asmeasuredbynumberofpatents.Inthatcase,theadditionoflow-qualityinnovationprojectsmaygeneratetheresultsratherthanarepositioningofresearchtolower-impacttopics.TheanalysisinTableVIIIaddressesthisquestionbyexploringchangesinthescaleofinnovation.15
[INSERTTABLEVIII]
Theendogenousmodelincolumn(1)indicatesthatIPO?rmsproducesigni?cantly
Accepted ArticlemorepatentsperyearfollowingtheIPO?ling,witha37.75%increaserelativetothepre-IPOaverage.Column(2),however,indicatesthattheabovee?ectisinsigni?cantwhenthereduced-formspeci?cationisestimated.The2SLSestimateincolumn(3)showsthatthecoe?cientontheIPOvariableisinsigni?cantandthemagnitudedeclinesto28.17%.Infact,whenusingtheIVPoissonspeci?cationincolumn(4),thecoe?cientontheIPOvariableisclosetozeroandinsigni?cant.
A.3.
RobustnessChecksandInterpretation
InthissectionIsummarizeseveralsupplementalanalysesthattesttherobustnessand
interpretationoftheabove?ndings.IstartbyexploringwhetherthedeclineintheaverageinnovationqualityofIPO?rmscanbedrivenbyalowerpatentingthresholdaftertheIPO.Thismayleadtotheadditionoflow-qualitypatentsandhenceloweraveragepatentingquality.Todoso,Iexamine?rms’best(i.e.,most-cited)patent,whichisunlikelytobea?ectedbytheadditionoflow-qualitypatent?lings.I?ndthatthequalityofIPO?rms’bestpatentdeclinesaswell,withacomparablemagnitudeasthedeclineinaverageinnovationqualityreportedinTableVI.Thisevidence,whichisreportedintheInternetAppendix,suggeststhatgoingpublica?ectstheentirepatentdistributionratherthansimplydrivingaverageperformancedownbytheadditionoflow-qualityprojects.16
Next,cash-rich?rmsmaybeassociatedwithfewercitationsbecauseciting?rmsfacehigherlitigationrisk,whichwouldmechanicallygeneratetheresultsinthepaper.Totest
20
This article is protected by copyright. All rights reserved.
thisconcern,IfocusonpatentsapprovedbeforetheIPO?lingandexaminewhetherpatents’annualcitationratechangesafterthe?rmsgopublic(relativeto?rmsthatwithdrawthe?ling).InresultsreportedintheInternetAppendix,I?ndthatchangesinthecitationrateofexistinginventionscannotbeexplainedbythetransitiontopublicequitymarkets.IalsoexplorewhethertheresultsaredrivenbytheInternetbubbleyears.Asillustrated
Accepted ArticleinTableIV,theinstrumentstronglypredictsIPOcompletionevenwhenall?rmsthat?ledduringtheInternetbubbleandthereafterareexcluded.InrobustnesstestsIreestimatetheinnovationnoveltyregressionsafterexcludingall?rmsthat?letogopublicasof1999.Theresults,reportedintheInternetAppendix,remainsigni?cantandqualitativelyunchanged.Inthemainanalysis,Icollectinformationontheinnovationtrajectoryof?rmsinthe?ve
yearssubsequenttotheIPO?ling.Ininterpretingtheresults,itisnaturaltowonderhowtheendogenoustransitionof18%ofthewithdrawn?rmstopublicequitymarketsa?ectstheestimates.Toexaminethisquestion,Irepeattheinstrumentalvariablesanalysisusinganendogenousvariablethatequalsoneifa?rmgoespublicinthetwo(orthree)yearsaftertheIPO?ling,regardlessofwhetheritwithdrawsinthe?rstattempt.Thus,ifa?rmwithdrawsits?lingandreturnstotheIPOmarketinthe?rsttwo(three)yearsafterthe?ling,this?rmwouldbeconsideredpartofthetreatmentgroup(andnotthecontrolgroup).17As
reportedintheInternetAppendix,thesespeci?cationsgenerateestimatesthataresimilartothosereportedabove.
Finally,itisimportanttounderstandwhethertheestimatesaredrivenbychangesin
?rmsthatwithdrawtheirIPO?lingandremainprivateorby?rmsthatgopublic.Forexample,withdrawn?rmsmaychoosetoincreasetheirinnovationinordertosuccessfullygopublicinasecondattempt,ratherthanIPO?rmsexperienceadeclineininnovation.However,I?ndthatallIPO?lersexperienceadeclineinscaledinnovationmeasures,both?rmsthatexperienceaNASDAQdrop(andthusaremorelikelytoremainprivate)and?rmsthatdonotexperienceaNASDAQdrop.18However,thedeclineismoresubstantialamong?rmsthatendupgoingpublic.Similarly,changesintheacquisitionofexternalinnovation,
21
This article is protected by copyright. All rights reserved.
asdiscussedinSectionIII.C,aredrivenbythose?rmsthatgopublicratherthanthosethatremainprivate.19
Accepted ArticleB.InventorMobilityandProductivityChanges
AsubstantialportionofR&Dinvestmentisintheformofwagesforhighlyeducated
scientistsandengineers,whoencompassthe?rm’sknowledge.ItisthereforenaturaltoexplorewhethergoingpublicleadstoinventormobilityandproductivitychangesfollowingtheIPO.
B.1.InventorLevelData
Thepatentdatabaseprovidesaninterestingopportunitytotrackinventors’mobility
across?rms,aseachpatentapplicationincludesboththenameoftheinventoranditsassignee(mostoftentheinventor’semployer).Analysisofinventor-leveldata,however,ischallengingforseveralreasons.First,patentsareassociatedwithinventorsbasedontheirnameandgeographiclocation,butinventors’namesareunreliableas?rstnamescanbeabbreviatedanddi?erentinventorsmayhavesimilarorevenidenticalnames.Second,whileitispossibletoinferthataninventorchanged?rms(forexample,aninventorisassociatedwithapatentforcompanyAin1987andapatentforcompanyBin1989),theprecisedateoftherelocationisunavailable.Additionally,transitionsforwhichinventorsdonotproducepatentsinthenewlocationarenotobservable.Hence,thismethodidenti?esrelocationsofthemorecreativeinventorswhopatentfrequentlyandpresumablymatterthemost.Toovercomethehurdleofnamematching,IusetheHBSpatentingdatabase,whichin-
cludesuniqueinventoridenti?ers.Theuniqueidenti?ersarebasedonre?neddisambiguationalgorithmsthatseparatesimilarinventorsbasedonvariouscharacteristics(Lai,D’Amour,andFleming(2009)).Whenpatentapplicationsincludemultipleinventors,Iattributeapatentequallytoeachinventor.Overall,Iobtaininformationonapproximately36,000in-ventorsinmysample.Irestricttheanalysistoinventorsthatproduceatleastasinglepatent
22
This article is protected by copyright. All rights reserved.
bothbeforeandaftertheIPO?lingandexaminethepatentingbehaviorofinventorsinthethreeyearsbeforeand?veyearsaftertheIPO?ling.Iidentifythreeinventortypes:1.Stayer–AninventorwithatleastasinglepatentbeforeandaftertheIPO?lingatthesamesample?rm.
2.Leaver–Aninventorwithatleastasinglepatentatasample?rmbeforetheIPO?ling,andatleastasinglepatentinadi?erentcompanyaftertheIPO?ling.20
3.Newcomer–AninventorwithatleastasinglepatentaftertheIPO?lingatasample?rm,butnopatentsbefore,andatleastasinglepatentatadi?erent?rmbeforetheIPO?ling.
Ofthe36,000inventorsinmysample,Icanclassify16,108inventorsintotheabove
Accepted Articlecategories.21Theseinventorsaccountforapproximately65%ofthesamplepatents.
B.2.
InventorLevelAnalysis
Iexplorechangesininventor-levelactivityusingtheinstrumentalvariablesapproach
introducedinSectionII.A.Istartbyinvestigatingchangesintheinnovationqualityofstayers.Next,Iexamineinventormobilitybystudyinginventors’likelihoodofleaving,spinningo?acompany,orjoiningthe?rmfollowingtheIPO?ling.
TheresultsarereportedinTableIX,wheretheunitofobservationisatthelevelofthe
inventor.Incolumn(1),Iexplorechangesinstayers’productivity.Ifocusonthesetofinventorsthatremainatthe?rm,andthedependentvariableistheaveragescaledcitationsproducedbyinventorsinthe?veyearsaftertheIPO?ling.Icontrolfortheinventor’spre-IPO?lingcitationsperpatent,aswellas?lingyearandindustry?xede?ectsandtheotherstandardcontrols.Standarderrorsareclusteredatthe?rmlevel,toallowforcorrelationsbetweeninventorsinthesame?rm.Iestimatethe2SLSincolumn(1),and?ndthattheIPOcoe?cientequals-1.094,whichisstatisticallysigni?cantatthe1%level.Themagnitudeislarge,correspondingtoa48Tclineininventors’innovationnoveltyinIPO?rmsrelativeto
23
This article is protected by copyright. All rights reserved.
