FirmsinInternationalTradeAndrewB.Bernard,J.BradfordJensen,StephenJ.Redding,andPeterK.Schottndiscussingtheoriginsandimplicationsofinternationaltrade,economistsusuallyemphasizecomparativeadvantage,increasingreturnstoscale,andconsumerloveofvariety,butpayrela - PDF document

FirmsinInternationalTradeAndrewB.Bernard,J.BradfordJensen,StephenJ.Redding,andPeterK.Schottndiscussingtheoriginsandimplicationsofinternationaltrade,economistsusuallyemphasizecomparativeadvantage,increasingreturnstoscale,andconsumerloveofvariety,butpayrelativelylittleattentiontotheÞrmsthatactuallydrivetradeßows.Yetengagingininternationaltradeisanexceedinglyrareactivity:ofthe5.5millionÞrmsoperatingintheUnitedStatesin2000,just4percentwereexporters.AmongtheseexportingÞrms,thetop10percentaccountedfor96percentoftotalU.S.exports.Sincethemid-1990s,alargenumberofempiricalstudieshaveprovidedawealthofinformationabouttheimportantrolethatÞrmsplayinmediatingcountriesÕimportsandexports.Thisresearch,basedonmicrodatasetsthattrackcountriesÕproductionandtradeattheÞrmlevel,demonstratesthattradingÞrmsdiffersubstantiallyfromÞrmsthatsolelyservethedomesticmarket.Acrossawiderangeofcountriesandindustries,exportershavebeenshowntobelarger,moreproductive,moreskill-andcapital-intensive,andtopayhigherwagesthannonexportingÞrms.Furthermore,theseAndrewB.BernardistheJackByrneProfessorofInternationalEconomics,TuckSchoolofBusiness,DartmouthCollege,Hanover,NewHampshire,andResearchAssociate,NationalBureauofEconomicResearch,Cambridge,Massachusetts.J.BradfordJensenisDeputyDirector,PetersonInstituteforInternationalEconomics,Washington,D.C.StephenJ.ReddingisReaderinEconomics,LondonSchoolofEconomics,andResearchFellow,CenterforEconomicPolicyResearch,bothinLondon,UnitedKingdom.PeterK.SchottisProfessorofEconomics,SchoolofManagement,YaleUniversity,NewHavenConnecticut,andResearchAssociate,NationalBureauofEconomicResearch,Cambridge,Massachusetts.Theire-mailaddressesare,respectively.JournalofEconomicPerspectives—Volume21,Number3—Summer2007—Pages105–130 othersledtothecreationofÒnewÓtrademodelsbyPaulKrugman(1980),ElhananHelpman(1981),andWilliamEthier(1982).Inthesemodels,acombinationofeconomiesofscaleandconsumerpreferencesforvarietyleadotherwiseidenticalÞrmstoÒspecializeÓindistincthorizontalvarieties,spurringtwo-wayorÒintra-industryÓtradebetweencountries.Incontrasttooldtradetheories,wherethewelfaregainsarisefromthedifferencesinopportunitycostsofproductionacrossTable1TradeTheoriesandTheirAbilitytoExplainFactsaboutTrade Facts“Old”tradetheory“New”tradetheoryrmsmodelrmsmodelRicardo(1817),Heckscher(1919),Ohlin(1933)Bernardetal.Redding,andSchott(2007)InterindustrytradeYesNoYesNoYesIntra-industrytradeNoYesYesYesYesExportersandnonexporterswithinNoNoNoYesYesTradeandproductivityExportersaremoreproductivethannonexporterswithinNoNoNoYesYesTradeliberalizationraisesindustryproductivitythroughNoNoNoYesYesTradeandlabormarketsNetchangesinemploymentacrossindustriesfollowingtradeliberalizationYesNoYesNoYesSimultaneousgrossjobcreationanddestructionwithinindustriesfollowingtradeliberalizationNoNoNoYesYesTradeliberalizationaffectsrelativefactorrewards(incomeYesNoYesNoYesInterindustrytradeoccurswhenacountryexportsinonesetofindustriesandimportsinanothersetofindustries;intra-industrytradeoccurswhenthereistwo-wayexportingandimportingwithinthesameindustry.AndrewB.Bernard,J.BradfordJensen,StephenJ.Redding,andPeterK.Schott107 ThethirdcolumnofnumbersinTable2showsthatexportingÞrmsshiparelativelysmallshareoftheirtotalshipmentsabroad.Here,too,substantialvaria-tionexistsacrossindustries,rangingfromahighof21percentincomputerandelectronicproductstoalowof7percentinbeverageandtobaccoproducts.AcrossallÞrms,theshareis14percent.TheinformationinTable2isconsistentwitholdandnewtradetheoriesinsomeways,butnotinothers.Forexample,exportingismorelikelyandexportintensityishigherinmoreskill-intensivesectorslikecomputersthaninmorelabor-intensivesectorslikeapparel.Thisaspectofthedataaccordswithendowment-drivenoldtradetheory:thatis,arelativelyskill-abundantcountryliketheUnitedStatesshouldberelativelymorelikelytoexportinskill-intensiveindustriesinwhichitpossessescomparativeadvantage.However,whileoldtradetheorycanexplainwhyacountryisanetimporterinonesetofindustriesandanetexporterinTable2ExportingByU.S.ManufacturingFirms,2002 NAICSindustryPercentofPercentofrmsthatMeanexportsasapercentoftotal311FoodManufacturing6.81215312BeverageandTobaccoProduct0.7237313TextileMills1.02513314TextileProductMills1.91212315ApparelManufacturing3.2814316LeatherandAlliedProduct0.42413321WoodProductManufacturing5.5819322PaperManufacturing1.4249323PrintingandRelatedSupport11.9514324PetroleumandCoalProducts0.41812325ChemicalManufacturing3.13614326PlasticsandRubberProducts4.42810327NonmetallicMineralProduct4.0912331PrimaryMetalManufacturing1.53010332FabricatedMetalProduct19.91412333MachineryManufacturing9.03316334ComputerandElectronicProduct4.53821335ElectricalEquipment,Appliance1.73813336TransportationEquipment3.42813337FurnitureandRelatedProduct6.4710339MiscellaneousManufacturing9.1215Aggregatemanufacturing1001814Dataarefromthe2002U.S.CensusofManufactures.TheÞrstcolumnofnumberssummarizesthedistributionofmanufacturingÞrmsacrossthree-digitNAICSmanufacturingindustries.ThesecondreportstheshareofÞrmsineachindustrythatexport.TheÞnalcolumnreportsmeanexportsasapercentoftotalshipmentsacrossallÞrmsthatexportinthenotedindustry.FirmsinInternationalTrade109 12and11percent,respectively.TheseÞndingsareemblematicofwhatistypicallyfoundinthisliterature.Theobserveddifferencesbetweenexportersandnonexportersarenotdrivensolelybysize.WhenwecontrolforÞrmsizeasmeasuredbylogemploymentaswellasindustryeffectsincolumn3,thedifferencesbetweenexportersandnonexport-erswithinthesameindustryonallothereconomicoutcomescontinuetobestatisticallysigniÞcantatthe1percentlevel.TheÞndingthatexportersaresystematicallymoreproductivethannonexport-ersraisesthequestionofwhetherhigher-productivityÞrmsself-selectintoexportmarkets,orwhetherexportingcausesproductivitygrowththroughsomeformofÒlearningbyexporting.ÓResultsfromvirtuallyeverystudyacrossindustriesandcountriesconÞrmthathighproductivityprecedesentryintoexportmarkets.TheseÞndingsaresuggestiveofthepresenceofsunkentrycostsintoexportmarketsthatonlythemostproductiveÞrmsÞnditproÞtabletoincur,asemphasizedinRobertsandTybout(1997).MoststudiesalsoÞndlittleornoevidenceofimprovedproductivityasaresultofbeginningtoexport;forexample,theworkofBernardandJensen(1999)onU.S.ÞrmsandtheworkofClerides,Lach,andTybout(1998)onÞrmsinMexico,Colombia,andMorrocoÞndnodifferentialgrowthinÞrmRecentestimatessuggestthatthesesunkcostsmaybesizable.Das,Roberts,andTybout(forthcoming)estimatevaluesofover$300,000forColumbianmanufacturingplantsduring1981Ð91.Table3ExporterPremiainU.S.Manufacturing,2002 Exporterpremia(1)(2)(3)Logemployment1.190.97Logshipments1.481.080.08Logvalue-addedperworker0.260.110.10LogTFP0.020.030.05Logwage0.170.060.06Logcapitalperworker0.320.120.04Logskillperworker0.190.110.19AdditionalcovariatesNoneIndustryÞxedIndustryÞxedeffects,logDataarefor2002andarefromtheU.S.CensusofManufactures.AllresultsarefrombivariateordinaryleastsquaresregressionsoftheÞrmcharacteristicintheÞrstcolumnonadummyvariableindicatingÞrmÕsexportstatus.Regressionsincolumn2includeindustryÞxedeffects.Regressionsincolumn3includeindustryÞxedeffectsandlogÞrmemploymentascontrols.Totalfactorproductivity(TFP)iscomputedasinCaves,Christensen,andDiewert(1982).ÒCapitalperworkerÓreferstocapitalstockperworker.ÒSkillperworkerÓisnonproductionworkerspertotalemployment.AllresultsaresigniÞcantatthe1percentlevel.AndrewB.Bernard,J.BradfordJensen,StephenJ.Redding,andPeterK.Schott111 ofthelate1970sandearly1980sisduetotherelativelygreatersurvivalandgrowthofhigh-productivityplants.SimilarÞndingsemergefromalargenumberofstudiesoftradeliberalizationreformsindevelopingcountries,assurveyedinTybout(2003).Thewithin-industryreallocationsofresourcesfoundbythesestudiesdominatetheacross-industryreallocationsofresourcesemphasizedbyoldtheoriesofcompara-tiveadvantage.Therefore,inthelabormarket,netchangesinemploymentbetweenindustriesimpliedbycomparativeadvantagearesmallrelativetogrosschangesinemploymentcausedbysimultaneousjobcreationanddestructionwithinindustries.Oneconcernisthatthelinkfromincreasedtradetotherelativeexpansionofhigher-productivityÞrmsindeveloping-countryresultsmightnotbedrivensolelybychangesintradepolicy,sincetradeliberalizationisoftenpartofabroaderpackageofeconomicreforms.However,similarpatternsofproductivitygainsfromtheexpansionofhigh-productivityexportingÞrmshavebeenfoundinresponsetoreductionsintradebarriersinbothCanada(Treßer,2004)andtheUnitedStates(Bernard,Jensen,andSchott,2006a).Forexample,Treßer(2004)ÞndseffectsofCanadiantariffreductionsonindustryproductivitythatareroughlytwiceaslargeasthoseonplantproductivity,implyingmarketsharereallocationsfavoringhigh-productivityplants.TheresourcereallocationeffectsofreductionsinU.S.tradecostsareexaminedbyBernard,Jensen,andSchott(2006a).Theyconsideranumberofdependentvariablesincludingtheprobabilityofplantdeath.Theirkeyexplanatoryvariableisameasureoftradecosts,includingbothtariffratesandshippingcostsattheindustrylevel.Control-lingforanumberofotherplantcharacteristics,theyÞndthatplantdeathismorelikelytooccurastradecostsfall,andthatreductionsintradecostshavethegreatestimpactonplantdeathforthelowest-productivityplants.Therelationshipbetweentradeliberalizationandaggregateproductivitygrowthisnotlimitedtotherelativegrowthandexpansionofhigh-productivityÞrms.InPavcnik(2002),one-thirdoftheincreaseinaggregateproductivityfol-lowingtheChileanliberalizationwasduetowithin-plantproductivitygains,poten-tiallyfromthereallocationofresourcesacrossactivitieswithinplants.QualitativelysimilarevidenceisreportedbyTreßer(2004),whoÞndsthattheCanadaÐU.S.FreeTradeAgreementraisedthelaborproductivityofCanadianmanufacturingplantsby7.4percentorbyanannualcompoundgrowthrateof0.93percent.Bernard,Jensen,andSchott(2006a)alsoÞndevidencesupportingalinkbetweenfallingtradecostsandwithin-plantproductivitygrowthinU.S.data.OneoftheirspeciÞcationsusesplantsÕtotalfactorproductivityasthedependentvari-able.Thekeyexplanatoryvariableisagainthechangesinindustrytradecostsdescribedabove.IntheirpreferredspeciÞcation(column3oftable6oftheirpaper),changesinindustry-leveltradecostsarenegativelyandsigniÞcantlyassoci-atedwithplant-levelproductivitygrowth,withaone-standard-deviationfallintradecosts(adropof1percentagepoint)implyingaproductivityincreaseof2.3percent.Standardtrademodelsemphasizingcomparativeadvantageandtheprolifer-ationofproductvarietyhavelittletosayaboutÞrmoraggregateproductivitygrowth.However,agrowingbodyofevidenceshowsthattradeliberalizationcausesFirmsinInternationalTrade113 productivityÞrmswhowerepreviouslymarginaltoexittheindustry.Aslow-productivityÞrmsexitandasoutputandemploymentarereallocatedtowardshigher-productivityÞrms,averageindustryproductivityrises.Heterogeneous-Þrmmodelsaddressanumberoftheempiricalchallengesfacingoldandnewtradetheory.TheycapturetheinteractionbetweenÞrmheterogeneityandinternationaltrade,withtheproductivityadvantageofexportersexplainedbytheself-selectionofthemostproductiveÞrmsintoexporting.Theshiftinresourcesfromlow-tohigh-productivityÞrmsgeneratesimprovementsinaggregateproductivity.Duringthisshift,exportersgrowmorerapidlythannonex-portersintermsofsizeandemployment.Themodelsfeaturesimultaneousjobcreationandjobdestructionwithinindustriesaslow-productivityÞrmsexitandhigh-productivityÞrmsexpand.InthemodelsofBernard,Eaton,Jensen,andKortum(2003)andMelitzandOttaviano(2005),themark-upofpriceovermarginalcostisendogenousanddecreasesasimportcompetitionintensiÞesfollowingreductionsintradecosts.InBernard,Redding,andSchott(2007),heterogeneousÞrmsareintegratedintothestandardtradeparadigmofHelpmanandKrugman(1985).Theresultingframeworkexplainswhysomecountriesexportmoreincertainindustriesthaninothers(endowment-drivencomparativeadvantage);whynonethelesstwo-waytradeisobservedwithinindustries(Þrm-levelhorizontalproductdifferentiationcom-binedwithincreasingreturnstoscale);andwhywithinindustriesengagedinthesetwoformsoftrade,someÞrmsexportandothersdonot(self-selectiondrivenbytradecosts).ConsistentwiththeempiricalÞndingsreportedinTable2,thefractionofexportingÞrmsandtheshareofexportsinÞrmshipmentsvariessystematicallyacrossindustriesandcountrieswithcomparativeadvantage.Althoughtradeliberalizationinthisframeworkinduceswithin-industryreal-locationandraisesaggregateproductivityinallindustries,productivitygrowthisstrongerinthecomparativeadvantageindustry.Thegreaterexportopportunitiesinthatindustryleadtoalargerincreaseinfactordemandthaninthecomparativedisadvantageindustry,whichbidsuptherelativepriceofthefactorusedintensivelyinthecomparativeadvantageindustry,andsoleadstogreaterexitbylow-productivityÞrmsthaninthecomparativedisadvantageindustry.Thisdifferentialproductivitygrowthacrossindustriesgivesrisetodifferencesinaverageindustryproductivitythatmagnifyfactor-abundance-basedcomparativeadvantageandsoprovidesanadditionalsourceofwelfaregainsfromtrade.Tradeliberalizationinthisframeworknotonlygeneratesaggregatewelfaregainsbutalsohasimplicationsforthedistributionofincomeacrossfactors.Increasesinaverageindustryproductivityarisingfromtradeliberalizationdrivedowngoodspricesandthereforeraisetherealincomeofallfactors.Ifproductivityincreasesarestrongenough,therealincomeofacountryÕsscarcefactormayevenriseduringtradeliberalization(acontradictionofthewell-knownStolperÐSamuel-sontheorem).Moregenerally,theproductivitygainsinducedbythebehaviorofAndrewB.Bernard,J.BradfordJensen,StephenJ.Redding,andPeterK.Schott115 largedifferencesinsales,aslow-pricedvarietiesareeasilysubstitutableforhigh-pricedvarieties.Alternativeexplanationsfortheconcentrationoftradeinvolverelativelysim-pleextensionsofexistingheterogeneousÞrmmodels.First,theremaybeecono-miesofscaleinoverseasdistributionandmarketingthatfavortheconcentrationoftradeamongasmallnumberofproducers.Second,iftherearesunkcostsspeciÞctoindividualdestinations,andifdestinationsvaryintermsoftheirproÞtability,relativelymoreproductiveexporterswillexporttomoredestinations.Thisexpan-sionalongtheextensivemarginofthenumberofdestinationsservedleadstomoreinequalityinexportvaluesthanifthenumberofdestinationsperÞrmwereconstant.Third,iftherearesunkcostsspeciÞctoindividualproducts,andifproductsvaryintermsoftheirproÞtabilityforaÞrm,relativelymoreproductiveexporterswillalsoexportawiderrangeofproducts.Thisexpansionalonganotherextensivemarginofthenumberofproductswillalsomagnifytheinequalityinexportvalues.WepresentempiricalevidencebelowontheimportanceofthesetwoextensivemarginexpansionsforindividualÞrmsandaggregatetrade.TradeisEvenScarcerThanThoughtObservedinternationaltradeßowsaresmallrelativetothelevelspredictedbybotholdandnewtradetheory.Inoldtradetheory,theamountoftradepredictedbycross-countrydifferencesinfactorendowmentsisagooddealgreaterthanobservedvaluesoftrade,asTreßer(1995)pointsoutinhisanalysisoftheÒmysteryofthemissingtrade.ÓInstandardnewtradetheorymodels,allvarietiesaretradedinequilibrium,apredictionthatisatoddswiththelargenumberofzerobilateraltradeßowsobservedintradedataateithertheaggregateorproductlevel.absenceoftradeßowscanbeexplainedbyoldtradetheoryintermsofprohibitivetradecostsandcompletespecialization.Buttheseexplanationsarenotfullyper-suasiveanddonotexplainwhy,whenpositivetradeoccurs,someÞrmsexportwhileothersdonot.TheexaminationofÞrm-leveldatadeepensthemysteryofthemissingtrade.TheaverageshareofexportsinÞrmoutputiswellbelowthelevelpredictedbystandardnewtradetheorymodels.Withnotradecostsandidenticalandhomo-theticpreferences,thesemodelspredictthattheshareofexportsinÞrmoutputequalstheshareoftherestoftheworldinworldGDP,avaluesubstantiallyhigherthanthosereportedinTable2.Similarly,thenumberofdestinationcountriesservedbytheaverageexportingÞrmissmall.Table4showsthat64percentofU.S.manufacturingÞrmsthatexportdosotoasingledestinationcountryin2000(toppanel,bottomofÞrstcolumn),thoughtheseexportsrepresentjust3.3percentofAslongasthedemandforvarietiesissufÞcientlystrong(aswithconstantelasticityofsubstitutionpreferences),allvarietiesaretradedinnewtradetheorymodelsforanyÞnitevalueoftradecosts.Forempiricalevidenceonthelargenumberofzerosinbilateraltradeßows,seeSchott(2004)andHelpman,Melitz,andRubinstein(2007).FirmsinInternationalTrade117 aggregateexportvalue(middlepanel,bottomofÞrstcolumn).Bycontrast,ÞrmsexportingtoÞveormoredestinationsaccountforjust13.7percentofexporters(toppanel,bottomofÞfthcolumn),but92.9percentofexportvalue(middlepanel,bottomofÞfthcolumn).Inrecentwork,Eaton,Kortum,andKramarz(2006)exploitvariationinthenumberofdestinationcountriesservedbyFrenchÞrmstoestimatedestination-speciÞcÞxedcostsofexportingwithinastructuralmodelofheterogeneousÞrmsandtrade.AnothermessagefromTable4istheimportanceofmultiproductexportersinoverallU.S.exports.In2000,42.2percentofexportingÞrmsexportedasingleproductabroad(toppanel,endofÞrstrow).Here,too,however,theseexportersrepresentedasmallshareofaggregateexports,just0.4percent(middlepanel,endofÞrstrow).FirmsexportingÞveormoreproductsaccountedfor25.9percentofÞrmsbut98percentofexportvalue(topandmiddlepanels,endofÞfthrows).Theseresultsprovidesupportforsomeoftheexplanationsfortheconcentra-tionoftradeadvancedabove.TheyrevealthattheverysmallshareofÞrmsthatdominateU.S.exportsarelargeinsizeinpartbecausetheyshipmanyproductstomanydestinations.Indeed,acrossexportingÞrmsin2000,weÞndapositiveandstatisticallysigniÞcantcorrelationbetweenthenumberofproductsthatÞrmsexportandthenumberofcountriestheyexportto(correlationcoefÞcientof0.81,signiÞcantatthe1percentlevel).NeitheroldnornewtradetheoryincludesconsiderationoftheseextensivemarginsofÞrmparticipationinexportmarkets.Yetadjustmentalongthesemar-ginsexplainsmuchofthevariationinaggregatetradepatterns.Furthermore,aÞrmÕsdecisionconcerningthenumberofexportdestinationstoserveandthenumberofproductstoexportissystematicallycorrelatedwiththecharacteristicsoftheÞrm,sothatÞrmheterogeneityisagainimportantforunderstandingaggregatetradeoutcomes.FromTable4,weseethatÞrmsthatexporttoÞveormoredestinationshaveemploymentlevelsÞvetimeslargerthanÞrmsthatexporttoasingledestination(69.2dividedby14.2inthethirdpanel);haveexportvaluesthatare30timesgreater(92.9dividedby3.3inthesecondpanel);andhencehaveexportsperworkerthatarearoundsixtimesgreater(30/5).MultiproductFirmsandExportingOneimplicationofTable4isthat,whenÞrmsexport,theytypicallyexportmultipleproducts.Thus,intheyear2000,Þrmsthatexportmorethanoneten-digitHarmonizedSystemproductcomprise57.8percentofexportingÞrmsandaccountformorethan99.6percentofexportvalue.SinceÞrmoutputequalsthenumberofproducts(theextensivemargin)timesaverageoutputperproduct(theintensivemargin),thedifferencesinsizebetweenexportersandnonexportersnotedearliercanbebrokendownintothesetwomargins.TheÞrstcolumnofTable5reportstheresultsoftworegressionsusingthe1997CensusofManufactures.AsinTable3earlier,thedependentvariablesarelisteddowntheleftsideofthetable.ThesevariablesarethenumberofÞve-digitAndrewB.Bernard,J.BradfordJensen,StephenJ.Redding,andPeterK.Schott119 importanceofÞrmentryandexitinspurringareallocationofeconomicresourcesacrossÞrmsastradebarriersfall.However,becausesurvivingÞrmscanenterandexitindividualproductmarkets,thisfocusonthecreationanddestructionofÞrmsmayunderstatethetrueextentofreallocationfollowingtradeEvidencesupportingthesigniÞcanceofwithin-ÞrmreallocationindrivingaggregateoutputgrowthisprovidedbyBernard,Redding,andSchott(2006a),whoÞndthatnetproductaddinganddroppingbysurvivingÞrmsaccountsforroughlyone-thirdofaggregateU.S.manufacturinggrowthbetween1972and1997,acontributionthatdwarfsthatofÞrmentryandexit.TogetherwiththepositivecorrelationsobservedbetweenthenumberofproductsÞrmsexport,exportsperproduct,andtotalexportsoftheÞrm,theseÞndingssuggestthatmoreattentionshouldbepaidtotheinteractionofinternationaltradeandÞrmscope.GravityReconsideredTheÒgravityequationÓforbilateraltradeßowsisoneofthemostsuccessfulempiricalrelationshipsininternationaleconomics.Earlyresearchonthegravityequationsupposedthattheaggregatevalueoftradebetweenapairofcountrieswasproportionaltotheproductoftheirincomesandinverselyrelatedtothedistancebetweenthem.Subsequentresearchhasconsideredawiderangeofothervariablesthatmayinßuencebilateraltradeanddevelopedmicrofounda-tions.Themicro-foundedformulationsofthegravityequationcontrolnotonlyforbilateralfrictionsbetweentradepartnersbutalsoformultilateralfrictionswithalltradepartners.Despitethisextensivebodyofresearch,empiricalandtheoreticalworkwithagravityequationtypicallyconcentratesontheaggregatevalueoftradeandignorestherolesofÞrmsandproducts.Inthissection,weuseabasicgravityequationtoexaminewhethertheeffectofdistanceonbilateraltradeoperatesthroughtheextensivemargin(thenumberofÞrmsandthenumberofproducts)ortheintensivemargin(valueperproductperÞrm).WedecomposetheaggregatevalueofU.S.exportstoaparticulardestinationintothreefactors:thecontributionofthenumberofÞrmsexportingtothedestination;thenumberofproductsexportedtothedestination;andtheaveragevalueofexportsperproductperÞrm.Thislastterm,theaveragevalueofexportsperproductperÞrm,willdependonboththepriceschargedfortheproductsandthequantitiesshipped.ToexaminewhethertheeffectofdistanceonbilateraltradeßowsoperatesthroughÞrmparticipation,thenumberofproductsexportedortheaveragevalueofaproductexportedbyaÞrm,weestimategravityequationsfortheaggregatevalueofexportsandeachofthesethreecomponents.Ingravityequations,itistypicaltohavedataonmanydifferentpairsoftradingpartnersandincludebothexporterandimporterincomeintheregression.Butsinceourdataareforasingleexportingcountry(theUnitedStates),exporterFirmsinInternationalTrade121 decreasinginimporterincomeisatÞrstsightpuzzling.Onepotentialexplanationinvolvestheideathatcostsofexportingdependonquantityorweightratherthanvalue(forexample,thecostsofexportingdependonthenumberofbottlesofwineratherthanthequalityoftheircontents).Inthiscase,increasesindistanceorreductionsinimporterincomemayleadtoachangeinthecompositionofexportstowardshigher-valuecommodities,forwhichitisproÞtabletoincurtheÞxedandvariabletradecostsofservicingtheremoteandsmallforeignmarket.Thediffer-encesinvalue-to-weightratioacrosscommoditiesmayinturnbeexplainedbydifferencesintheirquality,anideatowhichwewillreturnbelow.Ifthechangeincompositiontowardshigher-valuecommoditiesissufÞcientlylarge,theaveragevalueofexportsperproductperÞrmmaybeincreasingindistanceanddecreasinginimporterincome.ImportingandExportingTheempiricalliteratureonÞrmsininternationaltradehasbeenconcernedalmostexclusivelywithexporting,largelyduetolimitationsindatasetsbasedoncensusesofdomesticproductionormanufacturing.Asaresult,thenewtheoriesofheterogeneousÞrmsandtradeweredevelopedtoexplainfactsaboutÞrmexportbehaviorandyieldfewpredictions(ifany)forÞrmimportbehavior.Inmostmodels,consumerspurchaseimportsdirectlyfromforeignÞrms,andnointerme-diateinputsexistÑthatis,Þrmsthemselvesdonotimport.Withthedevelopmentoftransactions-leveltradedata,informationondirectTheseideasrelatetotheso-calledÒAlchian-AllenhypothesisÓthatgoodsexportedareonaverageofhigherqualitythanthosesolddomestically(HummelsandSkiba,2004).Table6GravityandAggregateU.S.Exports,2000 LogoftotalexportsvalueLogofnumberofexportingrmsLogofnumberofexportedproductsLogofexportvalueperproductperrmLogofGDP0.980.710.52(0.04)(0.04)(0.03)(0.04)Logofdistance1.060.84(0.17)(0.16)(0.15)(0.19)1751751751750.820.740.640.25Dataarefromthe2000Linked-LongitudinalFirmTradeTransactionDatabase(LFTTD).Eachcolumnreportstheresultsofacountry-levelordinaryleastsquaresregressionofthedependentvariablenotedatthetopofeachcolumnonthecovariatesnotedintheÞrstcolumn.Resultsfortheconstantaresuppressed.StandarderrorsarenotedbeloweachcoefÞcient.ProductsaredeÞnedasten-digitHarmonizedSystemcategories.AllresultsarestatisticallysigniÞcantatthe1percentlevel.AndrewB.Bernard,J.BradfordJensen,StephenJ.Redding,andPeterK.Schott123 InTable8,wecomparethecharacteristicsofexportingandimportingÞrms.TheÞrmcharacteristicsdataarefromtheCensusofManufactures,theidentiÞca-tionofexportingandimportingcomesfromtheLFTTDbasedoncustomsdocu-ments.Again,weuseillustrativeregressions.Thevariableslistedontheleftarethedependentvariablesintheseregressions.IntheÞrstcolumnofnumbers,theregressionincludesadummyvariableforwhethertheÞrmisanexporterornot,alongwithvariablescontrollingforindustryÞxedeffectsandforsizeofemploy-ment.(Ofcourse,theemploymentregressionsreportedintheÞrstrowofthetabledonotincludeemploymentasacontrolvariableontheright-handsideoftheregressions.)Thesecondcolumncarriesoutaparallelsetofregressions,exceptthatinthiscaseadummyvariableforwhethertheÞrmisanimporterreplacestheexportervariable.TheÞnalcolumninsteadincludesadummyvariableforÞrmsthatarebothexportersandimporters.Thesedummyvariablesareindicatedbythelabelsacrossthetopofthetable.FirmsthatareexportersshareavarietyofpositiveattributeswithÞrmsthatareimporters.Theyarebothbiggerandmoreproductive,payhigherwages,andaremoreskill-andcapital-intensivethannonexportersandnonimporters.Again,theseresultssuggestthatÞrmcharacteristicsaresystematicallyrelatedtoparticipationininternationaltrade,whetherimportingorexporting.ReductionsintradecostsarelikelytobeneÞtthelargest,mostproductive,mostskill-andcapital-intensiveÞrmsinanygivensector,bothbecausetheyexportandbecausetheyimport.Onepossibleexplanationforthepresenceofimportinginallmanufacturingindustries,forthecorrelationbetweenimportingandexporting,andhenceforthesimilarityofimporterandexporterpremia,istheÒinternationalfragmentationofproduction,Ówherestagesofproductionarespreadacrossnationalboundaries.ThispracticeisalsoreferredtoasÒoffshoringÓorÒslicingthevalue-addedchain.ÓTable8TradingPremiainU.S.Manufacturing,1997 (1)Exporterpremia(2)Importerpremia(3)Exporter&importerpremiaLogemployment1.501.401.75Logshipments0.290.260.31Logvalue-addedperworker0.230.230.25LogTFP0.070.120.07Logwage0.290.230.33Logcapitalperworker0.170.130.20Logskillperworker0.040.060.03Dataarefor1997andareforÞrmsthatappearinboththeU.S.CensusofManufacturersandtheLinked-LongitudinalFirmTradeTransactionDatabase(LFTTD).AllresultsarefrombivariateordinaryleastsquaresregressionsoftheÞrmcharacteristiclistedontheleftonadummyvariablenotedatthetopofeachcolumnaswellasindustryÞxedeffectsandÞrmemploymentasadditionalcontrols.EmploymentregressionsomitÞrmemploymentasacovariate.Totalfactorproductivity(TFP)iscomputedasinCaves,Christensen,andDiewert(1982).FirmsinInternationalTrade125 countryincome.WhilethenumberofÞrmsandthenumberofproductsareincreasinginsourcecountryincomeanddecreasingindistance,theaveragevalueperproductperÞrmisagaindecreasinginsourcecountryincomeandincreasingindistance(thoughthecoefÞcientisnotstatisticallysigniÞcantforimports),againsuggestingthepotentialimportanceofproductqualityconsiderations.Onenota-blefeatureoftheresultsisthatthemagnitudesofthecoefÞcientsondistancearequitedifferentforimportsandexports.VerticalDifferentiationMuchoftheresearchonnewandheterogeneous-Þrmtheoriesofinternationaltradehasemphasizedthehorizontaldifferentiationofvarieties.ButanalysisofU.S.importsrevealsthatasubstantialnumberofimportproductsoriginateincountrieswithverydissimilarrelativeendowments.ThisfactholdsfornarrowlydeÞnedproducts,suchasthoseoftheten-digitHarmonizedSystemclassiÞcation,ofwhichthereareroughly10,000categories(forexample,menÕscottonshirts).ThisfactisatÞrstsightconsistentwiththeideathatcountriesexportuniquehorizontalvarieties.However,thedataalsoshowthatpriceswithinproductcate-goriesvarysubstantiallyandsystematicallyacrosscountries,withimportsfromcapital-andskill-abundantcountriesgenerallycommandingmuchhigherpricesthanimportsfromlabor-abundantcountries(forexample,Schott,2004;HummelsandKlenow,2005;Hallak,2006;HallakandSchott,2006).Forexample,Schott(2004)ÞndsthatacrossallU.S.manufacturingimportsin1994,themedianratioofhightolowunitvalueswas24.Thispricevariationsuggeststheimportanceofverticaldifferentiation,withhigherpricesreßectinginparthigherproductquality.Therelationshipbetweenexportpricesandexporterrelativeendowmentsechoesakeyimplicationofoldtradetheory,albeitoneworkingatamuchmoredisaggregatelevel,withinnarrowproductsratherthanacrossbroadindustries.TheseÞndingsareconsistentwiththeideathatdevelopedcountriesusetheirendowmentadvantagetoproducehighunit-value,high-qualityvarietiesevenwithinnarrowproductcategories.ProductqualitywasÞrstemphasizedininternationaltradebyLinder(1961),whoarguedthatwealthycountrieshavebothahighertasteforqualityand,giventheirÞrmsÕproximitytorelativelywealthycustomers,acomparativeadvantageinproducingit.ProductqualityisalsocentraltotheoriesofproductcyclesfollowingVernon(1966),wherethemostadvancedproductsareproducedbydevelopedeconomiesuntiltheyaresuccessfullycopiedandtakenoverbydevelopingecono-mies,whereproductioncostsarelower.RecentempiricalresearchusingÞrm-leveldatasuggeststhatcountriesÕmovementsthroughproductcyclesmaytakeplacewithincontinuingÞrms.Bernard,Jensen,andSchott(2006b),forexample,ÞndthatU.S.manufacturingplantsinindustrieswithrelativelyhighexposuretolow-wagecountryimportssystematicallyswitchintoindustriesfacinglessexposure.TheyalsoshowthatplantsÕsurvivalwithinindustriesisassociatedpositivelywithplantsÕcapitalintensity.TheseresultssuggestthatU.S.manufacturingÞrmsescapeAndrewB.Bernard,J.BradfordJensen,StephenJ.Redding,andPeterK.Schott127 Alvarez,Roberto,andRicardoA.Lo´pez.2005.ÒExportingandPerformance:EvidencefromChileanPlants.ÓCanadianJournalofEconomics,38(4):1384Ð1400.Bernard,AndrewB.,JonathanEaton,J.Brad-fordJensen,andSamuelS.Kortum.ÒPlantsandProductivityinInternationalTrade.ÓAmericanEconomicReview,93(4):1268Ð90.Bernard,AndrewB.,andJ.BradfordJensen.1995.ÒExporters,Jobs,andWagesinU.S.Man-ufacturing:1976Ð87.ÓBrookingsPapersonEco-nomicActivity:Microeconomics,vol.1995,pp.67ÐBernard,AndrewB.,andJ.BradfordJensen.1999.ÒExceptionalExporterPerformance:Cause,Effect,orBoth?ÓJournalofInternational,47(1):1Ð25.Bernard,AndrewB.,andJ.BradfordJensen2007.ÒFirmStructure,Multinationals,andMan-ufacturingPlantDeaths.ÓReviewofEconomicsand89(1):1Ð11.Bernard,AndrewB.,J.BradfordJensen,andPeterK.Schott.2006a.ÒTradeCosts,Firms,andJournalofMonetaryEconomics53(5):917Ð37.Bernard,AndrewB.,J.BradfordJensen,andPeterK.Schott.2006b.ÒSurvivaloftheBestFit:ExposuretoLow-WageCountriesandthe(Un-even)GrowthofU.S.ManufacturingPlants.ÓJournalofInternationalEconomics,68(1):219Ð37.Bernard,AndrewB.,J.BradfordJensen,andPeterK.Schott.(Forthcoming.)ÒImporters,Ex-portersandMultinationals:APortraitofFirmsintheU.S.thatTradeGoods.ÓInProducerDy-namics:NewEvidencefromMicroData,ed.T.Dunne,J.B.Jensen,andM.J.Roberts.Univer-sityofChicagoPress.Bernard,AndrewB.,StephenJ.Redding,andPeterK.Schott.2006a.ÒMulti-productFirmsandProductSwitching.ÓNationalBureauofEconomicResearchWorkingPaper12293.Bernard,AndrewB.,StephenJ.Redding,andPeterK.Schott.2006b.ÒMulti-productFirmsandTradeLiberalization.ÓNationalBureauofEconomicResearchWorkingPaper12782.Bernard,AndrewB.,StephenJ.Redding,andPeterK.Schott.2007.ÒComparativeAdvantageandHeterogeneousFirms.ÓReviewofEconomic74(1):31Ð66.Caves,DouglasW.,LauritsR.Christensen,andW.ErwinDiewert.1982.ÒTheEconomicTheoryofIndexNumbersandtheMeasurementofInput,OutputandProductivity.Ó50(6):1393Ð1414.Clerides,Sofronis,SaulLach,andJamesTy-1998.ÒIsLearningbyExportingImpor-tant?Micro-dynamicEvidencefromColumbia,MexicoandMorocco.ÓQuarterlyJournalof113(3):903Ð47.Das,Sanghamitra,MarkRoberts,andJamesForthcoming.ÒMarketEntryCosts,Pro-ducerHeterogeneityandExportDynamics.ÓEaton,Jonathan,andSamuelKortum.ÒTechnology,Geography,andTrade.Ó70(5):1741Ð79.Eaton,Jonathan,SamuelKortum,andFrancis2006.ÒAnAnatomyofInternationalTrade:EvidencefromFrenchFirms.Óhttp://Eckel,Carsten,andPeterNeary.2006.ÒMulti-ProductFirmsandFlexibleManufacturingintheGlobalEconomy.ÓCentreforEconomicPol-icyResearchDiscussionPaper5941.Ethier,WilliamJ.1982.ÒNationalandInter-nationalReturnstoScaleintheModemTheoryofInternationalTrade.ÓAmericanEconomic,72(3):389Ð405.Grubel,HerbertG.,andPeterJ.Lloyd.ÒIntra-IndustryTrade:TheTheoryandMeasure-mentofInternationalTradeinDifferentiatedTheEconomicJournal,35(339):646Ð48.Hallak,JuanCarlos.2006.ÒProductQualityandtheDirectionofTrade.ÓJournalofInterna-tionalEconomics,68(1):238Ð65.Hallak,JuanCarlos,andPeterK.Schott.2006.ÒEstimatingCross-CountryDifferencesinProductQuality.Óhttp://www.som.yale.edu/Heckscher,Eli.1919.ÒTheEffectofForeignTradeontheDistributionofIncome,Ó,vol21,pp.1Ð32.Helpman,Elhanan.1981.ÒInternationalTradeinthePresenceofProductDifferentia-tion,EconomiesofScale,andMonopolisticCompetition:AChamberlinÐHeckscherÐOhlinJournalofInternationalEconomics,11(3):Helpman,Elhanan.1999.ÒTheStructureofForeignTrade.ÓJournalofEconomicPerspectives,13(2):121Ð44.Helpman,Elhanan.2006.ÒTrade,FDIandtheOrganizationofFirms.ÓJournalofEconomicLiter-44(3):589Ð630.Helpman,Elhanan,andPaulKrugman.FirmsinInternationalTrade129