencesforthepatternswedocumentinTable2.19 Table2:High-versuslow-import-intensityexporters Exporters ImportNotimportNon-exportersintensiveintensive Shareoftotalimportsintotalcost0.370.170.02Shareofnon-Euroimportsintotalcost('f)0.170.010.00 Employment(#full-timeequiv.workers)270.6112.220.7Averagewagebill(thousandsofEuros)48.842.334.9Materialcost103.328.23.0TotalFactorProductivity(log)0.330.07 Totalexports(manufacturinggoods)66.514.1tonon-EuroOECDcountries14.42.4Totalimports(allproducts)49.36.8outsideEuroZone20.80.5#ofimportsourcecountries14.46.6#ofHS8-digitproductsimported79.338.2 Note:Theexportersub-sampleissplitatthemedianofnon-Euroimportintensity(shareofnon-Euroimportsintotalcosts)equalto4.2%.Thenon-exportersubsampleisallnon-exportingmanufacturing rmswith5ormoreemployees.Materialcost,importandexportvaluesareinmillionsofEuros.28%oflowimportintensity rmsdonotimportatall,and33%ofthemdonotimportfromoutsidetheEuroZone.TheconstructionofthemeasuredTFPisdescribedinthedataappendix.intensiveexportersand13timeslargerthannon-exporters;theypaya15percentwagepremiumrelativetonon-import-intensive rmsanda40percentwagepremiumrelativetonon-exporters.Similarly,import-intensiveexportershavemuchlargertotalmaterialcosts,totalfactorproductivity,andmarketshare.These rmsalsoexportandimportonamuchlargerscale,andaremorelikelytotradewithmoredistantcountriesoutsidetheEuroZone.These rmsimportmoreintermsoftotalvalue,intermsofnumberofimportedgoods,andfromalargersetofimportsourcecountries.Theseresultshighlightthatbothtypesofexporting rmsareactiveinimportingfromarangeofcountriesbothwithinandoutsidetheEuroZonebutthatthetwotypesof rmsdiersubstantiallyinimportintensity,consistentwiththepredictionsofourtheoreticalframework.WenowprovidemoredetailsonthedistributionofimportintensityoutsidetheEuroZone('f)amongtheexporting rmsanditsrelationshipwithother rm-levelvariables.WeseethatthedistributionofimportintensityamongexportersinTable3,althoughsomewhatskewedtowardzero,hasawidesupportandsubstantialvariation,whichweexploitinourregressionanalysisinSection4.Over24%ofexportersdonotimportfromoutsidetheEuroZone,yettheyaccountforonly1%ofBelgianmanufacturingexports.Forthemajorityof rms,theshareofimportedinputsintotalcostsrangesbetween0and10%,whilethe20 Table3:Distributionofimportintensityamongexporters fractionfractionof# rmsof rmsexportvalue 'f=071724.9%1.4%0'f0:11,47851.3%38.2%0:1'f0:234812.1%23.8%0:2'f0:31555.4%8.8%0:3'f0:4943.3%23.0%'f&#x-278;0:4893.1%4.9% Note:Importintensity,'f,istheshareofimportedintermediateinputsfromoutsidetheEuroZoneinthetotalvariablecostofthe rm,averagedoverthesampleperiod.export-value-weightedmedianofimportintensityis13%.Atthesametime,nearly28%ofexportsalesaregeneratedbythe rmswithimportintensityinexcessof30%.27Wefurtherdepictthecumulativedistributionfunctionofimportintensity'finFigureA1intheappendix,whichalsoprovidesacumulativedistributionfunctionforourmarketsharevariableSf;s;k;t.Table4displaysthecorrelationsofimportintensitywithother rm-levelvariablesinthecross-sectionof rms.Con rmingthepredictionsofSection2.3,importintensityispositivelycorrelatedwithmarketshare,aswellaswith rmTFP,employment,andrevenues.Thestrongestcorrelateofimportintensityisthetotalmaterialcostofthe rm,consistentwiththepredictionsofProposition2.Overall,thecorrelationsinTable4broadlysupportthevariouspredictionsofourtheoreticalframework.Atthesametime,althoughimportintensityandmarketsharearepositivelycorrelatedwithproductivityandother rmperformancemeasures,thereissu cientindependentvariationtoenableustodistinguishbetweenthedeterminantsofincompletepass-throughinthefollowingsubsections.Weclosethissectionwithabriefdiscussionofthepatternsoftime-seriesvariationinimportintensityforagiven rm.Importintensityappearstobearelativelystablechar-acteristicofthe rm,movinglittleovertimeandinresponsetoexchangerate uctuations.Speci cally,thesimpleregressionof'f;ton rm xedeectshasanR2ofover85%,imply-ingthatthecross-sectionalvariationintime-averaged rmimportintensity'fisnearly6timeslargerthantheaveragetime-seriesvariationin'f;tforagiven rm.Whenweregressthechangein'f;ton rm xedeectsandthelagsofthelogchangein rm-levelimport-weightedexchangerates,thecontemporaneouseectissigni cantwiththesemi-elasticity 27Whiletheunweighteddistribution( rmcount)hasasinglepeak,theexport-value-weighteddistributionhastwopeaks.Thisisduetothefactthatoneexporterwith'f=0:33accountsforalmost14%ofexportsales.Ourresultsarenotsensitivetotheexclusionofthislargestexporter,whichaccountsforonly134observationsoutofatotalofover90,000 rm-destination-product-yearobservationsinoursample.21 Table4:Correlationstructureofimportintensity ImportMaterialintensityTFPRevenuesEmpl'tcost Marketshare0.160.200.280.250.27Materialcost0.230.700.990.83Employment0.100.600.86Revenues0.210.72TFP0.15 Note:Cross-sectionalcorrelationsof rm-levelvariablesaveragedovertime.Materialcosts,employment,revenuesandTFPareinlogs.ImportintensityistheshareofimportedintermediateinputsfromoutsidetheEuroZoneinthetotalcostofthe rm.MarketsharecorrespondstoourmeasureSf;s;k;t,de nedin(20),aggregatedtothe rm-levelbyaveragingacrosssector-destination-time.ofonly0.057,andwithosetting,albeitmarginallysigni cant,lageects.Thatis,a10%depreciationoftheeurotemporarilyincreasesimportintensityby0.57ofapercentagepoint.Furthermore,we ndthatthe rmhardlyadjustsitsimportsontheextensivemargininresponsetochangesinitsimport-weightedexchangerate.28AllofthisevidenceprovidessupportforourassumptioninSection2thatthesetofimportedgoodsisasunkdecisionatthehorizonsweconsider,andhencetheextensivemarginplaysaverylimitedroleintheresponseofa rm'smarginalcosttoexchangeratemovements,justifyingtheuseof'fasatime-invariant rmcharacteristicintheempiricalregressionsthatfollow.Tosummarize,we ndsubstantialvariationinimportintensityamongexporters,andthisheterogeneityfollowspatternsconsistentwiththepredictionsofourtheoreticalframework.Next,guidedbythetheoreticalpredictions,weexploretheimplicationsofthisheterogeneityfortheexchangeratepass-throughpatternsacrossBelgian rms.4EmpiricalEvidenceThissectionpresentsourmainempiricalresults.Westartbyintroducingandestimatingourmainempiricalspeci cation.Wethenprovidenonparametricevidenceandexploretheforcesbehindourempiricalresults,con rmingthespeci cmechanismsidenti edbythetheory.Weconcludewithabatteryofrobustnesstests. 28Wemeasuretheextensivemarginasthechangein rmimportsduetoaddinganewvarietyordroppinganexistingvarietyatCN8-digit-countrylevel.22 4.1Empiricalspeci cationWenowturntotheempiricalestimationoftherelationshipbetweenimportintensity,marketshareandpass-throughinthecross-sectionofexporters(Proposition3).Thetheoreticalpass-throughregressionequation(15)cannotbedirectlyestimatedsincepass-through k;iisnotavariablethatcanbeobservedinthedata.Therefore,westepbacktothedecompositionofthelogpricechangeinequations(12)(14),whichweagainlinearizeinimportintensityandmarketshare.Afterreplacingdierentialswithchangesovertime,wearriveatourmainempiricalspeci cation,whereweregresstheannualchangeinlogexportpriceonthechangeinthelogexchangerate,interactedwithimportintensityandmarketshare:pf;i;k;t= s;k+'f;t�1+~ Sf;s;k;t�1ek;t+s;k+b'f;t�1+cSf;s;k;t�1+~uf;i;k;t;(21)wherepf;i;k;tisthelogeuroproducerpricetodestinationk(asopposedtolocal-currencyprice),andanincreaseinthelognominalexchangerateek;tcorrespondstothebilateraldepreciationoftheeurorelativetothedestination-kcurrency.29Inouranalysisweestimateparametersand~ withvaluesaveragedacrosssector-destinations-time.Inordertokeepthepoolofaveragedcoe cients(s;k; s;k)relativelyhomogenous,wefocusonmanufacturingexportstohigh-incomeOECDcountriesinourbenchmarkspeci cations.Theregressionequation(21)isastructuralrelationshipthatemergesfromthetheoreticalmodelofSection2,andSf;s;k;t�1correspondstoourmeasureofmarketsharede nedinequation(20).Underamildassumptionthatek;tisuncorrelatedwith('f;t�1;Sf;s;k;t�1),weproveintheappendix:Proposition4TheOLSestimatesofand~ in(21)identifytheweightedaveragesacrosssector-destination-yearsofs;kand s;kSs;k;t�1respectively,whereSs;k;t�1isthesector-destination-time-speci ccumulativemarketshareofallBelgianexportersand(s;k; s;k)arethetheoreticalcoe cientinthepass-throughrelationship(15).Thisresultshowsthat,despitethefactthatwecannotdirectlyestimatethetheoreticalregression(15),wecannonethelessidentifythetheoreticalcoe cientsintherelationshipbetweenpass-through,importintensityandmarketshare.Furthermore,itformallycon rmsthevalidityofourmeasureofthemarketsharerelativetootherBelgianexporters.Equation(21)isourbenchmarkempiricalspeci cation.Notethatitisverydemandinginthatitrequiresincludingsector-destinationdummiesandtheirinteractionswithexchangeratechangesataverydisaggregatedlevel.Therefore,westartbyestimatingequation(21) 29ThenominalexchangeratesareaverageannualratesfromtheIMF.TheseareprovidedforeachcountryrelativetotheUSdollar,whichweconverttoberelativetotheeuro.23 withacommoncoe cient ,andlaterreplacesector-destination xedeectss;kwithsector-destination-year xedeectss;k;twhichabsorb s;kek;t.Wealsoestimate(21)forexportstoasingledestination(UnitedStates)only.Inourmainregressionswereplace'f;t�1withatime-invariant'ftoreducethemeasurementerror,andalsotomaximizethesizeofthesamplesincesomeofthelagged'f;t�1wereunavailable.Thishaslittleeectontheresultssince,asweshow,'f;tisverypersistentovertime.Inthemainspeci cationswealsoreplaceSf;s;k;t�1withthecontemporaneousSf;s;k;t,asbothgivethesameresults.30Intherobustnesssectionwereporttheestimatesfromthespeci cationwiththelagged'f;t�1andSf;s;k;t�1.4.2Mainempirical ndingsToexploretheunderlyingmechanismsbehindtheequilibriumrelationshipbetweenpass-through,importintensity,andmarketshares,webeginwithamoresimplespeci cationandbuilduptothespeci cationinequation(21).Table5reportstheresults.Weincludeindustry-destinationspeci ceects(whereindustryisde nedattheHS4-digitlevel)tobeconsistentwiththetheory,andyeareectstocontrolforcommonmarginalcostvariation.First,incolumn1,wereportthatattheannualhorizontheunweightedaverageexchangeratepass-throughelasticityintoproducerpricesinoursampleis0.2,or,equivalently,0.8(=1�0:2)intodestinationprices.Werefertoitas80%pass-through.Incolumn2,weincludeaninteractionbetweenexchangeratesanda rm'simportinten-sity.Weseethatthesimpleaveragecoe cientreportedincolumn1masksaconsiderableamountofheterogeneity,as rmswithdierentimportintensitieshaveverydierentpass-throughrates.Firmswithahighshareofimportedinputsrelativetototalvariablecostsexhibitlowerpass-throughintodestination-currencyexportpricesa10percentagepointhigherimportintensityisassociatedwitha6percentagepointlowerpass-through.Atypical rmwithzeroimportintensityhasapass-throughof87%(=1�0:13),whilea rmwitha38%importintensity(inthe95thpercentileofthedistribution)hasapass-throughofonly64%(=1�0:13�0:600:38).Next,weexplorewhetherimportintensityoperatesthroughthemarginal-costchannelorthroughselectionandthemarkupchannel.Incolumns3and4,weaddcontrolsforthemarginalcostofthe rmtoseewhethertheeectofimportintensityonpass-throughpersistsbeyondthemarginalcostchannel.Incolumn3,wecontrolforthechangeinmarginalcostmcf;t,measuredastheimport-weightedchangeinthe rm'simportpricesofmaterialinputs(seeequation(19)),whichislikelytobesensitivetoexchangeratechangesifthe 30Wedonotusethetime-averagedmarketshareas rmsmoveinandoutofsector-destinationsovertime.24 Table5:Importintensity,marketshare,andpass-through Dep.var.:pf;i;k;t(1)(2)(3)(4)(5)(6)(7) ek;t0.2030.1270.1570.1490.0980.057(0.026)(0.027)(0.028)(0.037)(0.030)(0.031)ek;t'f0.6040.3700.3410.2630.4730.470(0.112)(0.117)(0.201)(0.115)(0.104)(0.236)ek;tSf;s;k;t0.2380.2840.299(0.060)(0.063)(0.100)mcf;t0.5120.506(0.030)(0.031) Fixedeects:s;k+tyesyesyesnoyesyesnos;k;tnonononononoyesf;i;t+knononoyesnonono#obs.93,39593,39593,39593,39593,39593,39593,395R20.0570.0570.0620.4870.0620.0570.344 Note:Observationsareatthe rm-destination-product-yearlevel.correspondstoannualchanges.Columns(2)(3)and(5)(7)includeacontrolforthelevelof'f,andcolumns(5)(7)alsoincludeacontrolforthelevelofthemarketshare,Sf;s;k;t.Fixedeects:s;k+tisthecombinationofsector-destinationandyear xedeects;s;k;taresector-destination-year xedeects;f;i;t+kisthecombinationof rm-product-yearandcountry xedeects(thesectorisde nedatHS-4andtheproductatHS-8-digitlevel).Theresultswithouttheyear xedeectsarealmostidentical.,andcorrespondto10%,5%and1%signi cancelevelsrespectively.Standarderrorsareclusteredatthecountry-yearlevel,reportedinbrackets.Alternativeclusteringatthe rmlevelandatthecountry-HS4-digitlevelyieldthesameconclusions. rmreliesheavilyonimportedintermediateinputs.Comparingcolumns2and3,weseethatthecoe cientontheimportintensityinteractionnearlyhalvesinsizeoncewecontrolformarginalcost,droppingfrom0.6to0.37,butstillremainsstronglysigni cantwithat-statof3.2.Wecon rmthis ndingwithanalternativecontrolformarginalcostchanges,byincluding rm-product-year xedeects(f;i;t)incolumn4.Inthisspeci cation,theonlyvariationthatremainsisacrossdestinationsforagiven rmandhence,amongotherthings,controlsforallcomponentsofthemarginalcostofthe rms.31Thecoe cientontheimportintensityinteractionincolumn4isalmostthesameasthatincolumn3,butmuchlesspreciselyestimatedwithat-statof1.7.Thisresultisimpressive,giventhatthisspeci cationissaturatedwith xedeects,andthesimilarityoftheresultsincolumns3and4providescon denceinourmeasureofmarginalcost. 31Although rm-product-year xedeects(f;i;t)arguablyprovidethebestpossiblecontrolformarginalcost,thedisadvantageofthisspeci cationisthatitonlyexploitsthevariationacrossdestinationsandthusexcludesallvariationwithinindustry-destinationsthatisthemainfocusofouranalysis.Consequently,wecannotuseourmeasureofmarketshareinaspeci cationswithf;i;tsinceourmarketsharemeasureonlymakessensewithinindustry-destinations,aswediscussinSection3.1.25 Theresultsincolumns3and4suggestthat,althoughthemarginalcostisanimportantchannelthroughwhichimportintensityaectspass-through(seeProposition2),thereisstillaconsiderableresidualeectoperatingthroughthemarkupchannelafterconditioningonthemarginalcost.This ndingisconsistentwiththetheoreticalpredictions,sinceimportinten-sitycorrelateswithmarketshareinthecross-sectionof rmsandmarketsharedeterminesthemarkupelasticity,causingomittedvariablebias.Toseethis,incolumn5weaugmentthespeci cationofcolumn3(controllingformcf;t)withamarketshareinteractionwiththelogchangeinexchangeratetoproxyformarkupelasticity,assuggestedbyProposition3.Giventhatwenowcontrolforbothmarginalcostandmarkup,weexpectimportintensitytostophavingpredictivepower.Althoughtheeectofimportintensitydoesnotdisappearcompletely,thecoe cientdoesfallinsize(from0.37to0.26)andbecomeslesssigni cant.32Finally,column6implementsourmainspeci cationinequation(21)byincludingtheimportintensityandmarketshareinteractions,withoutcontrollingformarginalcost.Propo-sition3suggeststhatimportintensityandmarketsharearetwoprimepredictorsofexchangeratepass-through,andindeedwe ndthatthetwointeractiontermsincolumn6arestronglystatisticallysigni cant.Interpretingourresultsquantitatively,we ndthata rmwithazeroimportintensityandanearlyzeromarketshare(correspondingrespectivelytothe5thpercentilesofbothdistributions)hasapass-throughof94%(=1�0:06).Complete(100%)pass-throughforsuch rmscannotberejectedatthe95%con dencelevel.Ahypotheti-calnon-importing rmwitha75%marketsharerelativetootherBelgianexporterswithinsector-destination(correspondingtothe95thpercentileofthe rm-leveldistributionofmar-ketshares)hasapass-throughof73%,thatis21percentagepoints(=0:280:75)lower.Holdingthismarketshareconstantandincreasingtheimportintensityofthe rmfromzeroto38%(correspondingagaintothe95thpercentileoftherespectivedistribution)reducesthepass-throughbyanother18percentagepoints(=0:470:38),to55%.Therefore,variationinimportintensityandmarketshareexplainsavastrangeofvariationinpass-throughacross rms.Themarginalcostchannel(proxiedbyimportintensity)andthemarkupchannel(proxiedbymarketshare)contributeroughlyequallytothisvariationinpass-through.Column7concludesthisanalysisbyestimatingourmainspeci cation(asincolumn6)controllingfortheindustry-destination-year xedeects(s;k;t).Therefore,theonlyvariationusedinthisestimationiswithinindustry-destination-year,assuggestedbyourtheoryinSection2.4.Inthisregression,thecoe cientontheexchangerateiseectivelyallowed 32Thecoe cientonmcf;tinbothspeci cationsofcolumns3and5isremarkablystableat0.51.Thetheorysuggeststhatthiscoe cientshouldbe1=(1+�),thatistheaveragepass-throughelasticityofidiosyncraticshocksintoprices,correspondingtoanaveragemarkupelasticityof�1,closetotheestimatesprovidedinGopinathandItskhoki(2011)usingverydierentdataandmethods.26 tovaryattheindustry-destinationlevel(i.e., s;k)butisabsorbedinthe xedeects.Comparingcolumns6and7,weseethatthepointestimatesontheimportintensityandmarketsharebarelychange,andtheestimatesremainstronglystatisticallysigni cantincolumn7despitethousandsof xedeects.4.3NonparametricResultsOurmainempirical ndingsinTable5providestrongsupportforthetheoreticalpredictionsdevelopedinSection2.However,wewanttoensurethattheseresultsaresmooth(e.g.,notdrivenbyoutliers)andarenotartifactsofthelinearizedspeci cation.Were-estimatethespeci cationsinTable5nonparametrically,bysplittingthedistributionofimportintensity'fintoquartiles.Speci cally,weestimateaseparatepass-throughcoe cientforeachquar-tileoftheimportintensitydistributionandplotthesecoe cientsinFigure1.Allestimatedcoe cients,standarderrors,andp-valuesarereportedinTableA1intheappendix.Thegraphshowsthatthecoe cientisestimatedtobemonotonicallyhigher(thuslowerpass-through)aswemovefromlowtohigherimportintensitybinswhenwedonotincludebothmarginalcostandmarketsharecontrols.Thesteepestlinecorrespondstotheunconditionalregression(acounterparttocolumn2ofTable5),andissomewhat atterwithcontrolsformarginalcost(column3),anditismuch atteraftercontrollingjointlyforthechangeinthemarginalcostandthemarketshareinteraction(column5).Thedashedlinecorrespondstoourmainspeci cation(column6),whichcontrolsforbothmarketshareandimportin-tensity,butnotmarginalcost,anditalsoexhibitsaconsiderableslopeacrosstheimportintensitybins.Furthermore,inallofthesecasesthedierencebetweenthepass-throughcoe cientinthe rstandfourthquartilesissigni cantwithap-valueof1%,withtheex-ceptionofwhenwecontrolforbothmarginalcostandmarkup(marketshareinteraction).Inthiscase,consistentwiththetheory,thepro leofpass-throughcoe cientsacrossthebinsoftheimportintensitydistributionbecomesnearly atwiththedierencesbetweenthepass-throughvaluesindierentbinsstatisticallyinsigni cant.Analternativenonparametricspeci cationistodividetheobservationsbybothimportintensityandmarketshare.InTable6,wesplitall rmsintolowandhighimportintensitybinsatthemedianimportintensity,andallobservationsintolowandhighmarketsharebinsdependingonwhetherthe rm-sector-destinationmarketshareisbeloworabovethe75thpercentileofthemarketsharedistributionwithinsector-destination.Withthissplitroughly50%of rm-product-destination-yearobservationsfallwithineachmarketsharebin.Foreachofthefourbins,weestimateasimplepass-throughregressionofthechangeinproducer27 Figure1:Pass-throughbyquartileofimportintensityNote:Equal-sizedbinsintermsof rm-product-destination-yearobservations,sortedby'f.Themeansof'finthefourbinsare1.3%,5.5%,13.1%and30.1%,respectively.The gurereportspass-throughcoe cientsofpf;i;k;tonek;twithineach'f-quartile,wheretheregressionsincludeadditionalcontrolsinlevelsandinteractedwithek;t,asindicatedinthelegendofthe gure,toparallelcolumns(2)-(3)and(5)-(6)ofTable5.Thepass-throughcoe cientinBin4issigni cantlydierentfromthatinBin1atthe1%levelforallspeci cationsexceptthethirdone(controllingforbothmcf;tandSf;s;k;t)forwhichitisnotsigni cantatthe5%.AdditionalinformationisreportedinTableA1intheappendix.exportpricesonthechangeintheexchangerate.Consistentwiththeresultsincolumn6ofTable5,we ndthatpass-throughintodestination-currencyexportpricesdecreasessigni cantlyeitheraswemovetowardthebinwithahighermarketshareortowardthebinwithahigherimportintensity.Thelowestpass-throughof66%(=1�0:34)isfoundforthebinwithhighmarketshareandabove-medianimportintensity,comparedwiththepass-throughof87%(=1�0:13)for rmswithbelow-medianimportintensityandlowmarketshare,quantitativelyconsistentwiththeresultsinTable5.FromTable6,weseethattherearemoreobservationsalongthemaindiagonal(around30%ineachbin)relativetotheinversediagonal(around20%ineachbin),whichisduetothepositivecorrelationbetweenthemarketshareandtheimportintensityinthecross-sectionof rms.Thisnotwithstanding,theshareofexportvalueinthe rstbinwithbothlowmarketshareandlowimportintensityisonly8%.Thefourthbinwithbothabove-medianimportintensityandhighmarketshareaccountsforthemajorityofexports,over61%.Thissuggeststhatthepass-throughcoe cientintodestinationpricesfromaregressionweightedbytheirrespectiveexportvaluesshouldbesubstantiallylowerthanfromaregressioninwhichobservationsareunweighted.Indeed,whenweightingbyexportvalues,we ndapass-28 0 0.03 0.08 0.18 0.82 0 0.1 0.2 0.3 0.4 Importintensitybins,Producerpricepass-through,ª Unconditional Cond'lon¢mcf;tonly Condl'lon¢mcf;tandf;s;k;t Cond'lonf;s;k;tonly Bin4Bin2Bin1Bin3 Table6:Pass-throughbyimport-intensityandmarket-sharebins Lowimportintensity Highimportintensity Lowmarketshare 0:131 0:194Fractionofobservations 30.0% 21.0%Shareinexportvalue 8.1% 9.6% Highmarketshare 0:214 0:339Fractionofobservations 20.0% 29.2%Shareinexportvalue 21.3% 61.1% Note:Coe cientsfromregressionofpf;i;k;tonek;twithinrespectivebinswithout xedeects;bin-speci cinterceptsincluded(allestimatedtobeclosetozero).Firmsaresortedbyimportintensity'fintobelowandabovemedianequalto8.1%;andbymarketshareSf;s;k;taveragedacrosstheyearsthe rmservedagivensector-destinationintobelowandabovethe75thpercentilewithineachsector(HS4)-destination,whichensuresaroughlyevennumberof rm-product-destination-yearobservationsineachhighandlowcategory.Thecoe cientinthehigh-highbinissigni cantlydierentfromtheothercoe cientsatleastata2%level.throughcoe cientof62%,muchlowerthanthe80%resultintheunweightedspeci cation(column1ofTable5).Ourevidencefurthershowsthatpartofthisdierenceisduetogreatermarkupvariabilityamongthelargeexporters,butofaquantitativelysimilarimportanceisthehigherimportintensityofthese rms.Finally,weexplorethepossibilityofnonmonotonicandnonlineareectsofmarketshareandimportintensityonpass-throughbyaugmentingthemainspeci cationincolumn6ofTable5withsecond-orderterms.We ndthatthecoe cientonthesquaredmarketshareterminteractedwithexchangerateisnegative,butinsigni cantandsmall.Evenusingthepointestimate,theestimatedrelationshipbetweenpass-throughandmarketshareremainsmonotonicallyincreasingoverthewholerange[0;1]ofthemarketsharevariable,whichcon rmsthetheoreticalpredictioninProposition1.Further,althoughthecoe cientontheinteractionofimportintensitywithmarketshareispositive,itisalsosmallandinsigni cant.Theseresultsjustifyourfocusonthelinearspeci cationofTable5(seediscussioninfootnote19).4.4DecipheringthemechanismNowthatwehaveestablishedthathigh-import-intensive rmshavelowerpass-throughintoexportprices,wedelveintotheunderlyingmechanisms.Accordingtothetheory,higherimportintensityisassociatedwithhighermarginalcostsensitivitytoexchangerates(Propo-sition2).Wetestthisdirectlybyregressingthechangeinthemarginalcostmcf;tonthechangeinthedestination-speci cexchangerateek;tandseparatelyonthechangeinthe29 rm-levelimport-weightedexchangerateeMf;t,withineachquartileoftheimport-intensitydistribution.33Figure2,whichplotsthesecoe cients,illustratesaverytightmonotonicallyincreasingpatternofmarginalcostsensitivitytothedestination-speci cexchangeratesacrossthebinswithincreasingimportintensity(andcolumns5and6inTableA1intheappendixreportadditionaldetails).Quantitatively,anincreaseinimportintensityfrom1%onaverageinthe rstquartileto30%onaverageinthefourthquartileleadstoanincreaseinmarginalcostsensitivitytotheexchangeratefrom0.02to0.17.Thisvariationisquanti-tativelyconsistentwiththeeectsofimportintensityonpass-throughwestudiedearlier(seeFigure1).Theresponseofthemarginalcosttotheimport-weightedexchangerateisalsomonotonicallyincreasingin'fandliesstrictlyabovetheresponsetothedestination-speci cexchangerate,rangingfrom0.05to0.21.ThetheorypredictsthepatternsdepictedinFigure2holdwhenboththecorrelationbetweenexportandimport-weightedexchangeratesandthepass-throughintoimportpricesarepositive(seeequations(14)and(16)).Wenowprovideevidenceforeachofthesetwokeystructuraldeterminantsoftherelationshipbetweenimportintensityandpass-through.Col-umn7ofTableA1showsthereisapositivecorrelationbetweenimportandexportexchangeratesbyreportingtheprojectioncoe cientsof rm-levelimport-weightedexchangerateseMf;tondestination-speci cexchangeratesek;tacrossthequartilesofimport-intensitydistribution.We ndtheseprojectioncoe cientstobestableataround0.45,withnostatistically-detectablevariationacrossbinsofimportintensity.Therefore,we ndnosys-tematicrelationshipbetweenimportintensityandtheextenttowhich rmsaligntheirimportsourcesandexportdestinationstohedgetheirexchangeraterisks(i.e.,realhedging).Wenextshowtheimportanceofthispositivecorrelationbetweenimportandexportexchangeratesforexchangeratepass-throughintoexportprices.Intuitively,wewouldexpecttheeectsofimportintensitytobestrongerwheninputsareimportedfromthesamecountrytowhichthe rmsellsitsproducts.Tocapturethisideasystematically,wesplitallsource-destinationpairsofcountriesintoahighandalowcorrelationbinsdependingonwhetherthecorrelationbetween(theannuallogchangesin)thetworespectiveexchangeratesisaboveorbelow0.7.Foreach rm-destinationwecreatetwomeasuresofimportintensity'Highf;kand'Lowf;kfromhigh-andlow-correlationsourcecountriesrespectively.34 33Theimport-weightedexchangerateeMf;tisaweightedaverageofbilateralexchangerateswithweightsequaltotheimportexpendituresharesfromoutsidetheEuroZoneatthe rmlevel.34Notethatouroverallmeasureofimportintensityequals'f'Highf;k+'Lowf;kforalldestinationskservedby rmf.TableA2intheappendixreportsinformationonthepairsofhighandlowcorrelationcountries.Ourresultsarerobusttoalternativecorrelationthresholds,howeverraisingittoohigh(e.g.,settingitat1,whichamountstoplacinginthehighbinonlytheimportsfromthedestinationcountryitselfinmostcases),leavesthehigh-correlationbintoothinforreliablestatisticalinference,whilereducingitbelow0.730 Figure2:MarginalcostsensitivitytoexchangeratesNote:The gureplotsthepass-throughcoe cientsfromregressionsofthelogchangeinourmeasureofthemarginalcostofthe rmmcf;tonbothbilateralexportexchangeratesek;tand rm-levelimport-weightedexchangerateeMf;t,byquartilesofthe'f-distribution(asinFigure1).AdditionalinformationisreportedinTableA1intheappendix.Theaveragecorrelationbetweenexchangeratesinthetwobinsis0.92and0.25respectively,andaccordingtothetheory(seeequation(16))thisdierenceshoulddirectlytranslateintothedierentialeectofthetwoimportintensitiesontheexchangeratepass-throughofthe rm.Thisisexactlywhatwe ndincolumn1ofTable7,whichestimatestheaugmentedspeci cation(21)splittingthe rmimportintensityintothetwocomponentsjustintroduced.Theestimatedcoe cientsonthehigh-andlow-correlationimportintensityinteractionsare0.86and0.38respectively.Wefurthershowthattheeectofimportintensityonexportpricepass-throughisstrongerthehighertheexchangeratepass-throughintoimportprices.Weconstructthedierentimportintensitymeasuresbysplittingallnon-Euroimportsourcecountriesinoursampleintothreegroups:highpass-through,lowpass-through,andother.Thegroupofothercountriescontainssourcecountriesforwhichthereareeitherinsu cientobservationsornotenoughvariationintheexchangeratetoestimatepass-throughaccurately.There-mainingcountriesareassignedtothehighbinifpass-throughexceeds0.5,andtheresultsarerobusttoalternativechoicesofthiscuto.Withthisprocedurewehave19countriesin doesnotallowustodiscriminateeectivelybetweenveryhighcorrelationcountries(nearpegs)andaveragecorrelationcountriesgiventhesubstantialnoiseassociatedwiththecorrelationmeasures.Withthethresholdof0.7,importsfromthehigh-correlationsourcecountriesaccountonaveragefor22%oftheoverallimportintensityacross rms.31 0 0.03 0.08 0.18 0.82 0 0.1 0.2 0.25 Importintensitybins,Marginalcostsensitivity mcf;ton¢k;t mcf;ton¢f;t Bin4Bin2Bin3Bin1 Table7:Whichimportsmatter? ExchangerateImportEuroAreaOECDvscorrelationpass-throughimportsnon-OECDDep.var.:pf;i;k;t(1)(2)(3)(4) ek;t0.0530.0580.0440.044(0.031)(0.031)(0.040)(0.040)ek;t'f0.481(0.110)ek;t'Highf;k0.8640.7780.472(0.277)(0.242)(0.154)ek;t'Lowf;k0.3760.3410.505(0.131)(0.241)(0.210)ek;t'Otherf0.027(0.316)ek;t'Eurof0.0600.057(0.126)(0.126)ek;tSf;s;k;t0.2840.2860.2820.282(0.063)(0.063)(0.063)(0.064) FE:s;k+tyesyesyesyes#obs.93,39593,39593,39593,395R20.0580.0580.0580.058 Note:Resultsfromtheaugmentedspeci cationpf;i;k;t= +High'Highf;k+Low'Lowf;k+Other'Otherf;k+ Sf;s;k;tek;t+:::+f;i;k;t:Column1:splitsimportsintohighandlowcorrelationbinsforeach rm-destinationdependingonwhetherthecorrelation(basedonannuallogchanges)betweenimportandexportexchangerateisabove0.7;de-compositionofimportintensity:'f'Highf;k+'Lowf;kforeverydestinationk.Column2:splitsimportsbypass-throughintoimportpricesintothreebinshigh(above0.5),low(below0.5)andother(impreciselyestimated),asexplainedinthetext('f'Highf+'Lowf+'Otherf).Column3:doesnotsplitimportsfromoutsidetheEuroArea('f),butinsteadadditionallyincludesimportintensityfromwithintheEuroArea('Eurof).Column4:splitsnon-Euroimportsintoimportsfromhigh-incomeOECDandlow-incomenon-OECDcountries('f'Highf+'Lowf),alsocontrollingfornon-Euroimports('Eurof).Allregressionsadditionallycontrolforthelevelsofallinteractiontermsandincludecountry-destinationandtime xedeects.OtherdetailsasinTable5.thehighpass-throughbinwithaveragepass-throughintoBelgianimportpricesof63%andaccountingfor38%ofBelgian rmimports.Thelowpass-throughbincontains32countrieswithaveragepass-throughof25%andaccountingfor42%ofBelgian rmimports.Thelistofcountries,theirpass-throughandimportsharesarereportedinTableA3intheappendix.Weusethethreecorrespondingimportintensityvariablestoestimateanaugmentedspeci -cation(21)incolumn2ofTable7.Aspredictedbythetheory,we ndahighercoe cient,equalto0.76,forimportintensityfromhigh-pass-throughcountries,whilethecoe cientfor32 thelow-pass-throughimportsis0.35andinsigni cant.35Incolumn3ofTable7,insteadofsplittingtheimportintensityfromoutsidetheEuroArea('f),weadditionallycontrolfortheimportintensityfromwithintheEuroArea(de-notedwith'Eurof).Asexpected,we ndthatimportintensityfromwithintheEuroAreahasnoeectonpass-through,sincethepricesoftheseimports,justlikethoseoftheBel-giuminputs,areinsensitivetotheEuroexchangerate.Thisregressioncanbeviewedasaplacebotestcon rmingthatourimportintensitymeasure'findeedpicksupthemarginalcostsensitivitytoexchangeratesratherthanproxyingforotherdimensionsofheterogene-ityacross rmsrelatedtoimportintensity.Theresultsincolumns1,2,and3ofTable7providedirectevidenceofthetheoreticalmechanismslinkingimportintensityandexchangeratepass-through.Lastly,weaskwhethertheremaybeadditionalforcescorrelatedwiththe rm'simportin-tensitythatcouldpotentiallyconfounditsrelationshipwiththeexchangeratepass-through.Anoticeabledierenceacross rmsisintheirshareofimportsthatcomesfromnon-OECDcountries,whichtendstobesubstantiallyhigherforthehigh-import-intensive rms.Theimportsharefromnon-OECDcountriesintotalnon-Euroimportsincreasesmonotonicallyfrom25%to45%acrossthequartilesofimportintensity.36Column4ofTable7explorestheeectsofthisheterogeneitybysplittingtheoverallmeasureofthe rm'simportinten-sity'fintotheimportintensitiesfromthehigh-incomeOECDandlow-incomenon-OECDcountries,bothoutsidetheEuroArea.Surprisingly,we ndnodierenceintheeectsofimportintensityfromOECDandnon-OECDcountries,withbothcoe cientsestimatedtobearound0.5.Asexpected,OECDcountriestendtohavehigherpass-throughintoimportpricesofBelgian rms:44%onaverageversus15%onaveragefromnon-OECDcountries,thoughthereissigni cantvariationwithinthesegroupsasre ectedinTableA3intheap-pendix.However,theeectsoftheseimportpass-throughdierencesarecounterbalancedbythecorrelationpatternbetweenimportandexportexchangerates.Indeed,almostallsourcecountriesinthehigh-correlationpairswithdestinationcountriesarenon-OECD,whichistoalargeextentdrivenbythefullorpartialexchangeratepegsadoptedbymanynon-OECDcountries.Overall,we ndthatthecompositionofimportsfromOECDversusnon-OECDcountrieshasnodetectableeectontheexportpass-throughofBelgian rms.37 35Whiletheshareofhigh-correlationimportsisstableataround22%across rmswithdierentimportintensity,theshareofhigh-pass-throughimportsincreasesfrom34%to44%acrossimportintensityquartiles,slightlyreinforcingtheroleofimportintensityinourmainspeci cation.36Bothpatternsareconsistentwiththepresenceofa xedcostofimportingwhichincreaseswiththegeographicandeconomicdistancetothesourcecountry,makingitworthwhileonlyforthelargestimporterstosourceinputsfromthedistantorigincountries.37Theothertwopossibilitiesweexploredwerewhetherpass-throughintoimportpricesofBelgian rms33 Tosummarize,thedatastronglysupportsthespeci cmechanismsidenti edbythethe-ory,andwe ndnoevidencethatourimportintensitymeasureproxiesforothermechanismsorvariablesthatmayaectpass-through.4.5RobustnessInthissection,weconsiderthreesetsofrobustnesstests:includingadditionalcontrols,consideringalternativede nitionsofimportintensity,andusingalternativesamplesintermsofincludeddestinations, rmsandproducts.Weconcludethesectionbyadiscussionofthepossibleselectionandmeasurementissues.AdditionalcontrolsInTable8,wecheckwhetherourresultsarerobusttoaddinginalternativeproxiesformarkupandmarginalcostvariabilitysuchas rmemploymentsizeandmeasuredTFP.38Theresultsreportedincolumns1and2ofTable8showthatthecoe cientsonthemainvariablesofinterestarehardlyaectedbytheinclusionoftheseadditionalinteractionterms.ControllingforemploymentandTFPinteractionsreducesslightlytheestimatedcoe cientsonimportintensityandmarketshareinteractions,buttheyremainlargeandstronglystatisticallysigni cant.Thecoe cientsonemploymentandTFPinteractionsarealsopositiveandsigni cant,indicatingthatotherfactorsoutsideourmodelalsoin uenceexchangeratepass-through.Column3ofTable8augmentsthespeci cationincolumn2bycontrollingforthelocalcomponentofthemarginalcostproxiedbythelogchangesinthemeasureofthe rm-levelwagerateandthe rmTFPtoisolatetheeectofimportintensitythroughtheforeign-sourcedcomponentofthemarginalcostofthe rm.Theseadditionalcontrolshaveessentiallynoeectontheestimatedcoe cients.Alternativede nitionsofimportintensityToensurethattheresultsarenotsensitivetoourde nitionof'f,weexperimentedextensivelywithalternativede nitions.WereporttheserobustnesschecksinTableA4intheappendix,whereweestimateourmainempirical variesbythetypeoftheproductimported(manufacturedornot)and/orbythetype(size)oftheimporting rm.Althoughtheshareofmanufacturedproductsinimportsdecreasesfrom95%to86%acrossthequartilesofimportintensity,we ndnosystematicdierenceinpass-throughformanufacturingandnon-manufacturingimports,oncewecontrolforthesourcecountryofimports.Similarly,controllingforthesourcecountry,we ndnosystematicdierenceinimportpass-throughbetweensmallandlarge rms.Section4.5presentsfurtherrobustnesstestscontrollingforthesizeofthe rmandtypesofimportedinputs.38Intheorieswhereproductivityistheonlysourceofheterogeneity,marketshare,employment,andproductivityitselfareallperfectlycorrelated.However,whenthereismorethanonesourceofheterogeneity,thesevariablesarecorrelatedpositivelybutimperfectly(aswedocumentinTable4),whichallowsustojointlyincludethesevariablesinonespeci cation.OurempiricalresultsareconsistentwithBerman,Martin,andMayer(2012),whichfocuson rmproductivityasameasureofmarkupvariabilitywealso ndthatmoreproductive rmshavelowerpass-through,butwesplitthiseectintothemarkupandmarginalcostchannelsbycontrollingseparatelyformarketshareandimportintensity.34 Table8:Robustnesswithadditionalcontrols Dep.var.:pf;i;k;t(1)(2)(3) ek;t'f0.4130.4330.418(0.106)(0.109)(0.119)ek;tSf;s;k;t0.2190.2490.245(0.065)(0.064)(0.065)ek;tlogLf;t0.044(0.012)ek;tlogTFPf;t0.0700.080(0.023)(0.024)logWf;t0.004(0.002)logTFPf;t0.035(0.007) FE:s;k+tyesyesyes#obs.92,57692,10687,608R20.0580.0580.061 Note:Thesamespeci cationasincolumn6ofTable5,augmentedwithadditionalcontrols.Lf;tis rmemployment,Wf;tis rmaveragewagerate,andTFPf;tistheestimateof rmtotalfactorproductivity.speci cationusingdierentde nitionsofimportintensity.First,incolumn1,weverifythatourresultsareunchangedwhenasinspeci cation(21)weuselaggedtime-varying'f;t�1andSf;s;k;t�1,assuggestedbyProposition4,insteadof'fandSf;s;k;trespectively.Remarkably,theestimatedcoe cientsarevirtuallythesameasinTable5.Next,incolumn2ofTableA4weincludeonlymanufacturingproductsintheconstructionoftheimportintensityvariable.Incolumns3and4,werespectivelyrestrictthede nitionofimportstoexcludeconsumergoodsandcapitalgoods.Inthesubsequentcolumns,weuseIOtablestoidentifya rm'sintermediateinputs.Incolumn5,weincludeonlyimportsidenti edasintermediateinputsintheIOtablesforallofthe rm'sexports,andincolumn6weonlyincludeIOinputsfora rm'sIOmajorexports.Finally,incolumn7,weexcludeanyimportattheCN-8-digitindustrialcodeifthe rmsimultaneouslyexportsinthiscategory.Inallcases,theresultsareessentiallyunchanged,exceptthatinthelastcasethecoe cientontheimportintensitysubstantiallyincreases,butitshouldbenotedthattheaverageimportintensitiesherearemuchloweraswedropalargeshareofimportsfromtheimportintensitycalculation.AlternativesamplesWefurtherchecktherobustnessofourresultswithinalternativesubsamplesofthedataset,bothinthecoverageofexportdestinationsandinthetypes35 ofproductsand rms.TableA5intheappendixreportstheresultsineightalternativesubsamples.Byandlarge,itrevealsthesamequalitativeandquantitativepatternswe ndinourbenchmarksample.Columns13ofTableA5reporttheresultsforthreealternativesetsofexportdestinationsallnon-Eurocountries,non-EuroOECDcountriesexcludingtheUS,andtheUSonly.ItisnoteworthythatfortheUSsubsampleweestimatebothalowerbaselinepass-through(for rmswithzeroimportintensityandmarketshare)andastrongereectofimportintensityonpass-through,thanforothercountries.39OuranalysisinSection4.4suggeststhatthelargereectofimportintensityonpass-throughintoexportpricetotheUSisatleastinpartduetothestrongercorrelationbetweentheeuro-dollarexchangerateandatypicalBelgian rm'simport-weightedexchangerate(seeTableA2).TheremainingcolumnsinTableA5considerdierentsetsofproductsand rms.Sofar,allofthespeci cationshavebeenrestrictedtothesubsampleofonlymanufacturing rmsbecauseour'fmeasureislikelytobeabetterproxyofimportintensityinmanufacturingthanforwholesalers,whomaypurchase nalgoodswithinBelgiumtoexportthemoral-ternativelyimport nalgoodsfordistributionwithinBelgium.Incolumn4,whichaddsinallwholesale rmstoourbaselinesample,weseethatalthoughtheimportintensityandmarketshareinteractionsarestillpositiveandsigni cant,theirmagnitudesandt-statsaresmaller.Thewholesalersrepresentaround40percentofthecombinedsample.Next,incolumn5,wedropallintra- rmtransactionsfromourbaselinesample(around15percentofobservations),andthishaslittleeectontheestimatedcoe cients.40Finally,oursamplehasincludedonlythe rm'smajorexportproducts,basedonitslargestIOcode,inordertoaddresstheissueofmultiproduct rms.Incolumns68,weshowthattheresultsarenotsensitivetothischoiceofmainproducts.Incolumn6,weincludeallofthe rm'smanufacturingexportsratherthanrestrictingitonlytoIOmajorproducts.Incolumn7,weadoptanalternativewaytoidentifya rm'smajorproducts,usingtheHS4-digitcategory,whichismuchmoredisaggregatedthantheIOcategories.Andincolumn8,weonlyincludea rmifitsHS4-digitmajorcategoryaccountsforatleast50percentofitstotalexports.Inallthreecases,we ndthemagnitudesontheimportintensityandmarketshareinteractionsareveryclosetoourmainspeci cation. 39Speci cally,smallnon-importing rmsexportingtotheUSmarketpass-throughonaverageonly82%oftheeuro-dollarexchangeratechanges,whilethe rmswithhighimport-intensityandhighmarketshare(atthe95thpercentile)pass-throughonly33%.Thislowpass-throughisconsistentwithpreviousworkusingUSdata.40UsingdatafromtheBelgiumNationalBank,weclassifyintra- rmtradeasanyexporttransactionfromaBelgium rmtocountrykinwhichthereiseitherinwardoroutwardforeigndirectinvestmenttoorfromthatcountry.36 StickypricesandcurrencychoiceWenowbrie ycommentontheinterpretationofourresultsinanenvironmentwithstickyprices,whereexporterschooseto xtheirpricestemporarilyeitherinlocalorinproducercurrency.Sincewecannotconditionourempiricalanalysisonapricechangeorsplitthesamplebycurrencyofpricing,ourresultsconfoundtogetherthechangeinthedesiredmarkupwiththemechanicalchangesinmarkupinducedbytheexchangeratemovementswhenpricesarestickyinagivencurrency.41Therefore,oneshouldkeepinmindthatourresultssuggestthatimportintensityandmarketsharecontributeeitherto exible-pricepass-throughincompletenessortotheprobabilityoflocalcurrencypricing,whichinturnleadstolowpass-throughbeforepricesadjust.Inreality,boththeseforcesarelikelytocontributetoincompletepass-throughinourdata,howeversuchadecompositionisbeyondthescopeofthispaper.Nonetheless,Gopinath,Itskhoki,andRigobon(2010)showthatthestickypricedeterminantsofincompletepass-througharelargelyshapedbythesameunderlyingprimitivesasthe exiblepricedeterminants,andtheyreinforceeachotherinthecross-sectionof rms(asweexplaininfootnote9).MeasurementandselectionWeconcludetheempiricalsectionwithabriefdiscussionofmeasurementerrorandselectionbias.Oneconcernisthatsome rms,particularlysmallones,mayimporttheirintermediateinputsthroughotherBelgian rms(e.g.,specializedimporters),whichwecannotobserveinourdata,andhencecannotadjustaccordinglyourmeasureofimportintensity.Note,however,thatthiswouldworkagainstour ndingssincesomeofthefundamentallyimport-intensive rmswouldbewronglyclassi edintolowimport-intensity.Thismeasurementerrorshouldcauseanupwardbiasintheestimateofourbaselinepass-through(coe cient in(21))andadownwardbiasinourestimateoftheimport-intensityeectonpass-through(coe cient),whichwe ndtobelargenonetheless.Similarly,weexpectthemeasurementerrorinimportintensityformultiproduct rmstoworkagainstour ndings.Themeasurementerrorforthemarketsharevariableislikelytohaveclassicalproperties,andhenceweexpecttohaveadownwardbiasintheestimatesofcoe cient aswell.Further,intheappendixweprovideaformaltheoreticalargumentthatsampleselectionisalsolikelytoleadtoanupwardbiasintheestimatesof andadownwardbiasintheesti-matesofand ,aswellasprovidecorroboratingempiricalevidence.Intuitively,the rmsthatdropoutfromthesampleinresponsetoanexchangerateappreciation(ek;t0)aremorelikelytobetheonessimultaneouslyhitbyanadversemarginalcostshock(e.g.,neg- 41Ourdatadonotallowustodoadecompositionintothesetwosources,asweexplaininfootnote21,butonecanmakesuchinferencebytakingastandonaparticularcalibratedstructuralmodelofincompletepass-throughwithstickyprices.37 ativeproductivityshock),andhencedesiringtoraisepricesmorethan rmsstayinginthesample.Censoringofthese rmsfromthesampleleadstoanupwardbiasin .Sincetheprobabilityofexitdecreaseswithimportintensityandmarketshare(proxyingfor rm'spro tability),weexpectthisupwardbiastobelessseverefor rmswithhighimportin-tensityandhighmarketshares,inotherwordsamoreshallowrelationshipbetweenthesevariablesandpass-through.Tosummarize,thisanalysissuggeststhatbothmeasurementandselectionissuesarelikelytoleadustoestimatelowerboundsonbothand ,andhenceourquantitativeaccountofthevariationinpass-throughshouldbeviewedasconservative.5ConclusionInthispaper,weshowthattakingintoaccountthatthelargestexporting rmsarealsothelargestimportersiskeytounderstandingthelowaggregateexchangeratepass-throughandthevariationinpass-throughacross rms.We ndthatimportintensityaectspass-throughbothdirectly,byinducinganosettingchangeinthemarginalcostwhenexchangerateschange,andindirectly,throughselectionintoimportingofthelargestexporterswiththemostvariablemarkups.Weuse rms'importintensitiesandexportmarketsharesasproxiesforthemarginalcostandmarkupchannels,respectively,andshowthatvariationinthesevariablesacross rmsexplainsasubstantialrangeofvariationinpass-through.Asmall rmusingnoimportedintermediateinputshasanearlycompletepass-through,whilea rmatthe95thpercentileofbothmarketshareandimportintensitydistributionshasapass-throughofjustover50%.Aroundhalfofthisincompletepass-throughisduetothemarginalcostchannel,ascapturedbyourimportintensitymeasure.Sinceimportintensityisheavilyskewedtowardthelargestexporters,our ndingshelpexplaintheobservedlowaggregatepass-throughelasticities,whichplayacentralroleinthestudyofexchangeratedisconnect.Finally,weshowthatthepatternswedocumentemergenaturallyinatheoreticalframework,whichcombinesstandardingredientsofoligopolisticcompetitionandvariablemarkupswithendogenousselectionintoimportingatthe rmlevel.Our ndingssuggestthatthemarginalcostchannelcontributessubstantiallyreinforcingandamplifyingthemarkupchanneltolowaggregatepass-throughandpass-throughvari-ationacross rms.Thedecompositionofincompletepass-throughintoitsmarginalcostandmarkupcomponentshasimportantimplicationsfortheanalysisofthewelfareconse-quencesofexchangeratevolatility(asemphasizedbyBursteinandJaimovich,2008)andthedesirabilityto xexchangerates,forexample,bymeansofintegrationintoacurrencyunion.Furthermore,pricesensitivitytoexchangeratesiscentraltotheexpenditureswitch-38 ingmechanismatthecoreofinternationaladjustmentandrebalancing.Acompleteanalysisofthesetopicsrequiresageneralequilibriumframeworkdisciplinedbythetypeofevidenceontheimportanceofmarginalcostandmarkupchannelsthatweprovideinthispaper.Controllingforthemarginalcostchannel,ourevidencestillassignsanimportantroleforthemarkupchannelofincompletepass-through.Inparticular,we ndthatlargehigh-market-share rmsadjusttheirmarkupsbymoreinresponsetocostshocks.Thisiscon-sistentwithamodelinwhichlarger rmsalsochoosehigherlevelsofmarkups,apatternthatcanrationalizetheevidenceonmisallocationofresourcesacross rms,as,forexam-ple,documentedinHsiehandKlenow(2009).Themarkupinterpretationofthisevidenceonmisallocationdiersfromthecost-sidefrictionsinterpretationconventionalintheliter-ature(anexceptionisPeters,2011).Ourevidence,therefore,isusefulforcalibrationandquantitativeassessmentofthemodelsofmisallocationatthe rm-level.39 ReferencesAckerberg,D.,K.Caves,andG.Frazer(2006):Structuralidenti cationofproduc-tionfunctions,http://www.econ.ucla.edu/ackerber/.Alessandria,G.(2004):InternationalDeviationsFromTheLawOfOnePrice:TheRoleOfSearchFrictionsAndMarketShare,InternationalEconomicReview,45(4),12631291.Amiti,M.,andD.R.Davis(2012):Trade,Firms,andWages:TheoryandEvidence,ReviewofEconomicStudies,79(1),136.Amiti,M.,andJ.Konings(2007):Tradeliberalization,intermediateinputs,andpro-ductivity:EvidencefromIndonesia,AmericanEconomicReview,97(5),16111638.Atkeson,A.,andA.Burstein(2008):TradeCosts,Pricing-to-Market,andInterna-tionalRelativePrices,AmericanEconomicReview,98(5),19982031.Auer,R.A.,andR.S.Schoenle(2012):MarketStructureandExchangeRatePass-Through,http://people.brandeis.edu/~schoenle/research.htm.Berman,N.,P.Martin,andT.Mayer(2012):Howdodierentexportersreacttoexchangeratechanges?,QuarterlyJournalofEconomics,127(1),437492.Bernard,A.B.,E.J.Blanchard,I.VanBeveren,andH.Y.Vandenbussche(2012):Carry-AlongTrade,NBERWorkingPapersNo.18246.Bernard,A.B.,J.B.Jensen,andP.K.Schott(2009):Importers,ExportersandMultinationals:APortraitofFirmsintheU.S.thatTradeGoods,inProducerDynamics:NewEvidencefromMicroData,pp.513552.NationalBureauofEconomicResearch.Blaum,J.,C.Lelarge,andM.Peters(2013):Non-HomotheticImportDemand:FirmProductivityandQualityBias,http://www.econ.brown.edu/fac/joaquin_blaum/.Burstein,A.,andG.Gopinath(2012):InternationalPricesandExchangeRates,inHandbookofInternationalEconomics,ed.byG.Gopinath,E.Helpman,andK.Rogo,vol.IV.forthcoming.Burstein,A.,andN.Jaimovich(2008):UnderstandingMovementsinAggre-gateandProduct-levelRealExchangerates,http://econ.as.nyu.edu/docs/IO/8836/productRER.pdf.Burstein,A.T.,J.C.Neves,andS.Rebelo(2003):Distributioncostsandrealexchangeratedynamicsduringexchange-rate-basedstabilizations,JournalofMonetaryEconomics,50(6),11891214.Chatterjee,A.,R.Dix-Carneiro,andJ.Vichyanond(2012):Multi-ProductFirmsandExchangeRateFluctuations,AmericanEconomicJournal:EconomicPolicy,forth-coming.Damijan,J.P.,J.Konings,andS.Polanec(2012):ImportChurningandExportPerformanceofMulti-ProductFirms,LICOSDiscussionPaper307.DeLoecker,J.,P.K.Goldberg,A.K.Khandelwal,andN.Pavcnik(2012):Prices,MarkupsandTradeReform,NBERWorkingPaperNo.17925.Engel,C.(2001):TheSixMajorPuzzlesinInternationalMacroeconomics:IsTherea40 CommonCause?:Comment,inNBERMacroeconomicsAnnual2000,vol.15,pp.403411. (2003):ExpenditureSwitchingandExchangeRatePolicy,inNBERMacroeco-nomicsAnnual2002,vol.17,pp.231272.Fauceglia,D.,A.Shingal,andM.Wermelinger(2012):`Naturalhedging'ofex-changeraterisk:Theroleofimportedinputprices,http://www.alexandria.unisg.ch/publications/citation/Martin_Wermelinger/.Feenstra,R.C.,J.E.Gagnon,andM.M.Knetter(1996):Marketshareandex-changeratepass-throughinworldautomobiletrade,JournalofInternationalEconomics,40(12),187207.Feng,L.,Z.Li,andD.L.Swenson(2012):TheConnectionbetweenImportedIn-termediateInputsandExports:EvidencefromChineseFirms,NBERWorkingPaperNo.18260.Fitzgerald,D.,andS.Haller(2012):Pricing-to-Market:EvidenceFromPlant-LevelPrices,WorkingPaper,StanfordUniversity.Froot,K.A.,D.S.Scharfstein,andJ.C.Stein(1993):RiskManagement:Coordi-natingCorporateInvestmentandFinancingPolicies,JournalofFinance,48(5),162958.Garetto,S.(2012):Firms'HeterogeneityandIncompletePass-Through,http://people.bu.edu/garettos/research.html.Goldberg,L.,andJ.Campa(2010):TheSensitivityoftheCPItoExchangeRates:DistributionMargins,ImportedInputs,andTradeExposure,ReviewofEconomicsandStatistics,92(2),392407.Goldberg,P.,andM.Knetter(1997):GoodsPricesandExchangeRates:WhatHaveWeLearned?,JournalofEconomicLiterature,35(3),12431272.Goldberg,P.K.,andR.Hellerstein(2008):AStructuralApproachtoExplainingIncompleteExchange-RatePass-ThroughandPricing-to-Market,AmericanEconomicReview,98(2),42329. (2011):AStructuralApproachtoIdentifyingtheSourcesofLocal-CurrencyPriceStability,ReviewofEconomicStudies,forthcoming.Goldberg,P.K.,andF.Verboven(2001):TheEvolutionofPriceDispersionintheEuropeanCarMarket,ReviewofEconomicStudies,68(4),81148.Gopinath,G.,andO.Itskhoki(2010):FrequencyofPriceAdjustmentandPass-through,QuarterlyJournalofEconomics,125(2),675727. (2011):InSearchofRealRigdities,inNBERMacroeconomicsAnnual2010,vol.25,pp.261310.Gopinath,G.,O.Itskhoki,andR.Rigobon(2010):CurrencyChoiceandExchangeRatePass-through,AmericanEconomicReview,100(1),306336.Gopinath,G.,andB.Neiman(2012):TradeAdjustmentandProductivityinLargeCrises,http://faculty.chicagobooth.edu/brent.neiman/research/.Gopinath,G.,andR.Rigobon(2008):StickyBorders,QuarterlyJournalofEco-41 nomics,123(2),531575.Halpern,L.,M.Koren,andA.Szeidl(2011):ImportedInputsandProductivity,http://miklos.koren.hu/view/kmwprs/32/.Hsieh,C.-T.,andP.Klenow(2009):MisallocationandManufacturingTFPinChinaandIndia,TheQuarterlyJournalofEconomics,124(4),14031448.Kugler,M.,andE.Verhoogen(2009):PlantsandImportedInputs:NewFactsandanInterpretation,AmericanEconomicReviewPapersandProceedings,99(2),501507.Levinsohn,J.,andA.Petrin(2003):Estimatingproductionfunctionsusinginputstocontrolforunobservables,ReviewofEconomicStudies,70(2),317342.Manova,K.,andZ.Zhang(2009):China'sExportersandImporters:Firms,ProductsandTradePartners,NBERWorkingPaperNo.15249.Martin,J.,andI.Méjean(2012):InvoicingCurrency,FirmSize,andHedging,http://www.isabellemejean.com/publications.html.Melitz,M.J.(2003):TheImpactofTradeonIntra-IndustryReallocationsandAggregateIndustryProductivity,Econometrica,71(6),16951725.Nakamura,E.,andD.Zerom(2010):AccountingforIncompletePass-through,ReviewofEconomicStudies,77(3),11921230.Obstfeld,M.,andK.Rogoff(2001):TheSixMajorPuzzlesinInternationalMacroe-conomics:IsThereaCommonCause?,inNBERMacroeconomicsAnnual2000,vol.15,pp.339390.Olley,G.S.,andA.Pakes(1996):Thedynamicsofproductivityinthetelecommuni-cationsequipmentindustry,Econometrica,64,12631297.Peters,M.(2011):HeterogeneousMark-UpsandEndogenousMisallocation,http://economics.mit.edu/grad/mipeters/papers.Pierce,J.R.,andP.K.Schott(2012):Onestimating rm-levelproductionfunc-tionsusingproxyvariablestocontrolforunobservables,JournalofEconomicandSocialMeasurement,forthcoming.Wooldridge,J.(2009):Onestimating rm-levelproductionfunctionsusingproxyvari-ablestocontrolforunobservables,EconomicsLetters,104,112114.42 AOnlineAppendixA.1TheoreticalAppendixA.1.1CostfunctionandimportintensityForbrevity,wedropthe rmidenti eriinthisderivation.GivenoutputYandthesetofimportedintermediategoodsJ0,theobjectiveofthe rmisTC(YjJ0)minL;X;fXj;Zjg;fMjgWL+10VjZjdj+J0�EmUjMj+Wfdj;Denoteby, and theLagrangemultiplieronconstraints(5),(6)and(7)respectively.The rstorderconditionsofcostminimizationarerespectively:W=(1� )Y=L; = Y=X; = jX=Xj;j2[0;1];Vj= (Xj=Zj)1=(1+);j2[0;1];EmUj= (ajXj=Mj)1=(1+);j2J0;withMj=0andXj=Zjforj2~J0[0;1]nJ0.Expressingout and ,takingtheratioofthelasttwoconditionsandrearranging,wecanrewrite:WL=(1� )Y;VjXj= jY(Xj=Zj)1=(1+);j2[0;1];EmUjMj VjZj=ajEmUj Vj�;j2J0:Substitutingthelastexpressioninto(7),weobtainXj=Zj1+aj(EmUj=Vj)�1+ forj2J0,whichtogetherwiththeexpressionforVjXjaboveyields:VjXj= jYbj;j2J0; jY;j2~J0;wherebj1+aj(EmUj=Vj)�1=:(A1)Basedonthis,weexpressLandXjforallj2[0;1]asfunctionsofYandparameters.Substitutingtheseexpressionsinto(5)(6),wesolvefor=1 0@expn10 jlogVj jdjo expnJ0 jlogbjdjo1A W 1� 1� =C B ;(A2)43 whereB=expJ0 jlogbjdj(A3)andCisde nedinfootnote13.Finally,wesubstitutetheexpressionforWL,VjZj=VjXj(Zj=Xj)andEmUjMj=VjZj(EmUjMj=(VjZj))intothecostfunctiontoobtainTC(Y;J0)=Y+J0Wfdj:(A4)ChoiceofJ0withoutuncertaintysolvesminJ0TC(YjJ0),givenoutputY.42Consideraddinganadditionalvarietyj0=2J0tothesetJ0.ThenetchangeinthetotalcostfromthisisgivenbyY@ @BB j0logbj0+Wf=� Y j0logbj0+Wf;since j0logbj0istheincreaseinlogBfromaddingj0tothesetofimportsJ0.Notethat Y=10VjZjdj+J0EmUjMjdjisthetotalmaterialcostofthe rm.Therefore,theoptimalchoiceofJ0mustsatisfythefollowing xedpoint:J0=8:j2[0;1]: C= expn J0 `logb`d`oY jlogbjWf9=;:Thisimmediatelyimpliesthatoncej'saresortedsuchthat jlogbjisdecreasinginj,thesetofimportedinputsisanintervalJ0=[0;j0]forsomej02[0;1].Furthermore,theconditionforj0canbewrittenas:j0=max8:j2[0;1]: C= expn j0 `logb`d`oY jlogbjWf9=;;(A5)andsuchj0isuniquesincetheLHSoftheinequalityisdecreasinginj.FigureA2providesanillustration.ProofofProposition2Thefractionofvariablecostspentonimportsisgivenby'=J0EmUjMjdj Y=J0 j(1�bj)dj;whereweusedthe rstorderconditionsfromthecostminimizationabovetosubstituteinforEmUjMj.Notethat'increasesinJ0,andinparticularwhenJ0=[0;j0],'increasesinj0.Therefore,from(A5)itfollowsthat'increasesintotalmaterialcostTMC= Y= [CY]=[B ]anddecreasesin xedcostWf.Fromthede nitionoftotalcost(A4),holdingJ0constant,themarginalcostequals 42We rstconsiderthecasewithoutuncertaintytoestablishthefundamentaldeterminantsofimportintensityinasimplersetup,andnextgeneralizetheresultstothecasewithuncertainty(cf.(A5)and(A6)).44 MC(J0)=de nedin(A2).Wehave:@logMC(J0) @logEm=@log @logB@logB @logEm=� J0 j@logbj @logEmdj=';sincefrom(A1)@logbj=@logEm=�(1�bj).A.1.2PricesettingandexantechoiceofJ0UndertheassumptionthatJ0isasunkdecisionchosenbeforeuncertaintyisrealized,wecanwritethefullproblemofthe rm(bringingbackthe rmidenti eri)as:maxJ0;iE(maxYi;(Pk;i;Qk;i)(Xk2KiEkPk;iQk;i�TCi(YijJ0;i)));subjecttoYi=Pk2KiQk;i,with(Pk;i;Qk;i)satisfyingdemand(1)ineachmarketk2Ki,andtotalcostgivenin(A4).WeassumethatJ0;iischosenjustpriortotherealizationofuncertaintyaboutaggregatevariables,andforsimplicityweomitastochasticdiscountfactorwhichcanbeaddedwithoutanyconceptualcomplications.Substitutingtheconstraintsintothemaximizationproblemandtakingthe rstordercondition(withrespecttoPk;i),weobtain:EkQk;i+EkPk;i@Qk;i @Pk;i�@TCi(YjJ0;i) @Y@Qk;i @Pk;i=0;whichwerewriteasEkQk;i(1�k;i)+k;iQk;ii Pk;i=0;wherek;iisde nedin(3)andi=MCi(J0;i)isde nedin(A2).RearrangingandusingPk;i=EkPk;i,resultsinthepricesettingequation(11).NowconsiderthechoiceofJ0;i.BytheEnvelopeTheorem,itisequivalenttominJ0;iEfTCi(YijJ0;i)g;whereYiistheequilibriumoutputofthe rmineachstateofnature.Therefore,thisproblemisnearlyidenticaltothatofchoosingJ0;iwithoutuncertainty,withtheexceptionthatnowwehavetheexpectationandYivariesacrossstatesoftheworldalongwithexogenousvariablesaectingTCi.Asaresult,wecanwritethe xedpointequationforJ0;iinthiscaseas:J0;i=(j2[0;1]:E( C= i expn J0;i `logb`d`oYi jlogbj)EfWfig):(A6)Therefore,J0;istillhasthestructure[0;j0;i],butnowweneedtosortgoodsjindecreasingorderbythevalueoftheLHSintheinequalityin(A6)(inexpectedterms).45 A.1.3EquilibriumRelationshipsToillustratetheimplicationsofthemodelfortheequilibriumdeterminantsofmarketshareandimportintensity,westudythefollowingsimplecase.Considertwo rms,iandi0,inagivenindustryandbothservingasingledestinationmarketk.The rmsfacethesameindustry-destinationspeci cmarketconditionsre ectedinEk,Pk,Dk,Cand .Weallowthe rmstobeheterogeneousintermsofproductivity i,demand/qualityshifterk;iandthe xedcostofimportingfi.Forasingle-destination rmwehaveYi=Qk;i,andwedropindexkinwhatfollowsforbrevity.Wewanttocharacterizetherelativemarketsharesandimportintensitiesofthesetwo rms.Inordertodoso,wetaketheratiosoftheequilibriumconditions(demand(1),marketshare(2)andprice(11))forthesetwo rms:43Yi Yi0=i i0Pi Pi0�;Si Si0=i i0Pi Pi01�andPi Pi0=Mi Mi0B i0 i0 B i i;whereMi=i=(i�1)andi=(1�Si)+Si.Log-linearizingrelativemarkup,wehave:logMi Mi0=� �1logSi Si0;where�ismarkupelasticitygivenin(4)evaluatedatsomeaverageS.Usingthis,welinearizetheequilibriumsystemtosolvefor:logSi Si0=1 1+�logi i0+�1 1+�log i i0+ logBi Bi0(A7)andtheinterimvariable(totalmaterialcost),whichdeterminestheimportchoice:logTMCi TMCi0=logYi Yi0�log i i0� logBi Bi0=1�� �1logSi Si0:(A8)AssumptionA1�(�1).Thisassumptionimpliesthatthe(levelof)markupdoesnotvarytoomuchwiththeproductivityofthe rm,sothathigh-market-share rmsaresimultaneouslyhigh-material-cost rms(aswedocumentisthecaseinthedata,seeTable4).44Consequently,underA1,high-market-share rmschoosetobemoreimportintensive,aswediscussnext.Denote (j) jElogbj,wheretheexpectationisoveraggregateequilibriumvariables(i.e.,aggregatestatesoftheworld),andsortjsothat 0()0on[0;1].Assumingthe 43Notethattakingtheseratiostakesouttheaggregatevariablessuchasthepriceindex.Intuitively,wecharacterizetherelativestandingoftwo rmsinagivengeneralequilibriumenvironment,andaggregateequilibriumvariablessuchasthepriceindex,whichaectoutputsandmarketsharesof rmsproportionately,dropout.44Thisassumptionisnotveryrestrictivefortheparametersofthemodel,asforamoderatevalueof=4,itonlyrequiresS0:8(giventhede nitionof�in(4)and1).46 choiceoftheimportsetisinternalforboth rms,wecanrewrite(A6)asaconditionforacutoj0(i):E( j0(i)logbj0(i) CYi B i i)=EfWfig;andlog-linearizeittoyield:� 0�j0 �j0�j0(i)�j0(i0)=ElogYi Yi0�log i i0� logBi Bi0�logfi fi0;wherej0issomeaveragecutovariety.Finally,usingde nition(A3),wehaveElogBi Bi0= �j0�j0(i)�j0(i0):(A9)Combiningtheabovetwoequationswith(A8),wehave:� 0�j0 �j02 ElogBi Bi0=1�� �1ElogSi Si0�logfi fi0:Combiningwith(A7),wesolvefor: ElogBi Bi0=1 0� 1+��1"1�� �1 1+�logi i0+(�1)log i i0�logfi fi0#;(A10)ElogSi Si0=1 0� 1+��10 1+�logi i0+(�1)log i i0��1 1+�logfi fi0;(A11)where0� 0�j0=[ �j02]�0.AssumptionA20� 0�j0 �j02� 1+��1.TheparameterrestrictioninA2isalocalstabilitycondition:thefunction (j)=E jlogbjmustbedecreasinginjfastenough,otherwisesmallchangesinexogenous rmcharacteristicscanhavediscontinuouslylargechangesintheextensivemarginofimports.Weviewitasatechnicalcondition,andassumeequilibriumislocallystable.Finally,werelateimportintensityofthe rm'itoBi.Fromde nition(9)itfollowsthatE'i�'i0 =�j0�j0(i)�j0(i0)=�j0 �j0ElogBi Bi0;(A12)where(j)= jEf1�bjgandthesecondequalitysubstitutesin(A9).Equations(A10)(A12)providethelog-linearcharacterizationof(expected)relativemar-ketshareandrelativeimportintensitiesofthetwo rmsasafunctionoftheirrelativeexoge-47 nouscharacteristics.Theseapproximationsarenearlyexactwhentheexogenousdierencesbetween rmsaresmall.Inotherwords,onecanthinkofthoserelationshipsasdescribingelasticitiesofmarketshareandsemi-elasticitiesofimport-intensitywithrespecttoexoge-nouscharacteristicsofthe rm(productivity,demand/qualityand xedcostofimporting),holdingthegeneralequilibriumenvironmentconstant.Therefore,wehave:PropositionA1UnderAssumptionsA1andA2,the(expected)marketshareandimportintensityofthe rmarebothincreasinginitsproductivityandquality/demandshifter,andarebothdecreasinginthe rm's xedcostofimporting,inagivengeneralequilibriumenvi-ronment(thatis,holdingthecompositionof rmsconstant).Asimilarresultcanbeprovedfor rmsservingmultipleanddierentnumbersofdestina-tions.A.1.4Pass-throughrelationshipandproofofProposition3MarkupGiven(2)and(3),wehavethefollowingfulldierentials:dlogMk;idlogk;i k;i�1=(�)Sk;i k;i(k;i�1)dlogSk;i=�k;idlogSk;i �1;dlogSk;i=dlogk;i�(�1)�dlogPk;i�dlogPk;where�k;iisasde nedin(4).Combiningthesetwoexpressionsresultsin(13).MarginalcostTakingthefulldierentialof(10),wehave:dlogMCi=dlogC i� dlogBi:Usingde nitions(A1)and(A3),andundertheassumptionthatJ0isasunkdecision(thatis,thesetofimportedgoodsisheldconstant),wehave:dlogbj=�(1�bj)dlogEmUj Vj; dlogBi= J0;i j�dlogbjdj=�'idlogEmU V� J0;i j(1�bj)dlogUj U�dlogVj Vdj;where'iisde nedin(9),anddlogV=10 j�dlogVjdjdjandsimilarlydlogU=10 j�dlogUjdjdj.Substitutingthisexpressionintothefulldierentialofthemarginal48 costaboveresultsin(14),wheretheresidualisgivenby:MCi=J0;i j(1�b�j)dlogUj U�dlogVj Vdj�dlog i ;wheredlog isthesectoralaveragechangein rm-levelproductivity.Combining(13)and(14)with(12),wehave:dlogPk;i=��k;i�dlogPk;i�dlog~Pk+dlogC +'idlogEmU V+k;i;(A13)wherek;iMCi+�k;i �1Mk;i;Mk;idlogk;i k;dlogkisthesector-destinationaveragechangeindemand/qualityacross rms,andwedenotewith~Pk1 �1kPkthesector-destinationpriceindexadjustedfortheaveragede-mand/qualityshifterforBelgian rms.Wemakethefollowing:AssumptionA3�MCk;i;Mk;i,andhencek;i,aremeanzeroandindependentfromdlogEmanddlogEk.Notethatk;ire ectsthe rmidiosyncraticdierencesinthechangeininputprices,produc-tivityanddemand/qualityshifter,andthereforeAssumptionA3isanaturalonetomake.Essentially,weassumethatthereisnosystematicrelationshipbetweenexchangeratemove-mentand rm'sidiosyncraticproductivityordemandchangerelativetoanaverage rmfromthesamecountry(Belgium)servingthesamesector-destination.Thisnonethelessal-lowstheexchangeratestobecorrelatedwithsector-destinationaverageindexesforcostsandproductivity(thatis, ,U,V,aswellas~Pk).SubstitutingdlogPk;i=dlogPk;i�dlogEkinto(A13)andrearranging,wearriveat:dlogPk;i=�k;i 1+�k;idlogEk+'i 1+�k;idlogEmUs Vs+�k;idlog~Ps;k+dlogCs ~ s;k+k;i 1+�k;i;(A14)wherewehavenowmadethesectoridenti ersanexplicitsubscript(eachiuniquelydeter-miness,hencewedonotcarryswheniispresent).Notethat�k;iisincreasinginSk;i.Wenowlinearize(A14)in'iandSk;i:LemmaA1Logpricechangeexpression(A14)linearizedin'iandSk;iisdlogPk;i�s;k 1+�s;kdlogEk+gs;k 1+�s;k~Sk;idlogEk+1 1+�s;k'idlogEmUs Vs(A15)+24�s;kdlog~Ps;k+dlogCs ~ s;k+0k;i 1+�s;k+gs;kdlog~Ps;k�'sdlogEmUs Vs�dlogCs ~ s;k+00k;i 1+�s;k~Sk;i35;49 where�s;k=�k;iSs;k,gs;k@log(1+�k;i)=@Sk;iSs;k,Ss;kissomeaveragestatisticoftheSk;idistribution,~Sk;i=Sk;i�Sk;i,and0k;iMCi+�s;k �1Mk;i,00k;i�s;k �1Mk;i�MCi.Proof:Giventhede nitionsof�s;kandgs;kinthelemma,wehavethefollowing rst-orderapproximations:1 1+�k;i1�gs;k~Sk;i 1+�s;k;�k;i 1+�k;i�s;k+gs;k~Sk;i 1+�s;kand'i 1+�k;i'i�'sgs;k~Sk;i 1+�s;k:Substitutetheseapproximationsinto(A14)andrearrangetoobtain(A15).ProofofProposition3Divide(A15)throughbydlogEkandtakeexpectationstochar-acterizethepass-throughelasticity: k;iEdlogPk;i dlogEk s;k+s;k'i+ s;kSk;i;where s;k=�s;k(1+ Ps;k)+ Cs;k 1+�s;k� s;kSs;k;s;k= Ms;k 1+�s;kand s;k=gs;k(1�'s Ms;k)+( Ps;k� Cs;k) 1+�s;k;andwith Pk;iE(dlog~Ps;k dlogEk); Cs;kE(dlog(Cs=~ s;k) dlogEk); Ms;kEdlog(EmUs=Vs) dlogEk:Notethatthetermsink;idropoutsince,duetoAssumptionA3,Ek;i=dlogEk =0.Finally,notethat s;kcov(;dlogEk)=var(dlogEk),thatis -termsareapproximatelyprojectioncoe cients.Theexpectationsandthede nitionsof -termsareunconditional,andhenceaverageacrossallpossibleinitialstatesandpathsoftheeconomy.A.1.5Empiricalspeci cationandproofofProposition4Westartfromthelinearizeddecomposition(A15)byreplacingdierentialdwithatimechangeoperator,makingthetimeindextexplicit,andrearranging:pi;k;t�s;k~ps;k;t+cs;t+0k;i;t 1+�s;k+gs;k�~ps;k;t�cs;t+00k;i;t 1+�s;k~Sk;i;t�1(A16)+�s;kek;t 1+�s;k+'i;t�1 1+�s;klogEm;tUs;t Vs;t+gs;k~Sk;i;t�1 1+�s;kek;t�'s;t�1logEm;tUs;t Vs;t;wherepi;k;tlogPk;i;t�logPk;i;t�1,ek;tlogEk;t�logEk;t�1,cs;tlog(Cs;t= s;t)�log(Cs;t�1= s;t�1),and~ps;k;tlog~Ps;k;t�log~Ps;k;t�1.Notethatwechoset�1asthepoint50 ofapproximationfor~Sk;i;t�1and'i;t�1.Wealsochosetheapproximationcoe cients�s;kandgs;knottodependontimebyevaluatingtherespectivefunctions(seeLemmaA1)atatime-invariantaverageSs;k.Nextconsiderourmainempiricalspeci cation(21)whichwereproduceas:pi;k;t= s;k+'i;t�1+~ Sk;i;t�1 Ss;k;t�1ek;t+s;k+b'i;t�1+cSk;i;t�1 Ss;k;t�1+~uk;i;t;(A17)whereSs;k;tisthecumulativemarketshareofallBelgianexporters.Ourgoalistoestab-lishthepropertiesoftheOLSestimatorofand~ inthisregression,givenapproximatestructuralrelationship(A16).Tothisend,weintroducetwoassumptions:AssumptionA4Foreveryk,logek;tismeanzero,constantvarianceandindependentfrom('i;t�1;Sk;i;t�1;Ss;k;t�1).AssumptionA5Thevarianceandcovarianceof('i;t�1;Sk;i;t�1=Ss;k;t�1)within(s;k;t�1)areindependentfrom(s;k; s;kSs;k;t�1),wheres;kand s;karede nedintheproofofPropo-sition3above.AssumptionA4isaplausiblemartingaleassumptionfortheexchangerate,whichwerequireintheproofofProposition4.Oneinterpretationofthisassumptionisthatthecross-sectiondistributionof rm-levelcharacteristicsisnotusefulinpredictingfutureexchangeratechanges.AssumptionA5,inturn,isonlymadeforconvenienceofinterpretation,andqualitativelytheresultsofProposition4donotrequireit.Essentially,weassumethatthecross-sectiondistributionof rm-characteristicswithinsector-destinationdoesnotdependontheaggregatecomovementpropertiesofsectoralvariableswhichaectthevaluesofs;kand s;k.BeforeprovingProposition4,weintroducethefollowingthreeprojections:8������&#x]TJ ;� -2;.51; Td;&#x [00;&#x]TJ ;� -2;.51; Td;&#x [00;&#x]TJ ;� -2;.51; Td;&#x [00;&#x]TJ ;� -2;.51; Td;&#x [00;&#x]TJ ;� -2;.51; Td;&#x [00;&#x]TJ ;� -2;.51; Td;&#x [00;:logEm;tUs;t Vs;tMs;kek;t+vMs;k;t;Ms;k=covlogEm;tUs;t Vs;t;ek;t var(ek;t);~ps;k;tPs;kek;t+vPs;k;t;Ps;k=cov(~ps;k;t;ek;t) var(ek;t);cs;tCs;kek;t+vCs;k;t;Cs;k=cov(cs;k;t;ek;t) var(ek;t)(A18)andtherefore(vMs;k;t;vPs;k;t;vCs;k;t)areorthogonalwithek;t.Notethat(Ms;k;Ps;k;Cs;k)aretheempiricalcounterpartsto( Ms;k; Ps;k; Cs;k)de nedintheproofofProposition3.51 ProofofProposition4Substituteprojections(A18)into(A16)andrearrange:pi;k;t26664�s;k(1+Ps;k)+Cs;k 1+�s;k| {z } s;k+Ms;k 1+�s;k| {z }s;k'i;t�1+[(1�'sMs;k)+(Ps;k�Cs;k)]gs;kSs;k;t�1 1+�s;k| {z }~ s;k;tSk;i;t�1 Ss;k;t�137775ek;t+vMs;k;t 1+�s;k| {z }bs;k'i;t�1+�vPs;k;t�vCs;k;t�'s;t�1vMs;k;tgs;kSs;k;t�1 (1+�s;k)2| {z }cs;k;tSk;i;t�1 Ss;k;t�1+uk;i;t;uk;i;t=�s;kvPs;k;t+vCs;k;t+0i;t 1+�s;k++gs;kSk;i;t�1 (1+�s;k)200k;i;t:Comparingthisequationwiththeempiricalspeci cation(A17),theresidualintheempiricalspeci cationisgivenby:~uk;i;t=uk;i;t+h(s;k�)'i;t�1+(~ s;k;t�~ )Sk;i;t�1 Ss;k;t�1iek;t+(bs;k�b)'i;t�1+(cs;k;t�c)Sk;i;t�1 Ss;k;t�1:De nexk;i;t=(10s;k;'i;t�1;~Sk;i;t�1)0,sothatwecanwriteourregressorsasz0k;i;t=(x0k;i;t;x0k;i;tek;t).FromAssumptionsA3andA4andpropertiesoftheprojection(A18),itfollowsthatx0k;i;tek;tisorthogonalwithx0k;i;t,andx0k;i;tek;tisuncorrelatedwithuk;i;t.Therefore,thepropertiesoftheestimatesof( s;k;;~ )areindependentfromthoseof(s;k;b;c).OLSidenti es( s;k;;~ )fromthefollowingmomentconditions:0=Ek;i;tfxk;i;tek;t~uk;i;tg=Ek;i;tfxk;i;tek;t(~uk;i;t�uk;i;t)g;wherethesecondequalityfollowsfromEk;i;tfek;txk;i;tuk;i;tg=0(duetoAssumptionA3andprojection(A18)).Wenowrewritethismomentconditionintheformofsummation(acrossthepopulationof rms,sector-destinations,andtimeperiods/states):0=Xk;i;txk;i;tek;t(~uk;i;t�uk;i;t)=Xk;i;te2k;txk;i;tx0k;i;t�00s;k;s;k�;~ s;k;t�~ 0;wherethesecondequalitysubstitutesintheexpressionfor~uk;i;t�uk;i;tandusesthefactthatek;tisorthogonalwithxk;i;t(AssumptionA4).Usingthesameassumptionfurther,wecanrewritethelastexpressionas:Xs;k;t2kns;k;ts;k;ts;k�~ s;k;t�~ =0;(A19)where2kisthevarianceofek;t,s;k;tisthecovariancematrixfor('i;t�1;Sk;i;t�1=Ss;k;t�1)within(s;k;t�1),andns;k;tistherespectivenumberofobservations.Equation(A19)alreadyestablishestheresultofthepropositionthatand~ identifygen-eralizedweightedaveragesoftherespectivecoe cients.UnderadditionalAssumptionA5,52 wehaveaparticularlysimpleexpressionsfortheseweightedaverages:=Xs;k;t!0s;k;ts;kand~ =Xs;k;t!00s;k;t~ s;k;t;!0s;k;t/2kns;k;tvars;k;t�1('i;t�1)and!00s;k;t/2kns;k;tvars;k;t�1(Sk;i;t�1=Ss;k;t�1)withvars;k;t�1()denotingthevarianceforobservationswithin(s;k;t�1).Finally,s;kand~ s;k;t= s;kSs;k;t�1arede nedabove,and(s;k; s;k)provide rst-orderapproximationstotheiranalogsinProposition3since(Ms;k;Ps;k;Cs;k)( Ms;k; Ps;k; Cs;k).A.1.6SelectionbiasInthisappendixweprovideabriefexpositionofthetheoreticalargumentsforthedirectionofthepotentialbiasofthecoe cientestimatesinequation(21)duetosampleselection.Wealsoprovidecorroboratingempiricalevidence.Forsimplicity,imagineanenvironmentwithnosunkcostandonly xedcostFk;iofsup-plyingeachmarketkfor rmi,anddenotewithk;itheoperatingpro tof rmifromservingmarketk.Thentheselectionequationisk;iFk;i,orequivalentlylog�k;i=Fk;i0.Ageneralapproximationtothepro tfunction,whichcanalsobederivedfromthestructureofthepro tmaximizationproblemintroducedinSection2.3,resultsinthefollowingselectioncondition:logk;i;t Fk;i;ts;k+t+k;i;t�1+ek;t+vk;i;t0;(A20)wheres;kandtaresector-destinationandyeardummies,k;i;t�1isacombinationof rm-destinationcharacteristics(suchasproductivity ianddemandshifterk;i)andvk;i;tisanidiosyncratic rm-destinationshockinperiodt.Inwords,approximation(A20)impliesthat rmsaremorelikelytostayinthesampleunderfavorableindustry-destination-yearconditions(highs;k+t),whenthedomesticexchangeratedepreciates(highek;t),when rmshavestrongfundamentals(largek;i;t�1),andwhen rmsfaceafavorableidiosyncraticshock(largevk;i;t).Forourpurposesweprojectk;i;t�1=a'i;t�1+bSk;i;t�1+k;i;t�1.Nextconsiderourempiricalspeci cation:pk;i;t=q+ qek;t+uk;i;t;(A21)estimatedwithinbinsqofimportintensityandmarketshare(analogoustoTable6).Weassumethatuk;i;tisnegativelycorrelatedwithvk;i;t.Intuitively,thisimpliesthatanadversemarginalcostshock(e.g.,negativeproductivityshock)bothreducesvk;i;tandincreasesuk;i;t.Thisassumptioncanbeformallyderivedfromthestructureofourmodel:wecandecomposevk;i;t=zk;i;t�uk;i;t,where�0anduk;i;tandzk;i;tareuncorrelated.Thentheselectionequation(A20)canberewrittenas:uk;i;t k;i;t+ ek;t;where k;i;t1 [s;k+t+a'i;t�1+bSk;i;t�1+k;i;t+zk;i;t]:Wedirectlyverifyinthedatathata;b;�0(seethetablebelow).53 Giventhiseconometricmodel,wecandirectlyevaluatethemagnitudeofthebiasofanOLSestimateof q.First,foreachbinqwecalculate:Epk;i;tjek;t =q+ qek;t+fq(ek;t);wherefq(ek;t)=Euk;i;tek;t;uk;i;t k;i;t+ ek;t:Withuk;i;tunconditionallymeanzero,wehavethefollowingproperties(provided0):fq()0;fq(1)=0;f0q()0;fq0()fq();f0q0()f0q():Thelasttwopropertiescomefromthefactthatinbinq0withhigher'i;t�1and/orSk;i;t�1thedistributionof k;i;tisshiftedtotheright( rstorderstochasticallydominates),relativetothatforbinq.Inthespecialcaseof=0,wehavefq()fq,aq-speci cconstant.Giventhiscalculation,wecanevaluatethebiasintheOLSestimateof qasthestandardomittedvariablebias:bias(^ q)=plim�^ q� q=cov�pk;i;t;ek;t var�ek;t� q=cov�fq(ek;t);ek;t var�ek;t0;sincef0q()0whenever�0.For=0,thebiasequalszero.Furthermore,thebiasis(weakly)smaller(closertozero)forbinq0thanforbinq,if'i;t�1and/orSk;i;t�1arehigherinbinq0thaninbinq.Tosummarize,whenevera;b;�0,theOLSestimateshaveanupwardbiasin (down-wardbiasinthelevelofpass-through),whichdiminisheswithimportintensityandmarketshare.Thisinturnimpliesadownwardbiasinand .ThetablebelowestimatesaProbitregressionfortheprobabilityofstayinginthesample(f;t=1): Pff;t=1jf;t�1=1g(1)(2)(3)(4) 'f0.0660.0840.0250.284(0.025)(0.025)(0.026)(0.027)Sf;s;k;t�10.5640.5580.7100.777(0.011)(0.011)(0.011)(0.013)ek;t0.7450.3540.0900.094(0.048)(0.065)(0.067)(0.068) FixedEectstt+kt+k+s#obs.172,988172,988172,988172,988 Note:t;k;sareyear,countryandindustry(HS2-digit) xedeectsrespectively.Thedependentvariableequals1in67.6%oftheobservations.Thiscon rmsthat rmswithhighimportintensityandmarketsharearelesslikelytodropoutofthesample(a;b�0).Further,thistableprovidesevidencethatexitismorelikelyinresponsetoexchangerateappreciation(ek;t0),thatis&#x]TJ/;༕ ;.9;Ւ ;&#xTf 1;.42; 0 ;&#xTd [;0.54 A.2DataAppendixTradeDataTheimportandexportdataarefromtheNationalBankofBelgium,withtheextra-EUtransactionsreportedbyCustomsandtheintra-EUtradebytheIntrastatInquiry.Thesedataarereportedatthe rmlevelforeachproductclassi edatthe8-digitcombinednomenclature(CN)invaluesandweightsorunits.NotethattheCNcodeisaEurope-basedclassi cationwiththe rst6-digitscorrespondingtotheWorldHamonizedSystem(HS).Weincludealltransactionsthatareconsideredastradeinvolvingchangeofownershipwithcompensation(codes1and11).Thesedataareverycomprehensive,coveringall rmswithatotalextra-EUtradewhosevalueisgreaterthan1,000eurosorwhoseweightismorethan1,000kilograms.Since2006,evensmallertransactionsarereported.However,forintra-EUtrade,thethresholdsarehigher,withtotalintra-EUimportsorexportsabove250,000eurosinayear,andin2006thisthresholdwasraisedto1,000,000eurosforexportsand400,000forimports.NotethatthesethresholdsresultinchangingcutosforcountriesthatjoinedtheEUduringoursampleperiodastheirtransactionsmovefrombeingrecordedbyCustomstotheIntrastatInquiry.Firm-leveldataThe rm-leveldataarefromtheBelgianBusinessRegistry,coveringallincorporated rms.Theseannualaccountsreportinformationfrombalancesheets,incomestatements,andannexestotheannualaccounts.Onlylarge rmsarerequiredtoprovidefullannualaccountswhereassmall rmshavetoonlyprovideshortannualaccountssothatsomevariablessuchassales,turnover,andmaterialcostsmaynotbeprovidedforsmall rms.Alarge rmisde nedasacompanywithanaverageannualworkforceofatleast100workersorwhenatleasttwoofthefollowhingthreethresholdsaremet:(i)annualaverageworkforceof50workers,(ii)turnover(excludingVAT)amountstoatleast7,300,000euros,or(iii)totalassetsexceeding3,650,000euros.Notethatthelasttwothresholdsarealteredeveryfouryearstotakeaccountofin ation.Althoughlessthan10percentofthecompaniesinBelgiumreportfullannualaccounts,for rmsinthemanufacturingsectortheseaccountformostofvalueadded(89percent)andemployment(83percent).Each rmreportsa5-digitNACEcodebasedonitsmaineconomicactivity.Thekeyvariableofinterestistheconstructionof'de nedastheratiooftotalnon-Euroimportstototalcosts(equaltowagesplustotalmaterialcosts).Thesetotalcostvariablesarereportedby58percentofexportersinthemanufacturingsector.Combiningthisinformationwiththeimportdata,wecanset'equaltozerowhentotalnon-Euroimportsarezeroeveniftotalcostsarenotreported,givingusa'for77percentofmanufacturingexporters,whichaccountfor98percentofallmanufacturingexports.Notethatinlessthanhalfapercentoftheobservations,totalimportsweregreaterthanmaterialcostsinwhichcasewetreated'asmissing.ProductConcordancesWeuseSITCone-digitproductcodes(5to8)toidentifyaman-uacturingexportasitisnotpossibletodosodirectlyfromtheCN8-digitclassi ca-tionsnorfromitscorrespondingHS6-digitcode.Weconstructaconcordancebetween55 CN8-digitcodesandSITCRevision3bybuildingonaconcordancebetweenHS10-digitandSITC5-digitfromPeterSchott'swebsite,whichtakesintoaccountrevisionstoHScodesupto2006(seePierceandSchott,2012).WeupdatethistotakeaccountofHS6-digitrevisionsin2007usingtheconcordancefromtheU.S.ForeignCensus(seehttp://www.census.gov/foreign-trade/reference/products/layouts/imhdb.html).Webe-ginbytakingthe rst6-digitsofthe8-digitCNcode,whichiseectivelyanHS6-digitcode,andweincludeonlythecorrespondingSITCcodewhenitisauniquemapping.SomeHS6-digitcodesmaptomultipleSITCcodes,sothatinthosecaseswedonotincludeacorrespondingSITCcode.ThishappensmainlywhenwegettothemoredisaggregatedSITCcodesandrarelyattheone-digitSITCcode.Second,weneedtomatchtheCNcodestoinput-output(IO)codes.Weusea2005BelgiumIOmatrixwith74IOcodesofwhich56arewithinthemanufacturingsector.TheIOcodesarebasedontheStatisticalClassi cationsofProductbyActivity,abbreviatedasCPA,whichinturnarelinkedtotheCN8-digitcodesusingtheEurostatcorrespondencetables.ThematchingoftheIOcodestotheCN8-digitwasnotstraightforwardaswehadtodealwiththemany-to-manyconcordanceissues.WeincludedanIOcodeonlywhenthematchfromtheCNcodewasclear.SampleOursampleisfortheyears2000to2008,beginningwiththe rstyearaftertheeurowasformed.Wekeepall rmsthatreporttheirmaineconomicactivityinmanufacturingde nedaccordingto2-digitNACEcodes15to36,thusexcludingwholesalers,mining,andservices.Werestrictexportstothosethatarede nedwithinthemanufacturingsector(SITCone-digitcodes5to8).Toaddressthemultiproduct rmissue,wekeeponlythesetofCN8-digitcodesthatfallswithina rm'smajorIOexport,whichweidentifyasfollows.WeselectanIOcodeforeach rmthatre ectsthe rm'slargestexportshareoverthesampleperiodandthenkeepallCNcodesthatfallwithinthatIOcode.Formostoftheanalysis,wefocusonexportstononeuroOECDcountriesthatarede nedasadvancedbytheIMFandhigh-incomebytheWorldBank.Wekeepallimportproductcodesandallimportsourcecountries.Forsomerobustnesschecks,welimitthesetofimportstointermediateinputsde nedeitherusingtheBelgium2005IOtableoraccordingtoBroadEconomicCodes(BEC).Seehttp://unstats.un.org/unsd/cr/registry/regcst.asp?Cl=10,wherewede neintermediateinputsasincludingcodes111,121,2,42,53,41,and521.TotalFactorProductivityMeasuresWemeasuretotalfactorproductivity(TFP)foreach rmby rstestimatingproductionfunctionsforeach2-digitNACEsectorseparately.Wenotethatakeyproblemintheestimationofproductionfunctionsisthecorrelationbetweeninputsandunobservableproductivityshocks.ToaddressthisendogeneityproblemweestimateTFPusingtwodierentmethodologies.The rstapproachisbasedonLevinsohnandPetrin(2003)(LP),whoproposeamodi cationoftheOlleyandPakes(1996)(OP)estimator.OPusesinvestmentasaproxyforunobservableproductivityshocks.However,LP ndsevidencesuggestingthatinvestmentislumpyandhencethatinvestmentmaynot56 respondsmoothlytoaproductivityshock.Asanalternative,LPusesintermediateinputs,suchasmaterials,asaproxyforunobservedproductivity.Inparticular,weassumeaCobbDouglasproductionfunction,f;t=0+llf;t+kkf;t+!f;t+f;t;(A22)wheref;trepresentsthelogofvalueadded,lf;tisthelogofthefreelyavailableinput,labor,andkf;tisthelogofthestatevariable,capital.Theerrortermconsistsofacom-ponentthatre ects(unobserved)productivityshocks,!f;t,andawhitenoisecomponent,f;t,uncorrelatedwiththeinputfactors.Theformerisastatevariable,notobservedbytheeconometricianbutwhichcanaectthechoicesoftheinputfactors.Thissimultaneityprob-lemcanbesolvedbyassumingthatthedemandfortheintermediateinputs,xf;t,dependsonthestatevariableskf;tand!f;t,andxf;t=xf;t(kf;t;!f;t):(A23)LPshowsthatthisdemandfunctionismonotonicallyincreasingin!f;tandhencethein-termediatedemandfunctioncanbeinvertedsuchthattheunobservedproductivityshocks,!f;t,canbewrittenasafunctionoftheobservedinputs,xf;tandkf;t,or!f;t=!(kf;t;xf;t).Atwo-stepestimationmethodisfollowedwhereinthe rststepsemi-parametricmethodsareusedtoestimatethecoe cientonthevariableinput,labor.Inthesecondstep,theco-e cientoncapitalisestimatedbyusingtheassumption,asinOP,thatproductivityfollowsa rst-orderMarkovprocess.However,aspointedoutbyAckerberg,Caves,andFrazer(2006),apotentialproblemwithLPisrelatedtothetimingassumptionofthefreelyavailableinput,labor.Iflaborischosenoptimallybythe rm,itisalsoafunctionoftheunobservedproductivityshockandcapital.Thenthecoe cientonthevariableinputcannotbeidenti ed.Wooldridge(2009)showshowthetwo-stepsemi-parametricapproachcanbeimplementedusingauni edone-stepGeneralizedMethodsofMoments(GMM)framework.ThisisthesecondmethodologythatweadoptforestimatingTFP.Inparticular!f;t=!(kf;t;xf;t)isproxiedbyalaggedpolynomialincapitalandmaterials,whichcontrolsforexpectedproductivityint.Weuseathird-orderpolynomialincapitalandmaterialinourestimation.Todealwiththepotentialendogeneityoflabor,weuseits rstlagasaninstrument.Abene tofthismethodisthatGMMusesthemomentconditionsimpliedbytheLPassumptionsmoree ciently.ThelogofTFPmeasuresarenormalizedrelativetotheir2-digitNACEsectormeantomakethemcomparableacrossindustries.Thecorrelationbetweenbothmeasuresisveryhighat99percent.57 A.3AdditionalFiguresandTables FigureA1:Cumulativedistributionfunctionsofimportintensity'fandmarketshareSf;s;k;tNote:Estimatedcumulativedistributionfunctions.Intheleftpanel,theuppercdfcorrespondstotheun-weighted rmcount,whilethelowercdfweights rmobservationsbytheirexportvalues.Theunweighteddistributionof'fhasamasspointof24%at'f=0,whilethismasspointlargelydisappearsinthevalue-weighteddistribution,whichinturnhasastep'f=0:33correspondingtothelargestexporterinoursamplewithanexportshareof14%.Intherightpanel,theuppercdfcorrespondstothecountof rm-sector-destination-yearobservations,andithassmallmasspointsatbothSf;s;k;t=0andSf;s;k;t=1,whichlargelycorrespondtosmallsectorsinremotedestinations.Thelowercdfweightstheobservationsbytheirexportvalue,andthisweighteddistributionhasnomasspoints,althoughthedistributionbecomesverysteepattheverylargemarketshares. FigureA2:Importcutoj0andcost-reductionfactorB(j0)Note:FC=Wfiisthe xedcostofimportinganadditionaltypeofintermediateinput.TMC(j)=CYi=[B(j) i]isthetotalmaterialcostofthe rm,decreasinginjholdingoutput xedduetocost-savingeectofimporting.Theintersectionbetween jlogbjandFC=TMC(j)de nestheimportcutoj0,andtheexponentoftheareaunder jlogbjcurvedeterminesthecost-reductionfactorfromimporting.58 0 0.1 0.2 0.3 0.4 0.5 0.6 0.05 0.1 0.25 0.5 0.75 0.9 0.95 1 Importintensity,Countof¯rmsExport-value-weighted 0 0.2 0.4 0.6 0.8 1 0.05 0.1 0.25 0.5 0.75 0.9 0.95 1 Marketshare,f;s;k;tExport-value-weightedCountofobservations 0 0.2 0.4 0.6 0.8 1 0 0.5 1 1.5 2 Area=loglogFC TMC TableA1:Pass-throughintoproducerpricesandmarginalcostbyquartilesofimportintensity Dep.variable:pf;i;k;tmcf;teMf;t (1)(2)(3)(4)(5)(6)(7) e`;t1;f0.1280.1150.0710.0770.0230.0540.399(0.034)(0.034)(0.035)(0.035)(0.007)(0.006)(0.065)e`;t2;f0.2030.1760.1300.1490.0470.0920.434(0.036)(0.034)(0.033)(0.035)(0.017)(0.011)(0.068)e`;t3;f0.2390.1850.1130.1540.0950.1450.466(0.053)(0.049)(0.048)(0.052)(0.022)(0.010)(0.070)e`;t4;f0.3210.2270.1520.2320.1650.2130.421(0.039)(0.040)(0.041)(0.035)(0.037)(0.018)(0.084)ek;tSf;s;k;t0.2050.238(0.053)(0.059)mcf;t0.5690.565(0.031)(0.031) p-valueBin1vs40.0000.0140.0810.0000.0000.0000.479#obs.93,39593,39593,39593,39593,39589,50489,504R20.0030.0100.0100.0030.0250.0450.214 Note:Nonparametricregressions: rm-product-destination-yearobservationssplitintofourequal-sizedbinsbyvalueofimportintensity'f,withq;fdenotingadummyforrespectivebins(quartilesq=1;2;3;4).No xedeectsincludedinnonparametricspeci cations.Speci cations(3)and(4)additionallycontrolforthelevelofthemarketshareSf;s;k;t.Incolumns15and7,e`;tek;tisthedestination-speci cbilateralexchangerate;incolumn6,e`;teMf;tisthe rm-levelimport-weightedexchangerate(excludingimportsfromtheEuroZone).Incolumn7, rm-levelimport-weightedexchangerateeMf;tisregressedonthedestination-speci cbilateralexchangerateek;t.p-valuefortheF-testofequalityofthecoe cientsforquartiles1and4.Standarderrorsclusteredatthedestination-yearlevel.59 TableA2:Highexchangeratecorrelationsource-destinationpairs #ofsourcecountriesShareofimportsfrom Destinationpegscorr0:7destinationcorr0:7 Australia160.5%5.2%Canada0792.5%58.7%Iceland060.1%2.3%Israel0770.5%41.2%Japan0225.1%16.0%Korea0241.6%33.9%NewZealand030.3%0.6%Norway011.2%1.3%Sweden045.0%6.8%Switzerland016.3%6.7%UnitedKingdom01223.0%30.3%UnitedStates307917.6%38.0% Note:Totalnumberofnon-Eurosourcecountries:210.Numberofcountriesinthe rsttwocolumnsexcludesdestinationitself,whiletheshareofimportsinthelastcolumnincludesimportsfromthedestinationcountry.TableA3:Highandlowpass-throughimportsourcecountries Highpass-through(0:50)Lowpass-through(0:50) Pass-ImportPass-ImportCountrythroughshareCountrythroughshare Peru1.200.5%Israely0.450.2%Bangladesh0.930.2%India0.421.0%Chile0.750.2%Brazil0.413.1%Taiwan0.740.5%Thailand0.411.0%Canaday0.711.8%SriLanka0.400.2%Australiay0.691.5%Malaysia0.400.3%SaudiArabia0.671.3%Egypt0.390.4%China0.673.8%Philippines0.390.5%UnitedStatesy0.6316.6%Venezuela0.360.4%Russia0.623.8%Singapore0.310.2%HongKong0.610.2%Swedeny0.3114.3%Japany0.555.4%SouthKoreay0.240.9%Colombia0.550.3%UnitedKingdomy0.1915.7%Switzerlandy0.531.5%Indonesia0.180.6%Mexico0.500.4%Ukraine0.150.2%Argentina0.080.3%Turkey0.021.5%Pakistan�0.020.2%Vietnam�0.030.3%SouthAfrica�0.091.0% Note:Non-OECDimportsourcecountrieswithashareinBelgianimportsabove0.2%andpreciselyes-timatedpass-throughintoimportpricesofBelgian rms,splitintohighandlowpass-throughbinswithathresholdpass-throughof0.5.ymarkshigh-incomeOECDcountries.High-pass-throughcountriesalsoincludeGuatemala,Macao,UgandaandUnitedArabEmirates;Low-pass-throughcountriesalsoincludeBelarus,Congo,DominicanRepublic,Ethiopia,Ghana,Honduras,Madagascar,NewZealandy,Paraguay,Uruguay,Zambia,Zimbabwe.Fortheremainingimportsourcecountries,whichformtheOtherbininTable7,thepass-throughestimatesaretooimpreciseeitherbecauseoftoofewobservationsortoolittlevari-ationintheexchangerateagainsttheeuro(thislattergroupconsistsmainlyofthenon-Euro-AreaEuropeansourcecountrieswhichaccountforthemajorityofimportsintheOtherbin).60 TableA4:Robustnesstothede nitionofimportintensity LaggedOnlyDropDropOnlyOnlyDroptime-varyingmanuf.consumercapitalIO-tableIO-tableinputsin('f;t�1;S;t�1)importsgoodsgoodsinputsinputs*exportCN8Dep.var.:pf;i;k;t(1)(2)(3)(4)(5)(6)(7) ek;t0.0540.0620.0680.0650.0570.0560.077(0.032)(0.030)(0.030)(0.032)(0.031)(0.031)(0.033)ek;t'f;0.4520.4590.4290.4500.4710.4861.062(0.154)(0.114)(0.135)(0.153)(0.106)(0.106)(0.376)ek;tSf;s;k;0.2780.2940.2920.2860.2870.2860.288(0.058)(0.064)(0.063)(0.062)(0.063)(0.063)(0.060) FE:s;k+tyesyesyesyesyesyesyes#obs.87,79993,39593,39593,39593,39593,39593,395R20.0590.0580.0570.0570.0570.0570.057 Note:Column1estimates(21)withlaggedimportintensityandmarketsharevariables.Speci cationsincolumns27arethesameasincolumn6ofTable5,butwithalternativemeasuresofimportintensity'f,droppingrespectivecategoriesofimportsfromthede nitionof'f:Column2keepsonlymanufacturingproducts;Columns3and4excludeconsumerandcapitalgoodscategoriesrespectivelyaccordingtotheBECclassi cation;Columns5and6keeponlyimportsthatcorrespondtointermediateinputcategoriesfortheexportsofthe rmaccordingtotheinput-outputtables,wherecolumn6additionallyfocusesonthemajorexportcategoryofthe rm;Column7dropsallimportsinthesameCN-8industrialcodesasexportsofthe rm.OtherdetailsappearinthetextandasinTable5. 61 TableA5:Robustnesswithdierentsamples DestinationsAll rmsDroppingProducts allw/outonlyincludingintra- rmallHS4-digit countriesUSUSwholesalerstradeproductsmajormajor*Dep.var.:pf;i;k;t(1)(2)(3)(4)(5)(6)(7)(8) ek;t�0.0110.0340.1840.0940.0700.0620.1020.090(0.016)(0.035)(0.062)(0.028)(0.033)(0.027)(0.042)(0.045)ek;t'f0.2630.4380.6520.3350.4790.5870.4000.505(0.064)(0.122)(0.385)(0.079)(0.120)(0.107)(0.175)(0.165)ek;tSf;s;k;t0.0970.2920.3120.1620.2110.2240.1950.198(0.029)(0.062)(0.110)(0.057)(0.071)(0.051)(0.070)(0.087) FixedEects:s;k+tyesyesnoyesyesyesyesyessnonoyesnonononono#countries551111212121212#obs.218,87982,43810,957158,80479,461143,91262,67953,037R20.0770.0580.0550.0410.0620.0430.0570.060 Note:Mainspeci cationfromcolumn6ofTable5estimatedwithalternativesubsamplesofthedata.Incolumn3,saresector xedeectsatHS-4digitlevel.AllotherdetailsareasinTable5.Column5excludesall rm-destinationobservationsiftheBelgian rmhaseitherinwardoroutwardFDIwiththatdestination.Column6keepsallofthe rms'smanufacturingexports(i.e.,notonlyitsmajorproducts).Column7usesanalternativede nitionofthe rm'smajorproductandkeepsonlythemajorgoodreportedbythe rmattheHS4-digitlevel,andcolumn8keepsonlytheseHS4-digitmajorgoodsifitsmarketshareinthe rm'sexportsisabove50%. 62 Importers,Exporters,andExchangeRateDisconnectMaryAmitiMary.Amiti@NY.FRB.orgFederalReserveBankofNewYorkOlegItskhokiitskhoki@princeton.eduPrincetonUniversityJozefKoningsJoep.Konings@KULeuven.beUniversityofLeuvenandNationalBankofBelgiumThisdraft:September23,2013Firstdraft:July2012Largeexportersaresimultaneouslylargeimporters.Inthispaper,weshowthatthispatterniskeytounderstandinglowaggregateexchangeratepass-throughaswellasthevariationinpass-throughacrossexporters.First,wedevelopatheoreticalframeworkthatcombinesvariablemarkupsduetostrategiccomplementaritiesandendogenouschoicetoimportintermediateinputs.Themodelpredictsthat rmswithhighimportsharesandhighmarketshareshavelowexchangeratepass-through.Second,wetestandquantifythetheoreticalmechanismsusingBelgian rm-product-leveldatawithinformationonexportsbydestinationandimportsbysourcecountry.Wecon rmthatimportintensityandmarketsharearekeydeterminantsofpass-throughinthecross-sectionof rms.Asmallexporterwithnoimportedinputshasanearlycompletepass-through,whilea rmatthe95thpercentileofbothimportintensityandmarketsharedistributionshasapass-throughofjustabove50%,withthemarginalcostandmarkupchannelsplayingroughlyequalroles.Thelargestexportersaresimultaneouslyhigh-market-shareandhigh-import-intensity rms,whichhelpsexplainthelowaggregatepass-throughandexchangeratedisconnectobservedinthedata.Keywords:F14,F31,F41JELclassi cation:exchangeratepass-through,pricing-to-market,importintensity WegratefullyacknowledgetheNationalBankofBelgiumfortheuseofitsresearchfacilitiesanddata,andinparticularValereBogaertsforhelpwiththedatacollection,andEmmanuelDhyneandCatherineFussforcommentsanddataclari cations.WethankGeorgeAlessandria,ArielBurstein,ElhananHelpman,DoireannFitzgerald,LindaGoldberg,PennyGoldberg,LoganLewis,NickLi,BenMandel,UlrichMüller,SteveRedding,EstebanRossi-Hansberg,DavidWeinstein,HylkeVandenbusscheandseminarparticipantsatmultiplevenuesforinsightfulcomments.WealsothankSydneeCaldwell,StefaanDecramer,CecileGaubert,DiegoGilsanz,PrestonMui,andMarkRazhevforexcellentresearchassistance.TheviewsexpressedinthispaperarethoseoftheauthorsanddonotnecessarilyrepresentthoseoftheFederalReserveBankofNewYorkortheNationalBankofBelgium.