do peer firms affect corporate financial policy

《do peer firms affect corporate financial policy》由会员上传分享，免费在线阅读，更多相关内容在工程资料-天天文库。

THEJOURNALOFFINANCE•VOL.LXIX,NO.1•FEBRUARY2014DoPeerFirmsAffectCorporateFinancialPolicy?MARKT.LEARYandMICHAELR.ROBERTS∗ABSTRACTWeshowthatpeerfirmsplayanimportantroleindeterminingcorporatecapitalstructuresandfinancialpolicies.Inlargepart,firms’financingdecisionsareresponsestothefinancingdecisionsand,toalesserextent,thecharacteristicsofpeerfirms.Thesepeereffectsaremoreimportantforcapitalstructuredeterminationthanmostpreviouslyidentifieddeterminants.Furthermore,smaller,lesssuccessfulfirmsarehighlysensitivetotheirlarger,moresuccessfulpeers,butnotviceversa.Wealsoquantifytheexternalitiesgeneratedbypeereffects,whichcanamplifytheimpactofchangesinexogenousdeterminantsonleveragebyover70%.MOSTRESEARCHONCORPORATEfinancialpolicyassumesthatcapitalstructurechoicesaremadeindependentlyoftheactionsorcharacteristicsoftheirpeers.Inotherwords,afirm’scapitalstructureistypicallyassumedtobedeterminedasafunctionofitsmarginaltaxrate,expecteddeadweightlossindefault,informationenvironment,andincentivestructure.Assuch,theroleforpeerfirmbehaviorinaffectingcapitalstructureisoftenignored,oratmostim-plicitlyassumedtooperatethroughitsunmeasuredimpactonfirm-specificdeterminants.However,peerfirmsplayacentralroleinshapinganumberofcorporatepolicies,andexistingevidencesuggeststhatthebehaviorofpeerfirmsmaymatterforcapitalstructure.1Surveyevidenceindicatesthatasignificant∗LearyiswithOlinBusinessSchool,WashingtonUniversityinSt.Louis,andRobertsiswithTheWhartonSchool,UniversityofPennsylvaniaandNBER.WethankAndyAbel,UlfAxelson,DanielBergstresser,PhilipBond,MurrayFrank,KenFrench,ItayGoldstein,RobinGreenwood,RichardKihlstrom,JiroKondo,ArvindKrishnamurthy,CameliaKuhnen,DoronLevit,UlrikeMalmendier,DavidMatsa,AtifMian,TobyMoskowitz,FranciscoPerez-Gonzalez,MitchellPe-tersen,GordonPhillips,NickRoussanov,KenSingleton,andMotoYogo,aswellasconferenceandseminarparticipantsattheNBERFallCorporateFinanceMeeting,MinnesotaCorporateFinanceConference,UniversityofWashingtonSummerFinanceConference,WesternFinanceAssociationMeeting,ColumbiaUniversity,CornellUniversity,HarvardBusinessSchool,NotreDameUni-versity,PennsylvaniaStateUniversity,PurdueUniversity,RiceUniversity,StanfordUniversity,TempleUniversity,UniversityofBritishColumbia,UniversityofCaliforniaBerkeley,UniversityofKentucky,UniversityofMaryland,UniversityofPennsylvania,UniversityofRochester,UniversityofSouthernCalifornia,andWashingtonUniversityinSt.Louis.RobertsgratefullyacknowledgesfinancialsupportfromaRodneyL.WhiteGrant.1Examplesincludeproductpricing(Bertrand(1883)),productoutput(Cournot(1838)),nonpriceproductfeatures,suchasadvertising,productdurability,andwarranties(Stigler(1968)),andlaborpractices(e.g.,seeManning(2005)forahistoricaldiscussionandBizjak,Lemmon,andNaveen(2008)forevidenceonexecutivecompensation).DOI:10.1111/jofi.12094139 140TheJournalofFinanceRnumberofCFOscitetheimportanceofpeerfirmfinancingdecisionsfortheirownfinancingdecisions(GrahamandHarvey(2001)).Furthermore,recentem-piricalworkshowsthatindustryaverageleverageratiosareaneconomicallyimportantdeterminantoffirms’capitalstructures(Welch(2004),MacKayandPhillips(2005),andFrankandGoyal(2009)).Thegoalofthispaperistoidentifywhether,how,andwhypeerfirmbehaviormattersforcorporatecapitalstructures.Toeasethediscussionandtoprovidesomecontextforpeereffectsincorporatecapitalstructure,considerpeereffectsarisingfromalearningmotive.Managersareunsureofhowtosetoptimalcap-italstructure.Theinputsarehardtomeasureandthetruemodelisunknown.Assuch,managersconsiderthefinancingdecisionsandcharacteristicsofpeerfirmsasinformativefortheirownfinancingdecisions.Forexample,whenafirm’speersincreasetheirleverageratios,thatfirm’sleverageratioishigherthanitotherwisewouldhavebeenhadpeereffectsnotbeenpresent.Likewise,firmsmayconsiderthegrowthopportunitiesorfinancialhealthoftheirpeersindeterminingtheirowncapitalstructure.Thus,peereffectsincapitalstruc-tureoccurwhentheactionsorcharacteristicsofpeerfirmsexplicitlyenterafirm’sfinancingobjectivefunction.Whiletheoreticallyintuitive,identifyingpeereffectsisempiricallychalleng-ingbecauseofthereflectionproblem(Manski(1993)).Thisproblemreferstoaspecificformofendogeneitythatariseswhentryingtoinferwhethertheactionsorcharacteristicsofagroupinfluencetheactionsoftheindividualsthatcomprisethegroup.Inthecurrentcontext,thisproblemiscreatedbyus-ingmeasuresofpeerfirmfinancialpolicy,suchasindustryaverageleverage,orpeerfirmcapitalstructuredeterminants,suchasindustryaverageprof-itability,asexplanatoryvariablesforindividualfirms’financialpolicies.Anycorrelationbetweenfirms’financialpoliciesandtheactionsorcharacteristicsoftheirpeerscanbeattributedtotwobroadexplanations.Thefirstexplanationisbasedontheendogenousselectionoffirmsintopeergroupsoranomittedcommonfactor.Thisselectionresultsinfirmsfromthesamepeergroupfacingsimilarinstitutionalenvironmentsandhavingsimilarcharacteristics,suchasproductiontechnologiesandinvestmentopportunities.Theinabilitytoaccuratelymodeltheselectionmechanismgeneratesaroleforpeerfirmmeasuresindeterminingfinancialpolicy.Thisrolearisesbecausepeerfirmmeasuresproxyforlatentfactorsthatarecommontofirmsinapeergroupanddeterminefinancialpolicy.Inessence,thecorrelationbetweenfirms’financialpoliciesandthepoliciesorcharacteristicsoftheirpeersreflectsanendogeneitybias.Thesecondexplanationisthatfirms’financialpoliciesarepartlydrivenbyaresponsetotheirpeers.Thisresponsecanoperatethroughtwochan-nels:actionsorcharacteristics.Thefirstchannelariseswhenfirmsrespondtotheirpeers’financialpolicies.Thesecondchannelariseswhenfirmsrespondtochangesinthecharacteristicsoftheirpeers—profitability,risk,etc.Thus,identifyingpeereffectsposestwoidentificationchallenges.Thefirstinvolvesovercomingtheendogenousselection.Thesecondinvolvesdistinguishingbe-tweenthetwochannelsthroughwhichpeereffectsoperate. DoPeerFirmsAffectCorporateFinancialPolicy?141Thefirstchallengecanbeovercomebyshowingthat,controllingforchar-acteristicsoftheirownfirm,firms’behaviorsaresignificantlycorrelatedwithexogenouscharacteristicsoftheirpeers.Weusepeerfirms’idiosyncraticeq-uityreturns(i.e.,equityshocks)asapossiblesourceofexogenousvariationinpeerfirmattributes.Motivationforthisapproachcomesfromexistingre-search.Substantialtheoreticalandempiricalevidencelinksstockreturnstofinancialpolicy(e.g.,Myers(1977,1984),Marsh(1982),LoughranandRitter(1995)),suggestingthatreturnshocksmayberelevantforfinancingdecisions.Thefirm-specificnatureofidiosyncraticreturnsandthelargeassetpricingliteratureaimedatisolatingthiscomponentsuggestthatreturnshocksofferausefulstartingpointforidentifyingexogenousvariation.Indeed,theseshockshaveanumberofdesirableproperties.First,theshockstodifferentfirmswithinapeergrouparelargelyuncorrelatedwithoneanother.Second,theshocksareseriallyuncorrelatedandseriallycross-uncorrelated,implyingthatfirms’shocksdonotforecastfutureshocksforthemselvesorforotherfirms.Finally,theshocksareuncorrelatedwithfirmcharacteristicstypi-callyusedtoexplainvariationincapitalstructure(e.g.,profitability,tangibility,size,andmarket-to-book).Whilethesefeaturesdonotguaranteeexogeneity,theyarereassuringbecausetheysuggestthatpeerfirmreturnshockscontainlittlecommonvariation.Ourresultsshowthatfirms’capitalstructuresaresignificantlyinfluencedbytheirpeers.Leverageisstronglynegativelyrelatedtopeerfirmequityshocks.Debtandequityissuancedecisionsare,respectively,negativelyandpositivelyrelatedtopeerfirmequityshocks.Furthermore,theseinferencesarerobusttoahostofmeasurementandendogeneityconcerns.Toensurethatlatentcommonfactorsarenotbehindourresults,weunder-takeaseparateanalysisinwhichweutilizeequityreturnshockstopeerfirms’customersthat(1)areinanindustrydifferentfromfirmi,and(2)arenotacustomeroffirmi.CohenandFrazzini(2008)showthatcustomerreturnspredictsupplierreturns,suggestingthattheremaybeinformationinreturnshockstopeers’customersthatisrelevantfortheirbehavior.Thisapproachenablesustoeliminateallwithin-industryvariationforthepurposeofidenti-ficationbecausewecannowcontrolforfirmi’sindustryaveragestockreturn.Theidentifyingvariationnowcomesfromreturnshockstofirmsinadifferentindustrywithnosupplychainlinktofirmi.Furthermore,thisvariationisorthogonaltofirmi’sreturn,firmi’sindustryreturn,andallotherincludedde-terminants.Aplacebotestusingthereturnshocksofrandomlyselectedfirmsinthecustomers’industriesthatarenotcustomersoffirmiori’speersrevealsinsignificantpeereffects.Thus,peerfirmsmatterforfinancialpolicy.Toaddressthesecondidentificationchallenge(i.e.,thechannelthroughwhichpeereffectsoperate),weshowthat,conditionalonpeerfirmfinancialpolicy,capitalstructureislargelyinsensitivetopeerfirms’idiosyncraticstockreturns.Inotherwords,firms’leverageratiosonlyrespondtopeerfirms’equityshockswhenthoseshocksareaccompaniedbychangestopeerfirms’leverageratios.Furthermore,peerfirmcharacteristics(otherthantheirreturnshocks)arelargelyirrelevantforfinancialpolicy,bothstatisticallyandeconomically. 142TheJournalofFinanceRWealsofindthatmostothercorporatepolicies—investment,dividends,re-search,anddevelopment—areinsensitivetopeerfirmequityshocks.Takentogether,thisevidencesuggeststhattheprimarychannelthroughwhichpeereffectsoperateisviaactions—firmsrespondtothefinancialpoliciesoftheirpeers.Toquantifytheimportanceofpeereffectsincapitalstructure,weestimatethemarginaleffectofachangeinpeerfirmleverageonfirmi’sownleverage,usingpeerfirms’idiosyncraticequityreturnshocksasaninstrumentfortheircapitalstructures.Wefindthataonestandarddeviationincreaseinpeerfirms’leverageratiosisassociatedwitha10%increaseinfirmi’sleverageratio,aneffectlargerthananyotherdeterminant.Peerfirms’decisionstoissueequity,andtheirchoicebetweenequityanddebt,haveasimilarlylargeeffectonafirm’sownissuancedecisions.Withtheseestimatesweareabletoquantifytheexternalitiesgeneratedbypeereffectssinceashocktoonefirmaffectsalloftheotherfirmsinthepeergroup.Toillustrate,considerashocktofirmA’sprofitability.ThisshockaffectsnotonlyfirmA’sfinancingchoice,butalsothatofeveryothermemberoffirmA’speergroup.Thisimpactonpeerfirms’financialpoliciesfeedsbackontofirmA’sfinancialpolicy,andsoon.Thislinkamongpeerfirmsimpliesthatthemarginaleffectofanyexogenouscapitalstructuredeterminantcannolongerbegleanedsolelyfromthatdeterminant’scoefficient,eveninlinearmodels.Instead,themarginaleffectisafunctionofanamplificationtermduetotheactionchannelofpeereffects,aspillovertermduetothecharacteristicschannelofpeereffects,andthesizeofthepeergroup.Wefindthattheamplificationtermvariesfromalowof8%inlargepeergroupstoahighofover70%insmallpeergroups.Inotherwords,inindus-trieswithfewfirms,theimpactofachangeinprofitability,forexample,onleverageis70%largerthanthatimpliedbymodelsignoringthepresenceofpeereffects.Wealsoshowthatthespillovereffectsfromchangingpeercharac-teristicscaneitheroffsetorfurtheramplifytheeffectofchangesinexogenouscharacteristics.Finally,weexamineheterogeneityinthepeereffectstobetterunderstandwhypeerfirmsinfluencefinancialpolicy.Smaller,lesssuccessful(e.g.,lowerprofitability),andmorefinanciallyconstrainedfirmsaresensitivetothere-turnshocksofindustryleaders(i.e.,larger,moreprofitablefirms).However,theoppositeisnottrue.Financialpoliciesofindustryleadersareinsensitivetothereturnshocksoftheirlesssuccessfulpeers.Theseresultsareconsistentwiththeimplicationsofmodelsbasedonlearning(e.g.,Conlisk(1980))andreputationalconcerns(e.g.,ScharfsteinandStein(1990)andZwiebel(1995)),thoughtheydonotruleoutalternativesbasedonfeedbackfromtheproductmarkets(e.g.,BranderandLewis(1986)andBoltonandScharfstein(1990)).Whilehelpingtoshedlightontheunderlyingmechanismbehindpeeref-fects,thisanalysisalsoreinforcesouridentificationstrategyasmostalter-nativehypothesesleavelittleroomforsystematicheterogeneityinthepeereffect. DoPeerFirmsAffectCorporateFinancialPolicy?143Ourstudyismostcloselyrelatedtothosedocumentingtheimportanceofindustryasacapitalstructuredeterminant.2However,paststudiesleavein-terpretationoftheseindustryeffectslargelyunresolved,apointexplicitlynotedbyFrankandGoyal(2008,2009).Oursisthefirststudytosiftthroughthesealternativemeanings,identifypolicyinterdependenceasasubstantialcompo-nentoftheindustryleverageeffect,andestimatetheexternalitiesinducedbythepresenceofpeereffects.OurstudyisalsorelatedtotheworkofMacKayandPhillips(2005)andAlmazanandMolina(2005),whoexamineintrain-dustryvariationincapitalstructures.Ourstudycomplimentstheirworkbyshowingthatthisvariationisaccompaniedbystronginterdependenciesinfinancialpolicy.3Animportantby-productofourstudyistohighlightthesalientempiricalissuesthatappearinobservationalstudiesofpeereffects,asopposedtoran-domizedexperiments(e.g.,DufloandSaez(2003),LernerandMalmendier(2013)).Ordinaryleastsquaresregressionstypicallydonotprovidemeaning-fulresultsbecauseofthereflectionproblem,andthusaclearidentificationstrategyisneededtoruleoutthenullofomittedormismeasuredcommoncharacteristics.Furthermore,feedbackandspillovereffectsarisingfromthepresenceofpeereffectsobscurethemarginaleffectsofexogenousvariables.Neitherthedirectionnorthemagnitudeoftheassociationbetweenacovari-ateandthedependentvariablecanbeinferredfromthatcovariate’scoeffi-cient,eveninlinearspecifications.Wepresentclosed-formexpressionsforthemarginaleffectsofexogenouscovariatesinagenerallinearsetting.Thepaperproceedsasfollows.SectionIintroducesthedataandpresentssummarystatistics.SectionIIdevelopstheempiricalmodelandhighlightstheidentificationchallenge.SectionIIIdiscussesouridentificationstrategy,focus-ingontheconstructionofourmeasureofpeerfirmbehavior,itseconomicandstatisticalproperties,andpotentialidentificationthreats.SectionIVpresentsourprimaryresultsandrobustnesstests.SectionVexaminescross-sectionalheterogeneityintheeffectstobetterunderstandtheeconomicmechanismsbehindthepeereffects.SectionVIconcludes.I.DataandSummaryStatisticsOurprimarydatacomefromthemergedCenterforResearchinSecurityPrices(CRSP)-Compustatdatabasefortheperiod1965to2008.Becauseofitspopularity,werelegateacompletediscussionofthedata,sampleconstruction,2Bradley,Jarrell,andKim(1984)showthat54%ofthecross-sectionalvarianceinfirmleverageratiosisexplainedbyindustrialclassification.GrahamandHarvey(2001)showthatalmostone-quarterofsurveyedCFOsidentifythebehaviorofcompetitorsasanimportantinputintotheirfinancialdecision-making.Welch(2004)findsthatdeviationsfromindustryleverageareamongthemosteconomicallysignificantdeterminantsofleveragechanges.3Otherstudiesexaminingpeereffectsincorporatefinanceinclude:mutualfundvoting(MatvosandOstrovsky(2010)),governance(JohnandKadyrzhanova(2008)),investmentdecisions(DufloandSaez(2002)),entrepreneurship(LernerandMalmendier(2013)),andcompensation(Shue(2013)). 144TheJournalofFinanceRTableISummaryStatisticsThesampleconsistsofallnonfinancial,nonutilityfirmsintheannualCompustatdatabasebetween1965and2008withnonmissingdataforallanalysisvariables(seeAppendixA).Thetablepresentsmeans,standarddeviations(SD),andmediansforvariablesinlevelsandfirstdifferences.PeerFirmAveragesdenotesvariablesconstructedastheaverageofallfirmswithinanindustry-yearcombination,excludingtheithobservation.Industriesaredefinedbythree-digitSICcode.Firm-SpecificFactorsdenotesvariablescorrespondingtofirmi’svalueinyeart.LevelsFirstDifferencesMeanMedianSDMeanMedianSDPeerFirmAveragesBookLeverage(Total0.2380.2290.0940.0040.0030.031Debt/BookAssets)MarketLeverage0.2740.2620.1370.0060.0040.058Log(Sales)5.0854.9321.2780.0910.0940.119Market-to-Book1.3621.2010.650−0.032−0.0190.310EBITDA/Assets0.1080.1200.070−0.002−0.0010.031NetPPE/Assets0.3170.2700.172−0.002−0.0020.020Firm-SpecificFactorsBookLeverage(Total0.2380.2170.1960.004−0.0000.098Debt/BookAssets)MarketLeverage(Total0.2740.2160.2460.0060.0000.123Debt/MarketAssets)Log(Sales)5.0855.0182.1720.0910.0890.357Market-to-Book1.3620.9661.244−0.032−0.0060.829EBITDA/Assets0.1080.1290.155−0.0020.0000.104NetPPE/Assets0.3170.2710.217−0.002−0.0020.060IndustryCharacteristicsNo.ofFirmsper13.2178.00018.344Industry-YearTotalNo.ofIndustries217SampleCharacteristicsObservations80,279Firms9,126andvariabledefinitionstoAppendixA.TableIpresentssummarystatisticsforourfinalsampleof80,279firm-yearobservationscorrespondingto9,126uniquefirms.Wedefinepeergroupsbasedonthree-digitSICindustrygroups.4Thereare217industriesrepresentedinoursample.Thetypicalindustrycontainsapproximately13firms,thoughthedistributionisright-skewedasindicatedbythemediannumberoffirms,eight.Wediscusspotentialmeasurementconcernsregardingthedefinitionofanindustry(HobergandPhillips(2009)),aswellasthedocumentedintraindustryheterogeneity(MacKayandPhillips(2005)),below.4Belowweexaminetherobustnessofourresultstochangesinthebreadthofindustrygroups. DoPeerFirmsAffectCorporateFinancialPolicy?145Summarystatisticsforanumberofvariables,inlevelsandfirstdifferences,usedthroughoutthisstudyarepresentedafterWinsorizingallratiosatthe1stand99thpercentiles.WeWinsorizetomitigatetheinfluenceofextremeobservationsandeliminateanydatacodingerrors.Variablesaregroupedintotwodistinctcategories:peerfirmaveragesandfirm-specificfactors.Theformercategoryincludesvariablesconstructedastheaverageofallfirmswithinanindustry-yearcombination,excludingtheithobservation.Thelattergroupin-cludesvariablesconstructedasfirmi’svalueinyeart.Atthispoint,wesimplynotethesimilarityofmanystatisticstothosereportedinpreviousempiricalstudiesofcapitalstructure,suchasFrankandGoyal(2009).II.TheEmpiricalModelOurempiricalmodelofcapitalstructureisageneralizationofthatusedthroughouttheempiricalcapitalstructureliterature(e.g.,RajanandZingales(1995)andFrankandGoyal(2009)),y=α+βy¯+γX¯+λX+δμ+φν+ε,(1)ijt−ijt−ijt−1ijt−1jtijtwheretheindicesi,j,andtcorrespondtofirm,industry,andyear,respectively.Wefocusonalinearspecificationtoemphasizetheintuitionandhighlightthesalienteconometricissues.Extensionsareexaminedbelow.Theoutcomevariable,yijt,isameasureofcorporatefinancialpolicy,suchasleverage.Thecovariate¯y−ijtdenotespeerfirmaverageoutcomes(excludingfirmi).Weuseacontemporaneousmeasurebecauseitlimitstheamountoftimeforfirmstorespondtooneanother.Thischoicemakesitmoredifficulttoidentifymimickingbehavior.Italsomitigatesthescopeforconfoundingef-fectsbyreducingthelikelihoodofothercapitalstructurerelevantchanges.TheK-dimensionalvectorsX¯−ijt−1andXijt−1containpeerfirmaverageandfirm-specificcharacteristics,respectively.Industryandyearfixedeffectsarerepresentedbytheerrorcomponentsμjandνt,respectively.Finally,εijtisthefirm-yearspecificerrortermthatisassumedtobecorrelatedwithinfirmsandheteroskedastic.Assuch,allstandarderrorsandteststatisticsarero-busttothesetwodeparturesfromtheclassicalregressionmodel(Petersen(2009)).Theparametervectoris(α,β,γ,λ,δ,φ).Werefertotheseparametersasstructuralparametersonlytodistinguishthemfromthecomposite,orreducedform,parametersthatappearinthecontextofinstrumentalvariables.Likethevastmajorityoftheempiricalcapitalstructureliterature,weleaveunspec-ifiedthepreciseoptimizationproblemundertakenbythefirm.Thecoefficientsδ,alongwithλandφ,capturethefirstexplanationforcommonindustrybe-havior:sharedcharacteristicsorinstitutionalenvironments.Peereffectsarecapturedbyβandγ,whichmeasuretheinfluenceofpeerfirmactionsandcharacteristics,respectively,onfinancialpolicychoices. 146TheJournalofFinanceRIII.IdentificationTheempiricalgoalistodisentanglethevariousexplanationsforindustrycommonalityincapitalstructurebystatisticallyidentifyingthestructuralpa-rameters.Theprimarydifficultyarisesfromthepresenceof¯y−ijtasaregressorinequation(1).Intuitively,iffirms’financingdecisionsareinfluencedbyoneanother,thenfirmi’scapitalstructureisafunctionoffirmj’sandviceversa.Thissimultaneityimpliesthat¯y−ijtisanendogenousregressorandthatthestructuralparametersarenotidentified.Thissectiondiscussestheidentifica-tionproblemandourstrategyforaddressingit.A.TheIdentificationProblemIgnoringtheyearfixedeffectsfornotationalconvenience,considerthepop-ulationversionofequation(1)5y=α+βE(y|μ)+γE(X|μ)+λX+δμ+ε.(2)jjjThecorrespondingmeanregressionofyonXandμjisE(y|X,μ)=α+βE(y|μ)+γE(X|μ)+λX+δμ.(3)jjjjTakingexpectationsofthisequationwithrespecttothefirmcharacteristics,X,conditionalonμjyieldstheequilibriumconditionE(y|μ)=α+βE(y|μ)+λE(X|μ)+γE(X|μ)+δμ.(4)jjjjjAssumingthatβ=1,thisequilibriumhasauniquesolutionαγ+λδE(y|μj)=+E(X|μj)+μj.(5)1−β1−β1−βPluggingtheequilibriumsolutionintoequation(3)yieldsthereduced-formmodel∗∗∗∗E(y|X,μj)=α+γE(X|μj)+δμj+λX,(6)wherethesuperscript“*”referstoreduced-formorcompositeparametersthatarefunctionsoftheunderlyingstructuralparameters.Specifically,∗α∗βλ+γ∗δ∗α=;γ=;δ=;λ=λ.1−β1−β1−βImmediatelyapparentisthatthestructuralparameterscannotberecoveredfromthecompositeparameterssincetherearefewerequationsthanunknowns.However,asdiscussedbyManski(1993),estimationofthereduced-formmodelinequation(6)cansolvethefirstidentificationchallenge,thatis,separating5TheillustrationoftheidentificationprobleminthissectioncloselyfollowsthatinManski(1993). DoPeerFirmsAffectCorporateFinancialPolicy?147someformofpeereffect—viaactionsorcharacteristics—fromanalternativeexplanationforcommonindustrycapitalstructuresbasedonendogenousse-∗lectionoranomittedcommonfactor.Ifγisnotequaltozero,theneitherβorγisnotequaltozero.Thus,areduced-formtestforthepresenceofpeereffects∗6isatestofthesignificanceofγ.B.TheIdentificationStrategy∗Toidentifyγinequation(6),werequireanexogenouspeerfirmcharacter-istic.Suchacharacteristicisnoteasytofind,evenwhencontrollingforfirmi’sowncharacteristics.Considerpeerfirms’averagemarket-to-bookratio.Be-causethemarket-to-bookratioisanoisymeasureofinvestmentopportunities,thepeerfirmaveragemaybeabettermeasureoffirmi’sinvestmentopportu-nitiesthanisfirmi’sownmarket-to-bookratio.Ataminimum,thepeerfirmaveragelikelycapturessomevariationincharacteristicsrelevantforfirmi’scapitalstructurethatisnotcapturedbyfirmi’sownmarket-to-bookratio.Inotherwords,thepeerfirmaveragemarket-to-bookratioisnotexogenouswith∗respecttofirmi’sfinancialpolicyandγisnotidentified.Tomotivateouridentificationstrategy,consideraneventstudyapproachtotheproblem.Thechallengeistoidentifyeventsthatarerelevantforpeerfirmsbutthatarerandom—conditionalonobservables—withrespecttofirmi’scapitalstructure.Onemightconsidereventssuchaslossesduetonaturaldisasters,accidentalCEOdeaths,accountingscandals,etc.However,therearetwoproblemswiththisapproach.First,eventssuchasthesearerareenoughtoraiseconcernsoverstatisticalpowerandexternalvalidity.Second,andmoreimportantly,itisunclearwhetherthese,oranyother,eventsareinfactexogenousbecauseofspillovereffects.Forexample,anaccidentalCEOdeathatapeerfirmmayberelevantforfirmi’sfinancialbehaviornotonlythroughthepeerfirms’financialresponsebutalsothroughtheevent’simpactontheCEOlabormarketoranticipatedshiftinproductmarketbehavior.Likewise,anoilspill,suchasthe2010spillintheGulfofMexicoattributedtoBritishPetroleum,hasbroaderimplicationsfortheindustryviaitsimpactontheproductmarket,futureregulatoryenvironment,andexpectedliabilities.Thus,onemayfindeventsthatarerelevantforpeerfirms,butitisunlikelythatthesesameeventsarealsoexogenouswithrespecttofirmi’scapitalstructure.Assuch,wetakeanalternativeapproachthataddressesthesetwoconcerns.Webeginwithaknowncapitalstructuredeterminant,stockreturns(e.g.,Marsh(1982)).Wethenextracttheidiosyncraticvariationinstockreturnsusingatraditionalassetpricingmodelthatalsoincorporatesanindustryfac-tortopurgecommonvariationamongpeers.Theresidualfromthismodelis6NotethatatleastonecovariateoftheXvectormustbecorrelatedwithytoensurethatλis∗notazerovector.Otherwise,γcouldbezeroevenifβisnonzero.Furthermore,werequirethatβ∈(0,1). 148TheJournalofFinanceRthereturnshock.Welagthisshock1yearanduseitasastartingpointforexogenousvariationinpeerfirms’characteristics.Thisapproachhasseveralpositiveaspects.First,themeasureisavailableforabroadpaneloffirmsandthusmitigatesstatisticalpowerandexternalvalidityconcerns.Second,stockreturnsarerelativelyfreefrommanipulationwhencomparedtoothercapitalstructuredeterminantssuchasearnings,sales,andotheraccountingmeasures.Third,stockreturnsimpoundmany,ifnotall,value-relevantevents.Fourth,avastassetpricingliteraturefocusesonestimatingtheexpectedandidiosyncraticcomponentsofreturns.Finally,thereistheoreticalandempiricalprecedentforarelationshipbetweenstockreturnsandcapitalstructurechoices.7Intuitively,ouridentificationstrategybuildsontheevent-studyapproachbyaddressingitsshortcomings.Stockreturnsimpoundtheeffectofvalue-relevanteventssuchasnaturaldisasters,CEOdeaths,accountingscandals,etc.Theproblemisthattheseeventsaffectboththeidiosyncraticandthecommoncomponentsthatcomprisestockreturns.Ouridentificationstrategyistopurgethiscommonvariationsothattheonlyvariationremainingforidentificationofthepeereffectisfirm-specific.Thus,ouridentificationstrategydoesnotrelyonparticularfirm-specificeconomicevents,which,asdiscussedearlier,arenotonlyrarebutalsovirtuallyimpossibletoidentify.Rather,ourstrategyreliesonisolatingthefirm-specificvariationinstockreturns.Theweaknessofthisstrategyisthatthetruedata-generatingprocessforequityreturnsisunknown.Assuch,anyestimatedequityreturnshockmaycontaintracesofcommonvariationthatwouldfailtobeexogenous.Addressingthisweaknessguidesmuchofouranalysis.C.ConstructionoftheReturnShockWeestimatereturnshockswiththefollowingaugmentedmarketmodelforstockreturns,rijt:r=α+βM(rm−rf)+βIND(¯r−rf)+η,(7)ijtijtijtttijt−ijttijtwhererijtreferstothetotalreturnforfirmiinindustryjovermontht,(rmt−rft)istheexcessmarketreturn,and(¯r−ijt−rft)istheexcessreturnonanequal-weightedindustryportfolioexcludingfirmi’sreturn.Aswithourpeergroups,industriesaredefinedbythree-digitSICcode.Whilenotapricedriskfactor,thislastfactorisincludedtoremoveanyvariationinreturnsthatiscommonacrossfirmsinthesamepeergroup.Weestimateequation(7)foreachfirmonarollingannualbasisusinghistor-icalmonthlyreturns.Werequireatleast24monthsofhistoricaldataanduse7Forexample,MyersandMajluf(1984)suggestthatfinancialpolicyislinkedtostockpricesbe-causeofinformationasymmetrybetweenmanagersandinvestors.Likewise,Myers(1977)suggeststhatfinancialpolicyislinkedtostockpricesbecauseofdebtoverhangconsiderations.Empirically,Marsh(1982),LoughranandRitter(1995),BakerandWurgler(2002),andWelch(2004),amongothers,showastrongcorrelationbetweenpastreturnsandissuancechoiceorleverageratios. DoPeerFirmsAffectCorporateFinancialPolicy?149TableIIStockReturnFactorRegressionResultsThesampleconsistsofmonthlyreturnsforallnonfinancial,nonutilityfirmsintheintersectionoftheannualCompustatandmonthlyCRSPdatabasesbetween1965and2008.ThetablepresentsmeanfactorloadingsandadjustedR2sfromtheregressionR=α+βM(RM−RF)+βIND(R¯−RF)+η,ijtijtijtttijt−ijttijtwhereRijtisthereturntofirmiinindustryjduringmontht,(RMt−RFt)istheexcessreturnonthemarket,and(R¯−ijt−RFt)istheexcessreturnonanequal-weightedindustryportfolioexcludingfirmi’sreturn,whereindustriesaredefinedbythree-digitSICcode.TheregressionisestimatedforeachfirmonarollingannualbasisusinghistoricalmonthlyreturnsdatafromtheCRSPdatabase.Werequireatleast24monthsofhistoricaldataanduseupto60monthsofdataintheestimation.Expectedreturnsarecomputedusingtheestimatedfactorloadingsandrealizedfactorreturns1yearhence.Idiosyncraticreturnsarecomputedasthedifferencebetweenrealizedandexpectedreturns.MeanMedianSDαit0.0080.0070.017βM0.3990.4220.803itβIND0.6160.5350.567itObs.perRegression59605AdjustedR20.2280.2070.170Avg.MonthlyReturn0.0130.0000.182ExpectedMonthlyReturn0.0150.0140.090IdiosyncraticMonthlyReturn−0.002−0.0110.174upto60monthsofdataintheestimation.Forexample,toobtainexpectedandidiosyncraticreturnsforIBMbetweenJanuary1990andDecember1990,wefirstestimateequation(7)usingmonthlyreturnsfromJanuary1985throughDecember1989.UsingtheestimatedcoefficientsandthefactorreturnsfromJanuary1990throughDecember1990,weuseequation(7)tocomputetheexpectedandidiosyncraticreturnsasfollows:ExpectedReturn≡rˆ=αˆ+βˆM(rm−rf)+βˆIND(¯r−rf),ijtijtijtijtttijt−ijttIdiosyncraticReturnijt≡ηˆijt=rijt−rˆijt.Toobtainexpectedandidiosyncraticreturnsfor1991,werepeattheprocessbyupdatingtheestimationsamplefrom1986through1990andusingfactorreturnsduring1991.Thisprocessgeneratesβsthatarefirm-specificandtime-varying,hencetheparametersubscriptsinequation(7),butconstantwithinacalendaryear.Thus,ourconstructionofidiosyncraticreturnsallowsforhet-erogeneoussensitivitiestoaggregateshocks.TableIIpresentssummarystatisticsfortheestimatedfactorregressions.Onaverage,eachoftherollingregressionshas59monthlyobservations,thoughthemajorityrelyonafull5-yearwindow.TheaverageadjustedR2isapprox-imately23%.Theregressionsloadpositivelyonbothmarketandindustryfactors,whosefactorloadingssumtoapproximatelyone.Theaverageidiosyn-craticreturnislessthan20basispointsinmagnitude—anartifactofrounding 150TheJournalofFinanceRandsampleselectiononnonmissingdatafortheaccountingvariables(seeAppendixA).Tomaintainconsistencywiththeperiodicityoftheaccountingdata,wecom-poundthemonthlyreturnstoobtainanannualmeasure.Wethenaveragethismeasureoverpeerfirmswithineachyearandlagit1yearwithrespecttotheoutcomevariables.Thus,oursourceofexogenousvariationforpeerfirms’characteristicsisthelaggedaveragepeerfirmequityreturnshock,η¯ˆ−ijt.Intuitively,ourstrategycanbeviewedasmatchingeachfirmtoeveryotherfirminitsindustry.Consideranindustrywithjusttwofirms,AandB.OuridentificationstrategyusesfirmB’sreturnshocktocapturetheeffectofitsbehavior—financingdecisionsandcharacteristics—onfirmA’sfinancingdeci-sion,andviceversa.Nowconsideranindustrywiththreefirms,A,B,andC.OuridentificationstrategyusesthereturnshockstofirmsBandCtocapturetheeffectoftheirbehavioronfirmA’sfinancingdecisions.Averagingprovidesaconvenienttooltoreducethedimensionalityoftheproblemandsummarizethesalientinformation.Averagingalsoensuresthatnonlinearitiesarenotre-sponsibleforouridentification.However,averagingdoesreducethenoiseinindividualreturnshocks,whichcanthreatenidentificationwhenindividualreturnsarenoisy.Wediscussthisconcernbelow.Notethat,conditionalonaproperlyspecifiedassetpricingmodel(equation(7)),theaveragepeerfirmreturnshockneednotbezero.Thismeasureisaconditionalaverage,conditionalonindustryandyear.Inaddition,themeasureisnotexactlytheindustryaveragesinceitexcludestheithobservation.PanelAofFigure1illustratesthevariationinpeerfirmaveragereturnshockswithahistogram.Theunconditionalmeaniszero,assuggestedbytheapproximatelyzeroaverageidiosyncraticreturnshownatthebottomofTableIIandthezerobalancepointinthefigure.PanelsBandCshowwhathappenstoourmea-sureastheindustrydefinitionbecomescoarserandthesizeofthepeergroupincreases.Weseethatthedistributioncollapsesaroundzero,andmoresofortheone-digit(PanelC)thanthetwo-digit(PanelB)industrydefinition.Thus,consistentwiththeeconomicnotionofapeergroup,werelyonarestrictiononthesizeofthegrouptoensuresufficientvariationinourmeasure.D.IdentificationThreatsIdentificationthreatscomefromcorrelationbetweenourmeasureofpeerfirmidiosyncraticreturnshocksandomittedormismeasuredfirmicapitalstructuredeterminants.Werefercollectivelytothesedeterminantsascom-monfactors.Thissubsectiontakesafirststeptowardaddressingthisconcernbyexaminingthestatisticalpropertiesofpeerfirmequityshocksandtheireconomicimplications.Beforedoingso,weemphasizethatthescopeforpotentialidentificationthreatsislimitedtothefractionofvariationremainingafterconditioningontheobservablecontrolvariables.Ausefultaxonomyofthevariationinourmeasureisbetweenindustries,withinindustries,andovertime.Theinclusionofcontrolvariablesinequation(1)eliminatesmuchofthisvariation.Industry DoPeerFirmsAffectCorporateFinancialPolicy?151PanelA:Three-DigitSICCodePeerGroupsPanelB:Two-DigitSICCodePeerGroups.25.25.2.2.15.15Percentage.1Percentage.1.05.0500−1−.50.51−1−.50.51PeerFirmAverageEquityShock(3−DigitSIC)PeerFirmAverageEquityShock(2−DigitSIC)PanelC:One-DigitSICCodePeerGroups.25.2.15Percentage.1.050−1−.50.51PeerFirmAverageEquityShock(1−DigitSIC)Figure1.Industryaverageidiosyncraticstockreturnsdistribution.Thesampleconsistsofallnonfinancial,nonutilityfirmsintheannualCompustatdatabasebetween1965and2008withnonmissingdataforallanalysisvariables(seeAppendixA).Thefigurepresentstheempiricaldistributionofourinstrument,peerfirmaverageidiosyncraticannualequityreturns,forthreedefinitionsofpeergroupsbasedonthree-digitSICcode(PanelA),two-digitSICcode(PanelB),andone-digitSICcode(PanelC).Peerfirmaveragesaredefinedasthepeergroupaverageexcludingtheithobservation.Thedatahavebeentruncatedat−1and+1toeasethepresentation.fixedeffectsremoveallbetween-industryvariation.Inclusionoffirmi’sshockasacontrolvariableeliminatesallwithin-industry-yearvariation.8Thisimpliesthattheidentifyingvariationiswithin-industrytime-seriesvariationinthecomponentofpeerfirmreturnshocksthatisorthogonaltoalloftheincludedcontrolvariables,Xijt−1andX¯−ijt−1.Thoughcorrelationwithunobservablesisalwaysaconcern,weshowthattheremainingidentifyingvariationreducesthescopeforalternativehypotheses.Previousempiricalworkshowsthatobservableleveragedeterminantsdoarelativelypoorjobofcontrollingforsystematicvariationincapitalstructures8Intuitively,thedifferencebetweentheindustryaverageshockandthepeerfirmaverageshockistheexclusionoffirmi’sshock.Sincetheindustryaverageshockdoesnotvarywithinanindustry-year,thevariationinthepeerfirmaverageshockwithinanindustry-yearisperfectlynegativelycorrelatedwithfirmi’sshock. 152TheJournalofFinanceR(e.g.,Welch(2004),Lemmon,Roberts,andZender(2008),andStrebulaevandYang(2012)).Therelevantissueinthecurrentcontextiswhethertheremain-ingomittedvariablesormeasurementerrorsarecorrelatedwithpeerfirmaverageequityshocksconditionalonotherobservablecharacteristics.Thus,wefocusonensuring,asmuchaspossible,thattheaverageidiosyncraticeq-uityshocktopeerfirms(1)isnotabettermeasureoffirmi’scapitalstructuredeterminantsthanaretheotherincludedfirmcharacteristics(includingfirmi’sownreturnshock),and(2)isnotcapturingacommonfactorsharedamongfirmswithinthepeergroup.Thefirstconsiderationhighlightstheimportanceofisolatingtheidiosyn-craticcomponentofstockreturnsratherthanusingtotalreturns.TableIIshowsthattheidiosyncraticcomponentaccountsforasignificantportionofthevariationinstockprices—theaverageR2isequalto23%.Thisresultsug-geststhattheaveragetotalreturnofotherfirmsinanindustrymayprovidealessnoisymeasureoftheinvestmentopportunities,forexample,facingeachindividualfirmthantheirownindividualstockreturnormarket-to-bookra-tios.Intuitively,theaveragingofreturnssmoothsoutanynoiseinindividualstockreturns.TableIIIexaminesthepartialcorrelationsbetweenpeerfirmaverageeq-uityshocksandfirmicharacteristics.Weexaminethecorrelationswithbothcontemporaneousandone-periodleadeffects,todeterminewhetherourmea-surecontainsinformationaboutcurrentorfuturefirmicharacteristics.Notethatcorrelationwiththecharacteristicsisnotproblematicbecausethechar-acteristicsareallincludedintheregressionascontrolvariables.However,economicallylargeassociationsbetweenourmeasureandobservablefirmchar-acteristicswouldraiseconcernsabouttheextenttowhichourmeasuremaybecorrelatedwithunobservablefactors,andtheextenttowhichwehaveremovedcommonvariationamongfirms’returnsviatheassetpricingmodel.Theresultsrevealonestatisticallysignificantcoefficientamongthefirmicharacteristicsinthecontemporaneousspecification,andnoneintheone-period-leadspecification.Theeconomicmagnitudesofthecoefficientesti-matesarealltinyaswell.Fortheonlystatisticallysignificantcoefficient,EBITDA/Assets,aonestandarddeviationincreaseinthiscovariateisassoci-atedwitha10basispointdeclineincontemporaneouspeerfirmequityshocks.Thischangeinequityshocksislessthan0.01standarddeviations.Thus,thepeerfirmequityshockscontainnosignificantinformationrelatedtofirmi’scurrentornear-futureobservablecapitalstructuredeterminants.Regardinganomittedcommonfactor,consideration(2),wenotethefollow-ingfindingsfromuntabulatedresults.Thecorrelationbetweenfirmi’stotalreturnandtheaverageindustrytotalreturnexcludingfirmi’stotalreturnis0.37.Thecorrelationbetweenfirmi’sidiosyncraticreturnshockandtheaveragepeerfirmshockis0.02.Thisdeclinesuggeststhattheassetpricingmodelpurgesmost,thoughnotall,oftheintraindustrycorrelationinreturns.Weincludefirmi’sshockinequation(1)tohelpabsorbthisremainingcorre-lation.Thepeerfirmreturnshocksarealsoseriallyuncorrelatedandserially DoPeerFirmsAffectCorporateFinancialPolicy?153TableIIIPeerFirmReturnShockPropertiesThesampleconsistsofallnonfinancial,nonutilityfirmsintheannualCompustatdatabasebetween1965and2008withnonmissingdataforallanalysisvariables(seeAppendixA).Thetablepresentsestimatedcoefficientsandt-statisticsrobusttoheteroskedasticityandwithin-firmdependenceinparentheses.Thedependentvariableistheaveragepeerfirmequityreturnshock.Allindependentvariablesareinlevelsandareeithercontemporaneouswithoraone-period-leadrelativetothedependentvariable,asindicatedatthetopofthecolumns.Firm-SpecificFactorsdenotesvariablescorrespondingtofirmi’svalueinyeart.PeerFirmAverageCharacteristicsarepeerfirmaveragesofthesamevariableslistedunderfirm-specificfactorsinthetable:logofsales,themarket-to-bookratio,theratioofEBITDAtoassets,andtheratioofnetPPEtoassets.Peerfirmaveragesareconstructedastheaverageofallfirmswithinanindustry-yearcombination,excludingtheithobservation.Industriesaredefinedbythree-digitSICcode.Statisticalsignificanceatthe5%and1%levelsisdenotedby*and**,respectively.PeerFirmAverageEquityShockContemporaneous1-Period-LeadIndependentVars.IndependentVars.Firm-SpecificFactorsLog(Sales)−0.000−0.000(−0.565)(−0.334)Market-to-Book−0.0010.000(−1.444)(0.104)EBITDA/Assets−0.009*−0.000(−2.336)(−0.048)NetPPE/Assets0.0080.004(1.934)(0.994)PeerFirmAverageCharacteristicsYesYesFirmiEquityReturnShockYesYesIndustryFixedEffectsYesYesYearFixedEffectsYesYesObs.80,27980,119Adj.R20.1280.127cross-uncorrelated,implyingthatfirms’shocksdonotforecastfutureshocksforthemselvesorforotherfirms.IV.TheRoleandImplicationsofPeerEffectsA.Reduced-FormResultsPanelAofTableIVpresentstheresultsofestimatingequation(6).Thedependentvariableisindicatedatthetopofthecolumns.Thebodypresentscoefficientestimatesandt-statisticsinparentheses.Wepresentresultsformarketandbookleverageinlevels(columns(1)and(2))andfirstdifferences(columns(3)and(4)).Thelatterspecificationshelptoaddressconcernsoveromittedfirmicharacteristics,sincetheyaresimilartolevelsspecificationsthatincludefirmfixedeffects.Thelevelspecificationsusethelevelsforall 154TheJournalofFinanceRTableIVPeerEffectsinFinancialPolicy:Reduced-FormEstimatesThesampleconsistsofallnonfinancial,nonutilityfirmsintheannualCompustatdatabasebe-tween1965and2008withnonmissingdataforallanalysisvariables(seeAppendixA).BothpanelspresentOLSestimatedcoefficientsandt-statisticsrobusttoheteroskedasticityandwithin-firmdependenceinparentheses.Thedependentvariableisindicatedatthetopofcolumns.Allindependentvariablesarelagged1yearandareinlevelsorfirstdifferences()forconsistencywiththedependentvariableindicatedatthetopofthecolumns.Theexceptionisstockreturns,whichareinlevelformacrossallspecifications.Equity(Debt)IssuanceisequaltooneifNetStock(Debt)Issuancesnormalizedbylaggedbookassetsisgreaterthan1%.ThelastcolumnofPanelAisolatesthesubsampleofobservationsinwhicheitheranequityordebtissuance,butnotboth,occurred.InPanelB,thechangeinmarketleverageisthedependentvariableinallspecifications,whichincludeallfirm-specificandpeerfirmaveragesusedinPanelAascontrolvariables.StockReturnControlsincludesfirmi’slaggedandcontemporaneoustotalstockreturn.AdditionalControlVariablesincludeslaggedfirm-specificandpeerfirmaveragesforchangesincashflowvolatility,adividendpayerindicator,Altman’s(1968)Z-score,Graham’s(2000)marginaltaxrate,capitalinvestment,R&Dexpenditures,andSG&Aexpendituresaswellastheintraindus-trystandarddeviationofleverage.SeeAppendixAforcompletevariabledefinitions.PolynomialsofControlsincludesquadraticandcubictermsofallindependentvariablesotherthanindustryaverageleverage.ContemporaneousControlsreplacesthelaggedfirm-specificandpeerfirmav-eragescontrolvariableswithcontemporaneousvalues.Statisticalsignificanceatthe5%and1%levelsisdenotedby*and**,respectively.(Issuers)MarketBookMarketBookEquityDebtDebtLeverageLeverageLeverageLeverageIssuanceIssuanceIssuance(1)(2)(3)(4)(5)(6)(7)PanelA:FinancialPolicyPeerFirmAveragesEquityShock−0.024**−0.016**−0.020**−0.008**0.021*−0.029*−0.034*(−4.965)(−3.997)(−6.312)(−3.109)(2.440)(−2.469)(−2.313)Log(Sales)−0.002−0.0020.030**0.009*−0.008*0.0080.014**(−0.869)(−0.994)(5.941)(2.398)(−2.295)(1.790)(2.810)Market-to-Book0.0010.001−0.001−0.0010.024**0.030**−0.011(0.314)(0.426)(−0.431)(−0.343)(4.817)(5.079)(−1.505)EBITDA/Assets0.0450.118**−0.026−0.0210.0380.357**0.306**(1.574)(4.666)(−1.475)(−1.377)(0.909)(7.130)(4.840)NetPPE/Assets0.078**0.0250.057*0.050*0.066−0.008−0.006(2.700)(1.005)(1.972)(2.392)(1.836)(−0.192)(−0.131)Firm-SpecificFactorsEquityShock−0.008**−0.0020.001−0.0020.062**0.020**−0.052**(−5.812)(−1.624)(1.317)(−1.844)(21.199)(6.282)(−12.219)Log(Sales)0.010**0.010**0.016**0.006**−0.012**0.014**0.024**(9.260)(10.812)(8.071)(3.274)(−9.353)(11.251)(13.173)Market-to-Book−0.053**−0.014**−0.002**−0.002**0.077**0.005*−0.071**(−43.272)(−11.384)(−3.500)(−3.057)(35.226)(2.421)(−28.260)EBITDA/Assets−0.308**−0.231**−0.033**−0.024**−0.258**−0.041**0.161**(−28.937)(−20.865)(−5.400)(−3.511)(−17.546)(−2.689)(7.991)NetPPE/Assets0.161**0.195**0.068**0.053**0.042**0.185**0.086**(12.007)(16.169)(6.851)(5.685)(3.094)(12.220)(4.422)IndustryFixedEffectsYesYesNoNoYesYesYesYearFixedEffectsYesYesYesYesYesYesYesObs.80,27980,27980,27980,27980,27980,27935,363Adj.R20.310.200.100.010.160.050.27(Continued) DoPeerFirmsAffectCorporateFinancialPolicy?155TableIV—ContinuedMarketLeverage(1)(2)(3)(4)(5)(6)PanelB:ChangeinLeverageRobustnessTestsPeerFirmAveragesEquityShock−0.014**−0.019**−0.014**−0.019**−0.015**−0.021**(−4.866)(−5.522)(−3.102)(−6.131)(−5.318)(−6.487)PeerFirmAveragesYesYesYesYesYesYesFirm-SpecificFactorsYesYesYesYesYesYesIndustryFixedEffectsNoNoNoNoNoNoYearFixedEffectsYesYesYesYesYesYesStockReturnControlsYesNoNoNoNoNoAdditionalControlVariablesNoYesNoNoNoNoBank×MarketReturnEffectsNoNoYesNoNoNoLaggedDependentVariableNoNoNoYesNoNoContemporaneousControlsNoNoNoNoYesNoPolynomialsofControlsNoNoNoNoNoYesObs.80,27969,57833,67480,23080,11980,279Adj.R20.260.100.110.100.180.10ofthevariablesonboththeleft-andright-handsidesoftheequation.9Thefirstdifferencespecificationsusesfirstdifferencesforallofthevariablesonboththeleft-andright-handsidesoftheequation.Theonlyexceptionaretheequityshocks,bothforfirmiandpeerfirms,whicharethesameacrossallspecifications.Theresultsincolumns(1)and(2)revealthattheaveragepeerfirmequityshockisstronglynegativelyassociatedwithbothmarketandbookleverage.Thenegativesignsuggeststhatequityshockstopeersaffectfirmiinasim-ilarmannerasfirmi’sequityshocks.However,weemphasizethatapreciseinterpretationofthesignormagnitudeofthiscoefficientisdifficultbecauseitrepresentsacompositeoftheunderlyingstructuralparameters(seeSectionIII).Columns(3)and(4)reinforcethesefindingsbyshowingsimilarresultsforchangesinleverageratios.Thisfindingisreassuringbecauseitshowsthattheunobservedfirm-specificheterogeneityisnotresponsibleforourfindings(Lemmon,Roberts,andZender(2008)).Theeffectsofotherpeerfirmcharacteristics(besidestheequityshock)oncapitalstructurearenotrobustandeconomicallysmall.Peerfirmassettan-gibilityisthemostrobustrelation,thoughitisstatisticallyinsignificantinthesecondspecification.Thisfindingissuggestiveevidencethattheprimarychannelthroughwhichpeerfirmsmayinfluencefinancialpolicyisviaactions9Allcontrolvariablesarelagged1yearrelativetothedependentvariable. 156TheJournalofFinanceR(i.e.,peerfirms’policychoices),asopposedtocharacteristics.WeexaminethisissueinmoredetailbelowinSectionIV.D.Incolumns(5)through(7)ofTableIV,PanelA,weexaminenetequity-andnetdebt-issuingactivitytounderstandwhetherpeersareinfluencingspecificfinancingdecisions.Whilealogitorprobitmodelmaybemoreappropriatefromaforecastingperspective,wepresentresultsusingalinearprobabilitymodel(equation(1))toeaseinterpretationandcomparisonwithotherfindings.Unreportedresultsusingaprobitmodelrevealqualitativelysimilarfindings.Column(5)presentsresultswherethedependentvariableisanindicatorequaltooneifthefirmissuesequitynetofrepurchasesinexcessof1%oftotalassets,andzerootherwise.Thisregressionmodelstheprobabilitythatfirmsissueequityrelativetonotissuingequity,whichincludesdebtissuances,debtretirements,stockrepurchases,andnofinancingactivity.Column(6)presentsanalogousresultsfortheprobabilityofissuingdebt.Inbothmodels,thepeerfirmreturnshocksarestatisticallysignificantlyassociatedwithissuancedeci-sions.Column(7)conditionsonanissuancedecision(debtorequity),therebyeliminatinganumberofinactiveperiods.Theresultsreinforcethoseinthefirsttwocolumns.Firmsaltertheirfinancingbehaviorinresponsetotheirpeers.Insum,theresultsinTableIV,PanelA,suggestthatpeereffectsplayasignificantroleindeterminingvariationincorporateleverageratiosandsecurityissuancedecisions.Whenmakingthesechoices,firmsrespondtotheircompetitors.B.RobustnessTests:PeerEffectsversusOmittedandMismeasuredCommonFactorsInthissection,wefurtherreducetheidentifyingvariationbyconditioningonadditionalcontrolvariablesmotivatedbyalternativehypotheses.PanelBofTableIVpresentstheresults.Forbrevity,weonlyreportresultsusingthechangeinmarketleverageasthedependentvariable.Inunreportedresults,werepeattheanalysisforthelevelofmarketleverage,aswellasthelevelandchangeinbookleverage.Theresultsarequalitativelysimilartothosepresentedhere.Allspecificationsincludefirm-specificfactorsandpeerfirmaveragesforlog(sales),themarket-to-bookratio,EBITDA/Assets,andNetPPE/Assets.Thepresenceoffixedeffectsandallcontrolvariablesareindicatedinthebottompartofthepanel.Werestrictattentiontothekeyvariableofinterest,namely,peerfirmequityshocks.Incolumn(1),wereplacethelaggedfirm-specificequityshockwithlaggedandcontemporaneousfirm-specifictotalstockreturns,rijt.Weseeaslightat-tenuationintheestimatedeffectcomparedtothebaselineestimateof−0.020inPanelA,thoughthecoefficientisstillhighlysignificant,bothstatisticallyandeconomically.Thisspecificationchangeensuresthattheidentifyingvari-ationfrompeerfirms’idiosyncraticreturnsisorthogonaltofirmi’sstock DoPeerFirmsAffectCorporateFinancialPolicy?157returns(laggedandcontemporaneous).Inotherwords,alternativehypothesesmustnowrelyonlaggedidiosyncraticstockreturnsofpeerfirmscontaininginformationaboutfirmi’scapitalstructuredeterminationthatisnotcon-tainedinfirmi’sstockreturns,aswellasanyoftheothercontrolvari-ables.Thisfactallaysanumberofidentificationconcernsrelatedtocorrelatedreturns.Onesuchconcernisthattheassetpricingmodel(equation(7))ismisspeci-fied.Inthiscase,commonfactorsmayremainintheestimatedidiosyncraticcomponentofstockreturns.Byincludingfirmi’stotalreturn,wemitigatethisconcernbecausemostcommoncomponentsinstockreturnsthatarerelevantforcapitalstructurearearguablybettercapturedbyfirmi’sstockreturns,asopposedtofirmj’slaggedidiosyncraticreturn.Anotherconcernisthatfirmsreceiveindustry-wideshockstotheirequityvaluationsandthattheseshocksareasynchronous,sothattheyearfixedef-fectsareinadequatecontrols.Forexample,industriesmayexperience“hot”and“cold”equitymarketsduetoshiftinginvestordemands,whichcauseeq-uityvaluationsforallfirmsinanindustrytomoveinthesamedirection(e.g.,thetechsectorinthelate1990s).Becausetheseshiftsininvestordemandarereflectedinprices,thisconcernislargelyeliminatedbyincludingfirmi’sstockreturnsinthespecification.Furthermore,byincludingboththecontempora-neousandlaggedstockreturn,weeliminateconcernsregardingthetimingofequitypriceshockswherebysomefirmsinanindustrygetshockedearlierthantheirpeers.10Column(2)ofTableIV,PanelB,examinesa“kitchensink”modelofcapitalstructureincludingadditionalexplanatoryvariablespreviouslyidentifiedasrelevantforcapitalstructure.Specifically,weincludelaggedfirm-specificandpeerfirmaveragesforanindicatoridentifyingwhetheradividendwaspaid,Altman’sZ-score,Graham’s(2000)marginaltaxrate,capitalinvestment,R&Dexpenditures,SG&Aexpenditures,andintraindustryleveragedispersion.Theresultsareunaffectedbytheirinclusion.Column(3)incorporatesbankfixedeffects,andbankfixedeffectsinteractedwiththeCRSPvalue-weightedmarketreturn.11Thisspecificationaddressestheconcernthatcommonalityamongfirms’capitalstructuresisduetotheuseofcommonbanks(commercialorinvestment)withintheindustryandthatfinancialadvicefromthesebanksvariesoverthebusinesscycle.Thischange10Likewise,thisspecificationalleviatesconcernsovercommonmovementsincreditprices.Ifstockreturnscontaininformationaboutthecostofdebt,thenanalternativebasedonshiftsininvestordemandforcreditwouldrequireademandshockthat(1)affectsthewholeindustry,yetisnotcapturedbytheindustryreturnintheassetpricingmodel,and(2)isreflectedinpeerfirms’idiosyncraticreturns,butisnotreflectedinfirmi’stotalreturn.Coupledwiththeadditionalevidencediscussedbelow,thisalternativeseemsunlikely.11SeeAppendixAforadescriptionoftheconstructionofbankfixedeffects.Inunreportedresults,weinteractthebankeffectswiththeyieldspreadonBaaoverAaacorporatebondsasanalternatemeasureofmarketconditions.Theresultsarequalitativelysimilar. 158TheJournalofFinanceRhaslittleeffectonourresults,despitethesharpdeclineinobservationsduetotheadditionaldatarequirements.12Column(4)ofTableIV,PanelBincorporatesfirmi’slaggedleverageratiotocaptureanytargetingbehaviorordynamicfeedbackfromtheexplanatoryvariablesintoleverageratios.Thisspecificationaddressestheconcernthatthepeerfirmreturnshockiscorrelatedwithachangeinfirmi’sleveragetarget,orwithaperturbationawayfromthattarget,inawaynotcapturedbytheotherincludedvariables.Thisspecificationalsoallowsfordynamictargetingbehaviorinleverage(e.g.,FlanneryandRangan(2006)andKayhanandTitman(2007)).Column(5)ofTableIV,PanelBreplacesthelaggedcontrolvariableswithcontemporaneouscontrolstoaddresstheconcernthatcapitalstructure-relevantshocksaffectourfirm-specificandpeerfirmcharacteristicswithalag.Finally,column(6)includesquadraticandcubicpolynomialsofeachfirm-specificfactorandpeerfirmaveragecharacteristicinourprimaryspecification(i.e.,firmsize,profitability,tangibility,market-to-book).Again,weseelittlechangeintheresults,suggestingthatfunctionalformmisspecificationinthecontrolvariablesisunlikelytobebehindourresults.C.Customer–SupplierLinksInTableV,wetakeadifferentapproachtodefiningpeergroupsandourmeasureofpeerfirmbehaviortoaddressremainingidentificationconcerns.Inparticular,thenoiseinindividualstockreturnsmayleaveroomforourmeasureofpeerfirmequityshockstoprovideadditionalinformationaboutfirmi’scapitalstructurethroughthesmoothingeffectofaveraging,orthroughtracesofcorrelationbetweenourmeasureandanindustryfactorthatisrelevantforallfirms’leveragebutforwhichindependentvariablesdonotadequatelycontrol.Assuch,wedefinethepeergroupforfirmiasthesubsetoffirmsinthesameindustryasfirmiwithatleastonecustomerfirmthatsatisfiesthefollowingthreecriteria:(1)thecustomerisinanindustrydifferentfromfirmi,(2)thecustomerisnotacustomeroffirmi,and(3)thecustomeraccountsforatleast10%ofthepeerfirm’ssales.ThemotivationforthispeergroupdefinitioncomesfromCohenandFrazzini(2008),whoshowthatshockstocustomerspredictequityreturnsandrealoutcomesforsupplierfirms,butnotforfirmsinthesamesupplierindustrywithoutanactivecustomer–supplierlink.Usingthisinsight,weusetheaverageequityreturnshocktothecustomersofpeerfirmsthatarenotalsocustomersoffirmiasameasureofpeerfirmbehavior.1312Wealsobelievethatcommoninstitutionalownershipisnotlikelytoberesponsibleforourfindings.Thelargemajorityofinstitutionalinvestorsarepassiveandunlikelytobedictatingfinancialpolicy.Bravetal.(2008)estimatethattheactivistshareoftotalinstitutionalequityownershiprangesfrom0.7%to2.3%from2000to2007.13WethankLaurenCohenforkindlysharinghisupdateddataonlinkingcustomersandsup-pliersintheCRSPdatabase.SeeCohenandFrazzini(2008)fordetailsonthesedata. DoPeerFirmsAffectCorporateFinancialPolicy?159TableVCustomer–SupplierTestsThesampleconsistsofallnonfinancial,nonutilityfirmsintheannualCompustatdatabasebetween1965and2008withnonmissingdataforallanalysisvariables(seeAppendixA).PanelApresentsOLSestimatesusingpeergroupsdefinedasthesubsetoffirmsinthesameindustryasfirmithatsatisfythefollowingtwocriteria:(1)theyhavecustomersinanindustrydifferentfromfirmi,and(2)theircustomersarenotcustomersoffirmi.PanelBpresentsOLSestimatesinwhichwereplaceeachcustomerfromtheanalysisinPanelAwitharandomlyselectednoncustomerinthesameindustryasthecustomer.Theaverageshocktotherandomlyselectednoncustomersisthenusedinplaceoftheshocktothecustomer.WeperformtherandomselectionandOLSestimation100timestoobtainadistributionofestimatedcoefficientsandt-statisticsonthenoncustomerequityreturnshocks.Allspecificationsincludefirm-specificandpeerfirmaveragesforfirmsize,profitability,tangibility,andthemarket-to-bookratio.Statisticalsignificanceatthe5%and1%levelsisdenotedby*and**,respectively.MarketMarketIssueLeverageLeverageDebt(1)(2)(3)PanelA:CustomerReturnShocksAvg.PeerCustomerEquityShock−0.012*−0.011**−0.036*(−2.398)(−3.535)(−2.027)IndustryAvg.EquityReturn−0.044**0.009**−0.042**(−10.898)(3.553)(−3.211)PeerFirmAveragesLog(Sales)0.004*0.018**0.002(2.339)(4.978)(0.589)EBITDA/Assets0.062**0.0020.130**(5.031)(0.278)(3.431)Market-to-Book0.001−0.002*−0.001(0.759)(−2.294)(−0.216)NetPPE/Assets−0.0130.055**−0.049(−0.842)(2.942)(−1.256)Firm-SpecificFactorsLog(Sales)0.009**0.024**0.027**(8.271)(12.219)(11.015)EBITDA/Assets−0.176**−0.035**0.134**(−20.583)(−6.440)(6.143)Market-to-Book−0.039**−0.003**−0.062**(−39.523)(−5.795)(−24.398)NetPPE/Assets0.199**0.063**0.148**(13.153)(5.366)(5.701)EquityShock−0.009**0.004**−0.044**(−6.719)(3.878)(−9.187)IndustryFixedEffectsYesNoYesYearFixedEffectsYesNoYesObs.54,59952,22221,410Adj.R20.280.080.26(Continued) 160TheJournalofFinanceRTableV—ContinuedPercentilesMean525507595PanelB:PlaceboTestsCoefficientEstimatesMarketLeverage0.004−0.0020.0010.0040.0060.010MarketLeverage0.001−0.004−0.0010.0000.0030.006IssueDebt0.001−0.019−0.0060.0010.0070.021PeerEffectt-statsMarketLeverage1.496−0.8310.2831.5372.4994.033MarketLeverage0.414−2.738−0.8870.2741.7043.475IssueDebt0.080−2.016−0.6380.1110.7632.398Thebenefitofthisapproachisamorecompellingidentificationstrategy.Becausethecustomersareinadifferentindustryanddonotshareasupplychainlinkwithfirmi,thereislessconcernoverlatentcommonfactorsdrivingtheresults.Furthermore,becausethemeasureisnowbasedonshocksfromadifferentindustry,wecanincludefirmi’sindustryreturn,inadditiontofirmi’sownstockreturn,asacontrolvariable.Thus,theidentifyingvariationnowcomesfromreturnshockstofirmsinanotherindustrythatareorthogonaltofirmi’sstockreturnandfirmi’sindustryreturn,aswellasalloftheotherincludedcontrolvariables.Thedrawbackofthisapproachisanoisydefinitionoffirms’peergroups.Infact,thesecondcriterionaboveensuresthatthemostsimilarfirmsfromademandperspectivearenotincludedinthepeergroup.Theconsequencesofthisnoiseareareductioninstatisticalpowerandapossibleattenuationoftheestimatedpeereffect.TheresultsinPanelAofTableVshowaslightattenuationinthecoeffi-cientrelativetotheestimatesofTableIV.However,peerfirmcustomerequityshocksarestillsignificantlynegativelycorrelatedwithbothleverageandnetissuancedecisions,thelatterofwhichconditionsthesampleoneitherneteq-uityissuanceornetdebtissuance.Unreportedresultsexaminingbookleveragearesimilar.Wealsofindasignificantrelationbetweenfirmi’sindustrystockreturnandfinancialpolicy,thoughthecoefficientsonpeerfirmcustomerequityshocksremainsignificant.Toensurethatthecustomer–supplierlinkisunique,PanelBpresentstheresultsfromaplacebotest.Wereplaceeachcustomeroffirmi’speerswitharandomlyselectedfirmfromthesameindustryasthecustomerbutwithnoeconomictiestofirmi’sindustry.Wecallthesefirms“noncustomers.”WethenconstructtheexogenouspeerfirmmeasureusingthereturnshockstothenoncustomersandrerunouranalysisfromPanelA.Werepeatthisprocessofreplacingeachcustomerwitharandomlyselectednoncustomer, DoPeerFirmsAffectCorporateFinancialPolicy?161constructingthereturnshockmeasure,andestimatingtheregressions100times.Thedistributionofthecoefficientestimatesonthereturnshockandthecorrespondingt-statisticsarepresentedinPanelBofTableV.Toaddressoutlierestimates,weWinsorizetheresultsatthe5thand95thpercentiles.Theresultsinthetophalfofthepanelshowthattheaverageandme-dianpeereffectestimatesareallsignificantlysmallerinmagnitudethanthoseinPanelA.Focusingonthemedian,weseethattheplaceboestimateforthelevelofleverageis0.004,comparedwith−0.012.Theplaceboesti-mateforthefirstdifferenceinleverageis0.000versus−0.011.Finally,theplaceboestimatefordebtissuancesis0.001versus−0.036ThedifferencesinWinsorizedmeansaresimilar.PanelBshowsthatmostoftheplaceboesti-matesarestatisticallyinsignificantaswell.Forthelevelandfirstdifferenceinleverage,thereappearstobeapowerdistortionbecausemorethan5%oftheestimatesarestatisticallysignificant.Nonetheless,theevidenceissup-portiveofthepreviousfindings,furthersuggestingidentificationofapeereffect.D.PeerEffectsChannels:ActionsversusCharacteristicsWhileourreduced-formresultsestablishthepresenceofsignificantpeereffects,theyaresubjecttotwolimitations.First,asdiscussedinSectionIIItheydolittletodistinguishbetweenthetwochannelsthroughwhichpeereffectsoperate.Inthissection,weprovideadditionalanalysestoaidwiththisdistinction.Second,sinceweestimatecompositeparameters,itisdifficulttoassesstheeconomicmagnitudeofthepeereffects.Weturntothisissueinthenextsubsection.Toillustratethechallengeofdistinguishingbetweenthetwopeereffectchannels,considerthefollowinghypotheticalexample.FirmAintroducesanewproduct,whichpositivelyimpactstheidiosyncraticcomponentofitsstockreturn.Inthefollowingperiod,firmAissuesequitytofinanceanin-creaseinproductionandreduceitsleverageratio.Inresponse,peerfirmBissuesequityandreducesitsleveragetoo.Thequestionis:isfirmBre-spondingtothechangeinfinancialpolicy,ortotheintroductionofthenewproduct(i.e.,theinformationabouttheircompetitorembeddedinthestockreturn)?Tohelpdistinguishbetweenthesealternatives,weexploitheterogeneityinfirms’capitalstructureresponsestotheirpeers’equityshocks.Wedosobyperformingadoublesortofthedatabasedonquintilesofourpeerfirmaverageequityshocksandpeerfirmleveragechanges.Withineachquintilecombina-tion,wecomputetheaveragechangeinleverageforfirmiandat-statisticofwhetherthischangeissignificantlydifferentfromzero.Weperformthisanalysisonbothbookandmarketmeasuresofleverage,butpresentonlythemarketleverageresultsforbrevity. 162TheJournalofFinanceRTableVILeverageChangesbyPeerFirmEquityShockandPeerFirmLeverageChangeThesampleconsistsofallnonfinancial,nonutilityfirmsintheannualCompustatdatabasebetween1965and2008withnonmissingdataforallanalysisvariables(seeAppendixA).Thetablepresentsaveragemarketleveragechangesfor25groupsofobservations.Thegroupsareformedbytheintersectionofquintilesbasedon:(1)peerfirmaverageequityreturnshockslagged1yearand(2)peerfirmaveragechangeinmarketleverage.Thecolumnlabeled“5−1”presentsthedifferenceinmeansbetweencolumns5and1.Therowlabeled“5−1”presentsthedifferenceinmeansbetweenrows5and1.t-statisticsrobusttoheteroskedasticityandwithin-firmdependenceareinparentheses.Statisticalsignificanceatthe1%levelisdenotedby**.LaggedPeerFirmPeerFirmAvgLeverageChangeQuantilesAvgEquityShock1(Low)2345(High)5−11(Low)−0.033**−0.008**0.007**0.021**0.062**0.095**(−14.176)(−4.026)(3.158)(9.857)(29.059)2−0.044**−0.014**0.007**0.020**0.062**0.106**(−18.302)(−6.574)(4.348)(9.665)(26.558)3−0.042**−0.014**−0.0000.023**0.066**0.108**(−18.608)(−7.284)(−0.253)(13.002)(25.001)4−0.047**−0.013**0.0030.017**0.066**0.114**(−22.532)(−7.628)(1.566)(8.202)(28.275)5(High)−0.046**−0.024**0.006**0.016**0.062**0.108**(−27.376)(−11.640)(2.644)(7.327)(24.489)5−1−0.014**−0.016**−0.000−0.005−0.000TheresultsarepresentedinTableVI,wherequintile“1”representsthelow-est20%ofthedistributionandquintile“5”thehighest.Forexample,theaver-agechangeinleverageamongfirmsinthelowestpeerfirmequityshockquintileandthehighestpeerfirmleveragechangequintileis6.2%withat-statisticof29.06.Wenoteamonotonicincreaseintheaverageleveragechangeacrosseachrow.Inotherwords,holdingfixedthepeerfirmequityshock,leveragechangesarestronglypositivelycorrelatedwithchangesinpeerfirmleverage.Theconverseisnottrue.Averageleveragechangesarelargelyuncorrelatedwiththepeerfirmequityshock,holdingfixedpeerfirms’averageleveragechange.Infact,incolumn(3),wheretheaveragepeerfirmleveragechangeisindistinguishablefromzero,thecellaveragesarealleconomicallysmallandtwoarestatisticallyinsignificant.Thus,firmsonlychangetheirleverageinresponsetoapeerfirmequityshockifitisaccompaniedbyachangeinpeerfirmleverage.Thesefindingsreinforcetheimplicationoftheregressionresultsandsuggestthatpeerfirmaverageequityshocksaremorelikelycapturingaresponsetopeerfirmfinancialpolicies,asopposedtocharacteristics.However,theycomewithacaveat.Itmaybethecasethatpeerfirmcharacteristicsonlymatterforfirmi’sfinancialpolicywhentheyareaccompaniedbyachangeinpeerfirmfinancialpolicyaswell.Thus,theresultsinTableVIdiminishthescopeforour DoPeerFirmsAffectCorporateFinancialPolicy?163measuretocaptureapeereffectoperatingthroughcharacteristics,buttheycannotcompletelyruleitout.Inunreportedanalysis,weexaminewhetherothercorporatepoliciesareaf-fectedbypeerfirms’returnshocks.Wefindnorelationbetweenpeerfirmsreturnshocksandtheinvestment,researchanddevelopment,ordividendpoli-ciesoffirmi.Thisfindingreinforcesouridentificationbecause,iflatentin-vestmentopportunitiesarebehindourearlierfindings,thenthisshouldshowupintheinvestmentregression.Thisfindingalsofurthersupportstheviewthattheprimarychannelthroughwhichpeereffectsinfinancialpolicyoperateisviaactions,thatis,peers’capitalstructuredecisions.Thenextsubsectioninvestigatesthisinferencemoreformally.E.TheEconomicImportanceofPeerEffectsToestimatethemagnitudeofpeereffectsinfinancingpolicy,weneedtoes-timatethestructuralparametersinequation(1).Thisrequiresaninstrumentfortheendogenouspeerfirmoutcomevariable,¯y−ijt.Inthissection,weesti-mateequation(1)viatwo-stageleastsquares,usingourmeasureofpeerfirmequityshockstoinstrumentforpeerfirmfinancialpolicies.14Inemployingthisstrategy,weacknowledgethattheidentificationassump-tionsarenowmorestringentthanwhenestimatingthereduced-formmodel.Therelevanceconditionrequiresthatpeerfirmequityshocksbesignificantlycorrelatedwithpeerfirmcapitalstructurechoices.Thisassumptionistestable.Moreimportant,ourestimatesmaybebiasediftheaveragepeerfirmreturnshocksarecorrelatedwitheither(i)anomittedfirmicapitalstructuredeter-minantor(ii)anomittedpeerfirmcharacteristicthatisrelevantforfirmi’scapitalstructure.Whiletheformerconcernismitigatedbyourpreviousanal-yses,wecannotcompletelyruleoutthelatter.However,theresultsdiscussedinSectionIV.Dsuggestalimitedroleforpeerfirmcharacteristicsincapitalstructurechoices,implyingthatanyremainingbiasislikelytobesmall.E.1.InstrumentalVariablesResultsTableVIIdisplaystheresultsoftwo-stageleastsquaresestimationofequa-tion(1).Thefirstfourcolumnsshowresultswithleverageastheoutcomevariable,bothmarketandbookinlevelsandchanges.Thecoefficientsontheinstrumentfromthefirst-stageregressionsareshownatthebottomofthetable.Thefirst-stageresultsrevealthattheaverageequityshockisstronglynegativelycorrelatedwithboththelevelandfirstdifferenceinaverageindus-tryleverageratios.Thesignoftheestimateisconsistentwithpreviousfindingsrelatingtotalreturnstoleverageandwiththeoreticalargumentsrelatingin-vestmentopportunitiesandrisktooptimalleverageandfinancingchoices(e.g.,14ThisstrategyfollowsDufloandSaez(2002)andCaseandKatz(1991),whosimilarlyuseaverageexogenouscharacteristicsofthepeergroupasaninstrumentforpeers’behaviorinthecontextofretirementsavingsplanparticipationandneighborhoodeffectsonsocioeconomicout-comes,respectively. 164TheJournalofFinanceRTableVIIPeerEffectsinFinancialPolicy:StructuralEstimatesThesampleconsistsofallnonfinancial,nonutilityfirmsintheannualCompustatdatabasebe-tween1965and2008withnonmissingdataforallanalysisvariables(seeAppendixA).Thetablepresentstwo-stageleastsquares(2SLS)estimatedcoefficientsscaledbythecorrespondingvari-able’sstandarddeviation,andt-statisticsrobusttoheteroskedasticityandwithin-firmdependenceinparentheses.Thedependentvariableisindicatedatthetopofeachcolumn.Theendogenousvariableisthepeerfirmaverageofthedependentvariable.Theinstrumentistheone-period-laggedpeerfirmaverageequityreturnshock.PeerFirmAveragesdenotesvariablesconstructedastheaverageofallfirmswithinanindustry-yearcombination,excludingtheithobservation.In-dustriesaredefinedbythree-digitSICcode.Firm-SpecificFactorsdenotesvariablescorrespondingtofirmi’svalueinyeart.Allvariablesareinlevelsorfirstdifferencesasindicatedatthetopofthecolumns.Allindependentvariables,includingtheinstrumentbutexcludingtheendogenousvariable,arelagged1yearrelativetothedependentvariableunlessotherwisespecified.Equity(Debt)IssuanceareindicatorvariablesequaltooneifNetStock(Debt)Issuancesnormalizedbylaggedbookassetsisgreaterthan1%.ThelastcolumnofPanelAisolatesthesubsampleofobser-vationsinwhicheitheranequityordebtissuance,butnotboth,occurred.Statisticalsignificanceatthe5%and1%levelsisdenotedby*and**,respectively.(Issuers)MarketBookMarketBookEquityDebtDebtLeverageLeverageLeverageLeverageIssuanceIssuanceIssuance(1)(2)(3)(4)(5)(6)(7)PeerFirmAveragesDependentVariable0.100**0.100**0.077**0.045*0.046*−0.6310.105*(4.803)(3.573)(5.431)(2.566)(2.423)(−1.249)(2.230)Log(Sales)−0.011**−0.015**−0.004*−0.001−0.0030.0950.001(−2.719)(−3.111)(−2.440)(−0.777)(−0.599)(1.387)(0.120)Market-to-Book0.032**0.011**0.0010.002−0.0020.0920.017(4.403)(2.791)(1.463)(1.602)(−0.278)(1.555)(1.410)EBITDA/Assets0.021**0.020**0.002*0.0010.008*0.1020.006(4.618)(4.949)(2.416)(1.151)(2.244)(1.632)(0.744)NetPPE/Assets−0.013−0.033**−0.001−0.0010.0070.083−0.001(−1.659)(−2.858)(−1.788)(−1.251)(1.095)(1.200)(−0.144)Firm-SpecificFactorsLog(Sales)0.021**0.021**0.005**0.002**−0.026**0.035**0.051**(9.113)(10.435)(6.541)(2.640)(−9.161)(7.432)(12.410)Market-to-Book−0.067**−0.018**−0.002**−0.002**0.095**0.015−0.097**(−42.570)(−11.257)(−3.661)(−3.081)(34.880)(1.871)(−27.751)EBITDA/Assets−0.048**−0.037**−0.003**−0.002**−0.040**0.0050.026**(−28.700)(−20.769)(−5.128)(−3.280)(−17.645)(0.514)(7.661)NetPPE/Assets0.034**0.042**0.004**0.003**0.009**0.040**0.018**(11.281)(15.105)(5.955)(4.680)(2.940)(8.654)(4.137)EquityShock−0.003**−0.0000.002**−0.0000.032**0.007−0.028**(−4.468)(−0.484)(3.030)(−0.713)(21.071)(1.535)(−12.039)First-StageInstrumentPeerFirmAvgEquity−0.033**−0.015**−0.015**−0.005**0.074**0.009−0.091**Shock(−13.126)(−7.350)(−8.713)(−4.383)(19.430)(1.530)(−8.146)IndustryFixedEffectsYesYesNoNoYesYesYesYearFixedEffectsYesYesYesYesYesYesYesObs.80,27980,27980,27980,27980,27980,27934,578 DoPeerFirmsAffectCorporateFinancialPolicy?165Scott(1976)andMyers(1977)).Statisticallyspeaking,theinstrumenteasilypassesweakinstrumenttests(e.g.,StockandYogo(2005)).Therowlabeled“Dependentvariable”atthetopofthetablereportsβˆ,theestimatedcoefficientontheinstrumentedpeerfirmaverageoutcomevariable.Foreachspecification,theresultsindicatethatfirms’leveragechoicesaresig-nificantlypositivelyinfluencedbytheleveragechoicesoftheirpeers.Toeaseinterpretationofmagnitudes,allcoefficientsarescaledbythecorrespondingvariable’sstandarddeviation.Thus,fromcolumns(1)and(2),aonestandarddeviationincreaseinpeerfirmaverageleverageleadstoa10percentagepointincreaseinfirmi’sleverage.Comparedtotraditionalfirm-specificdetermi-nants,peerfirmfinancialpolicieshaveasignificantlylargereffect.Forexam-ple,inthemarketleverageregression(column(1)),thenext-mostimpactfuldeterminantisthemarket-to-bookratio,whosescaledcoefficientis−6.7%—almost40%smaller.Forbookleverage,theeffectofassettangibilityislessthanhalfthatofpeerfirmaverageleverage.Theresultsincolumn(5)reinforcethesefindingsbyshowingthatfirms’equityissuancedecisionsaresignificantlyinfluencedbytheirpeers’issuancedecisions.Thefirststageindicatesastrongpositiveassociationbetweenpeerfirms’returnshocksandtheirequityissuancedecisions.Second-stageresultsshowthataonestandarddeviationincreaseintheprobabilityofissuingequitybypeerfirmsleadstoa4.6%increaseintheprobabilityoffirmiissuingequity.Thepeereffectisoneofthemosteconomicallyimportantdeterminants,sec-ondonlytofirmi’sownmarket-to-bookratio.Column(6)presentsanalogousresultsforthedecisiontoissuedebt.Netherfirst-norsecond-stageestimatesarestatisticallysignificant.Column(7)showsthatthisresultisduelargelytothecomparisonset.Whenwerestrictattentiontothesubsampleofactivefinancingdecisions(netdebtissuanceornetequityissuance),wefindaneco-nomicallylargepeereffect.Specifically,thefirst-stageestimateisstatisticallysignificantlynegativebecausewearemodelingthedebt,asopposedtoequity,decision.Thesecond-stageestimaterevealsastatisticallyandeconomicallylargepositivepeereffect.Thus,conditionalonfinancingactivity,peerfirmsplayanimportantroleinthelikelihoodofnetissuingactivity.Insum,thepeereffectsplayaneconomicallysignificantroleindeterminingvariationincorporateleverageratios.Thisvariationinleverageisdrivenbypeereffectsinfinancingchoices.Theseeffectsareeconomicallylarge,signifi-cantlylargerthanalmostanyotherestimatedeffect.E.2.Amplification,Spillover,andMarginalEffectsAnimportantimplicationoftheempiricalmodelinequation(1)isthepres-enceofexternalitieswherebychangestoonefirmaffecttheoutcomesatanotherfirm.Theseexternalitiesimplythatthetotalderivativeisnolongerequaltothepartialderivative,eveninalinearmodel,becauseofthepresenceofthepeerfirmoutcomevariableontheright-handsideoftheequation.Sincethetotalderivativeistheeconomicquantityofinterest,theeffectofachangeinanyexogenouscapitalstructuredeterminantcannotbeinferredsolelyfromits 166TheJournalofFinanceRcoefficient.Rather,thederivativesofinterestare⎧β2βdyi⎨λm1+(N−1+β)(1−β)+γm(N−1+β)(1−β)fori=l=(8)dxlm⎩λβ+γ1fori=l,m(N−1+β)(1−β)m(N−1+β)(1−β)whereiandldenotefirm-yearobservationsandmdenotestheregressor.Thus,dyi/dxlmmeasuresthechangeinyforfirmigivenaoneunitchangeinxmforfirml.ThenumberoffirmsinthepeergroupisdenotedbyN.(SeeAppendixBforaderivation.)Inthetypicallinearmodelwithoutpeereffects,bothβandγareequaltozeroandthederivativereducesto∂yi/∂xlm=λmforalliandl.Withpeeref-fects,thereareseveraldistinctions.Wheni=l,thepeerfirmaverageleveragecoefficient,β,amplifiestheeffectofachangeinanexogenousvariableony.Thisamplificationmechanismisrepresentedbytheparentheticalexpressionmultiplyingλm.ForβintheopenunitintervalandN>1,thisexpressionisstrictlygreaterthanone.Becauseofthepresenceofpeerfirmcharacteristics,thisamplificationmaybeeitherfurtheramplifiedoroffsetdependingonthecorrelationbetweentheoutcomevariableandthepeerfirmcharacteristics,γm.(Thesecondterminthecasei=l.)Wheni=l,thederivativeisnolongerzero.Instead,cross-observationeffectsaredeterminedbytherelativeimportanceofpeerfirmactions(β)andcharacteristics(γ).TheestimationofthestructuralparametersinTableVIIallowsustoes-timatetheseeffects.TheresultsarepresentedinTableVIII.Thefirsttwocolumnsrepeatthescaledcoefficientsonthefirm-specificandpeerfirmaver-agecharacteristicsfromcolumn(2)ofTableVII(marketleveragespecification).Theexceptionistheestimateofβ,whichispresentedunscaledtoverifythatitisbetweenzeroandone.Intheremainingcolumns,thebottompanelpresentsestimatesoftheam-plificationterm(thefirstparentheticalexpressioninthecasei=l),spilloverterm1(thesecondparentheticalexpressioninthecasei=l),spilloverterm2(thesecondparentheticalexpressioninthecasei=l),andtheircorrespondingχ2teststatisticsinparentheses.Theupperpanelreportsthemarginaleffectofaonestandarddeviationchangeinownfirmandpeerfirmcharacteristics.15Becausethesizeoftheindustry,N,playsacentralroleinthederivativeex-pressions,wepresentestimatesforthreedifferentsizeindustriesbasedonthe10th(threefirms),50th(eightfirms),and90th(26firms)percentilesoftheindustrysizedistribution.Wenoteseveralfindings.First,theamplificationterm,thoughnoisilyesti-mated,variesdramaticallyacrossindustrysizecategoriesandiseconomicallylarge.Changesincapitalstructuredeterminantsaremagnifiedby71%insmallindustriesand8%inlargeindustries.Intuitively,eachfirmhasasmallereffectonitspeersthelargeristhepeergroup.15Forthederivativesandspilloverterms,thenullhypothesisisthatthesetermsareequaltozero.Fortheamplificationterms,thenullhypothesisisthatthesetermsareequaltoone.Standarderrorsarecomputedusingthedeltamethod. DoPeerFirmsAffectCorporateFinancialPolicy?167=).xms)σβ−)and×)(1)i0.0000.499)2ββyxkm−−β+−∂∂1−)(1ββpercentileN(+x1σth+−N(×yiim0.065**32.156)0.0020.047**(1.161)23.794)0.002(1.266)∂x∂−−−−xσ×i0.0020.500)0.021**(8.252)(yxkmshowsthechangetotheoutcome∂∂−−ximshowsthechangetotheoutcomeofx/∂ikmσyx×∂/∂yiyiim0.063**13.751)0.0060.045**(1.178)11.231)0.005((1.285)(∂x∂∂−−−−xσ×i0.0040.502)0.020**(5.803)(yxkm∂∂−−x).Thederivativeσ).Bothderivativesarescaledbythestandarddeviationofthexim×26firms).Thederivative(xkm4.597)(1.225)3.678)((1.337)(i(yiim0.055**0.0160.038**0.0150.977(1.414)0.345(1.346)=k∂x∂−−(1.092)(2.004)(1.051)(1.870)−−)xσ×TableVIIIpercentileγthforobservationmforobservation0.011**2.719)−0.018*(2.564)0.013−(1.659)−0.042**(5.298)0.011(1.257)0.036**(8.988)0.004(1.241)0.035**(10.984)(1.235)0.001xm−xobservation.Industriesaredefinedbythree-digitSICcode.Firm-SpecificFactorsdenotesthi.Derivativesarecomputedforthreepeergroupsdifferingintheirsize:small(10)ScaledCoefs(tσx×λeightfirms),andlarge(90=Factor(9.113)0.067**Average42.570)0.048**28.700)((11.281)Size:Small0.032**(4.403)(4.803)0.021**(4.618)Size:Medium(((Size:Large−(−−(−Firm-SpecificPeerFirmPeerGroupPeerGroupPeerGroup’svalueinyeari).AmplificationTermisthemultiplicativefactorduetothepeereffectactionvariableandisequalto(1percentileσx)0.727**βthExogenousVariableDerivatives,MarginalEffects,andLeverageMultipliers)followingaoneunitchangetovariableyi()followingaoneunitchangetovariablevariable,(iyi(xi),respectively.Statisticalsignificanceatthe5%and1%levelsisdenotedby*and**,respectively.)β−)(11β+1−NThesampleconsistsofallnonfinancial,nonutilityfirmsintheannualCompustatdatabasebetween1965and2008withnonmissingdataforallanalysisvariables(seeAppendixA).Thetablepresents2SLSestimatedcoefficientsscaledbythecorrespondingvariable’sstandarddeviation,and-statisticsrobusttoheteroskedasticityandwithin-firmdependenceinparenthesesfromaregressionofmarketleverageonpeerfirmleverage,peertfirmcharacteristics,andfirm-specificfactors.Allvariablesareinlevels.PeerFirmAveragesdenotesvariablesconstructedastheaverageofallfirwithinanindustry-yearcombination,excludingthevariablescorrespondingtofirmthreefirms),medium(50ofobservationobservationcorrespondingThetermsSpillover1andSpillover2aretheadditivefactorsduetothepeerfirmactionsandcharacteristics,andareequalto(((VariableLog(Sales)Market-to-BookEBITDA/AssetsNetPPE/AssetsScaledCoefs(PeerFirmAvg.Leverage(AmplificationTerm0.021**Spillover1Spillover20.034** 168TheJournalofFinanceRSecond,forsomedeterminants,thetruemarginaleffectdifferssignificantlyfromthatimpliedbythefirm-specificcoefficient.Forexample,themarginaleffectofassettangibilityis30%smallerinlargeindustriesrelativetosmallindustries.Theoppositeistrueoffirmsize,whichplaysamoreimportantroleinlargerindustriesthansmallerindustries.Thesedifferencesaredrivenbydifferencesinsignsandmagnitudesbetweenthefirm-specific(λ)andpeerfirmaverage(γ)coefficients.Wealsonotethatthecross-observationderivativesareallstatisticallyin-significant.Thisisconsistentwiththesmallerimpactofpeerfirmcharacteris-ticsoncapitalstructurerelativetofirm-specificcharacteristicsandpeerfirmactions.Inotherwords,achangeinfirmA’sprofitability,forexample,hasalargerimpactonfirmAthanonfirmB.However,cautionmustbetakenwheninterpretingthesederivatives.Theyisolatetheimpactofonlyonefirmonan-other.Iftheindustryasawholereceivesaprofitabilityshock,affectingallormanyofthefirms,thenthespillovercanbesubstantial.V.WhyDoFirmsMimicOneAnother?Giventheimportanceofpeerfirmbehaviorforfirms’capitalstructures,wenowturntowhyfirmsmimiconeanother.Webeginwithabriefdiscussionofthepotentialmechanismsbehindtheestimatedpeereffects,whichweusetoguidethesubsequentempiricalanalysis.A.TheoreticalMotivationPeereffectsincapitalstructurecanariseforavarietyofreasons.Forex-ample,interactionsbetweenfinancialstructureandproductmarketcompe-titioncanleadtofinancialpolicymimicking.BoltonandScharfstein(1990)presentamodelinwhichhighleverageinvitespredatorypricecompetitionfromless-leveredrivals.Iftheexpectedcostofthispredatorybehaviorisse-vereenough,highlyleveredfirmswillmimicthecapitalstructuresoftheirless-leveredrivals.Similarly,ChevalierandScharfstein(1996)presentamodelinwhichfirmswithhighleverageunderinvestduringanindustrydownturnandlosemarketsharetomoreconservativelyfinancedcompetitors.Thislosscanmotivatefirmstomimicthemoreconservativeleveragepoliciesoftheirpeers.16Additionalmotivationformimickingbehaviorincapitalstructurecomesfromrationalherdingmodels(DevenowandWelch(1996)).17Zeckhauser,Patel,16AnotherrelatedmodelistheduopolymarketmodelofBranderandLewis(1986),inwhichfeedbackbetweenproductmarketsandfinancialpolicyleadstocapitalstructuremimickingamongcompetitors.MaksimovicandZechner(1991)alsoexaminetheinteractionbetweenproductmar-ketsandfinancialpolicy.However,theimplicationsofthisstudyaregearedmoretowarddiffer-entialpositioningwithintheindustry,asopposedtomimickingbehavior.SeePhillips(1995)andMacKayandPhillips(2005)forempiricalexaminations.17AsDevenowandWelch(1996)discuss,therealsoexistmodelsofirrationalherdinginwhichagentsblindlyfollowoneanotherandforgorationalanalysis.Webelievethatsuchtheoriesareless DoPeerFirmsAffectCorporateFinancialPolicy?169andHendricks(1991)suggestthatfree-ridingininformationacquisitionorrelativeperformanceevaluationformanagersmayleadtoherdbehaviorincapitalstructurepolicies.Bothoftheseexplanationshavetheoreticalprecedentinthefinanceliterature.AsshownbyBanerjee(1992)andothers,whenafirm’sownsignalisnoisyandoptimizationiscostlyortime-consuming(Conlisk(1980)),managersmayrationallyputmoreweightonthedecisionsofothersthanontheirowninformation.Thisisespeciallylikelywhenotherfirmsintheindustryareperceivedashavinggreaterexpertise(Bikhchandani,Hirshleifer,andWelch(1998)).Indeed,DevenowandWelch(1996)notethatinformationalcascadesmayexplainthedecisionsofmanagerstoassumedebtbecause,withoutagoodmodelofwhyfirmsdoso,managersmayinferthebestchoicefrompeercompanies.Inaddition,managersneednotcompletelyignoretheirowninformation,asoccursinthelimitinsequentialinformationalcascademodels.Rather,itissufficientthattheyupdatetheirpriorsinaBayesianmannerbasedontheobservedactionsofotherfirms(e.g.,Romer(1993)andTrueman(1994)).Asaresult,theirdecisionwillbepulledtowardthoseoftheirpeers,relativetowhatitwouldbeiftheyreliedsolelyontheirowninformation.18Managersmayalsomimicotherfirms’policiestoinfluencetheirperceivedrelativequalityinthelabormarket.InthemodelofScharfsteinandStein(1990),higherqualityinvestmentmanagersreceivecorrelatedsignalsaboutinvestmentopportunities,whilelowerqualitymanagersreceiveindependentsignals.Managersthereforemimictheinvestmentchoiceofothersinordertoincreasetheirperceivedtype.Inthisenvironment,herdingismoreimportantthanmakingefficientinvestmentchoicesbecauseblameissharedintheeventofabadoutcome.Zwiebel(1995)showsthatcorporatemanagers’typesareinferredfromtheirrelativeperformance.Becausemanagersperceivedtobebelowacutofftypearefired,theyprefertomimictheinvestmentchoicesofotherstominimizethevolatilityoftheirrelativeperformance.B.EmpiricalResultsToshedlightonthepotentialmechanismsbehindpeereffectsinfinancialpolicy,weexamineheterogeneityinthecoefficientonpeerfirmleverage,β,fromequation(1).Toavoidredundancy,wefocusourattentiononthechangeinmarketleverageastheoutcomevariableofinterest.Specifically,weinteractpeerfirmreturnshockswithindicatorvariablesidentifyingthelowerandrelevantinthecurrentsetting.Rather,theunderlyingmechanismbehindanyherd-likebehavioramongcorporatemanagersismorelikelyduetoinformationorincentivedistortions,orlimitedcognitiveabilitiesofmanagers.18Becauseoursourceofidentifyingvariationisfirm-specific,onemustacknowledgeanaddi-tionalassumptionforalearningmechanismtobebehindourresults.Specifically,onemustassumethatmanagerscannotdisentanglethevariationinpeerfirms’actionsthatcomefromcommonandidiosyncraticvariationinpeerfirms’stockreturns.Iftheycould,thentheywouldrationallyre-spondonlytothevariationthatcontainsinformationabouttheirownfirm.Webelievethatitisunlikelythatnonfinancialcorporatemanagersareperformingsuchadecomposition. 170TheJournalofFinanceRupperthirdsofeachinteractionvariable’sdistribution.Forbinaryvariables,theinteractionisdirectlywiththebinaryvariable.Ourinferencescomefromanydifferencesintheestimatedscaled(bystandarddeviation)coefficientsacrosstheseareasofthedistribution.InTableIX,weexaminewhethersomefirmswithintheindustryaremoreorlesssensitivetotheirpeers’financialpolicies.Foreachindustry-yearcom-bination,werankfirmsintothreegroupsbasedonfirm-specificcharacteristicsandfocusonthelowandhighthirdsofthedistributionofcontinuousinter-actionvariables.Theresultsshowthatsmaller(marketshare),nondividendpaying,unratedfirmsaremoresensitivetotheirpeersthanaretheircounter-parts.Similarly,firmsdefinedasmorefinanciallyconstrainedaccordingtotheWhited–Wu(2006)indexaremoresensitivetopeerfirms.Theseresultssug-gestthatmimickingbehaviorisstrongestamongthosefirmswiththegreatestlearningmotiveandperhapsthegreatestneedtobuildreputation.InTableX,weexaminemoredirectlywhetherpeerfirmrelevanceisdrivenbyaleader–followermodelinwhichlesssuccessfulfirmsaresensitivetomoresuccessfulfirmsbutnotviceversa.Todoso,wecategorizefirmswithineachindustry-yearintotwogroupsthatwecallleadersandfollowers.Wedefinethesetwogroupsbysortingfirmswithineachindustry-yearintothreegroupsbasedonvariousmeasuresofsuccess—profitability,marketshare,andearn-ingsgrowth.Followersarethosefirmsinthebottomtwo-thirdsandleadersarethosefirmsinthetopthirdofthedistribution.InPanelAofTableX,weexcludethetopthirdofthedistribution(i.e.,theleaders)fromthesample.Wethenestimateviatwo-stageleastsquaresequa-tion(1)onthissubsampleusingthepeerfirmleverageoftheleaderfirmsinplaceofthefollowerfirms.Inessence,weareestimatingtheextenttowhichfollowerfirmsaresensitivetothefinancialpoliciesofleaderfirms.Thedepen-dentvariableisthechangeinmarketleverage,thoughresultsarequalitativelysimilarifweuselevels.Theresultsshowthatthefinancialpoliciesofsmaller(marketshare),lessprofitablefirmswithlowearningsgrowtharesensitivetotheleveragechangesoftheirmoresuccessfulcounterparts.Theresultsprovideausefulinterpreta-tionofthefindingsinTableIX,whichsuggestthatfinanciallyconstrainedfirmsaremoresensitivetopeersthanunconstrainedfirms.Thisfindingmaybeoddsinceonewouldexpectmimickingtobemorecostlyforfinanciallyconstrainedfirms,giventheirhighercostofexternalfinancing.However,theresultsheresuggestthatthiscostmaybeswampedbytheperceivedbenefitassociatedwithmimickingthebehaviorofleaderfirms.Asarobustnesscheckonthesefindings,weperformafalsificationtestbyrerunningtheanalysisusingthesampleofleadersandthepeerfirmleveragechangeofthefollowers.Thisanalysisaskswhetherleadersaresensitivetofollowerreturnshocks.TheresultsarereportedinTableX,PanelBandshownosignificantresults,statisticallyoreconomically.Inotherwords,leaderfirms’financialpoliciesappearinsensitivetothereturnshocksoffollowerfirms.Whileinsightful,wenotethattheseresultsdonotrejectaparticulartheoryperse.Theevidenceisconsistentwiththebroadimplicationsof DoPeerFirmsAffectCorporateFinancialPolicy?171tytedetherCnstrd)=-stat).IndustriesFHigh)(2=High)(2=Large)(21,marketshare,profitability,market-to-bookratio,andtheWhited–Wu=−tTableIX)statistictestingforweakinstruments(First-StageMultivariateWhichFirmsMimic?1993Yes)(2=Yes)(2=YesYesYesCreditRating(20.073**Dividend(5.332)Payer0.027**(3.387)0.061**Market(5.312)0.046**Share(4.933)0.050**(3.811)0.041**Profitability(2.861)0.056**(4.808)to-Book0.055**Market-(4.047)0.054**37.987**38.149**19.528**23.618**-statstatisticalsignificanceimplyinglessthan15%or10%sizedistortionisdenotedby*and**,respectively.F××-statF)Indexoffinancialconstraints.Weexcludethemiddlethirdofthedistributionforeachoftheseregressions.ThecoefficientestimatesarescaledLeverageChangeGroup1LeverageChangeGroup2MultivariateCharacteristicsThesampleconsistsofallnonfinancial,nonutilityfirmsintheannualCompustatdatabasebetween1965and2008withnonmissingdataforallanalysisvariables(seeAppendixA).Thetablepresents2SLSestimatedcoefficientsforthepeerfirmaveragemarketleveragechangeinteractedwithindicatorvariablesidentifyingthelowerandupperthirdofthewithin-industry-yeardistributionoflaggedvaluesforfirm-specificmeasuresofwhthefirmhasacreditrating,whetherthefirmpaidadividendinyear(2006bythecorrespondingvariablestandarddeviation.Thedependentvariableisthechangeinmarketleverageratio.Allmodelsareestimatedbylinear2SLSwheretheendogenousvariablesarethepeerfirmaverageleverageratiochangesinteractedwithindicatorvariables,andtheinstrumentsaretheone-period-laggedpeerfirmaverageidiosyncraticcomponentofstockreturnsinteractedwiththesameindicatorvariables.Thetablealsopresentstheheteroskedasticity-correctedCragg–Donald(aredefinedbythree-digitSICcode.Allvariablesareinfirstdifferencesexcepttheinstrument.Allindependentvariables,includingtheinstrumenbutexcludingtheendogenousvariable,arelagged1yearrelativetothedependentvariableunlessotherwisespecified.Allteststatisticsarecompuusingstandarderrorsthatarerobusttowithin-firmcorrelationandheteroskedasticity.Statisticalsignificanceatthe5%and1%levelsisdenotedb*and**,respectively.PeerFirmAvgPeerFirmAvgFirst-StagePeerFirmAverageFirm-SpecificFactorsIndustryFixedEffectsYearFixedEffectsObs.YesNoYes 172TheJournalofFinanceRTableXWhichFirmsAreMimicked?Thesampleconsistsofallnonfinancial,nonutilityfirmsintheannualCompustatdatabasebetween1965and2008withnonmissingdataforallanalysisvariables(seeAppendixA).Thetablepresentsestimatedcoefficientsscaledbythecorrespondingvariable’sstandarddeviation,andt-statisticsrobusttoheteroskedasticityandwithin-firmdependenceinparentheses.Allmodelsareestimatedbylinear2SLSwheretheendogenousvariableistheindustryaverageleveragechangeandtheinstrumentistheone-period-laggedindustryaverageidiosyncraticcomponentofstockreturns.Industriesaredefinedbythree-digitSICcode.Allspecificationsincludeone-period-laggedpeerfirmaveragesandfirm-specificeffectsforthefollowingcharacteristics:firmsize,profitability,tangibility,andthemarket-to-bookratio.Firmsareclassifiedaseither“Leaders”or“Followers”basedontheirwithin-industry-yearrankingby:profitability,marketshare(salesasafractionofindustrysales),stockreturns,andearningsgrowth.Thetablerestrictsattentiontothesubsampleoffirmsinthemiddleandlowerthirdsofthewithin-industry-yeardistribution(i.e.,Followers)ofeachclassificationvariableandregressestheirchangeinmarketleverageratioontheaveragechangeinmarketleverageoffirmsintheupperthird(i.e.,Leaders),aswellasthecontrolvariablesindicatedtowardthebottomofthetable.F-statstatisticalsignificanceimplyinglessthan10%or15%sizedistortionisdenotedby*and**,respectively.ChangeinMarketLeverageMarketStockProfitabilityShareReturnPanelA:DoFollowerFirmsRespondtoLeaders?LeaderFirmAvgLeverageChange0.076**0.084**0.087**(5.055)(6.045)(3.830)First-StageUnivariateF-stat62.913**92.725**34.530**PeerFirmAverageCharacteristicsYesYesYesFirmSpecificFactorsYesYesYesIndustryEffectsNoNoNoYearFixedEffectsYesYesYesObs.51,58843,67751,590PanelB:DoLeaderFirmsRespondtoFollowers?FollowerFirmAvgLeverageChange−0.019−0.002−0.088(−0.890)(−0.042)(−1.711)First-StageUnivariateF-stat17.688**5.235*6.021*PeerFirmAverageCharacteristicsYesYesYesFirmSpecificFactorsYesYesYesIndustryEffectsNoNoNoYearFixedEffectsYesYesYesObs.53,55445,52953,552reputationalandlearningmodels.Itisalsoconsistentwithfinancetextbookssuggestingthat“[F]irmsinabusinesstendtofollowtheleader...Whenthisfirmchoosesafinancingmix,presumablybaseduponitsfundamentals,otherfirmsinthatsectorthenimitatetheleader,hopingtoimitateitssuc-cess”(Damodaran(2010,p.443)).Likewise,Ross,Westerfield,andJaffe(2010, DoPeerFirmsAffectCorporateFinancialPolicy?173p.549)notethat“Afterall,theexistingfirmsinanyindustryarethesurvivors.Thereforeweshouldpayatleastsomeattentiontotheirdecisions.”Wehopefutureresearchwillprovideadditional,andmorepowerful,evidenceontheprecisemechanismbehindthepeereffects.Alternatively,sharperpredictionsfromtheorymayleadtomorepowerfultests.VI.ConclusionsThisstudyhasshownthatfirmsdonotmakefinancingdecisionsinisola-tion.Rather,thefinancingdecisionsand,toalesserextent,thecharacteristicsofpeerfirmsareimportantdeterminantsofcorporatecapitalstructuresandfinancialpolicies.Interdependenciesamongdebtandequityissuancesdriveinterdependenciesamongleverageratios.Indeed,peerfirmbehaviorhasare-markablyrobustandlargeimpactoncorporatecapitalstructure,largerthananyotherobservabledeterminant,onaverage.Aninterestingimplicationofthesefindingsisthepresenceofexternalities,whichweshowcansignificantlyamplifyordampentheimpactofchangesincapitalstructuredeterminants.Whilesomewhatsuggestive,ourcross-sectionalevidencepointstolearningandreputationalconcernsaspotentialmo-tivesforthesepeereffects.Mimickingbehaviorisconcentratedamongsmaller,younger,lesssuccessful,andmorefinanciallyconstrainedfirms.Bycontrast,industryleadersarenotinfluencedbythefinancialpolicychoicesoftheirlesssuccessfulpeers.Ourhopeisthatthisstudyinspiresfutureworkonbetterunderstandingthemechanismsdrivingthestronginterdependenciesamongfinancialpolicies.Furthermore,anopenempiricalquestioniswhetherthismimickingbehaviorisoptimalinavalue-enhancingsense.Finally,wehopethatthefindingsofthisstudyshiftthedirectionofcapitalstructureresearchtowardsmodels,boththeoreticalandempirical,thatexplicitlyrecognizetheinteractionsamongfirms.Initialsubmission:July30,2010;Finalversionreceived:June14,2013Editor:CampbellHarveyAppendixA:VariableDefinitionsCorporateaccountingdatacomefromthemergedCenterforResearchinSe-curityPrices(CRSP)-CompustatdatabaseavailableontheWhartonResearchDataServicesserver.Wedrawasampleoffirm-yearobservationsfortheperiod1965to2008.Wechoose1965asthestartyeartomitigatetheselectionbiastowardlarge,successfulfirmsthatexistsintheearlypartoftheCompustatsample.Tomaintainconsistencywithpreviousempiricalstudiesandtoavoidcapitalstructuresdictatedbyregulatoryconsiderations,weexcludefinancialfirms(SICcodesbetween6000and6999)andutilities(SICcodesbetween4900and4999),aswellasgovernmententities(SICcodesgreaterthanorequalto 174TheJournalofFinanceR9000).19StockreturndataforoursampleoffirmscomefromtheCRSPmonthlystockpricedatabase.Toensureconsistencythroughoutourprimaryanalysis,werequireeachfirm-yearobservationtohavenonmissingdataforthelevelsandfirstdifferencesofthefollowingvariables:netequityissuances,netdebtissuances,bookleverage,marketleverage,sales,market-to-bookratio,profitability,tangibility,stockreturns,andtheidiosyncraticcomponentofstockreturns.Variabledefinitionsarebelow.CompustatvariablenamesaredenotedbytheirXpressfeedpneumonicinbold.Timeperiodsaredenotedby(t)or(t−1)suffixes.TotalBookAssets=at.TotalDebt=Short-TermDebt+Long-TermDebt=dltt+dlc.BookLeverage=TotalDebt/TotalBookAssets.MarketValueofAssets(MVA)=prcc_f*cshpri+dlc+dltt+pstkl−txditc.MarketLeverage=TotalDebt/MVA.NetDebtIssuances=[(dltt(t)+dlc(t))−(dltt(t−1)+dlc(t−1))]/at(t−1).DebtIssuanceIndicator=1ifNetDebtIssuances>1%;0otherwise.NetEquityIssuances=(sstk(t)−prstkc(t))/at(t−1).EquityIssuanceIndicator=1ifNetEquityIssuances>1%;0otherwise.FirmSize=Log(Sales)=Log(sale).Tangibility=NetPPE/Assets=ppent/at.Profitability=EBITDA/Assets=oibdp/at.Market-to-BookRatio=MVA/TotalBookAssets.CommonDividends=dvc.CommonDividendIndicator=1ifdvc>0;0otherwise.Sales,General,andAdministrativeExpenses=xsga/FirmSize.ResearchandDevelopmentExpenses=xrd/FirmSize.CapitalExpenditures=capx.CapitalInvestment=CapitalExpenditures(t)/NetPPE(t−1).Altman’s(1968)Z-Score=(3.3*pi+sale+1.4*re+1.2*(act−lct))/at.EarningsVolatilityiscomputedeachyearasthehistoricalstandarddevia-tionofEBITDA/Assets.Werequireatleast3yearsofnonmissingdata.MarginalTaxRatesareobtainedfromJohnGraham’swebsite.Weconstructbankfixedeffectsforeachfirmwithavailableissuancedatabyassumingthatthefirmusesthesamebankeachyearuntileithertheendofthesampleorwefindadifferentbankbeingused,regardlessofthesecuritybeingissued.Resultsobtainedassumingthatthefirmusedthesamebankinallyearspriortotheissuanceuntileitherthebeginningofoursampleoranewbankwasfoundaresimilar.19Weincludefirmsthatundertookasignificantacquisitionduringthesampleperiodasindi-catedbyCompustatvariableaftnt1equalto“AB.”However,allofourresultsareinsensitivetotheirexclusion,whichaffectslessthan3%ofthesampleobservations. DoPeerFirmsAffectCorporateFinancialPolicy?175WeuseThompson’sSDCandReutersLoanPricingCorporation’sDealscandatabasetoidentifyleadunderwritersandarrangersoragentsfor(publicandprivate)debtandequityissuances.Specifically,SDCprovidesunderwriterinformationforpublicdebtandequityofferings,aswellasRule144aofferings.WerelyonDealscantoidentifytheleadbank(orarranger)onsole-lenderandsyndicatedloans.WematchSDCtoCompustatbymatchingcusipsanddatesofissuanceinSDCtocusipsanddatesintheCompustathistoricalcompanyinformationfile.WematchDealscantoCompustatusingthelinkfilefromChavaandRoberts(2008).AppendixB:ExogenousVariableDerivativesToeasethepresentation,consideraparticularindustryjandyeart.Rewrit-ingourmodel,equation(1),inmatrixnotationproducesβ1y=Qy+Xλ+QXγ+Zδ+ε,(B1)N−1N−1wherey=(y,...,y)isavectorofoutcomesfortheNfirmsinanarbitrary1Nindustry-yearcombination,QisanN×Nmatrixwithzerosonthediagonalandoneseverywhereelse,XisanN×k1matrixofexogenousvariablesthatappearasbothfirm-specificfactorsandpeerfirmaveragesinourmodel(i.e.,sales,profitability,market-to-book,andtangibility),ZisanN×k2matrixofallotherexogenousvariables(e.g.,industryandyearfixedeffects),andεisanN×1vectorofresiduals.Solvingequation(B1)foryyields−1β1y=I−QXλ+QXγ+Zδ+ε.(B2)N−1N−1Ofinterestisthemarginaleffectorderivativeoftheoutcomeforfirmi=1,...,N,yi,withrespecttoachangeineachm=1,...,k1exogenousvari-ablesforallfirmsl=1,...,N,xlm.Toderiveaclosed-formsolutionforthesederivatives,weneedexpressionsforthetwoN×NmatricesmultiplyingX:−1−1ββ1I−QandI−QQ.N−1N−1N−1InductionandmatrixalgebrashowsthatthefirstmatrixissymmetricandN−1−β(N−2)hastwodistinctelements.Thediagonalelementsequal,andtheoff-(N−1+β)(1−β)βdiagonalelementsequal.Thesecondmatrixisalsosymmetricwith(N−1+β)(1−β)βtwodistinctelements.Thediagonalelementsequal,andtheoff-(N−1+β)(1−β)diagonalelementsequal1.Therefore,thederivativeofanarbitrary(N−1+β)(1−β)elementyiinthevectorywithrespecttoanarbitraryelementxlminthematrix 176TheJournalofFinanceRXisequalto⎧β2β∂yi⎨λm1+(N−1+β)(1−β)+γm(N−1+β)(1−β)fori=l=∂xlm⎩λβ+γ1fori=l,m(N−1+β)(1−β)m(N−1+β)(1−β)whereweusetheequalityN−1−β(N−2)β2=1+.(N−1+β)(1−β)(N−1+β)(1−β)REFERENCESAlmazan,Andres,andCarlosA.Molina,2005,Intra-industrycapitalstructuredispersion,JournalofEconomicsandManagementStrategy14,263–297.Altman,EdwardI.,1968,Financialratios,discriminantanalysisandthepredictionofcorporatebankruptcy,JournalofFinance23,589–609.Baker,Malcolm,andJeffreyWurgler,2002,Markettimingandcapitalstructure,JournalofFinance57,1–30.Banerjee,AbhijitV.,1992,Asimplemodelofherdbehavior,QuarterlyJournalofEconomics107,797–817.Bertrand,Joseph,1883,TheorieMathematiquedelaRichesseSociale,JournaldesSavants67,499–508.Bikhchandani,Sushil,DavidHirshleifer,andIvoWelch,1998,Learningfromthebehaviorofothers:Conformity,fadsandinformationalcascades,JournalofEconomicPerspectives12,151–170.Bizjak,JohnM.,MichaelL.Lemmon,andLalithaNaveen,2008,Hastheuseofpeergroupscontributedtohigherlevelsofexecutivecompensation?JournalofFinancialEconomics90,152–168.Bolton,Patrick,andDavidScharfstein,1990,Atheoryofpredationbasedonagencyproblemsinfinancialcontracting,AmericanEconomicReview80,93–106.Bradley,Michael,GreggA.Jarrell,andE.HanKim,1984,Ontheexistenceofanoptimalcapitalstructure:Theoryandevidence,TheJournalofFinance39,857–878.Brander,JamesA.,andTracyR.Lewis,1986,Oligopolyandfinancialstructure:Thelimitedliabilityeffect,AmericanEconomicReview76,956–970.Brav,Alon,WeiJiang,FrankPartnoy,andRandallThomas,2008,Hedgefundactivism,corporategovernance,andfirmperformance,JournalofFinance63,1729–1775.Case,AnneC.,andLawrenceF.Katz,1991,Thecompanyyoukeep:Theeffectsoffamilyandneighborhoodondisadvantagedyouths,NBERWorkingPaper3705.Chava,Sudheer,andMichaelR.Roberts,2008,Howdoesfinancingimpactinvestment?Theroleofdebtcovenants,JournalofFinance63,2085–2121.Chevalier,JudithA.,andDavidS.Scharfstein,1996,Capital-marketimperfectionsandcounter-cyclicalmarkups:Theoryandevidence,AmericanEconomicReview85,703–725.Cohen,Lauren,andAndreaFrazzini,2008,Economiclinksandpredictablereturns,JournalofFinance63,1977–2011.Conlisk,John,1980,Costlyoptimizationversuscheapimitators,JournalofEconomicBehaviorandOrganization1,275–293.Cournot,Augustin,1838,RecherchessurlesPrincipesMathematiquesdelaTheoriedesRichesses(Hachette,Paris).Cragg,JohnG.,andStephenG.Donald,1993,Testingidentifiabilityandspecificationininstru-mentalvariablemodels,EconometricTheory9,222–240.Damodaran,Aswath,2010,AppliedCorporateFinance,3rdedn.(JohnWiley&Sons,Hoboken,NJ). DoPeerFirmsAffectCorporateFinancialPolicy?177Devenow,Andrea,andIvoWelch,1996,Rationalherdinginfinancialeconomics,EuropeanEco-nomicReview40,603–615.Duflo,Esther,andEmmanuelSaez,2002,Participationandinvestmentdecisionsinaretirementplan:Theinfluenceofcolleagues’choices,JournalofPublicEconomics85,121–148.Duflo,Esther,andEmmanuelSaez,2003,Theroleofinformationandsocialinteractionsinretire-mentplandecisions:Evidencefromarandomizedexperiment,QuarterlyJournalofEconomics118,815–842.Flannery,Mark,andKasturiRangan,2006,Partialadjustmenttowardtargetcapitalstructures.JournalofFinancialEconomics79,469–506.Frank,MurrayZ.,andVidhanK.Goyal,2008,Tradeoffandpeckingordertheoriesofdebt,inEspenEckbo,ed.:TheHandbookofEmpiricalCorporateFinance(Elsevier,Amsterdam).Frank,MurrayZ.,andVidhanK.Goyal,2009,Capitalstructuredecisions:Whichfactorsarereliablyimportant?FinancialManagement38,1–37.Graham,JohnR.,2000,Howbigarethetaxbenefitsofdebt?JournalofFinance55,1901–1941.Graham,JohnR.,andCampbellHarvey,2001,Thepracticeofcorporatefinance:Evidencefromthefield,JournalofFinancialEconomics60,186–243.Hoberg,Gerald,andGordonPhillips,2009,Dynamicproduct-basedindustryclassificationsandendogenousproductdifferentiation,Workingpaper,UniversityofMaryland.John,Kose,andDalidaKadyrzhanova,2008,Relativegovernance,Workingpaper,UniversityofMaryland.Kayhana,Ayla,andSheridanTitman,2007,Firms’historiesandtheircapitalstructures,JournalofFinancialEconomics83,1–32.Lemmon,Michael,MichaelR.Roberts,andJaimeZender,2008,Backtothebeginning:Persistenceandthecross-sectionofcorporatecapitalstructures,JournalofFinance63,1575–1608.Lerner,Josh,andUlrikeMalmendier,2013,Withalittlehelpfrommy(random)friends:Successandfailureinpost-businessschoolentrepreneurship,ReviewofFinancialStudies26,2411–2457.Loughran,Tim,andJayRitter,1995,Thenewissuespuzzle,JournalofFinance50,23–51.MacKay,Peter,andGordonM.Phillips,2005,Howdoesindustryaffectfirmfinancialstructure?ReviewofFinancialStudies18,1433–1466.Maksimovic,Vojislav,andJosefZechner,1991,Debt,agencycosts,andindustryequilibrium,JournalofFinance46,1619–1643.Manning,Alan,2005,MonopsonyinMotion:ImperfectCompetitioninLabourMarkets(PrincetonUniversityPress,Princeton,NJ).Manski,Charles,1993,Identificationofendogenoussocialeffects:Thereflectionproblem,ReviewofEconomicStudies60:531–542.Marsh,Paul,1982,Thechoicebetweenequityanddebt:Anempiricalstudy,JournalofFinance37,121–144.Matvos,Gregor,andMichaelOstrovsky,2010,Heterogeneityandpeereffectsinmutualfundproxyvoting,JournalofFinancialEconomics98,90–112.Myers,StewartC.,1977,Determinantsofcorporateborrowing,JournalofFinancialEconomics5,147–175.Myers,StewartC.,1984,Thecapitalstructurepuzzle,JournalofFinance39,575–592.Myers,StewartC.,andNicholasS.Majluf,1984,Corporatefinancingandinvestmentdecisionswhenfirmshaveinformationthatinvestorsdonothave,JournalofFinancialEconomics13,187–221.Petersen,Mitchell,2009,Estimatingstandarderrorsinfinancepaneldatasets:Comparingap-proaches,ReviewofFinancialStudies22,435–480.Phillips,Gordon,1995,Increaseddebtandindustryproductmarkets:Anempiricalanalysis,Jour-nalofFinancialEconomics37,189–238.Rajan,RaghuramG.,andLuigiZingales,1995,Whatdoweknowaboutcapitalstructure:Someevidencefrominternationaldata,JournalofFinance50,1421–1460.Romer,David,1993,Rationalasset-pricemovementswithoutnews,AmericanEconomicReview83,1112–1130. 178TheJournalofFinanceRRoss,StephenA.,RandolphW.Westerfield,andJeffreyJaffe,2010,CorporateFinance,9thedn.(McGraw-HillIrwin,NewYork,NY).Scharfstein,DavidS.,andJeremyC.Stein,1990,Herdbehaviorandinvestment,TheAmericanEconomicReview,80,465–479.Scott,JamesH.,1976,Atheoryofoptimalcapitalstructure,BellJournalofEconomics7,33–54.Shue,Kelly,2013,Executivenetworks,andfirmpolicies:EvidencefromrandomassignmentofMBApeers,ReviewofFinancialStudies26,1401–1442.Stigler,George,1968,Priceandnon-pricecompetition,JournalofPoliticalEconomy76,149–154.Stock,James,andMotohiroYogo,2005,Asymptoticdistributionsofinstrumentalvariablesstatis-ticswithmanyweakinstruments,inD.W.K.AndrewsandJ.H.Stock,eds.:IdentificationandInferenceforEconometricModels:EssaysinHonorofThomasRothenberg(CambridgeUniversityPress,Cambridge).Strebulaev,Ilya,andBaozhongYang,2012,Themysteryofzero-leveragedfirms,Workingpaper,StanfordUniversity.Trueman,Brett,1994,Analystforecastsandherdingbehavior,ReviewofFinancialStudies7,97–124.Welch,Ivo,2004,Capitalstructureandstockreturns,JournalofPoliticalEconomy112,106–131.Whited,ToniM.,andGuojunWu,2006,Financialconstraintrisk,ReviewofFinancialStudies19,531–559.Zeckhauser,Richard,JayenduPatel,andDarryllHendricks,1991,Nonrationalactorsandfinan-cialmarketbehavior,TheoryandDecisions31,257–287.Zwiebel,Jeffrey,1995,Corporateconservatismandrelativecompensation,JournalofPoliticalEconomy103,1–25.