La lecture en ligne est gratuite
Le téléchargement nécessite un accès à la bibliothèque YouScribe
Tout savoir sur nos offres
Télécharger Lire

Conflicts of interest [Elektronische Ressource] : articles on common benefits and private profits / Johannes Koenen

167 pages
Conflicts of InterestArticles on Common Benefitsand Private ProfitsInauguraldissertation zur Erlangung des akademischen Gradeseines Doktors der Wirtschaftswissenschaften der UniversitätMannheimJohannes Koenenvorgelegt im Frühjahrs- und Sommersemester 2010Abteilungssprecher: Professor Tom Krebs, Ph.DReferent: Konrad Stahl, Ph.D.Korreferent: Professor Dr. Martin PeitzTag der mündlichen Prüfung: 23. Juni 2010To my one great discoveryiiiAcknowledgementsMore people have supported me in the course of the past years than could beincluded in this whole volume, therefore I can only thank a selection of themhere.I was tremendously fortunate in the academic environment I was allowedto work in. My supervisor Konrad Stahl, though I prefer the German termDoktorvater, is an inspiring role-model as an academic with his indefatigableintellectual curiosity and his passion for Science and the Arts. I have greatlyenjoyed our discussions and collaboration and will continue to do so. I wouldliketothankMartinPeitzforhisever-astuteanswerstoquestionsinmanyfieldsand for his momentous support in important choices. And I am very gratefulto Elu von Thadden for his invaluable academic input and his dedication to thequality of our graduate program. I am extremely happy and proud to have hadthe opportunity to work together withLeonardo Felli fromwhom I havelearnedmuch about how to approach problems as an economist.
Voir plus Voir moins

InterestofConflicts

BenefitsCommononticlesAr

andPrivateProfits

InauguraldissertationzurErlangungdesakademischenGrades

Doktorseines

orgelegtv

derWirtschaftswissenschaftenderUniversität

im

Mannheim

enenoKJohannes

rühjahrs-F

und

Sommersemester

2010

her:Abteilungssprec

t:Referen

t:erenorrefK

agT

der

henündlicm

Prüfung:

Professor

Professor

Professor

23.

Juni

omT

Krebs,

Ph.D

Ph.D.Stahl,onradK

Dr.

2010

Martin

eitzP

oT

my

one

great

overydisc

iii

Acknowledgements

Morepeoplehavesupportedmeinthecourseofthepastyearsthancouldbe
includedinthiswholevolume,thereforeIcanonlythankaselectionofthem
here.IwastremendouslyfortunateintheacademicenvironmentIwasallowed
toworkin.MysupervisorKonradStahl,thoughIprefertheGermanterm
Doktorvater,isaninspiringrole-modelasanacademicwithhisindefatigable
intellectualcuriosityandhispassionforScienceandtheArts.Ihavegreatly
enjoyedourdiscussionsandcollaborationandwillcontinuetodoso.Iwould
liketothankMartinPeitzforhisever-astuteanswerstoquestionsinmanyfields
andforhismomentoussupportinimportantchoices.AndIamverygrateful
toEluvonThaddenforhisinvaluableacademicinputandhisdedicationtothe
qualityofourgraduateprogram.Iamextremelyhappyandproudtohavehad
theopportunitytoworktogetherwithLeonardoFellifromwhomIhavelearned
muchabouthowtoapproachproblemsasaneconomist.Iwouldfurtherliketo
thankAnnamariaLusardiforgivingmethewonderfulopportunityofworking
inthebeautifulandinspiringsurroundingsofDartmouthCollegeinFall2009.
Ihavegreatlybenefittedfrommycolleagues,bothpresentandpast:My
graduateschoolclassmateshavebeenfantasticcompanionsandourweekly
lunchhasbeenoneofmyfavoriterituals.FrankRosar,myiCERNco-founder,
wasalwayswillingtoassistmeinseminars,aswellasgame-andauction-
theoreticmatters.Finally,Itrulyenjoyedworkingtogetheranddiscussing
problemswithIsabelRuhmer,HeikoKarle,TobiasKlein,MichalKowalik,
ChristianLambertz,PaoloMassela,SteffenReik,DongsooShin,HannesUllrich
andFrankWachtler–Ihopewewillhavemanyopportunitiestocollaborate
inthefuture.IwouldalsoliketothankChristinaandKayWilkefortheir
assistanceinthelatestagesofmydissertation.FurtherIamverygrateful
forthehelpoftwooutstandingresearchassistants,PascalBuschandAndré
Stenzel.IgratefullyacknowledgefinancialsupportoftheUniversityMannheim,the
DeutscheForschungsgemeinschaftDFGandtheGermanIsraeliFoundationfor

iv

ScientificResearchandDevelopment.Partofthisworkwascarriedoutwhile

enjoyingthewonderfulhospitalityoftheStudyCenterGerzenseeforwhichI

wouldliketothanktheSwissNationalBank.Further,Iamindebtedtothe

GermanAutomobileIndustryAssociation(VDA)fortheircooperation.The

viewsexpressedinthisdissertationaresolelythoseoftheauthorandnotof

theseorganizations.

Finally,yetforemost,Ihavetothankmyfamily:MywonderfulwifeTabea,

whowithhernever-endingsupport,brillianceandjoyhasturnedthisendeavor

intosomethingfulfillingandfun.Myparents,whowiththeirgenerosity,pa-

tienceandlovehavebeentruepillarsthroughoutmylife,andonwhoseeco-

nomicintuitionIwasalwaysabletorely,aslongastherewereneithernumbers

norformulasinvolved.Andfinallymysiblings,whowerenevershortofen-

couragement,andwhoonsomelevelwerehappythatoneofuswentforhis

PhD–andonalllevelswerehappyitwasnotthem.

v

tstenCon1.GeneralIntroduction1
1.1.PrivateProfitsandPublicBenefits-HownottoReformthe
PatentSystem...........................3
1.2.Individual(Ir)rationality?BehaviorinanEmergingOnlineSo-
cialNetwork............................4
1.3.TrustandInvestment-ATheoreticalandEmpiricalAssessment5
2.PrivateProfitsandPublicBenefits-HownottoReformthe
PatentSystem7
2.1.Introduction............................7
2.1.1.Motivation.........................7
2.1.2.RelatedLiteratureandContribution...........10
2.2.ComparisonoftheUSandtheEuropeanPatentSystem-Styl-
izedFacts.............................14
2.2.1.PatentLegislationandExaminationthroughPatentOffices14
2.2.2.CourtsandPatentTrials.................16
2.3.ModelSetting...........................17
2.3.1.Therulesofthepatentsystem..............17
2.3.2.StrategicPlayersand“Ideas”...............17
2.3.3.PatentandTrademarkOffice...............18
2.3.4.TheCourtSystem.....................21
2.4.Analysis..............................25
2.4.1.Equilibria.........................25
2.4.2.ComparativeStatics-PotentialApproachestoReform.39
2.5.DiscussionandOutlook......................45
vi

tsentCon3.Individual(ir)rationality?Behaviorinanemergingsocial
48orkwonline-net3.1.Introduction............................48
3.1.1.Motivation.........................48
3.1.2.RelatedLiteratureandContribution...........51
3.2.TheSocialOnline-NetworkunderScrutiny...........56
3.2.1.FunctionandFeaturesoftheNetwork..........56
3.2.2.NetworkSizeandDevelopment..............58
3.3.AnalyticalFramework.......................60
3.3.1.CreationandSeveranceofIndividualLinks.......61
3.3.2.ProvisionofClubGoodsinNetwork...........63
3.4.EmpiricalAnalysis........................67
3.4.1.DataDescription.....................67
3.4.2.StrategicLinkingBehaviorandNegativeReciprocity..70
3.4.3.FreeRidingandPositiveReciprocity...........75
3.5.ConclusionandOutlook......................78
4.TrustandInvestment-ATheoreticalandEmpiricalAssess-
80tmen4.1.Introduction............................80
4.1.1.Motivation.........................80
4.1.2.RelatedLiteratureandContribution...........82
4.2.AnalyticalFramework.......................87
4.2.1.StructureandTimingoftheProcurementGame....87
4.2.2.Results...........................89
4.2.3.ASimpleNotionofTrust.................97
4.3.EmpiricalAnalysis........................101
4.3.1.SourceofData.......................101
4.3.2.DescriptiveStatistics...................103
4.3.3.MeasuresofTrust:WhoTrustsWhom-andWhy?...110
4.3.4.MistrustandUnderinvestment..............119
vii

tenConts

4.3.5.TrustandSourcingDecisions

4.4.ConclusionandOutlook.......

endixAppA.

endixAppB.

Results

yBibliograph

viii

to

to

reChapt

Chapter

Regression-3

4:

eDescriptiv

......

......

Results

Statistics

...

...

and

......122

......125

127

Regression

331

144

List

2.1.2.2.2.3.2.4.2.5.2.6.2.7.

3.1.3.2.

4.1.4.2.

Figuresof

Timelineoftheentiremodel...................
GameTreefortheChallengeLitigationSubgame........
GameTreefortheInfringementLitigationSubgame......
+Expectedprofitsofpatenteeswithvalidpatents(π),invalid
A−patents(π)andthecompetitor(π),dependingontheprior
BAprobabilityofpatent(in)validity..................
Changesinthepayoffstructureofthelitigationgamegivenan
increaseintheprivatevalueofthepatentυ...........
IllustrationofLemma3depictinganexampleinwhichall4cases
occur................................
Changesinthepayoffstructureofthelitigationgamegivena
decreaseinthecostsofgoingtocourtκ.............

DevelopmentofAct30Usersovertime..............
Greetingscreenwithinformationoffriends’activity.......

OEM’sexpectedprofitsdependingonn.............
OEM’sexpectedprofitsdependingonnfordifferentvaluesofλ

17262833353734

6066

69100

ix

ablesTofList

x

2.1.Equilibriumoutcomesandexpectedprofitsofthelitigationsub-
games...............................30

3.1.Descriptivestatisticsforvariousactivitymeasures.......59
3.2.Theaveragecharacteristicsofvariousgroupsofusers:.....71
3.3.Meancharacteristicsofusersinvolvedinseveredfriendships:..73

4.1.Investmentdecisions,expectedsurplusandexpectedprofitsof
theOEMint=0.........................95
4.2.ImportanceofInnovationandCostshareR&Dbyproducttype.105
4.3.(OLS):Differenceinfrequencyofrevealingoriginalresearchdata114
4.4.FactorAnalysisResults......................116
4.5.DeterminantsofTrustmeasures..................117

A.1.FreeRidingandNegativeReciprocitya):Logisticregressionson
theprobabilitythatagivenfriendshipissevered.........127
A.2.FreeRidingandNegativeReciprocityb):Logisticregressionson
theprobabilitythatagivenfriendshipissevered.........128
A.3.FreeRidingandPositiveReciprocityforInformationProvi-
sion,FE-regressionscontrollingforpotentialuser-clusterhet-
eroskedasticity,entiresample...................129
A.4.FreeRidingandPositiveReciprocityforInformationProvision,
activeusersub-sample......................130
A.5.FreeRidingandPositiveReciprocityforInformationOrganiza-
tion,entiresample.........................131

ablesTofList

A.6.FreeRidingandPositiveReciprocityforInformationOrganiza-
tion,activeusersub-sample....................132

B.1.RelationshipCharacteristics:Pre-Development(Suppliers’
view)................................133
B.2.RelationshipCharacteristics:Development(Suppliers’view).134
B.3.RelationshipCharacteristics:SeriesProduction(Suppliers’
view)................................134
B.4.ProcurementDecisions:Pre-Development(Suppliers’view).134
B.5.ProcurementDecisions:Development(Suppliers’view)....135
B.6.ProcurementDecisions:SeriesProduction(Suppliers’view).135
B.7.PairwiseCorrelationsofTrustMeasures.............136
B.8.OLS-regressionresults(Model1)forfrequencyofqualityproblems137
B.9.Probit-regressionresults(Model2),probabilityofnotobserving
qualityproblems..........................138
B.10.OLS-regressionresults(Model1)forfrequencyofproduct-
relatedrecalls...........................139
B.11.OLS-regressionresultsfornumberofsupplierspre-development140
B.12.OLS-regressionresultsfornumberofsuppliersdevelopment.141
B.13.OLS-regressionresultsfornumberofsuppliersseriesproduction142
B.14.OLS-regressionresults:Relationshipofdifferentrent-extraction
devicesandnumberofsuppliersduring(1)predevelopment,
(2)developmentand(3)seriesproduction.........143

xi

ductiontroInGeneral1.

Whydoachievementsdiffersowidelyfromaspirations?[...]Sociallifeisnot
onlyatrialofstrengthbetweenopposinggroups:itisactionwithinamoreor
lessresilientorbrittleframeworkofinstitutionsandtraditions,anditcreates
–apartfromanyconsciouscounter-action–manyunforeseenreactionsin
thisframework,someofthemperhapsevenunforeseeable.Totrytoanalyse
thesereactionsandtoforeseethemasfaraspossibleis,Ibelieve,themain
taskofthesocialsciences.-KarlPopper,TheOpenSocietyandItsEnemies.

Inthecourseoftheresearchformydissertation,Ihadthelucktoencounter
threefascinatingexamplesforwhichtheintroductoryquotemightaswellhave
beenspecificallytailored.Socialandeconomicbehaviortakesplacewithin
institutionsandisstronglyaffectedbysocialnormsandpastchoices.The
resultingcomplexityofissuesoftenappearsdaunting.Takeforexamplethe
currentlyexistingUSpatent-regime.Sinceitsinception,theUnitedStates
ConstitutionprovidesthatPatentLawwascreatedinorderto“promotethe
ProgressofScienceandusefulArts,bysecuringforlimitedTimestoAuthors
andInventorstheexclusiveRighttotheirrespectiveWritingsandDiscoveries”.
Twocenturieshavepassed,andwhiletheFoundingFathersmostcertainlyhad
tangiblematterslikenewtoolsorpracticalmethodstobeusedinthenascent
stateinmind,inthemeantimethesamelawmustgovernthetreatmentof
genomesamplingorthesourcecodeforcomputerprogramswhilecompeting
withthepatentsystemsofEurope,Japan,Chinaandmanyothers.This
adaptationwouldnothavebeenpossible–andsomestateforcefullythatit
hasfailedmiserably–withoutconstanttweakstothesystemandcontinuous
marginalreformprocesses.Itistheroleofeconomiststoaccompanysuch

1

1.GeneralIntroduction

processesofchangeandtousethetoolsatourdisposaltobetterunderstand
theunderlyingforcesandtoforeseetheireffectsasfaraspossible–suchas,
forexample,thecurrentroleofpatentexaminersandjudgesfortheinnovation
behaviorandcreativeeffortsinsideacountry.

Thisdissertationisconcernedwiththreedifferenttopics,eachofwhichis
treatedinaseparate,self-containedarticle.Theirunderlyingconnectionis
containedinthequoteabove,astheyapplythemethodsofsocialscience–mi-
croeconomictheoryandmicro-econometrics–toquestionsrelatedto“grown”,
orinonecase“growing”institutions.InChapter2,Iaddresstroublesthat
patentsystemsarefacingtoalargeandincreasingdegree,suchasthedeval-
uationofpatentsduetobadlyspecifiedclaimsandpropertyrightsandthe
ensuingcentralrolethatcourtsplayinthepatentsysteminthecontextofa
game-theoreticmodel.Whilethepatentsystemcanbeconsideredavenerable
institution,inChapter3Iturnmyattentiontoayoungandbuddingone:
Onlinesocialnetworks.Makinguseofauniquedataset,incollaborationwith
SteffenReikItrytoexplorewhethertheexistingeconomictheoryonsocial
networkscanassistusinunderstandingtheforcesthatshapeuserbehavior
within.Oneofthecentralfindingsistheimportanceofsocialnormsinthisen-
vironment.InChapter4,whichwasdevelopedworkingtogetherwithLeonardo
FelliandKonradStahl,theaimistoshowhowaspecificsocialnorm,trust,
shapestheinteractionbetweenupstreamsuppliersanddownstreammanufac-
turersintheautomotiveindustry.Ipresentagame-theoreticmodelandtest
itspredictionsusingdatafromanextensiveindustrysurveyweconducted.

Inthefollowing,Ibrieflyintroducethethreerespectivearticleswhichcom-
posetheremainderofthisdissertation.TheAppendixincludesmostofthe
tablesreferredtointhetextaswellasthecompletebibliography.

2

1.1.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

1.1.PrivateProfitsandPublicBenefits-How
nottoReformthePatentSystem

TheneedforareformoftheUSpatentsystemhasbeenwidelyexpressed,
bothbyscientistsandpractitioners.Therearewidespreadcomplaintsabout
thequalityofgrantedpatents,especiallythelackof“clearlydrawnproperty
lines”,totheextentthatsomerefertothepatentsystemas“failed”.Atthe
sametime,duetotheinvestmentsmadeintoexistingpatentsandtheresulting
rightsofpatentees,afundamentalreformofthepatentsystemisnotfeasible
intheshort-ormediumterm.Fortheforeseeablefuture,therefore,thesystem
willhavetoplodonwithrelativelyminortweaks;butduetoitscomplexity,
evenapparentlyminortweakscanhavequiteseriousconsequences.

Inordertobetterunderstandthetradeoffsinvolvedinthecontemplated
“improvements”,Ipresentamodelthatincludesthethreepillarsofcurrent
patentsystems:legislation,patentandtrademarkoffices(PTOs)andcourts.
Legislatorsdecidetheobjectivecriteriaconcerningthematterthatcanbe
patented,suchasnoveltyorutility.ThePTOinspectspatentapplicationsin
compliancewiththeseregulations.Thecourts,whoseimportancehasincreased
substantiallyinthepasttwodecadesintheareaofpatenting,thendecide
whetherclaimsgrantedinexistingpatentsaretrulyenforceable,orwhether
thepatentshouldperhapshavenotbeengrantedinitially.

IncontrasttotheexistingliteratureonpatentlitigationstartingwithMeurer
(1989),Iintroducetwodifferenttypesoflawsuits:Bothclaimsfordamagesby
thepatenteeagainsttheallegedinfringeraswellaschallengestothevalidityofa
patentbycompetitors.Ishowthatitdependsonthecompetitivesettingaswell
astheexpectedqualityofapatentwhetherlitigationcanhaveacomplementary
functiontotheexaminationeffortsofthePTO.Againstthisbackground,Ithen
showthatsomeofthecontemplated(ordemanded)reformsteps,especially
increasingthefeesforpatenting,mayactuallydecreasetheaveragequalityof
patentapplicationsinrelevantcases.

3

1.GeneralIntroduction

1.2.Individual(Ir)rationality?Behaviorinan
EmergingOnlineSocialNetwork

Thelastdecadehasseenimportantadvancesinthetheoreticalliteratureon
theeconomicsofsocialnetworks,spearheadedbytheworkofJacksonand
Wolinsky(1996).Thisarticle,anditssuccessors,depictsocialnetworksusing
agraph-theoreticenvironmentinwhichindividualsaredepictedasnodesand
theirlinks–forexamplebusinesspartnershipsorfriendships–areresembledby
thearcsofthegraph.Theexistingtheorycomestoverypreciseandrelatively
homogenouspredictionsconcerningstructuresofnetworksthatarestable,in
thesensethattheyaretheresultsofequilibriumbehavior.Butmorerecently,
experimentaleconomistshavestartedtoexaminethequalityofpredictions
derivedfromthetheoryinhighlycontrolledsurroundingsandtheresultshave
beenmixedatbest,sothatdoubtshavebeenraisedwhetherindividualstruly
behaveinarationalmannerinsocial-networksettings.

Inthischapter,Itakeanempiricallookatindividualbehaviorwithinanactu-
allyexistingsocialonline-networkwithadditionalutilitygeneratingfunctions.
Tobespecific,usersareabletouploadmusictoonline-libraries,fromwhich
boththeyandtheirfirst-degree“friends”canaccessthemusicwheneverthey
areonline.Standardeconomictheoryprovestobeveryhelpfulinpredicting
withwhomindividualusersaregoingtoformnewfriendships,asthisbehavior
iscompatiblewithutilitymaximization.Butwealsofindbehaviorthatpoints
inthedirectionofsocialnormsthataremutuallyenforced:Usersdeclineto
enterintoorseverpurelybeneficiallinkswhenthereisindicationoffreeriding,
forexample.Further,weshowthatinthefaceofpublic-goodlikeprovision
andorganizationofmusic,usersdoslightlyundersupplytheseformsofeffort
forhigherlevelsofotherusers’efforts,asstandardpublic-goodorclub-good
theorywouldpredict.Butusersimmediatelyreacttootherusers’additional
provisionofpublicgoodsinapositivelyreciprocalmannertoadegreethat
morethancompensatesforthefirsteffect.

4

1.3.TrustandInvestment-ATheoreticalandEmpiricalAssessment

1.3.TrustandInvestment-ATheoreticaland
AssessmenEmpiricalt

Inthefourthchapter,Ipresentasimplemodelofcontractualstructuresin-
volvingupstreamsuppliersanddownstreamproducers.Theupstreamsup-
pliersexertefforttodeterminethejointsurplusfromcooperatingwiththe
downstreamfirm.Sincethelatterhastheentirebargainingpower,thisisa
classicalholdupsituation.Definingtrustastheexpectationofthesupplier
thatthedownstreamfirmwillhonorhispropertyrights,itisstraightforward
toshowthathigherlevelsoftrustwillalleviatethetypicalunderinvestment
problem.Whatismoresurprisingistherelationshipbetweentrustandcom-
petitionamongsuppliers,whichthedownstreamfirmisabletoinduce.As
opposedtotheliteratureonrelationalcontracting,whichgenerallyfindsthat
arms-length(market)interactionsandfunctioninginformalrelationshipsareto
agreatextentmutuallyexclusive,inthissettingmoreintensecompetitioncan
beassociatedwithhigherlevelsoftrust.Theintuitionbehindthisresultisthat
competitionisonemechanismtoextractapartofthesupplier’ssurplusthat
isutilizedinsteadofexploitingthehold-upsituation.
Wethentakethesepredictionstothedata,auniquedatasetcollectedduring
atwoyearsurveyoftheGermanautomotiveindustry.Eachofourobserva-
tionsisarelationshipbetweenasupplierandacarmanufacturerwithregard
toonepartthatisprocured.Thisallowsustofocusontherelationshipspecific
natureoftrust,whileexistingempiricalandexperimentalstudiesmostlytreat
trustasacharacteristicofindividuals.Weareabletostudytheeffectsand
determinantsoftrustintheserelationshipsandshowthatthesuppliers’trust,
capturedbydifferentmeasureswithindividualconnotations,canbedamaged
byrent-extractingbehaviorofthedownstreamfirm.Ontheotherhand,aspre-
dictedbyourtheoreticalmodel,trustisnotnegativelyassociatedwithstronger
competitionbetweensuppliersinducedbythecarmanufacturer.Emphasizing
theimportanceoftrustinverticalrelationships,wethendemonstratethattrust
canmitigatetheunderinvestmentissueresultingfromthehold-upsituation:

5

1.

In

ductionotrInGeneral

particular,lowerlevels

withciatedassohigher

xypro

6

for

t.estmenvin

of

trust

failure

of

rates

the

of

pliersup

the

in

curedpro

the

car

part,

ufacturerman

hwhic

we

use

are

as

a

2.PrivateProfitsandPublicBenefits-
HownottoReformthePatent
System

ductiontroIn2.1.

ationMotiv2.1.1.

Apatentisnotapropertyright.Patenteescannotdirectlyexcludeothersfrom
marketsthatarecoveredbytheclaimsoftheirpatent(s).Instead,apatent
onlyconveystotheholdertherighttogotocourtagainstallegedinfringers.In
recentyears,thenumberofcourtcasesinvolvingpatentshasgrownextremely
rapidlytoimpressivelevels.1Inathought-provokingstudy,BessenandMeurer
(2008b)estimatethatlitigationcostsrelatedtopatentshaveexceededprivate
profitsfrompatentseversincethelate1990s.Accordingtotheirestimates,
globalprofitsdirectlyconnectedtoUSpatentsin1999accruedto9.3billion
USD,whiletheirestimateforthedomesticlitigationcoststocompaniesis
roughly16billionUSD.Thesefiguresarealarming.
TheneedforareformoftheUSpatentsystemhasbeenwidelyexpressed,
bothbyscientistsandpractitioners.Therearewidespreadcomplaintsaboutthe
qualityofgrantedpatents,especiallythelackof“clearlydrawnpropertylines”,
totheextentthatsomerefertothepatentsystemas“failed”.Currently,there
areeffortsunderwayinbothEuropeandtheUStoimprovethewaypatents
aregrantedandenforced.2Researchers’moreaudaciousrecommendationsfor

1SeeCook(2007)forananalysisofthisphenomenon,withafocusoftheroleofspecializedcourts.
2RecenteffortsinEuropeincludethedevelopmenttowardsaunified“EuropeanPatent”agreeduponin

7

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

thishaverangedfromadvisingfundamental,game-changingreformsuchas
BessenandMeurer(2008a),todemandingthepatentsystemasawholebeing
scrappedsuchasBoldrinandLevine(2002,2008).
Comparedtothesedemands,theactualreformsthathavebeenattempted
areextremelytame,focussingon(seemingly)minorproceduralmeasures.And
alsothebulkoftheoreticalresearchhasfocusedonmoremarginalshifts.The
reasonforthisisfairlyobvious-despitethemanyshortcomingsofthecurrent
systems,theexistingpatentsheldbyindividualsandcorporationsaretremen-
douslyvaluable.Theinvestmentsthatleadtotheprotectedinnovationswere
(arguably)sunkwiththegoalofpatentinginmind,whichmeansthatany
legislaturetamperingwiththesystemwillfaceneverbeforeseenclaimsfor
damages.Astheyoungestcurrentpatentshaveanother20yearsontheclock,
anyfundamentalreformstepsareallbutcertaintobepostponeduntilthen,
3least.atSofortheforeseeablefuture,anyrevolutiontothepatentsystembyneeds
willbeextremelymarginal,infusedwithapparentlyharmlesstweakstothe
patentapplicationfeeortheresponsiblejurisdictionforpatentclaims;butin
fact,duetotheimmensecomplexityofthecurrentpatent-machinery,pitfalls
loomevenfortweaksthatappeartobeharmless.Thegoalofthisarticleis
toprovideatheoreticalframeworkthatisassimpleaspossible,whileitstill
spansallthreepillarsofcurrentpatentsystems.Ouraimistodemonstrate
anumberofperhapsunexpectedtradeoffsinvolvedintamperingwiththem.
Whatweconsiderthethreepillarsare:

(1)Patentlegislation,whichdefinestheprerequisitesforapatenttobe
grantedaswellastherightofpatentholders.
theCouncilofMinistersoftheEuropeanUniononDecember4,2009,aswellastheintroductionof
aspecializedEuropeanPatentCourtasapartofthe(currentlystalled)EuropeanPatentLitigation
Agreement.IntheUS,introducingnewpatentlegislationhasbecomeabi-annualtradition,ofwhich
thePatentReformActsof2009,2007,2005and2003bearwitness.Themaingoalofthelatestwasto
reducetheburdenofthePatentandTrademarkOfficeintheinspectionprocess.
3Withthispointintimemovingalonguntilanyfirststepsareatseriousreformaretaken.Andthe20-year
timeframeisactuallyratheroptimistic,becauseresearchprogramsthatarecurrentlyunderwaywith
patentspotentiallyloomingyearsfromnowmayalreadyhavegeneratedclaimsbasedontheinvestors’
trustinthecurrentsystem.

8

2.1.ductionotrIn

(2)Patentandtrademarkoffices(PTOs),whichapplytheexistingrulesand
regulationsinexaminingpatentapplications.

(3)Courts,whichupholdtherightsofpatenteesagainstallegedinfringers
and,bygivingthirdpartiesthechancetochallengeexistingpatents,
havetheabilitytoinvalidatepatentsthatshouldnothavebeengranted.

Thebasicdilemmainpatentingisthatthepatentabilityofanideafromthe
perspectiveoflegalrequirementsorsocialwelfaredoesnotnecessarilycoincide
withtheprivatebenefitsaninventorobtainsfrombeinggrantedpatentclaims.
Inourmodel,wedistinguishbetweentheprofitanindividualorfirmmay
derivefromanideaifpatentedandthepatentabilityofanideafromthelegal
perspective.Unliketheexistingliterature,whichmostlyfocusesonbinarygood
vs.baddistinctions,weallowforacontinuumalongbothdimensions-ideas
mayrangefrompatentableover“almost”patentableto“clearlynot”patentable,
asisarguablythecaseinreality.ItistheroleofthePTOtomakethis
distinction,withitsdecisionnecessarilybeingimperfect-asPTOsfacetime,
budgetandstaffingconstraints.Werepresentthisinahighlyreducedformin
themodel.Finally,thereisthepossibilityofpatentholdersandcompetitors
becomingentangledincourtproceedings,wherewecandistinguishbetween
damagesuitsagainstallegedinfringersontheonehandandchallengesofthe
validityofgrantedpatentsontheother.Weattempttorepresentbothkinds
ofproceedingsinourmodel.
Theoccurrence,typeandoutcomeofcourtinteractionsdependonvarious
factors:Themostimportantaretheaveragequalityofpatents,asthereis
asymmetricinformationaboutthevalidityofindividualpatents,andthecom-
petitivesettingbetweenpatenteesandpotentialinfringers.Weshowthatone
ofthemostfrequentlycontemplated(ordemanded)reformsteps,increasing
feesforpatentapplicationsatthePTO,mayactuallydecreasetheaverage
qualityofpatentapplications.Further,wepointoutsomeunintendedand
potentiallycostlyconsequencesofincreasingthelevelofscrutinyexertedby
thePTOs.Further,wefindthatareductionoflitigationcostsforpatentcosts

9

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

hasmultipledesirableeffects-whichisespeciallyrelevantinthefaceofspe-
cializedcourtsofappealsbeingintroducedintheUSandtheEuropeanPatent
t.AgreemenLegislationTheremainderofthearticleproceedsasfollows:Afterabriefreviewofthe
relatedliterature,section2sketcheslegalandempiricaldifferencesbetween
theEuropeanandtheUSpatentsysteminordertoderivestylizedfactsand
motivatetherelevantdeterminantsofourmodel.Section3developsthemodel
ofpatentlegislation,examinationandlitigation.Insection4wederivethe
equilibriaofthemodelandperformcomparativestaticsexercisestodetermine
theeffectsofvariousapproachestopatentsystemreform.Section5empha-
sizesempiricallyrelevantpredictions,proposesvenuesforfurtherresearchand
concludes.

2.1.2.RelatedLiteratureandContribution

Thefirststrainoftheliteraturethatourarticleisrelatedtostudiestheeffects
ofpatentlitigation.Theissueofpotentiallyinvalidpatentsandresultingliti-
gationhasgarneredrecurringinterestinthepasttwodecades.Meurer(1989)
proposesthefirstmodelthatexplicitlytakesthepossibilityofpatentinvalid-
ityandresultinglitigationintoaccount.Patenteesfaceasinglecompetitor,
withbothpartiesawareofthefactthatcourtsmayoverturnthepatent.The
competitorcaneitheracceptthepatenteesnon-cooperativebargainingoffer,do
nothingorchallengethepatentthroughthecourts.Thestudycarefullyshows
howsettlementandlitigationprobabilitiesdepend,amongotherthings,onthe
strengthofthepatentbothundersymmetricandasymmetricinformation.We
adaptthemodelforoneofourlitigationsubgames.
CrampesandLanginier(2002)studytheoppositecompetitivesetting:Here,
thecompetitordoesnotchallengethepatent,butsimplyentersthemarket,
thereby(potentially)infringingthepatentee’srights.Thepatenteedoesnot
necessarilynoticethis,insteadhemustinvestinacostlymonitoringtechnology,
whichdeterminestheprobabilitywithwhichinfringementisobserved.Ifob-

10

ductionotrIn2.1.

served,thepatenteemustmakethedecisionwhethertoaccommodate,litigate
orsettlewiththeinfringer,wherethesettlementresultsfromcooperativeNash
Bargaining.Theotherwiseefficientbargainingpotentiallybreaksdownasthe
authorsimposecostlyfrictionsinthebargainingprocess.Monitoringisshown
tobeaneffectivewaytoprecludeentry.BessenandMeurer(2006)’sapproach
isveryissimilartoCrampesandLanginier(2002)-theygivethepotential
infringerthestrategicopportunitytoinvestintoresearch,whichaffectsthe
probabilityofbeingfinedbycourtsintwopossibleways:Thisactivitycould
begeneralR&D(e.g.addingadditionalfeaturestotheproduct),whichwould
potentiallyincreasetheriskofbeingfoundinfringing.Orthecompanycould
exertefforttospecificallyinventaroundexistingpatents,whichwouldreduce
theprobabilityofinfringing.InsteadofmonitoringactivityasinCrampes
andLanginier(2002),thepatentholdercaninvestinstrengtheninghispatent;
theprobabilitythatthecourtwillfindtheinfringerguiltyifatrialarisesin-
creasesinthisinvestment(thiscanbeinterpretedasapplyingforadditional
patentstocreateathicketoremployingbetterandmoreexpensivelawyersto
phrasethepatentapplicationandclaimsoptimally).Intheempiricalpartof
theirproject,theauthorsfindthatcompetitors’investmentisoverwhelmingly
h.researconcusedfoMorespecificissueswithintheareaofpatentlitigationhavealsobeendis-
cussed:Anumberofpapersinvestigatetheroleofdifferentliabilityrulesand
howtheseaffectthelevelofprotectiongrantedbyintellectualpropertyrights,
seee.g.SchankermanandScotchmer(2001)andChoi(2006)forcomparisonsof
lostprofitandunjustenrichmentrulesindifferentcompetitivesettings.Anton
andYao(2007)studytheinfringementdecisioninthecaseofprocessinnova-
tionsforlostprofitsdamages.Antitrustissuesandanticompetitiveeffectsof
thesettlementofpatentlitigationaretakenintoaccountinShapiro(2003)and
05).(20ShapiroandLemleyAsecondgroupofarticlesisconcernedwiththeorganizationofandincen-
tiveswithinpatentofficesandtheireffectsonthepatentsystem.Caillaudand
Duchene(2006)scrutinizethecapacityofthepatentofficetodealwithappli-

11

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

cationsandthesocalledoverloadproblem.Intheirmodel,ifthePTOwere
toemployastrictexaminationstandard,ahypotheticalequilibriumexistsin
whichbadapplicationsaredeterred.Aproblemarises,though,ifthePTOis
overloadedwithapplications-asitcannolongerenforcethestrictstandard,de-
terrencebecomeslesseffectiveandtheseparatingequilibriumcannolongerbe
upheld.InPrady(2008),lowqualityinventorscaninduceshirkingofthepatent
examinerbysendingsignalsthatrequiremoreefforttodisentangle.Schuett
(2009)modelstheexaminationprocessasacombinedmoral-hazard/adverse-
selectionproblemandtriestoexplainandarguefordifferentincentiveschemes
examiners.for

Tworecentstudiesarecloselyrelatedtoours.BothFarrellandShapiro
(2008)andChiou(2008)presentmodelswhichallowforpatentexamination
andpotentiallitigation.InFarrellandShapiro(2008),anupstreaminnovator
ownsapatentthatmaybe“weak”inthesensethatacourtwillonlyuphold
itifchallengedwithacertainprobability.Theupstreaminnovatorfacesa
setofdownstreamfirmswhichcanapplyhisinnovation.Dependingonthe
competitivesetting,theyshowthatevenveryweakpatentscanhaveastrong
price-shiftingeffect–whenthedownstreammarketisverycompetitive,then
thereisnexttonoprivateincentivetochallengethepatent,whichgiveseven
weakpatentsgreatpower.Insuchasetting,morestringentPTOreviews
arewelfareenhancing.Ourstudyisinmanywayscomplementarytothis
approach:WhileinFarrellandShapiro(2008)thereisnoroomforlitigation
inequilibrium,despitezerolitigationcosts,ourmainfocusistodetermine
underwhichconditionswhichkindoflitigationwillappear,andhowthisis
affectedbycontemplatedreforms.Theirmainanalysisfocusesontheroleof
theintensityofdownstreamcompetitioninaspecificsetup-wetreatthisfactor
inareducedform,whichtoacertaindegreeencompassestheiranalysis.Finally,
theytreatthequalityofgrantedpatentsasuncertainevenfortheholders.We
demonstratethatsimilareffectscanbegeneratedwithasymmetricinformation
andtwokindsofpatents–ironcladorvoid–andthattheprobabilisticnature
t.requiremennois

12

ductionotrIn2.1.

MotivatedbyLemley(2001)’sprovocativethesisthatincreasesinthedili-
genceofthepatentofficeareinefficientandoneshouldletthecompetitionsort
outbadpatentsthroughlitigation,Chiou(2008)setsupatwostagemodel,
inwhichfirstthepatentofficeexertseffortinordertofindpriorartthatal-
lowstodenyalatentlyinvalidpatent.Ifthepatentofficeisunabletodestroy
thepatent,inthesecondstageaprivatecompetitorcanexerteffortinthe
samewaytohaveitrepealed.Whileforrelativelygoodpatents(highpriors
ofpatentquality)hefindsacrowdingoutofprivateeffortsthroughthepatent
officealongthelinesofLemley(2001),forlow-qualitypatentpopulationhe
findstheoppositeeffect:Stricterenforcementbythepatentofficeactuallyen-
couragesthecompetitortohimselfinvestinresearcheffortstotrytoinvalidate
thepatent.Thepaperreliesonanewwayofmodelingcourtinteractions–
higherexertionof(costly)effortbythechallenger(orbythePTO)leadsto
thedestructionofthepatentwithahigherprobabilityandthe“quality”ofa
patentaffectsthemarginaleffectofeffort.Again,ourapproach,usingamore
classicalwaytomodelsuitsandcountersuitsappliedtoapopulationofpatent
applications,canbeseenascomplementary.

Thisarticlecontributestobothstrandsoftheliterature.Ourmodelencom-
passesboththeexaminationandthelitigationphaseofthepatentingprocess.
Regardinglitigation,weintegratethecompetitors’choicebetweenchallenging
anexistingpatentorenteringthemarketdirectlyandpotentiallyinfringing.
Competitors’beliefsconcerningpatentvalidityaredeterminedbythepolicy
adoptedbythepatentofficeaswellassettlementoffersproposedbythepatent
holders.Further,weallowforadelaybetweenpatentapplicationandthefinal
decisionofthepatentoffice-aformofpatentofficeoverloadrelatedtotheone
studiedbyCaillaudandDuchene(2006),whichcanbeconsidereda“cost”of
demandinghigherlevelsofdiligencefrompatentexaminersandwhichawaits
empiricalexploration.Finally,westudyapopulationofideaswhichvaries
continuouslyregardingprivatevaluetotheinventorandobjectivepatentabil-
ity,whichallowsustoshedlightontradeoffsduetothecompositionofthe
populationofpatentapplications,whichhavenotbeenstudiedbefore.

13

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

2.2.ComparisonoftheUSandtheEuropean
PatentSystem-StylizedFacts

Beforewedescribethetheoreticalmodel,wefirstprovideaglimpseofsomeof
thedifferencesbetweentwoofthemostimportantpatentregimestoday:the
USandtheEuropeanpatentsystem.Thegoalisnottoexhaustivelydisplay
andanalyzetheseissues,buttomotivatethefactorsandvariablesweintegrate
del.moourtoin

2.2.1.PatentLegislationandExaminationthrough

OfficestatenP

Eventhoughtherearemanycallsforandinitiativesinthedirectionofharmo-
nization,theUSandtheEuropeanpatentlegislationsretainsomeverydistinct
features.Generallyspeaking,therearestarkdifferencesbetweenEuropeanand
USpatentlaws,whilethedifferencesamongthecorecountriesoftheEUare
negligible.comparablyWhilethe“firsttofile”systemisprevalentinEurope,thatis,thefirsttoap-
plyforapatentisawardedtheright,traditionallytheUSawardpatentrights
tothefirstpersontodiscoveraninnovation(“firsttoinvent”).Afurtherdiffer-
enceregardingthepatentingrequirementsconcernswhatintheUSistermed
“novelty”and,e.g.,inGermanythe“inventivestep”.Bothrequirementsstate
thatforanideatobepatentableitmaynotyethavebeenmadepublicprevi-
ouslyinanyform.InGermanythisholdsabsolutely.IntheUS,inventorsare
granteda“graceperiod”,i.e.theygenerallyonlyhavetofiletheirapplication
withinayearoftheiridea’spublication.
Theothergeneralrequirementsforsomethingtobepatentablearesimilarin
bothsystems-patentablesubjectmatter,non-obviousness,andapplicability.
Still,thereisquitealotofevidencethattheserequirementsareinterpreted
differentlybytheUSandEuropeanpatentoffices,whichisclosetothefocus
ofourstudy.Forexample,StrausandKlunker(2007)citethefollowingnum-

14

2.2.ComparisonoftheUSandtheEuropeanPatentSystem-StylizedFacts

bersoftheUSandtheEuropeanpatentoffices:In2005,therewere409,532
applicationsforpatentsintheUS.Outofthese,165,485weregranted,i.e.a
shareof40.4percent.Ontheotherhand,inEurope,thenumberofappli-
cationsinthesameyearwas197,391with53,256patentsbeinggranted,i.e.
asignificantlysmallershareof27,0percent.Fromthesenumbersalone,it
appearsthatthattheGermanpatentofficesapplyastricterstandardthanthe
Americanones.Thisiscorroboratedbyotherobservations:Theaveragetime
fromfilingapatenttoitbeinggrantedwas45,3monthsinEurope,whileit
wasonly24monthsintheUS,asHallandHarhoff(2004)report.Theyalso
lookatthegrant-ratesofpatentsattheEuropeanPatentofficeforUSpatents
seekingEuropeanapprovalontheoneandpatentsfromothercountriesseeking
approvalinEuropeontheotherhand.Theyfindthattheapprovalratefor
USpatentswassubstantiallylower,witha16percentdifferencein1995.The
trendpointstowardsadeteriorationoftheUSstandards,asin1979therewas
parityconcerningthesenumbers.
StrausandKlunker(2007)furtherciteastudybytheconsultingfirmRoland
Bergerthatfoundthatinordertoapplyforandmaintainapatentoverits
entirelifetime,costsbetween32,000and47,000EURaccrueinEurope,while
theaveragefigureisonly10,250EURintheUS.4
Fromtheseobservations,wederivethefollowingstylizedfactsregardingthe
patentapplicationprocessintheUSandinGermany:

(F1)PatentapplicationsarescrutinizedmorestrictlybyEuropeanthanby
tpatenUSoffices.

(F1’)Asaresult,aEuropeanpatentispotentiallyabetterindicatorofthe
strengthofthepatentee’sclaimsthanaUSpatent.

(F2)ItissignificantlymoreexpensivetoobtainapatentinEuropethanin
theUSintermsofthefeesrequired.

4Formonthsincomparison:ChinaandRegularUS.pPerioatendtfromcostsfilingaretoestimatedgrant:to31,6rangemonthsfrominaboutJapan,2,40045,3tomon4,000thsinEURinEurope,China.24

15

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

(F3)Thedurationofpatentpendingvariesbetweenpatentingsystemsbya
margin.considerable

2.2.2.CourtsandPatentTrials

Afterbrieflydiscussingthedifferencesconcerningthelawsinthetwojurisdic-
tions,wewillnextconsiderhowpatentsuitsfareinthecourts.
Cook(2007)notesthatinthecourseofthelastdecade,thenumberofpatent
casesfiledintheUShasroughlydoubled,fromabout1,250in1990toabout
2,500in2000.Heshowsthatthisisonetheonehandaresultofmorepatents
beingfiledingeneral(theratioofcasespergrantedpatentremainsrelatively
unchanged,asthelatternumberincreasesfromabout90,000toabout180,000
peryearinthesameperiod).Ontheotherhanditresultsfromcourtsbeing
moreaccommodatingtoplaintiffs.Inhisempiricalmodelthedecisiontogoto
courtdepends,amongotherfactors,ontheshareofcasesthatweresuccessful
intherespectivedistrictinpreviousperiods.Oneimpressivemeasureinthis
regardisthattheprobabilityofreceivingapatentrewardof1millionUSDor
morein2001constantdollarsincreasedfromlessthan10percentin1976to
morethan30percentin2000.
Concerningtheaveragecostsofgoingtocourtoverpatentinfringementin
theUS,estimatesrangefrom500,000USDto3,000,000USDforeachparty.
Muchlargersumsarementionedinthecontextofcomplexorhigh-stakescases,
especiallyintheareaofpharmaceuticals.ForaEuropeanpatenttobelitigated,
courtfeesof70,000EURariseaccordingtoEuropeanPatentOffice(EPO)
reports.Theirestimatesofthelawyers’feesbournbythepartiesinaddition
amountstoapproximatelythesamefigure,sothateachpartywouldhaveto
expectcostsaround150.000EURexante.
Wegleanthefollowingstylizedfactfromthesebriefobservations:

(F4)Thereisasignificantcostdifferencebetweenbringingacasetocourtin
theUSandEurope,i.e.thecostsoflitigationarevastlyhigherintheUS
e.Europinthan

16

SettingdelMo2.3.

2.3.1.Therulesofthepatentsystem

2.3.ModelSetting

Legislators(exogenously)defineaminimumstandardforutility,novelty,non-
obviousness,etcwhichmakesanideaobjectivelypatentable.Forthismodel,
weassumethatthesevariouscriteriacanbereducedtoonedimension,andthe
minimumobjectivestandardsetbylegislatorsisdenotedµ.

Figure2.1.:Timelineoftheentiremodel

2.3.2.StrategicPlayersand“Ideas”

Weconsideragamewithtworisk-neutralstrategicplayers,AandB.Player
Aactivelygeneratesideas.Ideasare“randomevents”,i.e.individualideas
areimbuedwithcertainrandomlydrawncharacteristics(moreonthisbelow).
∂C(n)Generatingn∂2ideasC(n)ina5givenperiod,AincurscostsC(n).Weassumethat
∂n>0and∂n2>0.
Eachindividualideathatisgeneratedisdefinedbythefollowingcharacter-
istics:5Thiscanbeinterpretedinthefollowingway:Aresemblesthepopulationofpotentialinventorsandthey
aresortedaccordingtothemarginalcostsoftheirideas.

17

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

(1)Thepatentabilityoftheideafromalegalstandpoint,i.e.itsutility,
novelty,non-obviousnessetc.Againweassumethatthesecanbereduced
toasinglescalar,whichwedenotebyι,withι[0,1].Notethatifι≥µ
anideaisobjectivelypatentablefromthelegalstandpoint.

(2)Ameasureoftheadditionalvaluetotheinventorderivedfrompatenting
theidea,whichwedenotebyυ.Tobemorespecific,thismeasureis
definedasthevaluethattheinventorcanappropriatefromexploiting
theideaoptimallyasamonopolistafterobtainingthepatent,i.e.given
no-oneinfringesuponthepatent.Asweareconcernedwiththeincentive
effectsofpatentsinthefaceofchallengeswithrespecttoenforcement,
wesetthevaluefromexploitinganideawithoutpatentprotectionto
zero.Thereforeinourmodel,fromanexanteperspective,anideais
onlyvaluabletoitsinventorinasfaritcanbepatented.Wenormalize
oncemoresuchthatυ∈[0,1].

Theinventoronlylearnstheactualcharacteristicsofagivenideaafterit
hasbeengeneratedandhehasincurredtheassociatedcosts.Specifically,the
parametersιandυarerespectivelydrawnfromcommonlyknownandinde-
pendent6distributionsF(ι)andG(υ)andrevealedtoA.Wefurtherassume
thatFandGarebothcontinuousandstrictlyincreasing.

2.3.3.PatentandTrademarkOffice

PatentApplicationsandthePatentOffice

Adecidesforwhichofhisideastosubmitpatentapplications.ItcostsA
thefixedamountτ1tosubmitanideatobeexaminedbythepatentoffice.
WeassumethatthePTOcanonlymakeabinarydecisionregardingagiven
patentapplication:eitherapproveorrejectit.Accordingtothelegalrules
statedabove,thePTOshouldapprovepatentapplicationsifandonlyifι≥µ.
6Wforeanonlyuseexpressthisionbeloassumptionw.Thema(indepjorityendenceofourofresultsdistributions)holdwithouttobeit.abletoderiveaclosedformsolution

18

2.3.ModelSetting

WeassumethatthePTOcannotdirectlyobservetheactualιofanindividual
application.Conceptuallytherefore,itcouldcommittwokindsoferrors:Errors
oftypeone,i.e.grantingapatentdespiteι<µ,anderrorsoftypetwo,
i.e.decliningpatentsdespiteι≥µandbothoftheseerrorsareassociated
withsocialcosts.Forthesakeofthismodel,weneglecterrorsoftypetwo-
discussionsbothwithscientistsandpractitionershaveconvincedusthatthey
areempiricallyclosetoirrelevant,sincethereisarelativelyfastandcheapway
toappealthedecisioninbothjurisdictionsinthiscase.ImaginethatAcan
incurtheadditional(small)costsτ2inordertohavetheapplicationreexamined
whichleadstoitbeinggrantedifι≥µ.Theonlyeffectofthiswillbetoshift
theexpectedprofitsfrompatentingofholdersofpatentableideasdownwards
sum.fixedaybWemodelthepatentofficepolicyinthemostsimplepossiblereducedform.
Thepatentofficesimplyimplementsanexogenouslygivenexaminationpolicy
Φwhichdeterminesthelikelihoodφ(ι)ofagivenideabeingpatented,including
theprobabilityofafirstordermistakebeingmadeifι<µ.Forexample,ifthe
PTOgrantseverypatentapplication,thenφ(ι)=1foranyι.Theinspection
policydoesnotdependonυ–forone,patentexaminers(aseveryoneelse)have
averyhardtimedeterminingtheexpectedvalueofapatent,andalso,more
importantly,guidelinesingeneralforbiddifferenttreatmentofapplicationsde-
pendingontheirsuspectedvalue.Weassumethatφ(ι)>0forallι,i.e.even
theworstqualityideaalwayshasapositivechancetobeawardedapatent.
Furtherweassumethat∂∂ιφ>0ifι<µ,i.e.thecloserιcomestotheobjective
patentingthreshold,themorelikelyapatentistobe(falsely)granted.There-
foretheprobabilitywithwhichagivenapplicationwillbegrantedisφ(ι)≤1
ifι<µ,andφ(ι)=1iftheideaisobjectivelypatentable.

[Considerthefollowingexampleforapossiblemicro-foundationasanillus-
trationoftheidea:Thepatentofficegeneratesanumberζofsignals,where
ζispartoftheexogenousscheduleΦ.ζcanforexamplebeinterpretedas
thetimethatapatentexaminerspendsresearchingpriorartorperusingthe
patentapplication.Ifindividualsignalsarenormallydistributedwithmeanι

19

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

andstandarddeviationσ,themeanofthesignalsιhastheexpectedvalueι
andthestandarddeviationς=ζ0σ.5.Theexaminationschedulefurtherincludes
acutofflevelµ∗.Ifthemeanofthegeneratedsignalsisbelowthiscutofflevel,
thePTOrejectstheapplication,anditgrantsthepatentotherwise.ζandµ∗
thereforeimplicitlydeterminetheprobabilityoftypeoneandtypetwoerrors
thatthePTOcommitsforanygivenι.Denotethenormaldistributionwith
meanιandvarianceςasNι,ς.Thentheprobabilityforatype1-errorgiven
thatι≥µissimplyNι,ς(µ∗),andequivalentlytheprobabilityforatype2-error
(whichwedonotconsider)givenι<µwouldbe1−Nι,ς(µ∗).]

Extension:PeriodofPatentPending

Asdiscussedabove,thedecisionwhetherornotapatentapplicationwillbe
grantedisbynomeansmadeimmediately.Instead,asubstantialperiodoftime
passesbetweenapplicationanddecision,duringwhichtheideashasthestatus
of“patentpending”.Littleinvestigationhasfocusedonthevalueofpatentsin
thecourseoftheapplicationprocesssofar.ButwhenSteveJobsexclaimed
hisnowfamous“And,boy,havewepatentedit!”duringthespeechintroducing
Apple’siPhoneattheMacWorldExpoinJanuary2007,thegreatestshare
ofthepatentsprotectingthetouch-screentechnologyinvolvedhadnotbeen
granted,yet.Infacteachofthe21iPhone-patentsconsidered“central”by
technology-afficionadoeswasstillpendingatthispointintime.7Partofthis
stancewassurelyjustifiedbytheknowledgethatthepatentswouldbegranted,
lateron,whichturnedouttobetrue.Buttosomeextent,pendingpatent
applicationsitselfarevaluable,especiallyinfast-movingindustries.
Thereareanumberofreasonstosuspectthatinventorsdobenefitfrom
theirideasduringthependingperiod.Manyifnotmostlicenseagreements
arenegotiatedpriortotheactualpatentbeingissued.Otherfirmsmaythink
twiceaboutenteringthemarketiftheproductisdesignatedwiththe“patent
pending”stamp,fearingfuturelawsuits.Finally,inoneofthefewempirical

7Seedetailedhdescriptionandlinkstottp://www.mad4mobilephones.com/the-21-most-imptheindividualclaims.ortant-iphone-patents/562/foranextremely

20

2.3.ModelSetting

studiesrelatedtothetopic,Häussleretal.(2009)findthatthechanceto
receiveventurecapitalriseswiththenumberofpatentapplicationsinthe
ortfolios.pfirms’Again,wereducethisconceptasfaraspossible.Wesimplydenotethe
patent-pendingvalueofanidea(ι,υ)asδ(Φ)υ,where0≤δ<1and∂∂φδ<0.
Thereforeincreasingthequalityofpatentofficescrutiny(orloweringφ(ι)is
“expensive”inthesensethatitincreasesthepatentpendingvalueofideas,for
examplebyincreasingthedurationbetweenapplicationandexaminationon
average.Wewillnotover-stretchthisconceptinthefollowing,especiallyin
thisoverlysimplelinearspecification,butitisusefultorememberthisfactor
whendiscussingreformsaimedatmorestringentinspectionpoliciesthrough
PTO.the

[Consideragaintheillustrativeexampleabove.Imaginethatthepatentpend-
ingdurationofagivenideaafterapplicationistperiods,wheret=t(ζ,n)and
∂∂ζt>0.Duringthistime,theperspectivepatenteereceivesthebenefitsΔ(υ,ι)
perperiod.]

2.3.4.TheCourtSystem

Wehavedescribedabovehowideasaregenerated,howAdecideswhetheror
nottoapplyforpatents,andwhetherornotpatentsaregrantedbythePTO.
Inthenextandfinalstageofthemodel,Bentersthepicture.WeconsiderBto
betheonlystrategicallyactingcompetitorofAineachofthemarketscovered
byapatentintheeconomy.8BobserveswhichpatentsweregrantedtoAand
choosesoneofthefollowingreactions:Hemayeitherleavethecorresponding
marketuncontested(U),challengethevalidityofthepatentbeforecourt(C)
orenterthemarketandtherebyinfringeA’spatent(I).
Choices(C)and(I)respectivelyinducedifferentlitigationgames:For(C),
8ofinvAnalogouslyentorstoA.aboMoreve,spBecificcanablley,inBisterpretedthemostastheefficienptopulationorofprofitablepotentialcompcompetitorofetitorsA-tothisthepmitigatesopulationthe
losswithofB’s,ygeneralitetlessyfromstrong.assumingonlyonecompetitor,asotherlessefficientfirms’incentivesarealigned

21

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

weadapttheapproachintroducedbyMeurer(1989).Forchoice(I),weuse
asetuprelatedtoCrampesandLanginier(2002),butwithnon-cooperative
bargaining.

Information

Whilethepatentholderisperfectlyinformedaboutthequality(ι,υ)ofany
givenpatent,Bcannotobserveι.Thisparameterhasthefollowingsignificance
inthecontextofcourtproceedings:Theprobabilitythatacourtupholdsa
givenpatentobviouslydependsonthelegalpatentabilityoftheunderlying
idea.Forsimplicity,weassumethatcourtsmakethe“right”decision,thatis,
theyupholdthevalidityofthepatentwheneverι≥µ.Onemightquestion
theassumptionthatcourtscandiscoverthetrueιofanidea,whilethePTO
cannot.Butseeingthatcourtcasestakeyearswhilepatentexaminershavedays
(ifthat!)tocometoadecisionregardingagivenapplication,theassumption
mayappearlesssevere.
Therefore,fromA’sperspective,thereisnouncertaintyregardingthecourt’s
decisiononthevalidityofagivenpatent:ifι≥µthecourtwillupholdits
claims,anditwillrepealthepatentotherwise.AsBcannotobserveι,though,
thecourtproceedingstakeonarandomcharacterfromhispointofviewand
hemustformthesubjectiveprobabilityofapatentbeingupheldgiventhe
informationavailabletohim.Inoursimplemodel,Bonlyobservesthree
things:First,thefactthatapatentwasgrantedbythePTO.Second,thevalue
υofthepatenttothepatentholder,whichhemaye.g.derivefromobserved
sales.Finallythird,eventualsettlementoffersSproposedinthecourseof
thelitigationgamebyA.Fromtheseobservations,Bformstheconditional
subjectiveexpectationpB(ι≥µ|υ,Φ,S)ofthevalidityofthepatent.We
willabbreviatetheprobabilityunconditionalonasettlementofferSaspBin
thefollowinganddenoteupdatedequilibriumbeliefsfollowinganinformative
signalSasp∗B.

22

StructureofthePost-PatentingLitigation-Game

2.3.ModelSetting

Combiningtheingredientsabove,asthefinalstepofthemodel,foreachpatent
thatwasgrantedtoA,thefollowingsubgameisplayed:

Binitiallydecideswhethertoleavethepatentuncontested(U),challengethe
validitybeforeacourtoflaw(C)orenterthemarketandinfringeA’spatent
(I).

(U)IfBleavesthemarketuncontested,thegameregardingthegivenpatent
endsandtheplayersgetthepayoffsπUA=υandπUB=0.

(C)Thiscaseistreatedanalogouslyto,forexample,Meurer(1989)andChiou
(2008).IfBannouncesherintentiontochallengethepatent,AandB
firsthavetheopportunitytonegotiateasettlement.Theinformedparty,
i.e.A,whoknowswhetherornotthepatentwillbeheldvalidbythe
court,canfirstproposeasettlementofferSC.Bobservesthesettlement
offerandinthisprocessupdateshersubjectiveprobabilitythatthepatent
isvalid.Thenshedecideswhetherornottogotocourt,withthefollowing
outcomes:Ifι≥µthecourtupholdsthepatentdespitethechallengeand
thepayoffsareυ−κforAand−κforB.Ifι<µthecourtrulesthat
thepatentisinvalidandthereforethemarketisnolongerprotected.By
definition,thepayoffofAthereforeis−κandBobtainsthepayofffrom
acompetitivemarketwithfreeentrynetofcourtfeeswhichwedenote
asυBC−κ.9

(I)ThefollowingsubgameensuesifBunilaterallyentersthemarketwitha
productthatpotentiallyinfringesA’spatent.Weassumethat(C)and
(I)aremutuallyexclusive,i.e.itisimpossibleto,forexample,challengea
patentandenterthemarket.10Again,weassumethatitistheinformed
9NotethatCanymarketpaCyoffalwaysdependsontheoverallprofitabilityofthemarket,thereforefor
exampleυBisactuallyυB(υ).Wecontinuetousetheformerasashorthand.
10Thereareanumberofsoundeconomicreasonsunderlyingthisassumption.Forone,challengingapatent
beforeintroducingaproductraisesthespecterofwillfulinfringementandpunitivelyhighercompensation
inthecaseofbeingfoundinfringing-weleavethisconsiderationoutofthemodel.Further,ifthecompany

23

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

partyAwhomakesthesettlementofferSI.IfBrejectstheoffer,A
decideswhetherornottolitigate.Asthepatentiseithervalid(ι≥µ)
ornotvalidandAisperfectlyinformedaboutthis,thecaseinwhichhe
suesandthecourtsinvalidatethepatentdoesnotarise(here,Awould
receiveapayoffof−κandBreceivesthecompetitivepayoffnetofcourt
feesυBC−κ;thereforeitisbetterforAtoaccommodategivenpatent
invalidity,whichgivestheduopolypayofftobothparties).Ifthepatent
isvalidandAsues,thecourtsgranthimlostprofitsorcompensatory
damages,thereforehispayoffsfromlitigationareυ−κ.11Ontheother
hand,thepayoffoftheinfringerinthiscaseisherduopolyprofitplusthe
differencebetweenA’sduopolyprofitandtheforgonemonopolyprofit
minuscourtcosts,i.e.υBD−(υ−υAD)−κ.

TheStructureofIndustryPayoffs

Wemakethefollowingassumptionswithregardtotheindustryprofits.First,
weassumethatυ>υAD+υBD,theprofitthatAcanoptimallyextractislarger
thanthesumoftheduopolyprofitsofAandB.12Second,weassumethat
υBD≥υBC>0.Bmakesweaklyhigherprofitsinduopolythanunderfull
entrycompetition,butstillmakesstrictlypositiveprofitsinthelattercase
(thoughtheymaybearbitrarilycloseto0).Third,weassumethatυAD,υBDand
υBCareallcontinuousandincreasinginυ-looselyspeaking,amorevaluable
marketingeneralleadstohigherduopolyandcompetitivepayoffsformarket
participants.EachofthesewouldholdforgenericspecificationsofCournot
competitionwithentryunderintegerconstraints,forexample.Fourthand

isnotfoundtobeinfringing,itmayitselfbenefitindirectlyfromthepatentprotectionofthemarket,as
wewillseebelow.
11Weonlyfocusonthisformofdamages,foranexcellentdiscussionoftheeffectsofdifferentkindsofdamage
awardsseeforexampleSchankermanandScotchmer(2001)orChoi(2006).Further,weabstractfrom
thefactthatanallegedinfringementofavalidpatentisnotnecessarilycoveredbytheclaimsofthe
patent.Onecouldsimplyrescalethepayoffsoftheplayersbytheexanteprobabilitythattheproduct
willbefoundinfringing,withoutsignificantchangestotheresults.Toreducethenotationalburden,we
exercize.thisfromabstain12Implicitly,wetherebydisregardsuchphenomenaas“patenttrolls”,companieswhichcannotexploittheir
patentseffectivelywithoutgoingtocourt(orthreateningtodoso)againstproducerswhoinadvertently
infringedupontheirpatent.SeeReitzigetal.(2007)forastudyfocusingonthistopic.

24

Analysis2.4.

finally,weassumethat(υ−υBC)and(υ−υAD)arenondecreasinginυ.Asthe
marketasawholebecomesmorevaluable,thedifferencesbetweenmonopoly-
andduopoly-,ormonopoly-andcompetitiveprofitsdonotgrowsmaller.

Analysis2.4.

Equilibria2.4.1.

Inthefollowing,wefirstderivetheBayes-NashEquilibriaforthelitigation
subgames,beforeembeddingtheminthelargermodelinordertolearnmore
abouttheincentivestogenerateideasandpatentthem.

SubgamesLitigation

Wefirstfocusonthecasethathasalreadybeenstudiedintheliterature,i.e.
subgame(C)inwhichBthreatenstochallengeA’spatentbeforecourt.13
Clearlythisthreatisnotcredibleunlesstheunconditionalexpectedpayoff
fromgoingtocourtisnon-negative,i.e.thefollowingconditionholds.

(2.1)

(1−pB)υcB−κ≥0

If(1)isviolated,neitherholdersofvalidnorofinvalidpatentsmakeasettle-
mentoffertoB,whointurndoesnotgotocourtovertheissue.Thisresembles
apoolingequilibriumwiththepayoffsπA=υandπB=0.
Theclearlymoreinterestingcaseistheoneinwhich(1)issatisfied.Notethat
ifι≥µ,AcannotcrediblysignaltoBthatheistheownerofavalidpatent
viahissettlementofferandkeephimfromgoingtocourtcompletely,assignals
arecostlessandcanbemimickedbyholdersofinvalidpatents.Intuitively,no
matterwhichsumholdersofgoodpatentsoffertotheircompetitor,theycan
13AsFortheaderivdetailededdisresultscussionforofthisthetesubgamechnicalhavaspebectseenoftheestablishedequilibrium,previouslywe,wreferecotoverMeurerthiscase(1989v).erybriefly.

25

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

Figure2.2.:GameTreefortheChallengeLitigationSubgame

alwaysbemimickedbytheholdersofbadpatentsandlitigationwilloccurwith
positiveprobabilityinequilibriumasaresult.Thefollowingisanequilibrium:
Holdersofgoodpatentsmakethetoughestsettlementofferpossible,i.e.SˆC=
0.HoldersofbadpatentsmixbetweenSˆCandofferingasettlementworth
SC=υBC−κ.

Lemma1:[Proposition1-Meurer(1989)]PerfectBayesianEquilibriumof
subgame(C):(i)If(1)isviolated,inequilibriumnosettlementofferismade
andnopatentischallenged.(ii)If(1)issatisfied,asemi-separatingequilib-
riumarisesinwhichholdersofgoodpatentsnevermakeasettlementoffer,
holdersofbadpatentsmixbetweenmakingnosettlementofferandoffering
SC=υBC−κandBupdateshisbeliefstop∗Buponreceivingornotreceivinga
settlementofferandmixesbetweenlitigationandinaction.

Webrieflysketchtheexistenceproofinthefollowinginordertoclearly
demonstratethemechanicsandintuitionoftheequilibrium.Withtherefine-
mentsofsequentialequilibriumandD1,thisequilibriumisalsounique-for

26

theseproofswereferthereadertoMeurer(1989).

Analysis2.4.

Proof.(i)AstheexpectedpayofffromlitigationgivenpBisnegativeand
nosignalisforthcoming,noupdatingtakesplaceandBchoosesinaction.(ii)
DenotetheprobabilitywithwhichBgoestocourtgiventhesignalSC=0
asλ.Thentheexpectedpayoffoflowqualitypatenteesfrommimickingthe
behaviorofhighqualitypatenteesis:(1−λ)υ+λ(−κ),whilethecertainpayoff
giventhesettlementofferisυ−(υBC−κ).ForlowqualityAstobewillingto
mixbetweenthesetwo,theyhavetobeidentical.Considernextthepayoffof
BifsheobservesthesignalSC=0.Letuscalltheprobabilitywithwhich
low-qualitypatenteesmimicgoodqualitypatenteesβ.Uponperceivingthe
signalSC=0,Bayesianupdatingofthepriorgivesusthefollowingcondition:
1−p∗B=pBβ+(1β−(1p−Bp)B).ForBtobewillingtomixbetweenlitigationandaccepting
thetoughsettlementoffer,thefollowingequalityhastohold:p∗B(−κ)+(1−
p∗B)(υBC−κ)=0.Forthefollowingvaluesofλandβ,alloftheseconditions
Carefulfilledsimultaneously:λ∗=υυB+−κκandβ∗=1−ppBBυBCκ−κ.Finally,specify
offtheequilibriumpathbeliefspB(ι≥µ|υ,Φ,S>0)=0.Thenitiseasyto
showthatnoprofitabledeviationexistsforeitherplayer.

Notethatlitigationariseshereasaresultofthecompetitornotbeingableto
distinguishbetweengoodandbadpatentsgiventhattheofferedsettlementis
0,whileshecanidentifythosepatentholdersasbadwhoofferamoregenerous
settlement.Forthese,though,Bneedsnotlitigate,asshereceivesthesame
expectedpayofffromthesettlementasshewouldfromlitigation.Notethat
theholdersofbadpatentsareexactlyindifferentbetweenthetoughsettlement
offerof0andthegenerousofferduetothechancethatthecompetitorwill
takethecasetocourtwithapositiveprobabilitygiventhetoughoffer,which
resultsinthemlosingtheirpatentprotectionentirely.

Letusnextconsiderthesubgame(I),inwhichBentersthemarket,thereby
infringingthepotentiallyvalidpatentofA.Fromthepayoff-structureaboveit
isclearthatitisneverprofitableforholdersofbadpatentstogotocourt,asthe
courtiscertaintodeemtheirpatentinvalidandclearlyυDA>−κ.Holdersof

27

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

Figure2.3.:GameTreefortheInfringementLitigationSubgame

validpatentswillpreferlitigationtoinactiononlyifthe(certain)courtoutcome
ofmonopolyprofitsnetofcourtcostsislargerthantheduopolyoutcomethey
wouldreceivegiveninaction,i.e.thefollowingconditionholds:

(2.2)

υ−υAD≥κ

If(2)isviolated,eventhepatenteewithavalidpatentcannotcredibly
threatentogotocourt,thereforeBwillnotbewillingtopayanycompensation
insettlementnegotiationsandSI=0.Thepayoffsofthetwopartiesare
πIA=υDAandπIB=υDB.Again,themoreinterestingcaseiswhen(2)is
satisfied.Here,itisprofitableforholdersofgoodpatentstosueforinfringement
damages,whileitisstillnotintheinterestofholdersofbadpatentstodoso.
Butasopposedtosubgame(C),nowitistheinformedpartywhodecides
whetherornottogotocourt.Unlikeinthepreviouscasethereforethereis
nopositiveprobabilityoflitigationtoinducepatenteesofthebadtypenot

28

Analysis2.4.

tomimicthegoodtypeandevenasemi-separatingequilibriumcannotbe
supported.Thereforeinequilibrium,noinformativesettlementofferexists,in
thesensethatitallowsBtoupdateherpriorprobabilityofpatentvalidity.
Goodpatenteesknowthattheywillreceiveagainof(υ−κ)−υADfromliti-
gationoverinaction.TheexpectedlossofBfromfailingtoreachasettlement,
ontheotherhand,isequaltotheexpecteddamagespaymentpluscourtcosts,
i.e.pB(υ−υAD+κ).Fromthiswederivethefollowingcondition:

(2.3)

pB(υ−υAD+κ)≥(υ−κ)−υAD

Ifthisconditionisviolated,thenholdersofvalidpatentsandthecompetitor
areunabletoobtainasettlementandthepatentholdersuesfordamages.
Theseconsiderationsallowustoformulatethefollowinglemma:

Lemma2:PerfectBayesianEquilibriumofsubgame(I):(i)if(2)isvio-
lated,nosettlementisreachedandAremainsinactiveinequilibrium.(ii)If
(2)and(3)aresatisfied,settlementisreachedwithcertainty.(iii)If(2)is
satisfiedand(3)isviolated,nosettlementisreached.Uponfailure,holdersof
validpatentssueandholdersofinvalidpatentsremaininactive.

Proof:(i)When(2)isviolated,thethreatoflitigationisnotcrediblefor
eitherkindofpatentee,thereforenonon-negativesettlementdemandwillbe
met.(ii)DefineasettlementofferSI∗[(υ−κ)−υAD,pB(υ−υAD+κ)].Then
thefollowingpairofstrategiesisaPerfectBayesianequilibrium:Bothtypes
ofAofferSI∗.BacceptsSI∗anddeclinesanyothersettlementoffer.Ifa
settlementofferhasbeendeclined,holdersofvalidpatentssueandholdersof
invalidpatentsremaininactive.14(iii)DenotegoodqualityA’ssettlementoffer
asSI+andbadqualityA’sofferasSI−,withSI≥0.Weknowthatgood
qualitypatenteeswilldemandnolessthantheircertaingainsfromlitigation,
sothatSI+≥(υ−κ)−υAD.NowassumethatSI+>SI−andBacceptsSI+
14Clearlyselect,thethereisequilibriumacontinmostuumofprofitableequilibria.toAInthewhenevfolloerwewing,asconsiderAcanpayoffsmakefromataktheeitorsubgames.leaveitoffer,we

29

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

withpositiveprobability.Thenbadqualitypatenteeswillwanttomimicand
deviatetoSI+.Thesameholdsfortheoppositecase,thereforeSI+=SI−in
equilibrium.Astheminimalsettlementofferofgoodinventorsislargerthan
theexpectedcourtoutcomeforB,i.e.SI=(υ−κ)−υAD>pB(υ−υAD+κ),
nofeasiblesettlementofferexiststhatBiswillingtoaccept.

ItisconvenienttosummarizetheresultsofLemma1and2inthetable
w.elobOutcomesπA(ι≥µ)πA(ι<µ)πB
(C)Case(1)violatednoS,noLυC*0
κ−υ(1)satisfiedall(mixed)υ−υB+κκυ−(υBC−κ)(1−pB)υBC−κ
(I)Case(2)violatednoS,noLυAD*υBD
(2)sat,(3)vioLifvalidυ−κυADυBD−pB(υ−υAD+κ)
(2)&(3)satSυAD+pB(υ−υAD+κ)*υBD−pB(υ−υAD+κ)

Table2.1.:Equilibriumoutcomesandexpectedprofitsofthelitigation
subgames(Sindicatessettlemensamet,Lprofits,litigation.theseWhenareholdersabbreviatedofinbvyalidanandasterisk.)validpatentsreceivethe

Inpassing,notethatwegetthefamiliarresultinthechallengesubgame(C)
thatweakerpatentsaremorelikelytoresultinthesettlementoutcome.More
interestingly,theinfringementsubgame(I)deliverstheoppositeresult-here,
iftheperceptionofpatentqualityishighest,settlementoccurs(bothforvalid
andinvalidpatents),whileintermediatepatentqualityletslitigationarise(only
forvalidpatents).Ananalysisfocusingononlyonetypeofpotentiallitigation
willnecessarilyoverlookthisfact.
ThecombinationofLemma1and2allowsustoderiveourfirstproposition:

Proposition1:InthePerfectBayesianEquilibriumofthelitigationsub-
game,Bwillpreferchallenginganexistingpatenttoinfringementandinaction
iffthefollowingDhold:C(i)Condition(1)issatisfied,(ii)condition(2)issatisfied
and(iii)pB>υυ−BυAD−υ−BυB+Cκ+κ.

30

Analysis2.4.

TheproofofthepropositionfollowsdirectlyfromtheprecedingLemma1and
2.Brieflynotethatif(1)isviolated,Bisindifferentbetweenalternative(C)
andinaction,andwhen(2)isviolatedBalwaysprefersinfringement,asthis
giveshimtheduopolypayoff.Condition(iii)ofthepropositionisperhapsthe
mostsurprising:Thisshowsthatonlyforpatentqualitypriorsaboveacertain
thresholdBprefersthechallengesubgameoverinfringement.Intuitively,as
thepatentqualityincreases,thehigherlikelihoodofhavingtopaydamages
makesinfringementrelativelylessattractive.
Thislowerthresholdincreases(i.e.challengesbecomelesslikely)asthe
costsofgoingtocourtκincrease-thisfavorsinfringement,whereitisthe
patentholderwhomustdecidewhetherornottoinitiatecourtproceedings.
Itfurtherincreasesinthedifferencebetweenduopolyandcompetitiveprofits
forB,υBD−υBC.ThehighertheprotectionthatBenjoyspassivelyfromA’s
patent,thelesslikelyheistochallengeit,asthiswouldthreatenhisowncozy
situation.Ontheotherhand,ifduopolyandfreeentryprofitsarerelatively
similar,sayinanindustrywithotherentrybarriers,thecompetitorismore
likelytochallengeexistingpatents,unlessstrategicconsiderationsoutsideof
ourmodelareatplay.Thepropositionalsocapturesthecentralfindingof
FarrellandShapiro(2008):Themorecompetitivetheindustryis,i.e.the
lowerυBC,thelesslikelyitisthatapatentwillbechallengedbycompetitors
andthemorepernicioustherolethatbadpatentscanplay.
Condition(1)givesusastraightforwardupperthresholdforpatentquality,
abovewhichBprefersinactiontothechallengingsubgame,ashisexpected
profitsfromthelatterbecomenegative.Thisthresholdisloweredasκincreases.
Thereby,thesetofpBshrinks.Forlargecourtcosts,thetwobounds(1)and(iii)
bypasseachotherandthesetofpriorsforwhichchallengestopatentvalidity
arisebecomesempty.15Thepriorregardingpatentqualitycanbeinterpreted
asameasureofpatentquality.Thereforewefindanalternativemechanismto
Chiou(2008)thatleadstotheresultthatcompetitorswillcontributetothe
controlofpatentqualityprivatelyonlyforintermediatepatentquality.Inour
15Thesetisgivenbyυ−υυBDAD−−υυBCBC++κκ<pB≤υυBCBC−κ.

31

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

model,ifthepatentqualityis“toohigh”,therewillbenochallengesandif
thepatentqualityis“toolow”.Ourmodelthereforeincorporatesthecentral
resultsfromFarrellandShapiro(2008)andChiou(2008).Letusnextconsider
theanaloguetoproposition1todiscernthecasesinwhichtheinfringement
arises:subgame

Proposition2:InthePerfectBayesianEquilibriumofthelitigationsub-
game,Bwillpreferinfringinganexistingpatenttochallengingitandinaction
iffoneofthefollowingholds:

(A)Condition(2)isviolated.

(B)(a)Condition(2)issatisfied,(b)υBD−pB(υ−υAD+κ)>0,and(c)
υD−υC+κ
pB<υ−BυAD−BυBC+κ

Again,theprooffollowsdirectlyfromlemma1and2.(B)issimplytheop-
positecasefromproposition1,where(b)ensurespositiveexpectedprofitsand
(c)letsinfringementbemoreattractivethanchallengingtheexistingpatent.
Thecase(A)appearsrathersimple,butitsignifiesoneofthepossibilitiesof
thepatentsystem“breakingdown”.Ifcondition(2)isviolated,courtcostsare
sohighinrelationtothemonopolyprofitsthatthe(certain)payoffofholders
ofvalidpatentsfromgoingtocourtislowerthanfromaccommodatingthein-
fringer.Asaresult,thecompetitorcaninfringewithimpunity.Thisproblem
isclearlyalleviatedwiththeBritishsystemofassigningcourtfees,inwhichthe
losingpartyhastocoverthewinner’scostsandthiswouldbeoneimportant
reformsuggestiontorejuvenatetheAmericanpatentingsystem.Butonehas
tobecarefulaboutatoonarrowinterpretationofκ.Forus,thistermdoesnot
onlycoverthefeesofthecourtsandlawyersthemselves,butinadditionalso
costsofthetimeforpreparationandthehassleofproceedings.Streamlining
andsimplifyingproceedings,asisthegoaloftheEPLA,wouldthereforebea
methodofreducingκevenintheEuropeanorBritishfeesystem.
Thesetwopropositionsallowustoconsiderhowtheprofitsofthevarious
playertypesdevelopastheaveragequalityofapatentdecreases,givenitsvalue

32

Analysis2.4.

Figure2.4.:Expected−profitsofpatenteeswithvalidpatents(π+A),invalid
patents(πA)andthecompetitor(πB),dependingontheprior
probabilityofpatent(in)validity.

υandthecorrespondingstructureofpayoffs.Figure2.4illustratesthemost
complexcase,forthecasethatcondition(2)isnotviolated.16Astheaver-
agequalityofthepatentdecreases(ortheshareofinvalidpatentsincreases)
startingat0,theequilibriumswitchesfromthepuremonopolycasetoachal-
lengingequilibrium(whencondition(1)issatisfied),thentotheinfringement
withsettlement(whencondition(iii)fromproposition1issatisfied),tothe
separatinginfringementcase(whencondition(3)issatisfied).Asdiscussed
above,theareabetween(1)and(iii)doesnotnecessarilyexist,asinfringement
candominatechallengingfromtheviewofthecompetitor.
Thefigurealsoallowsustoanalyzeunderwhichcircumstancesprivateefforts
bycompetitorsmayactascomplementstotheexaminationprocessofthe
PTOs.Thiswillbethecaseinasociallydesirablewayifthedifferentforms
oflitigationactasadeterrenttoholdersofnot-patentableideasspecifically,
i.e.iftheexpectedprofitoftheholderofaninvalidpatentisstrictlylower

16Ifcondition(2)isviolated,allpartiesobtainexpectedprofitsofυDA,B.

33

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

thantheexpectedprofitoftheholderofavalidpatent.Suchawedgebetween
theexpectedprofitsofholdersofvalidandinvalidpatentsonlyoccursinthe
sectionstotherightof(3),wherethepatentholdersareforcedtogotocourt
againstinfringers,andbetween(1)and(iii),wherecompetitorsweedouta
shareofthebadpatentsthroughchallenges.
Overall,theprofitsofpatentholders–bothvalidandinvalid–areweakly
decreasingintheshare−ofinvalidpatentsingeneral.17Inthefollowing,we
willfocusonthecase∂∂pπBA≥0,inordertoavoidhavingtodealwithmultiple
equilibrialateron.Allofthefollowingresultscanalsobederivedwithoutthis
assumption,aslongastheprofitsofbothkindsofpatenteesareweaklydecreas-
ingbeyondsomepBandreachtheirminimumbeyondthispoint,whichisalways
thecaseduetoourassumptionsregardingthespecificationofindustryprofits:
AsthepriorprobabilityofpatentvaliditypB→0,(πA−,πA+)→(υAD,υ−κ),
whichistheminimumleveltherespectiveexpectedprofitscanreach.Therefore
focussingonmonotonouslydecreasingexpectedprofitsgreatlysimplifiesproofs
inthefollowing,withoutaffectingthegeneralityoftheargument.
Figure2.5displaystheeffectsofanincreaseintheprivatevalueυofagiven
patentontheprofitsofthevariousplayers.Weobservetwothings:First,profits
increasewithineachoutcomeforallplayers.Second,theboundariesofthe
outcomesshift,mostnotably,(1)isrelaxedsothatthethresholdforchallenging
apatentdecreases.Theothercleareffectisthat(3)shiftsdownwards,so
thatfewercasesreachsettlementgiveninfringementanditismorelikelythat
infringerswillbesued.Theeffectontheboundarybetweenchallengingand
infringingpatentsisuncleargiventheassumptionsmadeanddependsonthe
signsandrelativemagnitudesofthechangesinυBD−υBCandυ−υAD−υBC.
Wecannoteasaresult,thatasthevalueofapatentincreases,patenteesare
lesslikelytoenjoythefullmonopolybenefitstherefrom.Further,ifpatentees
sufferlossesfromanincreaseinυ,thisiseitherduetothefactthatwhileless

17Theonlyexceptionisifthesumofcompetitiveandduopolyprofitsmorethanexceedsthemonopoly
profits,tobeprecise,wheneverυ<υAD+υBD+υBCυ−υυBDAD−−υυBCBC++κκ.Then,theholderofinvalidpatents
receiveahigherexpectedprofitsunderthepoolingandinfringementcasethanunderchallengesfrom
etitors.comp

34

Analysis2.4.

Figure2.5.:increaseChangesininthetheprivpayateoffvastruluectofuretheofpatenthetυ.litigationgamegivenan

profitablepatentswerenotchallengedmoreprofitablepatentsare,orbecause
whilelessvaluablepatentswerechallenged,morevaluableonessufferfrom
infringement.Mostimportantly,though,asnotedabove,theexpectedprofits
oftheholdersofvalidpatentsareboundedfrombelowatυ−κ(orυAD,if(2)
isviolated)andtheexpectedprofitsofholdersofinvalidpatentsarebounded
frombelowatυAD.Theseboundsarestrictlyincreasinginυ.

IncentivesforPatentingandIdeaGeneration

LetπA(pB,ι,υ)denotetheexpectedoutcomefromthelitigationsubgameforA
foragivenideaandgivenpriorbeliefsregardingthepatentqualityifgranted.
Obviously,Awillpatentthisideaonlyifthefollowingconditionholds:

(2.4)

φ(ι)πA(pB,ι,υ)+δ(Φ)υ≥τ1

Abenefitstwofoldfromapatentapplication.First,therearetheexpected

35

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

profitsfromobtainingthepatent,takingtheprobabilityofdoingsoφ(ι)into
accountiftheideaisnotobjectivelypatentable.Second,Areceivesthepatent
pendingstatustohisideainanycase,whichispotentiallyvaluable.Ifthesum
ofthesetwofactorsexceedsthecostsofthepatentapplicationτ1,Awilldecide
topatent.Atthisstage,therearetwopotentialsourcesofdifferencebetween
theexpectedgainsfrompatentingobjectivelypatentableandnon-patentable
ideas:

1)φ(ι<µ)≤1,i.e.nonpatentableideasare(weakly)lesslikelytoobtain
tpatena

2)andπ−A≤πA+,i.e.giventhatapatentwasawarded,thebenefitsderived
fromthepatentareweaklysmallerforholderoflatentlyinvalidthanfor
holdersofvalidpatents.Wedescribedaboveindetailwhenthissecond
inequalityisstrict,asonlyinthiscaseprivatelitigationiscomplementary
totheinspectioneffortsofthePTO.

Upuntilnow,wehavenotdiscussedwherethepriorprobabilityofpatent
validitypBoriginates.Weimposethatthisprioriscorrectinthesensethatit
resemblestheactualshareofinvalidpatentsinthepatentpopulation.Based
onthis,weareabletostatethefollowingLemma:

Lemma3:Foragivenstructureofmarketpayoffsandlevelofυ,eitherno
ideaispatented,everyideaispatented,oracutofflevelιˆ(υ)existswiththe
followingproperty:Foreachidea,ifι≥ιˆ,Aappliesforapatentandrefrains
fromdoingsootherwise.

Proof:Rearrange(4)toφ(ι)πA(pB,ι,υ)≥τ1−δ(Φ)υ.Notethattheright
handsideoftheinequalityisascalarforgivenvaluesofυ.Denotetheshare
ofvalidideasamongallideasforagivenυaspmin.Considerthelefthand
sideoftheinequality.AsπA+≥πA−anditisnon-decreasinginpB,obviously
ifπA+(pB=1,υ)<τ1−δ(Φ)υ,noideawillbepatented.Analogously,if
φ(ι=0)πA−(pmin,υ)≥τ1−δ(Φ)υ,everyideawillbepatented.Ourrequire-
mentwithregardtothepriorspBbeingcorrectmeansthattheoriginalprior

36

Analysis2.4.

probabilityoffacingavalidpatentpB(Φ,υ)isdeterminedbytheactualshares
inthepatentingdecision,1i.e.thefollowingconditionmustholdinequilib-
dι)ι(frium:pB(Φ,υ)=ιˆ1φµ(ι)f(ι)dι∀υ.ItissimpletoshowthatifBappliesfor
apatentforanideawithι1,andthereisasecondideawithequalvalueυ
andgreaterqualityι2>ι1,thenBwillalsoapplyforapatentforthesecond
idea–theexpectedvaluefromobtainingapatentisstrictlyincreasinginι.
Analogously,ifBderivesanegativeexpectedvaluefrompatentinganidea
withaobjectivequalityι1andthereisasecondideawiththesamevalueand
objectivequalityι2<ι1,i.e.φ(ι1)πA(pB,ι1,υ)<τ1−δ(Φ)υandι2<ι1,then
φ(ι2)πA(pB,ι1,υ)<τ1−δ(Φ)υandBwillalsodecidenottoapplyforapatent
forthesecondidea.Combiningthesestatementscompletestheproofofthe
Lemma.

Logicallythefollowingfourcasesmayarise:(i)itisnotevenprofitableto
patentvalidideas,(ii)itisevenprofitabletopatenttheleastpatentableideas,
(iii)itisprofitabletopatentinvalidideas,aslongastheirqualityisabovea
certainthresholdaswellasvalidideasand(iv)itisonlyprofitabletopatent
objectivelypatentableideas.

Figure2.6.:IllustrationofLemma3depictinganexampleinwhichall4cases
cor.cu

Figure2.6illustratesLemma3anddepictsanexampleforhowpatenting

37

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

decisionsaremadein(ι,υ)space.Itisworthwhiletoconsiderwhichofthe
depictedcasesaregenericandwhichonlyariseundercertaincircumstances.
Firstnotethatforeverypatentapplicationfeeτ1>0,case(i)willarisefor
lowenoughvaluesυofvalidideas,i.e.noideathatyieldsrelativelylowprivate
benefitswilleverbepatented.Nextletusfocusoncase(iv),inwhichonly
validideasareconferredtothepatentoffice.Thiscasewillariseifι→limµφ(ι)<1,
forthenthereisawedgebetweenthevalueofpatentableandnon-patentable
ideas.Notethatiftheshareofinvalidpatentsis“smallenough”πA+=πA−=υ,
onlyastheshareexceedsthethresholdgivenby(1)adifferencebetweenthe
twovaluesarises.Case(ii)occursifφ(ι=0)πA−(pB=0,υ)≥τ1−δ(Φ)υ,
thereforeitdisappearsifτ1islargeenough.Thiscaseisproblematicfroma
policyperspective,asthismeansthatthepatentofficeisunabletodeterany
badapplications.Itismorelikelytooccurforhigherprivatevaluesandlarger
returnsfrompendingpatents.Intuitively,ifobtainingapatentforanideawill
enableafirmtomakeanabsolutekillinginthemarket,thefirmisgoingto
takethecomparativelycheapgamble–itonlyincursthefixedpatentfee–of
applyingforapatent,nomatterhowsmallthechanceisofobtainingit.The
higherthegainsfromthepatent-pendingstatus,themorerelevantthiscase
becomes.(iii)arisesintheremainderofconstellations.Thefrontierin(iii)is
generallynotlinearasdepicted.18Duetotheshiftsintheboundariesofthe
litigationoutcomesasυincreasesasdiscussedabove,thefrontiercanevenbe
increasing,butonlylocally.
Finallytoclosethemodel,Ageneratesthe(n+1)thideaaslongasthe
expectedprofitislargerthanitscosts,i.e.thefollowingconditionholds:

11(2.5)0ιˆ(υ)

[φ(ι)πA(υ,ι)+δ(Φ)υ−τ1]f(ι)g(υ)dιdυ≥C(n+1)−C(n)Thereforealsopatentsthatshouldformallyandlegallynothavebeengranted
18InsteadthisresemblesacaseinwhichF(ι)isindependentofυandtheexpectedprofitsfromreceiving
apatentareeitherconstantoralinearfunctionofυ.Thisoccursforexampleifthesamelitigation-
equilibriumarisesindependentofυ.

38

Analysis2.4.

giveinventorsanincentivetoexertinnovationeffortsaslongastheygenerate
positiveexpectedpayoffs.Forexample,alaxerinspectionpolicywillresult
inthegenerationofmoreideas.Thisisanobvioussimplification,asitdoes
nottakesecondordereffectssuchaspatent-thicketsetc.intoaccount.Never-
theless,wehavecreatedarelativelycomplexsystem.Forthisgeneralsetting
withgeneraldistributionfunctions,aclosed-formsolutiondoesnotexist.But
despitethiscomplexity,inthefollowingsectionwewillbeabletogeneratea
numberofimportantinsightsconcerningtheproposedpiecemealreformsand
adaptationsofexistingpatentregimes,focussingonpointingouttrade-offsand
problemsthathavebeenmainlyoverlookedsofar.

2.4.2.ComparativeStatics-PotentialApproachesto
Reform

eesFApplication

OneobviousdifferencebetweentheEuropeanandtheUSpatentregimeis
thesignificantdifferenceinfeesthatareimposedforpatentapplicationand
maintenance.Variousauthors,mostrecentlyBessenandMeurer(2008a)and
Chiou(2008)havesuggestedtoincreasepatentingfeesinordertodeterlower
qualityinventorsandraisetheaveragequalityofpatentapplications.
Oursetupallowsustodissecttheresultsofsuchareforminanextremely
straightforwardandsimplemanner.Letusconsiderachangefromτ1toτ1.This
affectsthepatentingdecisionattwomargins.Asshownabove,thelefthand
sideoftheconditionthatimplicitlydefinesˆι,i.e.φ(ι)πA(pB,ι,υ)+δ(Φ)υ≥τ1,
isstrictlyincreasinginιgiventhatι<µforeachvalueofυ.Thereforeifone
addsaconstanttotherighthandside,thecutoffvalueιˆ(υ)mustincreaseas
well,whichresemblesashiftofthefrontierinFigure6totheright.Wewill
denotethenewcutoffvalueasιˆ(υ).Intuitively,thisisthedesiredeffect,as
fewerinvalidideasaresubmittedtothePTOofficeforscrutiny,andcaptures
theintentionofthoseproposingthiskindofreform.
Thereisaverysimplesecondeffectthough,onethatdirectlyaffects

39

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

patentableideasonly.Considerthecaseinwhichonlyvalidideasarepatented,
thereforepB=1andπA+=υ.Herethepatentingdecisionisdefinedby
(1+δ(φ))υ≥τ1,whichdirectlyimpliestheminimalprivatevalueofanidea
thatispatentedisυ=1+δτ1(Φ).Anincreaseintheapplicationfeeτ1movesthis
frontierupward,whichistheeffectatthesecondmargin.Thissecondeffectis
extremelyundesirable,asitprecludesadditionalobjectivelypatentableideas
frombeingpatented,becausethecostsoutweightheprivatebenefits.
Further,botheffectsleadtoastrictdecreaseintheexpectedprofitstobe
derivedfromgeneratingnewideasandshouldthereforeleadtolessgenerated
ideas(andpatentapplications)ingeneral.Asashortsidenote:Duetothe
staticcharacterofourmodel,weareoverlookingafurtherpotentialeffect:
Thesmallernumberofapplicationsmayleadtoadecreaseinthedelaybe-
tweenapplicationandpatentingandtherebyreducethebenefitsfrompending
patents,whichwouldbeafurtherpositiveeffect.Arguably,thisshouldnot
happenintheshortrun,though,duetothe“stockpile”ofapplicationsthathas
accumulatedovertheyearsandstillhastobedealtwith.Wesummarizethese
considerationsinthefollowingproposition:

Proposition3:Anincreaseinthepatentapplicationfeefromτ1toτ1leads
toanincreaseintherelativeshareofvalidapplicationsonlyifthefollowing
holds:

υµf(ι)g(υ)dιdυυµf(ι)g(υ)dιdυ
1111
υιˆf(ι)g(υ)dιdυυιˆf(ι)g(υ)dιdυ
(2.6)11−11>0

Proof:Thepropositionfollowsdirectlyfromlemma3andthedefinitionof
.υWhentheconditioninthepropositiondoesnothold,theshareofvalid
patentsactuallydecreaseswithachangeoftheapplicationfee!Asthecondition
isnotnecessarilyintuitiveinthisform,wemaybenefitfromconsideringthe
marginaleffectsonly.Fromourpreviousdiscussionitisclearthatachangein

40

Analysis2.4.

τ1willaffectthemassofvalidaswellasthetotalmassofapplications.Letus
denotetheoriginalshareofvalidpatentapplicationasγ(τ1)andthetotalmass
ofapplicationsasα(τ1).Thensimplecalculusallowsustoderivethefollowing
muchmoreinterpretablecorollarytoproposition3:

CorollarytoProposition3:Themarginaleffectofachangeoftheap-
plicationfeeτ1toτ1ontherelativeshareofvalidapplicationswillbepositive
onlyifthefollowingholds:

(2.7)

|α|>(1−1F+(δµ))(Φ)g(υ)(γ(τ1))−1

Fromthisonecaneasilydeterminethattheaveragequalityofpatentap-
plicationsismorelikelytodecreaseforincreasesintheapplicationfee,ifthe
densityofpatentswiththemarginalprivatevalueυisrelativelyhigh.This
stemsfromthefactthatonthelowermarginonlyvalidideasarepatented.
Thiseffectmechanicallybecomesstronger,thelowertheoriginalpatentqual-
ityis,asthedecreaseinvalidpatentsthenhasastrongerweight,aswellasfor
laxerpatentregimes(lowerlevelsofµ).Asadirectresultofthis,inregimes
witharelativelyhighaveragequalityofpatentapplicationssuchastheEuro-
peanpatentoffice,adecreaseofthepatentingfeecouldincreasetheaverage
qualityofapplications.Ontheotherhand,fortheUS,inwhichthecomplaints
aboutthequalityofapplicationsarewidespread,anincreaseinthepatenting
feeshouldbelessproblematic.
Intuitively,thereisakindof“positiveselection”atworkhere.Asthepri-
vatevalueofpatentsdecreases,thequalitycutofflevelforwhichpatentsare
profitableincreases,untilatthemarginonlythebestpatentssurvive-ifan
inventorexpectsapatenttobeofrelativelylowprivatevaluetohim,hemust
berelatively(orcompletely)sureofitspatentabilitytobewillingtoincurthe
costsofpatenting.Raisingpatentfeesthereforegenericallykillsoffhigh-quality
patentapplicationover-proportionally.

41

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

CostsLitigation

VariousdevelopmentsintheUSandEuropehaveaffected(orwillaffect)the
waypatentlitigationfunctions.IntheUS,theevolutionofaspecializedap-
pellatecourthasreducedlegaluncertainty(whichcanbeinterpretedasaform
ofcostinoursimplifiedmodel)andpossiblyledtooratleastencouragedan
increaseinpatentlitigation.19Concurrently,inEuropetheEuropeanPatent
LitigationAgreementhasbeenseekingtoestablishaunifiedEuropeanPatent
Courtwiththeexpressedgoalofreducingpatentlitigationcosts.20Ourmodel
setupallowsustodissecttheeffectsofsuchadecreaseinsomedetail.
Letusfirstconsiderthechangesinthelitigationsubgame.Wecollectour
qualitativeresultsinFigure2.7.Againweobservetworelatedkindsofshifts:
Thefirstaffectsthethepayoffsoftheindividualplayersdirectly,andthesec-
ond,asaresult,movestheboundariesbetweenthedifferentlitigationscenarios.
Intheinfringementgame,wegetthe(wellestablished)result,thatthelike-
lihoodofsettlementswilldecreaseifthelitigationcostarereduced.Froma
competitionpolicyandwelfarestandpoint,thisisahighlydesirableresult,as
itreducesthescopeforcollusion.Inourfigure,thisisrepresentedbyashiftof
boundary(3)totheleft.
Next,thereductionmakeschallengingexistingpatentsmoreattractiveto
thecompetitor.Thiseffectisunambiguousatbothmargins.Forrelativehigh
priorsofpatentquality,challengesoccurwheretheydidnotbefore.Thisis
representedbyashiftoftheboundaryat(1)totheleft.Butalsoatthe
marginbetweenchallengingandinfringement/settlement,theeffectisclear,
withchallengingbecomingrelativelymoreattractivecomparedtoinfringement.
Thereforetheboundaryat(iii)movestotheright.Eachoftheseeffectsis
arguablydesirablefromasocialwelfarepointofview.
Thepayoffsofthevariousplayer-typesdevelopinaninterestingpattern.The
competitorisweaklybetterofascourtfeesdecrease–thethreatoflitigation

19SeeCook(2007)formoredetails.FurtherseeBessenandMeurer(2008a)foraverycriticalassessment
oftherecentdevelopmentsofthepatentjudiciaryintheUS.
20Seeforexamplehttp://www.epo.org/patents/law/legislative-initiatives/epla.htmlfordetails.

42

Analysis2.4.

Figure2.7.:Changesinthepayoffstructureofthelitigationgamegivenade-
creaseinthecostsofgoingtocourtκ

bearsslightlylessweightintheinfringementcase(whichalsomakessettlement
lesslikely)andthepayoffofachallenge,whethersuccessfulorunsuccessfulis
strictlyhigher.Thechangetotheexpectedpayoffofaholderofavalidpatent
issomewhatambiguous.Giventhathispatentischallenged,thelowercourt
feesmakehimbetteroff,yethewillbechallengedinmorecases.Itisthiseffect
thatBessenandMeurer(2008a)worryaboutinthefaceoflitigationreformin
theUS.Patentlitigationbecomesmorelikelyalsointheinfringementsubcase,
asarguedabove-assettlementisnotnecessarilydesirablefromanantitrust
perspective,ashasbeenbroughttoattentionbyShapiro(2003),thiscouldbea
beneficialeffect,butitdecreasestheexpectedpayoffofpatentholders.Finally,
inthecasesinwhichtheholdersofvalidpatentshavetosuetoprotecttheir
investment,i.e.totherightof(3),thedecreaseincostsstrictlyincreasestheir
profits.

Theeffectsontheprofitsofholdersofinvalidpatentsarelessambiguous.The
factthatmorepatentswillbechallengedstrictlydecreasestheareainwhich
poolingequilibriaarise:Theshifttotheleftof(1)impliesthatmorepatents

43

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

willbesubjecttochallenges,fromwhichbadpatentssuffermorestronglythan
goodones.Further,theshiftof(iii)totherightimpliesthatpatentsthat
wouldotherwisehavebeeninfringedandsettledarenowsubjecttochallenges
aswell.Whiletheeffectontheprofitsofholdersofbadpatentsinthisarea
mightevenbepositive,asdiscussedabove,neverthelessthisincreasesthearea
inwhichawedgeexistsbetweengoodandbadpatentsandtherebyincreases
thelikelihoodthatthecourtsmaybeabletoaidtheworkofthePTO.Finally,
theshiftoftheboundary(3)totheleftimpliesthatholdersofgoodpatents
mustdefendmoreclaimsincourtthattheywouldotherwisehavesettled.As
theholdersofbadpatentsnevergotocourt,thisreducestheirincomefrom
settlementsandagainincreasesthewedgebetweenthepayoffstothetwotypes.
Tosummarize,thisanalysisshowsthatstreamliningcourt-proceedingscan
bepoisonormedicine,dependingonwhatyoubelieveailsthepatentsystem.
If,likeBessenandMeurer(2008a)onebelievesthatthemainproblemofthe
systemistheerosionofpayoffstopatentingfromtoomuchlitigation,thenthe
effectofloweringcostsisunclear–notstrictlynegative,asonemighthave
expected.Ontheonehand,itshouldleadtomorelitigation,butontheother
handthecoststopatenteesdecrease,sotheneteffectisunclear.Atthesame
time,thereformstepsinthisdirectionunambiguouslystrengthentherolethat
litigationandcourtscanplayinsiftingoutbadpatentsfromthesystemand
mightthereforereducetheburdenonthePTO.

Extension:DiligenceofthePatentOffices

TheeffectsofanincreaseinthediligenceofthePTOappearobviousatfirst
sight.Anincreaseintheinspectionintensityshouldleadtomorebadappli-
cationsbeingrejected,sothatitbecomeslessattractivetotrytopassofbad
ideasaspatentable.Consideranadjustmentofthepatentoffice’spolicysuch
thatΦchangestoΦandtherebyφ(ι)<φ(ι)∀ι<µ.Byitself,thiseffect
wouldbepurelybeneficial.Butasassumedabove,thisimprovementcomesat
acost,asthereturnsfrompendingpatentsincreasetoδ(Φ)>δ(Φ).Note
thatanincreaseoftherequireddiligenceinthePTO’sassessmentswouldin

44

2.5.okOutloandDiscussion

realitynotonlyaffectnewpatentapplicationsbutalsotheexistingbacklogof
pastapplicationsthathasaccumulated.
Revisitingcondition2.4,wehadestablishedthefollowingregardingthede-
cisionwhetherornottoapplyforapatentwithagivenidea:

(2.8)

φ(ι)πA(pB(Φ),ι,υ)+δ(Φ)υ≥τ1

Theeffectsofthisreformthenarefarlessclearcutthanwewouldhave
expected.Notefirstthatinanycasethecutofflevelυdecreases,asthereturn
fromvalidpatentsisincreasinginδ,whichmaybeconsideredapositiveside-
effect.Theoveralloutcomeontheothermargindependontherelativesizes
inthechangesinpatentingprobabilityandprofitsreceivedduringthepending
phase.Insteadofover-stretchingourmodelatthispointwithregardstothe
assumptionswehavemade(forexamplethelinearityofpending-revenuesin
theprivatevalueofanidea),wewouldliketopointoutthattheprofitsduring
thependingphasehaveafundamentaleffectonthecurrentpatentingsystem,
yettherearenexttonoempiricalfindingswithregardtothistopic,whilethey
needed.sorelyclearlyare

2.5.DiscussionandOutlook

Thecurrentregimeofissuingandenforcingpatentsandtheresultingrightsof
patentownershasfacedincreasinglyharshcriticismintherecentpast.There
havebeencallsforcompletelyabolishingthesystemoratleastcarryingouta
fundamentalreformthereof.Duetotheenormousinvestmentsoftheholdersof
existingpatents,itappearsrelativelyunlikelythatsuchafundamentalreform
isgoingtohappensoon.
Whatappearsfarmorelikely,instead,isacontinuationofthecurrentpraxis
ofminorpiecemealimprovements(orratherchanges,tophraseitmoreneu-
trally).Weaddtotheunderstandingoftheeffectsofthreeofthemostconsid-

45

2.PrivateProfitsandPublicBenefits-HownottoReformthePatentSystem

ered–orcurrentlybeingimplemented–steps:Anadjustmentoftheapplication
fees,simplifiedandcheaperadjudicationofpatentconflictsandfinallyapolicy
ofstricterscrutinybythePTOs.Ourapproachisnewintwocentralregards
-first,wetakethreemajoraspectsofthecurrentsystemintoaccountinour
theoreticalmodel,i.e.policysetting,theroleofthePTO,andtheinterplayof
patenteesandpotentialchallengers/infringersinthemarketandinthecourts.
Second,wetakeintoaccountthatideascomeinaplethoraofformandshape
andthattheymayvarycontinuouslyintheirprivatevaluetothepatentee
andintheirpatentability(whichtoacertainextentreflectstheirvaluetothe
y).unitcommFromourapproach,wegainedbothnewinsightsandnewquestions.Allow-
ingthecompetitortobothchallengeorinfringeuponanexistingpatent,we
foundthatbelowacertainthresholdofpatentquality,thecompetitorlosesher
interestinpolicingpatentqualitythroughchallengesbutwillprefertoinfringe
upontherightsofthepatenteeinstead.Thiseffectisstronglyexacerbatedin
industriesinwhichtherearerelativelyfewbarrierstoentry-apatentdoesnot
onlyprotectitsholder.Inthiscontext,thereisclearlyathinlinebetweentol-
eratinganinfringerandanticompetitivesettlementsofweakpatents-focusing
onexplicitagreementsfromanantitrustperspectivedoesnotdojusticetothis
issue.Allowingforcontinuoustypesallowedustoderiveasurprisingresultwith
regardtoanoftencontemplatedreformstep:Raisingapplicationfeesmayac-
tuallydecreasetheaveragequalityofpatentapplications.Thiswillespecially
bethecaseifthedistributionofprivatepatentvaluesisskewedtowardslower
values-whichappearstoreflectrealityverywell-andifthestandardsfor
patentingarerelativelylax.Further,weshowedthatadecreaseinlitigation
coststhroughspecializedcourts(oracentralizedcourtinthecaseoftheEU)
mayactuallymaketheworkofthepatentofficeeasier,byforcingawedgebe-
tweenthepayoffsfromgoodandbadpatents(orincreasinganexistingwedge).
Furtheritcandestabilizetheinfringement“agreements”discussedabove.
Finally,weintroducedaproblemintooursetting,whichissofarverylittle

46

2.5.okOutloandDiscussion

understood-patentsyieldbenefitstotheirownersbeforetheyareapprovedby
thePTO.Priortothedecision,evenapplicationsthatarelaterrejectedmay
beofconsiderablemonetaryorstrategicvaluetotheirowners.Westrongly
appealtoempiricaleconomistsworkinginthisfieldtohelpbetterunderstand
theissuesconnectedtothe“patentpending”stamp.Thesecondempirical
issueraisedistheinterplaybetweenmarketstructure(e.g.barrierstoentry)
andtheoutcomesofpatentdisputes.Ourmodelpredictsthattheopennessof
anindustryshouldbenegativelyrelatedtotheprobabilityofapatentbeing
hallenged.cTheapproachofthisarticlewastomakerelativelyfewassumptionsandsee
whichgeneralresultscouldbederivedtherefrom-clearly,onelogicalnextstep
wouldbetobreathemorelifeintothemarketsettingbyimposingtestedforms
ofcompetitionthatarewellunderstoodinthespiritofFarrellandShapiro
(2008)togetherwithspecificdistributionsinordertobeabletoworkwith
closedformsolutions.Webelievethatthismoreflexiblesettingisavaluable
complementtotheexistingliterature,asitenabledustopointoutanumber
oftradeoffsthatareworthstudyinginmoredetail.

47

3.Individual(ir)rationality?Behavior

inanemergingsocialonline-network

ductiontroIn3.1.

ationMotiv3.1.1.

Actionsandinteractionsinsocialnetworksinvolvehighlycomplexdecisions.
Theresultingpayoffscanbeimmense,suchasacquiringajobofferthatone
wouldotherwisenothavereceivedorfoundinglong-lastingmutuallybeneficial
partnerships.Regardingthetiming,thesebenefitsordetrimentscanbeim-
mediateorgradual,lastingorshort-lived.Theirsourcemaybelinkedtoan
individualdirectlyorthroughanynumberofintermediaries.Whiletheim-
portanceofnetworksinhighlyrelevantareas,suchasthespreadofmedical
knowledge(seeColemanetal.(1957))orthelikelihoodofobtainingadesirable
occupation(seeGranovetter(1974)),hasattractedtheattentionofsocialscien-
tistsmanydecadesago,researchersforalongtimegrappledwiththeempirical
andtheoreticalproblemsraisedbytheircomplexity.1
Thelastdecadehasseenmajorbreakthroughsinthewayeconomictheorists
thinkaboutandapproachthesubjectofnetworks.WorkssuchasJacksonand
Wolinsky(1996),BalaandGoyal(2000)andJacksonandWatts(2002)have
mademanyoftheseconceptstractableinagraph-theoreticsetup.Inthese
articles,rationalagentsmaximizethepayoffsgeneratedfromlinkstoothers,
whichinturnallowspredictionsconcerningthestructureoftheunderlying
networks.Yetinmanysocialnetworksoneencounterspotentiallyemotionally

1SeeGranovetter(1976)foranearlydiscussionontheempiricalissuesraisedbythedimensionalityofdata.

48

ductionotrIn3.1.

chargedterms,suchas“friendships”insocialonline-networksor“partnerships”
inbusinessrelations.Consequentially,amorerecentstrandoftheliterature
hasrealizedthatconceptssuchasinequalityaversionortrustmayplayan
importantroleininteractionsbetweenagentsinsocialnetworks.Experimental
economistshavethereforeintroducedtheseconceptsintosimplenetworkmod-
elsandtestedtheirbehavioralpredictionsunderhighlycontrolledsettings,
providingevidencethatoftenstronglycontradictsthe“rational”predictionsfor
equilibriumnetworkstructure.2
Thecontrasttothesecontrolledsettingshighlightssomeofthechallenges
thatnon-experimentalempiricalworkonindividualbehaviorinsocialnetworks
faces:e.g.,itisimpossibletodisentanglethemotivesofindividualpersonsor
entitiesenteringintorelationships,costsandutilitiesofdecisionsaremostly
unobservableandmaydiffersubstantiallybetweenindividuals,etc.Mostim-
portantly,though:Thereisnexttonodatathatcombinesmicro-(individual
decisionsandcharacteristics)andmacro-level(structureofthenetworkoverall,
locationsoflinksinsideanetwork)observations.
Ourstudyaimsexactlyatthisgap,usingacompletelyuniquedata-set.We
wereabletoobtaindatafromanemergingonline-socialnetwork:ontheone
handonuserbehaviorofmorethan30,000registeredusers,forexamplewith
regardtoprovisionofapublicgoodtypeeffort,numberofprivatemessages
sent,numberofpubliccommentsposted,entryintonewandseveranceofex-
isting“friendships”,andontheotherhandonthestructureoftheunderlying
networkitself,i.e.numberandlocationoffriendshipsbetweenusersandoverall
networksize.3Weusethisdatafirstofalltoperformwhatcanbeconsidered
“basic”research:Addressingthechallengesbyexperimentalstudies,wedemon-
strateclearlythatusersbehaveinwaysthatarecompatiblewiththerational
paradigm.Basedonthesefindings,weaddresstwofurtherissuesthatcould

2SeeKosfeld(2003)forasurveyoftheearlyliteratureinthisvein.
3Aspartoftheagreementconcerningthedata,thenameofthecompanyanditsserviceisnotusedinthis
study.Alluser-relatedinformationwasprovidedtousinawaytocompletelyassureusers’confidentiality
andprotectionoftheirpersonaldata.Wewereonlyprovidedwithrandomizeduser-IDsthatcannotbe
linkedtoindividualsinanywayandwehavenoaccesswhatsoevertoindividuals’personalcharacteristics.

49

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

barelybestudiedempiricallysofar:
Thefirstistheprovisionoflocalpublicgoodsbyusers.Usersinthenet-
workcanprovideandorganizeinformation–asdescribedbelowindetail–
twokindsofactivityfromwhichboththeythemselvesandtheirdirectfriends
deriveutility.Standardtheory,suchasBramoulléandKranton(2007)inthe
specificcontextofanetworkwithinformationprovision,wouldleadustoex-
pecttofindevidenceoffreeridinginthesensethatusersshouldprovideless
effortthemselvesiftheyhavefriendswhoprovidemoreeffort,allelsegiven.
Interestingly,wefindonlyweakevidenceforthistypeoffreeriding.Instead–
andfarmorepronouncedly–usersseemtoreactinapositivelyreciprocalway
totheirfriends’effortprovision-thatistheyreacttotheirfriends’provisionof
effortbyexertingmoreoftheirown,aphenomenonthathasbeennotedbefore
inthecontextofpeer-to-peernetworksamongothersbyGuetal.(2009).4
Thisleadsustothesecondissueofourstudy:Direct,negativelyreciprocal
behaviorasanenforcementdevice.Studyingtheformationandstabilityof
individuallinks,wefindstrongevidencethatusersarewillingtoincurcosts
toactivelypunishfreeridingbytheirfriendsbysevering(otherwisepurely
beneficial)relationships.
Ourfindingsarehighlyrelevant,astheprovisionofpublicgoodsinWeb2.0
servicesiscreatingtremendous,constantlyincreasingvalues.Microsoft’spar-
tialacquisitionofFacebookpricedthelattercompanyasawholeatmultiple
billiondollars5-whichissomewhatsurprising,becauseapartfromtheinfras-
tructure,theentirecontentofthesocial-networkingsiteaswellasmostapplets
arecreatedandprovidedbyusers.Thesameholdstruefor3otherpurelysocial
networksthattogetherwithFacebookform4ofthe20mostfrequentedsites
ontheinternet.6Ifonefurtherconsidersmassivelypopularsiteswithastrong
social-networkcharactersuchasYoutube.comorflickr,itbecomesobviousthat
socialonline-networksareencroachingonmanypeople’sdailylivesandthat
amajorbuildingblockofthisdevelopmentisfreeuser-providedcontent.In
54Foranintroductiontotheeconomicsofreciprocity,werefertoFehrandGächter(2000).
6See,TheforsitesexaaremplehMySpace,FaceBook,ttp://www.time.com/time/busiOrkutandHi5.com,accordingness/article/0,8599,16756toalexa.com58,00.hasoftmlApril2010.

50

ductionotrIn3.1.

thecourseofthisarticle,weseektoprovideneweconomicinsightsintothe
behavior-andpotentialmotivationandpayoffs-ofindividualsinthesekinds
orks.wnetofWeproceedasfollows:Inthefollowingsubsectionwepresentthemostclosely
relatedliterature.Firstthetheoreticalfoundationsofthetheoryofsocial
networks,thenthe“experimentalattack”andfinallywebrieflycoverexisting
empiricalstudies.InSection2weintroduceanddescribetheonline-network
wewishtostudyaswellasthefeaturesthatmakeitespeciallyinteresting
fromaneconomicstandpoint.Section3relatesthesefeaturestosomesimple
theoreticalconceptsthatwillformthebasisforourfurtherinquiry.Section4
containsthedatadescriptionandourempiricalanalysis.Section5concludes
andproposesresultingnewvenuesforfutureresearch.

3.1.2.RelatedLiteratureandContribution

Theliteraturerelevanttoourquestionscanbeseparatedintotheoretical,ex-
perimentalandempiricalstudies.
Fromthetheoreticalperspective,asetofbynowcanonicalarticleshas
enabledeconomiststocometotermwiththecomplexityofthesubject,allowing
themtomakepredictionsaboutthestructureofnetworksandtojudgetheir
efficiency.PerhapstheinitiatorsofthisdevelopmentareJacksonandWolinsky
(1996).Theirgraphtheoreticframework,depictingindividualsinanetwork
asthenodesinagraphandtheconnectionsorlinksbetweentheseindividuals
asthearcs,hasbecomethede-factostandardapproach.Individualsderive
utilityfromthelinkstheyareinvolvedin(andindirectlyalsofromthelinks
ofthosetheyareconnectedto),whileitiscostlytomaintaindirectlinks.A
networkis“stable”iftheutilityofagentscannotbeincreasedbycreatinganew
orseveringanexistinglink.Thepredictionsconcerningstablenetworksfrom
thismodelrangefromcompleteinterconnection(highbenefitstocostsratio)
overa“starnetwork”,inwhichonenodeisconnectedtoeveryoneelse,while
theperipheralnodesmaintainonlyasingleconnectioneach,toacompletely

51

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

disconnectednetwork(lowestbenefitstocostsratio).BalaandGoyal(2000)
enrichthesetupbyexplicitlymodelingthelinking-strategiesofplayersina
non-cooperativegameanddistinguishingbetweencasesinwhichinformation
(andtherebybenefits)flowinonlyoneorbothdirections.Inadditiontothe
stablenetworkstructuresmentionedabove,theyfindthatasymmetriccircle
orwheelnetworkinwhicheachagentislinkedtoexactlytwootherplayers.
Offurtherimportanceintheirsetupistheconceptofdecayofbenefits-the
moreintermediatelinksseparatetwoindirectlyconnectedplayers,thelower
thebenefitstheyderivefromeachothers’information.7
Thereareanumberofimportantextensionstothesebasicmodels.Goyal
(2005)notesthatdepictingtheexistenceofalinkbetweenresearchaffiliates,
forexample,asapurelybinaryvariablemayoftenbetoomuchofasimplifi-
cation.Instead,partnerscaninvestspecificallyintothestabilityofindividual
links,sothatheterogenouslinkstrengthsmayariseinequilibrium.Blochand
Dutta(2009)adoptthisconceptandmodelitexplicitly.Theyfindthatunder
relativelybroadconditions,thestableandefficientnetworkagainwillhavea
structure.starThreepapersofwhichweareawareoftrytomodeltheprovisionofpublic
goodsinsocialnetworks.InCho(2006)ex-antesymmetricagentsagreetoa
bindingcontractthatcoversboththeamountofalocalpublicgoodeachagent
providesaswellastheformofthenetworkthroughwhichthebenefitsofthe
publicgoodaretransmitted.Inthissetup,equilibriaexist,inwhichagents
almostsurelyprovideefficientlevelsofeffortiftheyarepatientenough.In
BramoulléandKranton(2007)agentsonlyspecifytheireffortlevels,taking
thestructureofthenetworkasgiven.Effortprovidedbyanagent’sneighbors
isasubstituteforherownefforts.Theyshowthatundertheseassumptionsfree
ridingexistsineverysocialnetworkinequilibriumandaddinglinksbetween
playersreducesindividualincentivestocontribute.InthemodelofO’Dea
(2008),dependingontheoutcome,onecanfindbothequilibriainwhichagents

7Foranexcellentsurveyofthetheoreticalliteratureonnetworkdevelopmentandstability,seeJackson
(2004).

52

ductionotrIn3.1.

voluntarilyprovidepublicgoodsandequilibriawhichresemblecostsharingfor
anexcludableclubgood.
Theseclearanddistinctpredictionsconcerningnetworkformationandstruc-
turehavepiquedtheinterestofanumberofexperimentaleconomists,who
findsignificantdeviationsfromtherationaldepictions.FalkandKosfeld(2003)
directlytestthehypothesesdevelopedintheBalaandGoyal(2000)modeland
showthatinsettingsinwhichtheNash-equilibriumprovidesforsymmetric
payoffsandlittlecoordinationefforts,thepredictionsarerelativelyaccurate.
Yetinsettingsthatshouldleadtostarnetworks,withrelativelyhighcostsper
linkandhighinformationbenefits,therationalequilibriumpredictionshave
almostnoexplanatorypower.Theyshowthatthisisonlytoasmallextent
explainedbythehigherrequirementswithrespecttocoordination.Instead
theyarguethatstarnetworksprovideuserswithextremelyunequalpayoffs
andthereforethedeviationfromtherationalequilibriummaybeexplainedby
subjects’inequalityaversion.CagnoandSciubba(2007)attempttodissect
thevariousdriversofindividualdecisioninnetworkformation.Theyidentify
twoconflictingdrivingforces:Optimizingbest-responsebehavioronlyexplains
theobservedbehaviortoacertaindegree.Insteadtheyfindthatreciprocity
andinertiaplayimportantexplanatoryroles.Inahighlysurprisingstudy,
CagnoandSciubba(2008)combineanetworkformationexperimentwitha
trustexperiment.Theyfindthatthelevelsoftrustinatrustgameplayed
afteranendogenousnetworkformationgamearesignificantlylowerthanin
thesimpletrustgame.Subjectsherebytendtotrustplayerslessonaverage
withwhomtheyhaveinteractedpreviouslythantheytrustcompletestrangers.
Theauthorsdeducefromthisthatpotentiallyexistingreciprocalbehaviorin
thenetworkformationgamedoesnotfostertrust.Thisindicatesthatthere
arepsychologicalmechanismsatplayinnetworkinteractionthatstillhaveto
explored.ebNotallexperimentalstudieshavebeenpurelydetrimentaltotherational
behaviorparadigm.CorbaeandDuffy(2008)testtherationalbehaviorhy-
pothesisinratherspecificexperimentalsetting,inwhichindividualscanform

53

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

risk-sharingpartnershipsthroughanon-cooperativeproposalgamewithid-
iosyncraticrisk.Theyfindthatindividualbehaviorisgenerallywellpredicted
byrationaltheory,whichagentachievingnear-efficientoutcomes.Goereeetal.
(2008)analyzeasettinginwhichagentswithheterogeneouslinkingcostscan
formlinksunderdifferentinformationalsettings.Whileforhomogenousagents
equilibriumpredictionscannotbevalidated,theheterogeneousagentpredic-
tionsarewellsuitedtoexplainindividuallinkingbehavior.Interestingly,star
networksdodevelopinthisexperimentalsettingasfrequentlyasexpected.

Thereisalargenumberofempiricalstudiesregardingnetworksingeneral
thatmainlyfocusonlocatingandquantifyingnetworkeffects,suchaspromi-
nentlySalonerandShepard(1995)orGowrisankaranandStavins(2004).For
asurveyofthisstrainoftheliterature,wereferthereadertoBirke(2008).
Thereisonlyalimitedamountofempiricaleconomicresearchonsocialnet-
works,though.PromptedbyNewman(2001)wholooksatthepropertiesof
researchcollaborationsfromaphysicist’sperspective,Goyaletal.(2006)and
RosenblatandMobius(2004)bothanalyzethestructureofresearchnetworks
ineconomics.Goyaletal.(2006)findacollaborationstructurecomposedof
anumberofstars(authorsthatwritearelativelylargenumberofpapersto-
getherwithcoauthors)andalargeperiphery(composedofauthorswhoonly
publishedpaperstogetherwithoneofthestar-authors),whichsuitsthestruc-
turalpredictionsoftheoreticalmodels.Basedonamodelthattheydevelop,
RosenblatandMobius(2004)lookattheriseoftheinternetandelectronic
communicationsasakindofnaturalexperimentwhichloweredthecostsof
communicationbetweenpotentialcollaborators.Theyfoundthatthedegrees
ofseparationbetweenindividualsonaveragedecreaseandeachauthorismore
likelytocoauthorapaperwithadistantauthorfromasimilarfield.Onthe
otherhand,thelikelihoodofcoauthoringapaperwithacoauthorfromadis-
similarfieldsignificantlydecreases.Further,KossinetsandWatts(2006)use
theemail-communicationsinsidealargeUScollegetodrawupasocialnetwork
structureanddiscovervariouscharacteristicsthatdeterminethelikelihoodthat
twoindividualswithinthecollegewillbelinkeddirectly.

54

ductionotrIn3.1.

Asmallwaveofmorerecentyettobepublishedpapersproceedsinadirection
thatisalmostahybridbetweeneconomicandinformationsystemsresearch-
intheirsubject,thesearemostcloselyrelatedtoourpaper,eventhoughthey
stronglydifferinmethodologyandfocus.
Kumaretal.(2006)trytocharacterizeusertypesintwohugesocialonline-
networks,FlickrandYahoo!360.Theycharacterize3kindsofusers:passive
members,inviterswhoenticeotherstojoin,and“linkers”whomakefulluse
ofthesocialnetworkingcapabilitiesoftheservices.Guetal.(2007)lookat
atpeer-to-peerIRCmusicsharingserviceandshowthatitresemblesatwo-
sidedmarketinmanyways,inwhichcontributionandconsumptionarehighly
.tarycomplemenThreestudiesspecificallyscrutinizepublicgoodproblemsandfreeriding:
Asvanundetal.(2004)trytoquantifytheextentoffreeridinginapeer-to-
peerfilesharingnetwork(Gnutella)onamacroscopiclevel.Theyfindthat
asubstantialshareofusers(42%)behaveasfreeriders,i.e.accessothers’
fileswithoutprovidinganyoftheirown.Xiaetal.(2007)discoverthatusers
aremorelikelytosharefilesinonlinesharingcommunitiesifhehashimself
benefitedfromthecommunityorhasarecognizedsocialstatus.FinallyGu
etal.(2009)considertheeffectsofindirectreciprocityonthepublicgood
provisionbehaviorofmembersofapeer-to-peerfilesharingnetwork.Theyfind
thatthepropensityofanindividualusertofree-ridedependsontheprevalent
behaviorofotherusersinthecommunity,whichtheyconstrueasevidence
foranindirectreciprocalbehavior.Furthertheyfindevidenceforreciprocity
asasocialnorm,asfreeridersarediscriminatedagainstthroughvoluntarily
enforcedsettingsoncommunalservers.Thesestudiesarenotcomparableto
ours,astheydonothavetheindividualcharacteristicsofusersattheirdisposal.
Thisallowsustoapplyindividualutilitymaximizationbehaviorandanalyze
userbehaviorinamicroeconomiccontext.Apartfromthisfact,pathological
incentivesmayarisethroughthefactthatfilesharingisillegalbehavior,while
theservicethatwestudyoperateswellwithintherealmsofcopyrightlaw.
Againstthisextensivebackground,ourcontributionisthefollowing:We

55

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

makeuseofauniquedata-setinwhichweobservebothindividualcharacteris-
ticsandbehavioraswellasthelinkstructureamongallindividualswithinan
actuallyexistingsocialonline-network.Incontrasttothefindingsofaseriesof
experimentalstudies,weprovideevidencethatnetworkformationandlinking
decisionsaremotivatedbystrategicbehaviorthatcanbeexplainedwithinthe
utilitymaximizingrationalagentframework.
Thenwefocusonpublic(ormorepreciselyclub-)goodprovisionwithin
thenetwork.Lookingattheentirebehaviorwithrespecttotheformingand
continuationoffriendshipsamongusersoverapproximatelyfourmonths,we
showthatnegativereciprocity,i.e.unilaterallyseveringexistingfriendships,
eventhoughtheyarepurelybeneficial,isusedtopunishfree-ridingbehavior.
Finally,welookattheoveralleffectsofthis.Usingauser-fixedeffectspanel
estimationprocedure,weshowthatthereisonlyveryweakevidencefortheex-
istenceoffreeridingbehavior,despiteanalmosttextbookpublicgoodproblem.
Onthecontrary,weareabletoidentifysignsofpositivelyreciprocalbehavior.
Theeffectofthelattermorethancompensatestheeffectoffree-ridingthatwe
e.observ

3.2.TheSocialOnline-NetworkunderScrutiny

3.2.1.FunctionandFeaturesoftheNetwork

Inthisarticle,westudyanemergingEuropeansocialonline-network,whichwe
willrefertofromnowonastheNetwork.VariousfeaturesmaketheNetwork
especiallyinterestingforeconomicanalysis.Thebasicinfrastructureisvery
similartowell-knownsitessuchasFacebook.Eachuserhasanindividual
profilepagewhichisaccessibletoothersandwhichallowshimtoprovideothers
withinformationabouthimself,suchasage,gender,hobbiesetc.Asafurther
featurefamiliarfromothernetworks,individualscanbecome“friends”with
otherusers-forthistooccur,oneoftheusershastoproposetheformationof
abilaterallinktoanotherandtheseconduserhastoacceptthisproposal.For

56

3.2.TheSocialOnline-NetworkunderScrutiny

legalreasons,thenumberoffriendshipsauserisallowedtoenterintoiscapped
at150,butthislimitisnon-bindingforallusersforthedurationofourstudy.
Whileittakestheconsentofbothuserstobecomefriends(link-formationis
bilateral),eitherofthetwocanendthefriendshipatanylaterpointoftime
erance).sevlink(unilateralThemaindifferentiatingfeatureoruniquesellingpropositionoftheNetwork
isthatuserscanuploaddigitalcopiesofmusicthattheylegallyowntoanonline
musiclibrary.Wecallthisuploadingactivityprovisionofinformation.Via
aplayerembeddedintheirweb-browser,usersthencanlistento(butnot
download)musicintheirlibrary.Further,andthisisthemainmotivationfor
ourstudy,userscanalsolistentomusicinthelibraryoftheirdirect,i.e.first-
degree,friends,aslongasthefriendshiplasts.Auserthereforehasaccessto
allthemusicthateitherheoroneofhisdirectfriendshasuploadedwhilehe
online.isAfeatureoftheNetworkthatissomethingofanuisancetousersisablessing
toourstudy:Theaccesstoindividualmusicfilesissomewhatcomplicated.
Forone,themusicavailabletoagivenuserisdistributedintodifferentmusic
libraries(e.g.musicuploadedbyhimselfisonelibrary,butmusicuploaded
byeachfriendisstoredinanindividualseparatelibrary).Themusicfiles
ineachoftheselibrariesaresortedalphabetically,whichmakesitpotentially
time-consumingtofindspecificsongsinlargelibrariesortofindsomething
interestingwhilebrowsingthroughthem.Testswithmediumsizedlibraries
ofaround1,000songssizeshowedthatittookmorethan6minutestolocate
asongbeginningwiththeletter“m”.Whiletheservicedoesprovideusers
withasearchfunction,searchesduetotheprogrammingofthewebsitealways
coverthelibraryoftheentireNetwork,i.e.bothfilesthatausercanaccess
andfilesthathecannot.Fromthesearchresults(itselfalistofhundredsof
songs,dependingonthepopularity),ausercanonlyestablishwhethershe
hasaccesstoagivenfilebyaddingittoherplayerandwaitingforanerror
messagetoappearuponhittingtheplay-buttonandwaitingforthesongto
load.Duetotheextremeusabilityissueswiththisprocedure,thesitewas

57

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

recentlycompletelyrefurbished,butthistookplaceoutsidethetime-window
e.observewthatItisexactlythisnuisancethatintroducesthesecondkindofeffortprovision
intotheequation.Thereisamechanismthatreducestheinconveniencefor
users.Theycancreate“play-lists”byaddingmultiplesongstotheembedded
musicplayer.Anynumberofsongscanbeaddedfromthelibrariesthatauser
hasfullaccessto.Theseplay-listsaretemporaryinprinciple,i.e.theyare
deletedwheneverthebrowsercacheiscleared.Butausercanalsosaveand
giveanametoaplay-listthathehascompiled.Thishastwoeffects:Itstores
thelistwiththeNetworkandmakesitavailabletotheuserwheneverhelogsin
againinthefuture,aswellastoanyonevisitinghiswebsite.Clearly,thiswill
bemostusefultotheuser’sfriends,whocanaccessmanyofthesamesongsas
herduetothegreateroverlapoftheirfirst-degreelibraries.Forthisreason,we
callthecreationandstoringofplay-listsorganizationofinformation.The
secondeffectisthatotherusers,whoarenotfriendsyet,cannotviewagiven
user’ssongs,buttheycanviewhisplaylists.Auser’splayliststhereforeare
themainindicationofhisvalueasasourceofsongstoothersinthenetwork.
Ausercanpreventoneofhisfriendsfromaccessingmusic-files(only)byend-
ingtheirfriendship.Asthisalsosignificantlydecreasesthevalueofhisplaylists,
provisionand(indirectly)organizationofinformationhavethecharacteristics
ofclassicalclubgoods.

3.2.2.NetworkSizeandDevelopment

Theweb-sitecameonlineinDecember2007with2,000registeredusers.This
numbergrewto9,000inMarch2008,morethan20,000inSeptember2008and
exceeded30,000inNovember2008,whenthedataforthisstudyweremainly
collected.Formany-maybeevenmost-questionsofinterestconcerningonline
networks,thenumberofregisteredusersisnotaverygoodindicator,asitmay
includeindividualswhoonlyregisteredatsomepointoftimebutneveractually
usedtheserviceorhaveceasedtodosoatsomepointinthepast;thereforewe

58

3.2.TheSocialOnline-NetworkunderScrutiny

willbrieflydiscussthemostwidelyusedalternativemeasures.
Accordingtothecommonpracticeofsitessuchasfacebook,myspaceor
similarsitessuchastheGermanstudi-vz,auserisconsideredactiveifshehas
loggedintoheraccountinthecourseofthepast30days.Atfirstglance,it
isnotobviouswhetherandhowthismeasureismeaningful.Table3.1below
shedssomelightonthisquestion.Init,wehavegathereddescriptivestatistics
forvarioussuchcutofflevelsofuseractivityasofOctober15th2008.8

AllUsersAct30Act20Act10
Friends0.922.162.503.91
SongsListened35.57100.82121.50208.63
Songsuploaded28.6250.2356.9396.88
Play-listscreated1.632.552.803.80
Messagessent0.381.301.673.02
Commentsleft0.030.120.160.28
Daysloggedin2.926.037.0110.57
#Observations:31,4556,0134,4662,291
Table3.1.:Descriptivestatisticsforvariousactivitymeasures

Someimportantobservationscanbegleanedfromthesenumbers,bothabout
theactivitymeasuresingeneralandabouttheNetworkinparticular.Firstof
all,eachmorerestrictivesub-sampleincludesusersthatareonaveragehighly
significantlymoreactivethantheusersinthemoreinclusivesample.For
example,usersthathavebeenactiveinthelast30days(Act30)uploadalmost
twiceasmanysongsonaverageasthegeneralpopulation,i.e.thoseusersthat
areleftoutofthesamplearesignificantlylessactive.Onelosesalargenumber
ofobservationsintheprocess,fromallregistereduserstoAct30,thedropis
precipitous,from31,455to6,013.Thesefindingsindicatethatthefrequently
employedmeasuredoeshavesomemerit.Figure3.1depictsthedevelopment
8Act30designatesuserwhohaveloggedintotheiraccountsinthelast30days,theothercolumnshave
analogousinterpretations.Byline,thesamplemeansofthefollowingarereported:numberoffriends
peruser,numberofsongslistenedtoperuser,numberofsongsuploadedperuser,numberofplay-lists
createdperuser,numberofmessagessentperuser,numberofpubliccommentsleftperuser,numberof
daysonwhichtheuserhasloggedintoheraccountatleastonceperuser.Allrespectivesamplemeans
differat1%significancelevel(t-tests).

59

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

Figure3.1.:DevelopmentofAct30Usersovertime.

ofregisteredvs.activeusersintheperiodfromDecember2007toNovember
2008.Whilethenumberofregisteredusersincreasesmorethan17-fold,the
numberofactiveusersincreasesonlybyafactorof2.5inthecourseofthis
d.eriopTwofurtherfindingsfromTable3.1concerningtheNetworkareparticularly
importantforourstudy.Ontheonehand,notethat“social”activities,such
aspostingcommentsorsendingmessagestootherusersappeartoplayavery
minorroleinthenetwork,witheventhemostactivegroupofusershaving
sentnomorethanatotalof3.02messagesandpostingnomorethan0.28
commentsonaverage.Ontheotherhand,listeningtoanduploadingsongs
takeupamuchlargershareofusers’activities,whichpointsinthedirection
thatthesemusic-consumptionrelatedactivitiesarethemainsourceofutility
forthoseinvolvedintheNetwork.

3.3.AnalyticalFramework

Oneoftheaimsofthisarticleistoprovideevidenceforstrategicandutility
maximizingbehavioramongusersintheNetwork.Inordertofacilitatethe
understandingofreadersandhelpusformulateourhypothesesmoreprecisely,

60

orkramewFAnalytical3.3.

wewillsketchasimpleformalmodeloflinkformationandindividualbehavior
withintheNetwork.Thiswillformthebasicframeworkforourempirical
analysis.WeherebydrawonthenotationandconceptsintroducedbyBalaand
Goyal(2000),JacksonandWolinsky(1996)andBramoulléandKranton(2007).
Aswedonotintendtoexplainoranalyzetheequilibriumnetworkstructure,
butinsteadindividuals’decisionbehaviorgiventhecontemporaneousstructure,
wewillkeepthecomplexityoftheexpositiontotheabsoluteminimum.
Followingtheestablishedgraph-theoreticapproach,wethinkoftheNetwork
asasetofnodesN={1,...,n}.Inourcontext,eachindividualagent(user)
isrepresentedbyonenode.Further,foreachagenti,thereisavectorgi=
{gi,1,...,gi,i−1,gi,i+1,...,gi,n},suchthateachgi,j∈{0,1}indicateswhetheror
notadirectlinkexistsbetweennodesiandj.Theselinkscanbeinterpreted
asfriendshipsbetweenusersinourcontext.Anetworkgthereforeisdefined
bythesetofvectorsgi,i∈N.Allfriendshipshavethesame“strength”inthis
framework,astheyaredepictedaspurelybinaryrelationships.

3.3.1.CreationandSeveranceofIndividualLinks

Assumethateachlinkgi,jbetweentwousersorplayersiandjisundirected,
i.e.bothcanbenefit.Toformanewlink,bothplayershavetoconsentto
theformation.Oneplayer(the“sender”)initiatesthelinkformationprocess
throughaninvitationwhiletheother(the“receiver”)decideswhethertoaccept,
ignoreordeclinetheinvitation.Ifthereceiveracceptstheinvitation,anewlink
isformedandgi,jbecomes1.Wedenotethischangeinthenetworkstructure
byg=g+gi,j.Ifthereceiverdeclinesorignorestheinvitation,thestructure
ofthenetworkdoesnotchange.
Conversely,eachuserinsideafriendshipcanseveranexistinglinkunilater-
ally.Ifuserioruserjcancelstheirfriendship,thengi,jbecomeszero.Analo-
gouslytoabove,wedenotethischangeinthenetworkstructureasg=g−gi,j.
Theutilitythatauserobtainsfromthenetworkdependsonitsstructure,
thereforeui=ui(g).Itappearssensibletoassumethattheformationofanew

61

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

friendshipisassociatedwithcosts,whichmayvarywiththecharacteristicsof
theindividualsinvolved:Thesendermustfirstlocatethereceiver,thenclick
onabuttonandcomposeamessagetosendoutafriendshipinvitation,and
finallyhelosesonefreeslotamonghisfriendshipsifthereceiveracceptsthe
invitation.Onceafriendshipofferhasbeenextended,thesearchcostsaresunk
andonlythecostsofthefreeslotremain.Inourcontext,allthesecostsare
mostlikelynegligible;ignoringthemwouldnotchangethefollowinganalysis.
Thereisacertainasymmetryhere:Thereceiveronlyhastoclickononebutton,
whetherheacceptsorrejectsafriendship(shehastotakenoactiontoignore
theinvitation)andsheloosesonefreefriendshipslotifsheaccepts.
Astheformationofnewfriendshipsrequiresmutualconsent,thefollowing
twoconditionshavetoholdatthesametimeforanewfriendshiptobecreated,
assumingthatiisthesenderandjisthereceiver,anddenotingtheirrespective
costsofinviting/acceptingasκi/j:

(3.1)

(3.2)

ui(g+gi,j)−κi>ui(g)⇔Δui>κi

uj(g+gi,j)−κj>uj(g)⇔Δuj>κj

ThewaytheNetworkissetup,additionallinksareclosetopurelybeneficial.
Theyallowuserstoaccessmoremusicandtomakebetteruseofadditional
play-lists.Asindicatedabove,usersintheNetworkaremainlyinterestedin
musicconsumption,thereforetheseconsiderationsshouldbetheirmainpriority.
Ontheotherhand,eachadditionalfriendshipcostsnexttonothingaslongas
usersarenotclosetotheupperlimitoffriends,whichisgivenfortheperiod
oftimethatwestudy.
Analogously,foruseritobewillingtoseveranexistinglink,denotingthe
costsofseveranceasλithefollowingconditionmusthold:

(3.3)

62

ui(g−gi,j)−λi>ui(g)⇔Δui>λi

orkramewFAnalytical3.3.

Forthesamereasonsasstatedabove,wewouldexpectΔuitobenegative
inthecaseoflinkseparation,unlessthenumberoffriendsis150(orcloseto
theupperlimitandusersanticipatereachingitwithcertainty).Asthecosts
oflinkseparationλiarenon-negative,inthissimplesettingusersareexpected
links.separatetonot

3.3.2.ProvisionofClubGoodsinNetwork

OneofthemainfeaturesoftheNetworkisthatusersprovidedifferentkindsof
effort,uploadingmusicorcompilingplay-lists,thatarepotentiallyusefulboth
tothemselvesandtotheirfriends.Wetreatthisinthesimplestimaginable
form:Considerthatusericanexertefforteiatcostc(ei).Wedenotetheset
ofuseri’sfirstdegreefriendsasJi,eachofwhomalsoexertseffort.Expanding
theutilityfunctionofuseriabovetoincorporatetheseconsiderations,itnow
becomesui(g,ei)=ui(ei,{ej}j∈Ji)−c(ei).Clearly,theoptimallevelofeffort
ofiisdeterminedby

(3.4)

∂ui(ei,{ej}j∈Ji)=c(e)
ie∂i

Therefore,iftheeffortofuseriandhisfrienduserjaresubstitutes(i.e.the
cross-partialsarenegative)andthecostsofeffortprovisionarenon-negative,
theoptimaleffortofishoulddecreasewiththeeffortthathisfirstdegreefriends
provide.Ifinadditiontheutilityofuseriincreasesintheeffortofhisfriend
userj,i.e.∂ui(ei,∂{eejj}j∈Ji)>0,wecantalkofalocalpublicgood.9Buthow
likelyisitthattheseassumptionsaresatisfiedinthesettingweencounter?

1)Utilityincreaseineffortprovisionoffriends:Moresongsprovidedbyhis
friendsgivesausermoresongstochoosefromandlikewisemoreplaylists
givehimacombinationofmorechoiceandbetterchoices.Standard
9Noteinformation,thatinthewhichNetarework,wcomplemenecants.distinguishtwodimensionsofeffort,provisionandorganizationof

63

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

theoryusuallystatesthatmorechoiceisagoodthing,recentadvances
inbehavioraleconomicsnon-withstanding.AstheusersoftheNetwork
displaymanycharacteristicsofcollectorsandmusicconnoisseurswefind
argumentstotheoppositeunconvincing.

2)Negativecrosspartials:Thispointislessclear-cut.Ifsomeoneusesthe
servicetolistentoagivensetofsongsthathehasdecideduponexante,
thenthereisnosubstituteforthesesongs.Ontheotherhand,ifusers
wanttocreateamixofmusictoentertainthem,thentheywillvaluevari-
ationaboveallelseandasonguploadedbyafriendisanalmostperfect
substitutetoasonguploadedmyself.Wedoobserveuserslisteningto
theirfriends’musicfrequently,asonecangleanfromtheanalysisbelow,
thereforewefindthisassumptiontobeequallyjustified.

Itisstraight-forwardtoderivetheexpectationoffree-ridingbyusersfrom
theseconsiderations.Infact,inasimilarsetup,BramoulléandKranton(2007)
findauniquestableequilibriumwithcompletespecialization,i.e.theagents
canbecleanlyseparatedintotwogroups,oneofwhichprovideseffort,theother
ofwhichfree-ridesentirely.
Therearetwostraightforwardreasonswhythisisnotnecessarilysoclear
cutintheNetworkweareconsidering.Forone,usersmaynottakethelink-
structureasgiven,butcaninsteadstrategicallyformfriendshipsandsever
existinglinks.Thereforesocialnormsmaybeformedandupheldthatfree-
ridingcanandistobepunishedbyseveringlinkswithfreeriders,whichwe
refertoasnegativereciprocity.Inaddition,usersmayonlyenterintonew
friendshipswithotherswhohavealreadydemonstratedthattheyaregood
citizens(strategiclinkformation),i.e.whohaveprovidedmeasurableeffortup
t.fronThesecondreasonwhyfree-ridingdoesnotnecessarilyhavetooccuristhe
factthatmanyofthepeer-groupswithinthenetworkarerelativelysmall.Even
themostactivegroupofusersdistinguishedabove,thosewhohaveloggedinto
thenetworkinthecourseofthepast7days,onlyhasaround4friendson

64

orkramewFAnalytical3.3.

average.Insuchaclose(ifstillpotentiallyanonymous)setting,onemayex-
pectthatsocialforcessuchasshameorguilt,combinedwiththerelativeease
ofmonitoringasmallnumberofconnections,maypreventusersfromengag-
inginfree-ridingbehavior.Experimentalevidenceshows,though,thateven
insmallgroupstheplayinrepeatedpublic-goodgamesconvergesneitherto
apurefree-ridingequilibrium(therationalequilibrium),nortopurelycoop-
erativebehavior(theefficientoutcome),butinsteadbehaviorismixed,with
playconvergingtowardsasituationinbetweenthetwoextremes.Seeforex-
ampletheseminalcontributionofAndreoni(1988).Wewillsimplytakethese
competinghypothesestothedata.
Fromalloftheconsiderationsabove,wederivethefollowinghypotheses:

Hypothesis1a-StrategicLinkingBehaviorregardingInvitations:Usersthat
areinvitedtojoinafriendshiphaveprovidedhighereffortsonaveragethanthe
peer-groupofactiveusers.

Hypothesis1b-StrategicLinkingBehaviorregardingAcceptance:Forfriend-
shipinvitationsthatareaccepted,theinviterhasprovidedhighereffortson
averagethanherpeer-group.Especially,userswhosefriendshipinvitations
havebeenacceptedhaveprovidedsignificantlymoreeffortthanuserswhose
friendshipinvitationshavebeendeniedorignored.

Notethatfortheformationofnewfriendships,bothusershavetoagree,
thereforethebehaviorandcharacteristicsofbothpartiesarerelevantforthe
formation.Asopposedtothis,fortheseveranceoffriendshipsitissufficient
foroneoftheuserstomakethedecision.Thisisreflectedinthefollowing
hypothesis,whichwillbemorecloselyspecifiedfurtherbelow:
Hypothesis2-NegativeReciprocity:Friendshipsfromwhichoneoftheusers
deriveslittleutilityaremorelikelytobesevered.Especiallyuserswhodis-
playbehaviorthatresemblesfreeridingaremorelikelytobeexcludedfrom
friendships.

Thehypothesesconcerningfreeridingbehaviorareverystraightforward:

65

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

Hypothesis3a-FreeRidingforInformationProvision:Usersuploadfewer
songsthemselves,themoresongsareavailabletothemthroughtheirfriends.

Hypothesis3b-FreeRidingforInformationOrganization:Usersgenerate
fewerplay-liststhemselves,themoreplay-listsareavailabletothemthrough
friends.their

Whileourfirstthreehypothesesfocusonfree-ridingandnegativelyreciprocal
punishmentbehavior,converselyonecouldalsoexpectasocialnormtoexist
thatrewardsdesirablebehaviororeffortprovisioninsomeform.Thedesignof
manywebsitestakesexactlythisformofrewardintoaccount:Forexample,in
manypopularspecializedinternet-forums,usersareawardedascendingtitles
basedonthenumberofcontributionstheyhavewritten.Similarly,Wikipedia-
contributorscanaccessstatisticsregardingthenumberandextentoftheir
ributions.tcon

Figure3.2.:Greetingscreenwithinformationoffriends’activity.

AlsothedesignoftheNetworkwestudyhasabuiltinperquisiteofthisform.
Wheneverauserlogsontotheservice,heisgreetedbyawelcoming-screen.
Placedcentrallyonthisscreenisamessageboxthatinformstheuserwhen-
everoneoftheirfriendshassuppliedadditionalplay-listsormusic,asdepicted

66

AnalysisEmpirical3.4.

inFigure3.2.Confrontedwiththisinformationregardingthebeneficialactiv-
ities,usersmightwanttopositivelyreciprocatebyprovidingadditionaleffortof
theirown,asforexamplesuggestedbyFehrandGächter(2000).Thepotential
reasonsforthiskindofbehavioraremanifold:Theyrangefromcompetitive
instincts(whohasthemostsongsinhislibrary),overimitationtogratitude
andreciprocation.Irrespectiveofthetruemotivation,wecansubsumeeachof
theseunderthefollowingtwohypotheses:

Hypothesis4a-PositiveReciprocityforInformationProvision:Usersre-
spondinkindwhenevertheirfriendsprovideadditionalinformation/upload
newsongs,i.e.usersaremorelikelytouploadnewsongsiftheirfriendshave
songs.newuploadedtlyrecen

Hypothesis4b-PositiveReciprocityforInformationOrganization:Users
respondinkindwhenevertheirfriendsprovideadditionalorganizationofinfor-
mationinformation/createnewplay-lists,i.e.usersaremorelikelytocompile
newplay-listsiftheirfriendshaverecentlygeneratednewplay-lists.

AnalysisEmpirical3.4.

DescriptionData3.4.1.

Weusetwodifferentdatasourcestoapproachthequestionsthatwewishto
study.Forthequestionsrelatedtolinkformationandseverance,wemainly
usedatacoveringeverychangeofuserfriendshipstatusbetweenJuly10thand
November11th2008.Fromthiswecreatetwodifferentdata-sets.Totest
Hypotheses1aand1b,welookatallfriendshipinvitationsthatwereissued
duringthistime,aswellaswhathappenedtothem(acceptance,declination,
revocation,nothing).Weobserve3,657friendshipinvitationsinthistime-span.
Importantlyforourtopicofinterest,wealsoobservecertainusercharacteristics
forthoseinvolvedintheseexchanges.Inordertoestimatetheprobabilityfor
afriendshipbeingseveredforHypothesis2,welimitouranalysistothose

67

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

invitationsthatwereacceptedinthistime-spanandwhichweobserveforat
least50days.Thislimitsournumberofobservationsto1,52.Wecallthisdata
.AsetFortheremaininghypothesesonfreeridingandpositivereciprocity,wemake
useofapaneldataset.Wehave10weeklycross-sectionsofuserdatainwhich
weobservevariouscharacteristicsandbehaviorvariablesasdescribedbelow.
ThedatacovertheperiodfromSeptember14thuntilNovember18th2008.
Duetoattritionandnewarrivals,thisresultsinanunbalancedpanelwithca.
25,000usersandanaverageof8.1observationsperuser.Wewillrefertothese
.BsetdataasInthefollowing,webrieflydescribethevariablesusedinourstudyand
indicateinwhichofthedatasetstheyareavailabletous:

DaysOnline(A,B)-Thiscountvariablelogseachdaythatauserlogged
intoheraccountorusedoneofthefunctionsontheNetwork.Thevalue
ofthisvariableistheaggregatednumberofdaysthatauserhasvisited
thesite.Notethatthisvariabledoesnotmeasurehowmanyactionsa
userhasperformedonagivendayorhowmuchtimehehasspentusing
service.the

DayssinceRegistration(A,B)-Thisvariabledescribeshowmanydays
agoauserregisteredwiththeservice.

LastLogin(A,B)-Measuredindays,thisvariabledenotesthetimesince
thelastactivityoftheuser.Anactivityisaloginoranyotheraction
terface.inuserthewithin

Friends(A,B)-Thenumberofcurrentlyactivefriendshipsauserisin-
volvedin.Thereisalimitof150friendshipsthatausercanenterinto,
butthislimitwasbindingforonlyoneuseratthetimeofouranalysis.10

MusicUploaded(A,B)-Thisvariablestatesthenumberofuploaded

10Duetoatechnicalglitchinthesystem,thisuserwasactuallyabletoenterintomorethan150friendships.

68

Empirical3.4.Analysis

songsbyaspecificuser.Thesesongsarecollectedintheusersownmusic
library,whichisaccessiblefortheuser’sfriends.

SongsListened(A,B)-Thisisthetotalnumberofsongstheuserhas
listenedtoontheNetwork.Alimitationtothisvariableisthatitincludes
bothsongsthatwereskippedaswellassongsthatwerelistenedtoinfull
length.Nevertheless,assumingthatusersderiveutilityfromlisteningto
musicthroughtheserver,thisisoneofthebetterproxiesforuserutility
us.toailableva

OwnPlay-lists(A,B)-Thenumberofplay-liststhatauserhascreated
andthatexistatthecurrentpointoftime,i.e.ifauserdeletesaplay-list,
thisvariabledeclinesbyone.

Play-listsFriends(B)-Thetotalnumberofcurrentplay-listsofallofa
friends.user’s

CommonFriends(A)-Allusersthatarefriendsofboththesender
andthereceiverofafriendshipinvitationatthegivenpointoftimeare
countedinthisvariable,illustratingtheoverlapofsenders’andreceivers’
friends.

CommonMusic(A)-Thisvariablecountssongsthatareownedbyboth
senderandreceiver.Importanttonoteisthatthisvariableisnotable
tocompletelycapturetheoverlapofthebothmusiclibraries,sincesome
songsmayvaryinqualityorlengthandarethustaggedwithdifferent
internalIDsbythesystem.

SongsfromFriend’slibrary(A)-Thiscountsthenumberofsongsfrom
theotheruserslibrarythatauserhasactuallylistenedto.

69

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

3.4.2.StrategicLinkingBehaviorandNegative
ycitRecipro

Onecriticismthatonecouldeasilypointatourprojectisthatusersdonot
actoutofrationalobjectives,i.e.thattheyinteractwithpeopletheycon-
siderfriendstotallyirrespectiveoftheireffortprovision.Theaimofthefirst
hypothesisistoifnotrepudiatethisargumentthenatleasttoputitintoper-
spective.Table3.2belowshowstheaveragecharacteristicsoffourdifferent
groups:First,thosewhoreceivedfriendshipinvitationsthroughouttheperiod
weobserve.Wepartitionthesendersoftheseinvitationsintotwosubgroups,
thosewhosefriendshipwasacceptedandthosewhosefriendshipwasdeclined
orleftpendingformorethan30days.11Wecomparethecharacteristicsof
thesegroupstothegroupofthosethathaveloggedinwithinthepast10days,
themostrestrictive“conventional”activitymeasure.
Theresultsarequitestriking.Peoplewhohavereceivedaninvitationareon
averagebyfarthemostactivegroup.Theyhaveuploadedfarmoremusic(3
timestheaverageofthesecondmostactivegroup),providedmoreplay-lists
andarealsomoreactiveconcerningthesocialnetworkfunctions.Bothmean-
andmedian-equalitytestsarerejectedatthe1%significancelevelforeachof
thesevariables.ThereforewecannotrejectHypothesis1aandconsiderthe
evidenceinsupportofthehypothesistobestrong.Theusersinthenetwork
appeartobeguidedbyrationalconsiderationsindecidingwhomtoinviteto
friendship.aformSimilarargumentsholdwithrespecttoHypothesis1b.Thoseuserswhose
invitationwasacceptedhaveprovidedtwiceasmuchmusicastheirrejected
(orignored)counterparts,almostthreetimesasmanyplay-listsandhavesent
twiceasmanymessagesorcommentsonaverage.Again,bothmean-and
median-testsshowthatthedifferencesaresignificantatthe1%level.Sothe
evidenceisstronglyinfavorofthehypothesis.Forthemoregeneralpoint,
wealsoconcludethatusersdobehaveinawaythatappearstobedrivenby
11Thehighlydatewrobustasctohosenmoasvingmorethecutoffthan95%date.ofacceptanceswereissuedwithin28days,buttheresultsare

70

Empirical3.4.Analysis

InvitersInviters
InviteesAcceptedNotAcceptedaAct10usersb
Ownmusic1936.02667.30316.28368.71
(1299.06)(1326.79)(1879.08)(5416.59)Ownplay-lists40.6520.957.7911.12
(47.00)(37.37)(90.41)(126.97)Friends38.0721.520.6114.48
(27.93)(26.69)(27.50)(39.76)Messagessent36.4110.405.6111.47
(88.03)(33.44)(62.61)(150.83)Commentsleft2.821.17.591.06
(4.78)(2.88)(4.16)(7.82)6029302,0763657obs.

aAninvitationwasconsiderednon-acceptedifitwasoutrightdeclinedorifitwasleftpendingformore
ys.da30thanbHere,weexcludethoseusersthathaveregisteredlessthan10daysago,whichexplainsthedifferencesto
table1.

Table3.2.:Theaveragecharacteristicsofvariousgroupsofusers:
inNetvitwees,ork’susersacceptedandaccordingnon-atothecceptedmostinviterswidelyandusedthedefinition,mostactive(meansgrouprepoofrted,the
std.deviationsinparentheses)

rational(utilitymaximizing)motives.
Next,wefocusontheunilateraldecisiontoseverexistingfriendships.We
distinguishbetweentwoparties,thepersonwhoseveredafriendship(“Severer”),
andthepersonwhopassivelyhadtoenduretheseverance(“Severed”).Evena
cursoryglanceatthedescriptivesdisplaysaveryclearpattern.InTable3.3,
wepresenttheaveragecharacteristicsofaSeverer,aSeveredandan“average”
userinthesub-sample.Comparingthelattertotheothermeasuresofuser
characteristicspresentedaboveshowsusthatthenewfriendshipsweobserve
wereformedbetweenrelativelyactiveandinvolvedusers.
ThecomparisonbetweenSevererandSeveredisverystark:Fromastatic
viewpoint,theSevererhadprovided22-timesthemusic(4472.15vs.206.14)
and48-timesthenumberofplay-lists(358.62vs.8.47)atthetimethatthe
friendshipwasformed.Lookingattheirindividualbehaviorduringthetime

71

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

thatthefriendshiplasted,theSevereruploaded110-timesasmanysongsper
day(33.71vs.0.28)and12-timesasmanyplay-lists(0.65vs.0.051)asthe
Severed.Finally,theSevererwasfarlesslikelytolistentoasongofthe
Severedthanvice-versa(0.02vs.1.29songsoftheotherlistenedtoperday).
Thesmall,yetpositive,numberofsongsthattheSevererlistenstofromthe
Severed’smusiclibraryisareminderthatthedecisiontoendthefriendshipis
costly:Youloseaccesstothosesongsthattheotherpersonhasprovided.In
astaticworld,whichtakestheutilityderivedfromafriendshipasgiven,this
appearsnottoberational,unlessauserderivesanimmediatesatisfactionfrom
linkseverance(“revenge”forexample).Ontheotherhand,itisrelativelyeasy
toimagineadynamicgameinwhichanequilibriumexists,inwhichpunishment
counteractsfreeriding.Bothconceptshaveincommonthatfreeridingisthe
causeofdis-utilitytotheotherparty.Torestateoursecondhypothesisin
t:argumenthiswithaccordanceHypothesis2-NegativeReciprocity:Friendshipsinwhichoneoftheusers
displaysbehaviorresemblingfree-ridingaremorelikelytobesevered.
Importantlyforouranalysis,nouserreachedthemaximumnumberoffriends
(150)duringthetimethatweobserve,themaximumnumberoffriendsthat
wasreachedintheseveredfriendshipswas125.
Fromthedescriptivestatisticsalone,itishardtodisentanglewhetheritis
thesheeramountofeffortprovidedbyauserortheadditionalprovisionof
effortovertimethattheotherpartyvalues.Inordertoshedmorelighton
thisquestion,weconsiderthefollowingdiscretedecisionmodel.Letuscallthe
probabilitythatagivenfriendshipisseveredp(x)≡P(y=1|x),wherexis
avectorthatcapturesbothparties’characteristics.Usingthelatentvariable
approach,weestimatelogit-models,seeTablesA.1andA.2intheAppendix
results.exacttheforInmodel1a,weonlyincludethe“static”characteristics,i.e.theeffortpro-
videdbybothusersatthebeginningoftheirfriendship(hereagain,wetalkof
asevererandasevereduser-ifthefriendshipwasnotsevered,thenthesetwo
roleswereassignedtotheusersinafriendshiprandomly).Wewouldexpect

72

AnalysisEmpirical3.4.

aOwnmusicbeginningSev4472.15ererSev206.14eredAverage885.84User
(1335.54)(987.39)(3111.36)Averagesongsaddedperday33.710.284.08
(11.05)(1.03)(34.44)Ownplay-listsbeginning358.628.4731.52
(71.19)(36.25)(317.54)Averageplay-listsaddedperday0.650.0510.29
Friendsbeginning12.82(1.80)15.61(0.11)28.96(0.84)
Friendsendb(29.15)22.27(22.90)17.63(22.68)34.48
(24.34)(23.24)(35.36)Numberoffriend’ssongslistenedperday.0205361.290.15
(0.74)(3.65)(0.13)114114obs.3,042

aAverageofusersintheentiresampleofnewfriendshipsbeingformed.
bFTheormaximnon-sevumerednumbfriendserofhips,friewndserepobservortedthenamongumbersevafterered50dafriendshipsys.was125,wellbelowthelimitof150.
Table3.3.:Meancharacteristicsofusersinvolvedinseveredfriendships:
Theseverersoffriendshipscomparedtothosewhowereseveredandtheaverage
ofallusersinvolvedin“new”friendships.)

thattheeffortprovidedbytheSeveredshouldlowertheprobabilitythatthe
friendshipwillbeended,asitraisesthecostsoftheSeverer(moreinformation
thathecannolongeraccess),andtheeffectdoeshavetheexpectedsignand
is(thoughweakly)significantfortheprovisionofmusic,whiletheprovisionof
play-listsisnotsignificant.Interestingly,theeffortprovisionoftheSevereris
significantandincreasestheprobabilityofseverancealongbotheffortdimen-
sions-thismeansthatusersthathavecontributedalotofeffortthemselves
aremorelikelytoendafriendship(whiletheyarealsomorelikelytoencounter
freeriders,accordingtotheanalyticalframeworkpresentedabove).
Inmodel2a,weonlyincludethe“dynamic”characteristics,i.e.thenumberof
songsandplay-liststhattheindividualusershaveaddedovertimeonaverage.
Weencounterthesamepattern:ThemoremusictheSevereradds,themore
likelyheistoendafriendship,whileaddingmoremusicmakesitlesslikelythat
auserwillhavehisfriendshipendedbytheotherparty.Theeffectofplay-lists

73

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

addedbytheSevererismoreinteresting.Addingmoreplay-listsmakesitmuch
lesslikelythatauserwillendafriendship.Onsecondglance,thisisrather
intuitive:Whenaddingaplay-listauserorganizestheinformationavailableto
herselfforeasieraccess.Thismay(ormaynot)includeinformationprovided
bytheotheruser.Inanycase,thismakestheinformationmorevaluable(asit
ismoreaccessible)andthereforeraisesthecostsofendingafriendship.
Inmodel3a,weincludeboththestaticanddynamiccharacteristics.En-
couragingly,theeffectsretaintheirsignsandsignificancelevels.
Inmodels1band2b,weaddthosevariablesthatcapturefreeridinginthe
potentiallymostintuitivelyappealingway,byincludingthenumberofsongsof
theotherpersonthattherespectiveusershavelistenedtoperday.Model1b
onlyincludesthesetwovariablesand,asonewouldexpect,songslistenedby
thesevererdecreasetheprobabilitythatafriendshipissevered.Further,the
numberofsongslistenedperdaybytheseveredhighlysignificantlyincreasethe
probabilitythatafriendshipwillbesevered.Thisresemblesthemostintuitive
descriptionoffree-riding,butitisnoteworthythatthesevererdoesnotknow
thathismusiclibrarywasaccessedbytheotherparty.Inmodel2bweadd
allexplanatoryvariablesusedinthepreviousmodelsandtheeffectthatthis
intuitiveformoffree-ridingstronglyincreasestheprobabilityofafriendship
beingseveredpersists.
Toconcludethesection,letusbrieflysummarizeourfindings.Thereare
strongindicationsthatusersmaketheirlink-formationdecisioncontingenton
theutilitytheyexpecttoderivefromagivenfriendship.Invitationsaresent
outtoveryactiveindividuals,andinvitationsbyuserswhohaveprovidedlittle
effortthemselvesarefarmorelikelytoberejected.Regardingtheseverance
oflinks,wefindthatusersarewillingtoincurseverancecosts(notbeingable
toaccesstheother’sinformation)inca.7.5%offriendshipsintheintervalof
timethatweobserve.Patternsandregularities,aswellasourlogitregression
results,suggestthatthisbehavioriscausedbythedesiretopunishtheother
users’freeriding.Thiscorrespondswithexperimentalfindingsespeciallyin
theliteratureontheultimatumgame.Interestingly,thispunishmentbehavior

74

AnalysisEmpirical3.4.

canitselfbeinterpretedasaformofpublic-goodprovision,e.g.,enforcement
norms.cialsoofWhatremainsnowistoshowhowprevalentfree-ridingisinthiscommunity,
giventheexistenceofthedescribedcontrol-andpunishmentmechanismson
el.levfriendshipindividualan

3.4.3.FreeRidingandPositiveReciprocity

Asdiscussedabove,weconsiderausertobefree-ridingifsheprovideslesseffort
ceterisparibusgiventhatherfriendshaveexertedthemselvesmore.Anaive
approachtothisproblemthereforewouldbetosimplyregressallusers’levels
ofuploadedsongsandgeneratedplay-listsontothelevelsoftheirfriendsina
cross-sectionalapproach.Atsecondglance,though,thisapproachisclearlynot
conductivetoansweringthequestionduetoareversecausalityissue.Assume
thatonefindsthatpeoplewhosefriendsprovidealotofeffortprovidealotof
effortthemselves.Onecanclearlyimagineastoryinwhichpeoplewhohave
providedalotofeffortinthepastattractmorefriends,especiallyconsidering
ourfindingsonstrategiclinkingbehavior.
Tocircumventthisissue,wethereforedon’tconsiderthetotallevelsofeffort
thatusersprovideasadependentvariable.Instead,welookattheirweekly
efforts,i.e.howmanysongs/play-liststheyaddinagivenweek,giventhe
explanatoryvariablesanddifferentsetsofcontrolsfromwhichweobtainthe
equations:regressionwingfollo

(3.5)

eit=β1Xit+β2Zit+uit

whereeitistheeffortanindividualexerts,Xitdescribestheexplanatory
variablesandZitdesignatesthecontrols.Clearly,theremaybeunobserved
individualeffectsthatarecorrelatedwiththeindividualerrorterms,e.g.some
individualsmaybecompulsivecollectorsofmusic,whileothersarecasuallisten-
ersonly.Inordertocapturethesedifferences,weestimateauserfixed-effects

75

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

model,andadjustthestandarderrorsforpotentialuser-clustereffects.
Weestimatethreedifferentmodels(withtwodifferentsamples)forinfor-
mationprovisionandorganizationrespectively,inordertoshedsomelighton
hypotheses3and4.Ineachspecification,Xiscomposedofthreevariables:
Thedeterminantsfortheeffortthatauserprovidesforinformationpro-
visioninagivenweekshouldbedeterminedbya)thetotalamountofmusic
thathisfriendshavemadeavailabletohim,b)thetotalamountofmusicthat
hehimselfhasuploadedinthepastandc)theamountofmusicthathisfriends
haveaddedintherecentpast.Forc),weusetheadditionalmusicfriendshave
uploadedinthecourseofthepastweek-ifanything,thisshouldunder-estimate
theeffectwearetryingtofind.Relatingthesevariablestoourhypotheseswe
wouldexpectthefollowing:a)shouldhaveanegativeinfluenceoneffortpro-
vision(freeriding,hypothesis3a),b)shouldhaveanegativeeffectoneffort
provisionduetodecreasingmarginalutilityfromeffortandc)shouldhavea
positiveinfluenceoneffortprovision(positivereciprocity,hypothesis4a).
Thethreemodelsdifferbythecontrolsthatweadd.Model1isbare,i.e.
withoutadditionalcontrols.Inmodel2weaddacontrolfortheuser’stime
sinceregistrationbyaddingadummythattakesthevalue1iftheuserhas
beenregisteredforlongerthan60days(experienceddummy).Furtherweadd
acontrolfortime.Inordertocontrolforuserswhosemainmotivationtojoin
thenetworkismainlysocial,weaddcontrolsfortheamountofpubliccomments
thatusershaveleftandprivatemessagestheyhavewritten.Inmodel3wegive
credittotheconsiderationthatthetwokindsofeffortsarecomplementary.One
wouldexpectuserstoaddmoremusiciftheirexistingmusiciswellorganized
andviceversa.Thereforeinmodel3weaddcontrolsforthenumberofown
play-listsandnumberoffriends’play-listsavailabletoauser.Weestimate
eachofthemodelsoncefortheentirepopulationofregisteredusersandonce
forthesub-sampleofactiveusersonly.
Analogously,forinformationorganizationeffortsinagivenweek,the
explanatoryvariablesinXarea)thetotalnumberofplay-liststhatauser’s
friendshaveprovided,b)thetotalnumberofplay-liststhathehimselfhas

76

AnalysisEmpirical3.4.

compiledandc)thenumberofplay-liststhathisfriendshaveaddedinthe
pastweek.Fromourhypotheses,wewouldexpectthesamesignsasabove.
Again,model1isbare,model2addscontrolsfortimeandsocialinteractions
andmodel3takesintoaccounttheamountofmusicthattheuserandher
friendshaverespectivelyuploaded.
Forthedetailedregressionresults,pleaserefertoTablesA.3throughA.6
intheAppendix.Wefindveryweakevidenceforhypothesis3a,freeridingwith
regardtoinformationprovision.Usersuploadweaklysignificantlylessmusic
onlyforthesub-sampleofactiveusersinmodel3,i.e.takingtheeffectsofthe
complementaryeffortprovisionintoaccount.Ontheotherhand,theevidence
forpositivereciprocitywithregardtoinformationprovision(Hypothesis4a)is
comparativelyrobust.Ineachspecificationofthemodelthereisasmallbut
significantandpositivereactiontootherusers’providingmoremusic.Some
otherobservationsfortheregressionsregardingmusicfilesaddedarehighly
interesting.Forone,theleveleffectoftheownmusicuploadedisneversig-
nificant-thiscouldeitherbeduetothefactthatthemarginalutilityfrom
additionalmusicisnon-decreasing,i.e.nowmatterhowmuchmusicisalready
available,moremusicisyetbetter.Orthereissomuchmusicavailablefreely
inanycase,thatadditionalownmusicfilesfromthebeginningonhavelittle
users.toaluevTheregressionsfortheadditionalprovisionofplay-lists(TablesA.5and
A.6)resembleourassumptionsratherclosely.Thereisasignificantnegative
leveleffectfortheamountofownplay-listsalreadycompiled,whichresembles
initiallypositiveanddecreasingmarginalutilityoforganizationofinformation.
Model3showsthatthemoremusicauserhimselfprovides,themorelikely
heorsheistoexertmoreeffortinorganization.Theeffectoffriends’mu-
sicontheotherhandissurprisinglysignificantandnegative(ifeconomically
notverymeaningful).Again,thismightresemblethefactthatmoremusicin
questionablyaccessiblelibrariesmayactuallyreducetheutilityauserderives
fromtheservicebecauseitmakesthingshardertofind.Regardingourcen-
tralhypotheses,wefindrobustevidenceforhypothesis4b,usersreacttotheir

77

3.Individual(ir)rationality?Behaviorinanemergingsocialonline-network

friends’organizationofinformationinapositivelyreciprocalmanner.Again,
theevidenceforfreeridingislimitedtooneofthesub-samplesandonlyap-
pearswhencontrollingforthecomplementarylevelofinformationprovisionin
model3.Finallynotethattheabsolutevalueofthecoefficientofthepositively
reciprocaleffectistwicethesizeofthefree-ridingeffect-thereforethenet
effectofafriendprovidingmoremusicispositiveintheshortrun.

okOutloandConclusion3.5.

Inthisarticle,weattemptedtoempiricallyassesstherationalityofuserbe-
havior,analyzinguniquedatafromarealsocialonline-network.Themajor
advantageofthisdataisthatweobservebothindividualusercharacteristics
aswellasthestructureoftheentirenetworkwhichallowsustostudyissues
thatwerenoteasilyapproachableupuntilnowempirically.
Wefindthatdespitemixedexperimentalevidence,theexistingtheoryon
link-formationallowsstraightforwardandusefulpredictions:Userssystemati-
callypursuemorevaluable“friends”intheprocessoflinkformation.Onthe
otherhand,individualsarewillingtoseveratfirstglancepurelybeneficiallinks
in7.5%ofthefriendshipsthatweobserve.Thereissubstantialevidencethat
thisislinkedtopunishment-behaviorwhenusersencounterfreeriding,i.e.a
directformofnegativereciprocitythatinthepasthasbeenmostprominently
encounteredinultimatumgames.Viewedstatically,thiskindofbehaviorwould
notbepredictedbynetworktheory(oreconomictheoryingeneral).Interest-
ingly,thiscaninitselfbeconstruedasaformofpublicgoodprovision,asitmay
helptoenforcesocialnorms,fromwhichusersingeneralbenefit.Finally,we
focusedonwhetherfree-ridingisdetectable,orwhetheronthecontraryweob-
serveaformofpositivelyreciprocalbehaviorinwhichindividualusersrespond
totheirfriend’sadditionaleffortprovisioninkind.Thereisclearevidencefor
thelatter,withthepositiveeffectsofreciprocalbehaviorovercompensatingthe
free-ridingeffectsintheshortrun.
Webelievethatourresearchraisessomeinterestingtheoreticalquestions.

78

okOutloandConclusion3.5.

Evidently,thereareinteractiveeffectsbetweentheformationoflinks,thepro-
visionofclubgoodswithinanetworkandunilaterallink-severance,whichmay
influencetheprevalenceoffree-ridingwithinasocialnetwork.Theseeffects
have,toourknowledge,notbeencapturedintheoreticalnetworkmodelsup
untilnow,eventhoughtheymayyieldpredictionsthatarefarclosertothe
empiricallyobservedfacts.Inadditiontotheindividualbehavior,itwouldbe
interestingtogeneratepredictionsconcerningnetworkformationandequilib-
riumnetworkstructureinaninthissenserichercontext.
Forthespecificnetwork,thereareanumberofissuesthatwewanttoaddress
infutureworkwithadditionaldata.Thequestionsinclude,butarenotlimited
to:Isthepunishmentmechanismthatweobserveeffective,i.e.doindividual
users,whohavebeen“kickedout”offriendshipschangetheirbehavior?What
aretheeffectsofthesizeofneighborhoods(orthenumberoffriendsofan
individualuser)onhisfree-ridingbehavior-e.g.aremembersoflargerneigh-
borhoodsmorelikelytoexhibitthiskindofbehavior,aspotentiallymonitoring
costsincreaseandpunishmentislesslikely?Finally,itwouldbewonderfulto
seewhetherourfindingsholdtrueforcomparablesocialnetworks,ifsimilar
databecomesavailableinthefuture.
Inclosing,wewouldliketorestatethatwearewellawareofthedrawbacks
thatempiricalworkinthisfieldsuffersfromingeneral.Wedonotobserve
andcannotperfectlycontrolforindividualusers’actualmotivationintheir
actions.Theremaybesubstantialandunpredictabledifferencesintheirutility
functions,“networksavvy”andcomputerliteracy,whichmaysystematically
biasourresults.Whileexperimentalstudiesarenotnearlyaspronetothese
problems,theyaresubjecttocertainshortcomingsoftheirown.Wehopethat
ourstudyisviewedasastronglycomplementarybuildingblockthatenriches
ourunderstandingofthebehaviorofindividualsincomplexsocialnetwork
settings.

79

4.TrustandInvestment-ATheoretical
tAssessmenEmpiricaland

troIn4.1.duction

ationMotiv4.1.1.

Inthelastcoupleofyears,theautomotivesector,oneofthemostinnovative
andimportantindustriesindevelopedcountries,haswitnessedunprecedented
turbulence.InJune2009GeneralMotors,thesecondbiggestcarmakerinthe
world,filedforchapter11bankruptcyproceedings.Itappearsthatmorethan
justbeingstrangledbymountingpensionobligationsandbeingunderminedby
risingfuel-costs,theformerindustryjuggernauthasbeenout-manouveredby
moreinnovativeanddesign-savvycompetitors.Sincethen,thefinancialcrisis
hasmadeconsumersevenmorelothtomakeextensiveinvestmentsintotheir
mobility.1Increasinglyharshcompetitionforaslowingdemandhaseroded
industrymargins,placinganevergreaterpremiumoncompaniesthatproduce
efficientlyandareabletodifferentiatethemselvesfromothersthroughtheir
highlevelofqualityandinnovativeproducts;buttoanevergreaterdegree,it
isupstreamsuppliers’know-howandeffortsthatareresponsibleforsuccessful
tiation.differenAsaresultofincreasedoutsourcingefforts,carmakerscontributeonlyami-
norshareofinnovationeffortsandthetotalvalueaddedoftheirproduct-

1pingDespiteuptheailingfactthatcarmakgoversbernmenytsguaranthroughoutteeingloansthewandorldprohavevidingrespondedliquiditydeconisivtheelyoandneexphand,ensivandely,givprop-ing
consumersadditionalincentivestoreplacetheiroldcarsthroughvariationsofthe“cashforclunkers”
program.

80

ductionotrIn4.1.

insomecasestheyarealmostreducedtopureassemblers.Upstreamsuppli-
ersareresponsibleformuchoftheground-breakingbasicresearch,whichis
thenadaptedtothespecificneedsofindividualcarmodels.Nevertheless,the
structureofthemarketplacesthecarmanufacturersinthesuperiorbargaining
positionvis-à-vistheirsuppliers.Thisleadstoaninherentdilemmabetween
attainingthedesiredqualitylevelsandshort-termrent-extraction/profitgen-
eration.Wedemonstratethattrustcanplayacentralroleinsuchahold-up
situationinatheoreticalmodel,whosepredictionswethentestusingaunique
data-set.Tobeabletodothis,wecontributetotheunderstandingoftrust:While
therehavebeensignificantadvancesineconomicresearchonthistopicinthe
pastdecade,manyofthemsharewhatcanbeconsideredadrawback:Both
experimentsaswellasempiricalapproachesfocusontrustasacharacteris-
ticofsubjects,eitherinvestigatingtheirgeneralattitudestowardsothers(“Do
youthinkingeneralotherscanbetrusted?”)oranalyzingtheirwillingnessto
contributefundstomoreorlessanonymousplayersinlabsettings.Incon-
trasttothis,wedefineanupstreamsupplier’strustinthedownstreamfirm
ashisexpectation(thesubjectiveprobability)ofwhetherthelatterisgoing
toexploitaholdupscenariotoextractrents.Thisisclosertothesociological
andcolloquialreadingoftheterm,inwhichinaspecificsettingAtrustsB
todosomething(orrefrainfromdoingsomething),whichisinB’srealmof
influence.Withthisunderstanding,trustcanalleviatetheunder-investment
problem.Wefurthershowthat,asopposedtomanyresultsintheliterature
onrelationalcontracting,trustandcompetitionbetweenupstreamsuppliers
inducedbythedownstreamfirmarenotmutuallyexclusive.Infact,compe-
titioncanbeasubstituteforotherformsofrentextractionandinthiscase
strongercompetitioninducedbythemanufacturerwouldbeassociatedwith
higherlevelsofsuppliers’trust.
Weareabletoapproachthesubjectoftrustempiricallyusingaunique
data-set,collectedfromanonline-surveyofsuppliersandmanufacturersinthe
Germanautomotiveindustryin2007-2008.Themeasuresoftrustthatwe

81

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

generatearerelationship-specific–insteadofsubjectspecific–andweanalyze
thedeterminantsofsuppliers’trustinthemanufacturerbylinkingthemto
reportedpastbehavior.Weshowthathigherlevelsofsuppliers’trustlead
tohigherrelationship-specificinvestmentproxiedbyfailure-ratesofsupplied
parts.Finallyweshowthatmoreintensesuppliercompetitionisnotmutually
exclusivewithtrustinthedownstreamfirm.
Theremainderofthearticleisorganizedasfollows.Afterbrieflyoutlining
therelatedliterature,wedevelopaverysimplemodelinSection2,fromwhich
wederivehypothesesontheeffectsoftrustonverticalrelationships.InSection
3,wefirstintroducethestudythatprovidedthedataonwhichwebaseour
empiricalanalysis,anin-depthsurveyinvestigationintothestructureofthe
Germanautomotiveindustry.Wepresentpotentialmeasuresoftrustandtry
tocarefullyevaluatewhattheycapture.Atthecenterofourempiricalanal-
ysis,weanalyzehowtrustbetweenmanufacturersandsuppliersisrelatedto
twoimportantquestions:Sourcingdecisionsandsupplier(under-)investment.
Finally,Section4concludesandraisessomenewresearchquestionsresulting
findings.ourfrom

4.1.2.RelatedLiteratureandContribution

InteractionsbetweensuppliersandOEMs2intheautomobileindustryareno-
toriouslycomplex,fraughtwithmoralhazardandhold-upproblems.Neverthe-
less,bothpartiesregularlyinvestsubstantialamountsoftime,know-howand
moneyintospecificrelationships.Apparentpuzzleslikethishavepiquedthe
interestofeconomistsforquitesometime—perhapsthemostprominentexpla-
nationapproachescanbesubsumedundertheheadingsofpropertyrightsthe-
ory(appliedmainlytohold-upproblems)andcontracttheory(appliedmainly
toasymmetricinformationandmoralhazard).3Beyondthat,ithasbeenwell

2WeusethetermOEM,i.e.originalequipmentmanufacturer,forthedownstreamautomobileproducer.
Wewillrefertotheupstreamfirmssimplyassuppliers.
3Duetothemoreappliedfocusofourstudy,werefrainfromdelvingdeeplyintotheintricaciesofthe
literatureontheholdup-problem.Instead,werefertotheseminalHartandMoore(1988),Grossman
andHart(1986)andHartandMoore(1990),aswellasthemorerecentHartandMoore(2007)and

82

ductionotrIn4.1.

establishedthatinsettingslikethese,relational(orinformal)contractscanplay
animportantroleingoverningrelationships.Asopposedtoformalcontracts,
whicharelinkedtooutcomesverifiablebythirdpartiesandcourts,thetermre-
lationalcontractreferstoself-enforcing,oftenimplicitagreements“sealedwith
e”.handshaka

TheoreticalApproaches

Thereisarichtheoryonrelationalcontractingindifferentcontexts,4beginning
withBull(1987)whoprovidestheoriginalrepeatedgames-frameworkforem-
ploymentrelationships.Alsoforemployment,Bakeretal.(1994)demonstrate
howthecombinationofformalandrelationalcontractscanleadtobetterre-
sultsthaneitherinstrumentcouldachievealone.Interestingly,therecanbea
substitutive(ifeitherworksalmostperfectly)orcomplementaryrelationship
betweenthetwo.Bakeretal.(1999)analyzetheinformaldelegationofdecision
rightswithinhierarchiesunderdifferentinformationalsettings.
Morerecently,researchhasfocusedonmoregenericsettings,searchingfor
optimalcontractdesign.Levin(2003)findsthatwhileundermoralhazard
optimalrelationalcontractsexistandarerelativelysimple,underhiddeninfor-
mationcasesariseinwhichagentsdonotrespondtotheincentivesprovided
thereinatall.CalzolariandSpagnolo(2009)furtherextendthisscenario:The
relationshipbetweenaprincipalandanagentinteractingrepeatedlycansuf-
ferfrombothmoralhazardandhiddeninformation-buttheprincipalhasa
furthertoolavailabletohimbybeingabletoselectfromvariouscompeting
agents(screening)whoareabletocollude.Theyfindtheintuitivelyappealing
resultthatincasesinwhichnon-contractiblefactorscontributemoretothe
principal’spayoff,thebestpackageofinstrumentswillrelymoreheavilyonre-
lationalcontractswithasmallersetofagents.Viceversa,in“simpler”settings
inwhichthemostimportantissuesarecontractible,theprincipalwillrelyon

Hart(2008),thelattertwowithmanyfurtherreferences.
4WerefertoMacLeod(2007)foracarefulsurveyoftheliteratureonrelationalcontracting,withaspecial
focusontheeffectsofthequalityoflegalssystems.

83

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

amorecompetitivesettingamongstagents.Brownetal.(2004)carryoutan
experimentalstudyinordertobeabletocontrolforthelevelofenforceability
ofcontracts.Theyfindthatasenforcementbecomesmoreeffective,theoriginal
long-termrent-sharingrelationshipsarereplacedwithshort-termarms-length
agreements,verymuchinlinewiththetheoreticalresultsdescribedabove.
Thiscanbetakenasevidenceofasubstitutiverelationshipbetweenformal
andrelationalcontractsintheirspecificsetting.
Enteringintoabusinessrelationshipwithoutbeingabletoresorttolegal
meansofenforcementwouldprobablybecalledaformof“trust”colloquially.
Themodelsdescribedabovehaveincommonthatrelationalcontractsandcom-
petitionareinacertainsensepolarconcepts—youeitherrelyonhandshakes
oryoupreferarms-lengthmarketinteractions.Wewilldemonstrateinour
analyticalmodel(aswellastheempiricalpartofourstudy)thattrustand
competitioncanactuallybeassociatedwitheachotherinahold-upsetting,
ifcompetitionisusedasa“legitimate”mechanismofextractingrentfromthe
individual,insteadof“illegitimately”extractingrentsthroughsuperiorex-post
bargainingpower.Astheconceptoftrustthatweproposeisnotcongruentwith
relationalcontracts,thisresultcanbeseenascomplementarytotheexisting
literature.

StudiesEmpirical

Trusthasforsometimedrawnconsiderableattentionandscrutinyfromexperi-
mentaleconomists.5Yetresearchersinappliedmicro-economicsandindustrial
organizationhavebeenrathercautiousaboutusingthisterm,evenactively
tryingtoavoidit(see,e.g.thediscussioninMacLeod(2007)).6Ontheother
hand,empiricalresearchersintheareasofmacroeconomicsandgrowthhave
beenlessreticentinthisregard,sothatsomeempiricalstrategiesdoalready
exist.

5SeeFehr(2009)forasweepingoverviewoftheexperimentalandneuro-economicliterature.
6Foracarefulsurveyofthedevelopmentoftheterm“culture”ineconomicsandtheeffectsofcultureon
economicoutcomes,seeGuisoetal.(2006).

84

ductionotrIn4.1.

Asabasisformanystudies,theanswerstothefollowingquestionfromthe
WorldValuesSurveyhasbeenused:“Generallyspeaking,wouldyousaythat
mostpeoplecanbetrustedorthatyouhavetobeverycarefulindealingwith
people?”Whileonemaydoubtthepowerofthisconstructatfirstglance,7it
hasbeenusedfrequentlytoobtainresults.ThebasichypothesisofLaPorta
etal.(1997)isthattrustisanintegralrequirementforthefunctioningoflarger
organizationsinwhichthelikelihoodofrepeatedinteractionsisrelativelysmall
andtherebytheestablishedmechanismsforensuringcooperativebehaviorare
lesseffective.Inacross-countrystudytheytrytoestablishthatpopulationsin
whichhigherlevelsoftrustareprevalentshouldfostermoreeffectivegovernance
aswellasrelativelylargerfirms.Aghionetal.(2008)performaninternational
comparativestudyinwhichtheyscrutinizetheconnectionbetweenlevelsof
socialcapital(ortrust/distrust)inpopulationswiththeamountofexisting
stateregulationaswellasthedemandforit.Thebasicintuitionisthatalack
ofcivicmindednessinone’sfellowcitizensmayleadtoastrongerdesireforthe
statetoregulateinteractions.Theyfindverystrongevidenceforthis,evenfor
societiesinwhichthegovernmentitselfisplaguedbycorruption.Thereforeit
appearsthattrustandregulationsaretosomeextentsubstitutes.Incontrast
tomostotherarticlescitedsofar,Butleretal.(2009)studytheeffectsoftrust
onindividual’seconomicoutcomesinsteadofaggregateeconomicperformance.
TheyusetheEuropeanSocialSurveyaswellasexperimentalevidencetoargue
thatamediumamountoftrustmaybeoptimalforindividuals:Withtoolittle
trust,toomanyopportunitiesforbeneficialinteractionsaremissed,withtoo
muchtrustthedangerofbeingtakenadvantageofbecomestoogreat.
Guisoetal.(2009)useaslightlymoreconcretemeasure,thetrustthat
citizensofagivencountryinEuropehaveforcitizensofanother.Theyfind
thatthelevelsoftrustareexplainedinpartbycharacteristicssuchasthe
distancebetweencountries,butalsobyfactorssuchassociologicalandgenetic
closenessandcommonhistory.Theyfindthatlesstrustinthecitizensofa

7SeeSapienzaetal.(2007)foranexperimentalstudyonthemeritsofthismeasureandadiscussionofthe
literature.previous

85

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

countryisassociatedtosignificantlyloweraggregatetradeandinvestment.In
asecondstudy,Guisoetal.(2004)suggestthatthedifferentcharacteristicsof
Italianregionsleadtheircitizenstodevelopdifferentlevelsofsocialcapital.
Theythenshowthatinhighsocialcapital/trustareaspeoplearemoreprone
toinvestinstocksinsteadofholdingcashreservesandhaveeasieraccessto
bankcredit.Theeffectismitigatedbylevelsofeducation.Alongsimilarlines,
Guisoetal.(2005)findthatindividualswhodisplayhigherlevelsoftrustbuy
moreriskyassetsrelativetotheirwealth.Theycounteracommoncriticismby
controllingforrisk-andambiguity-aversion,whichdoesnotmaketheiroriginal
ear.disappresult

Finally,Bottazzietal.(2009)studythewillingnessofventurecapitaliststo
performnon-contractibleservicesinamicro-economicenvironment.Inpartic-
ular,theyanalyzetheinfluenceofmoreeffectivelegalsystemsinthiscontext.
Bothintheirtheoreticalmodelandtheirempiricalanalysisofadata-setwith
Europeanventure-capitaldealstheyfindthatamoreefficientlegalsystemhas
twoeffects.Ontheonehanditiscomplementarytotrust,inthesensethatit
makesventurecapitalistsmorewillingtograntnon-contractiblesupport;but
ontheotherhand,theyrequiremoreprotectionforthecaseoffailureofthe
8ture.env

Webelievethattheempiricalpartofourstudyaddsanewangletotheway
economiststhinkabouttrust.Wedefinetrustnotasacharacteristicspecificto
individuals.Instead,weinterprettrustastheexpectationsofonepartytowards
futurebehavioroftheother,basedonrelationship-specificobservableandun-
observablecharacteristics,includingthejointhistory.Basedonthisapproach,
weattempttoanalyzehowhigherorlowerlevelsoftrustinadownstreamfirm
byanupstreamfirmaffectupstreaminvestmentlevelsaswellasthechoiceof
contractualsettingintheGermanautomobileindustry.

8Inthebarometerworkingmeasurepaperofontrustthebetsameweendata,nationsBottazziareetassoal.ciated(2007)withalsohighershowinvthatestmenhigherts.scoresontheEuro-

86

orkramewFAnalytical4.2.

orkramewFAnalytical4.2.

Weintroduceasimplemodelasananalyticalframeworkforourstudywith
twomaingoalsinmind:Ontheonehandtoclearlydefinetheconceptoftrust
thatwewishtoworkwithandontheotherhandtointegratesomeindustry
specificobservationsthatweencounteredinthecourseofthequalitativepart
ofourresearchprojectintothecanonicalmodel.Inparticular,wewantto
analyzehowcompetitionbetweensuppliers—inducedbytheOEM—relatesto
issuesofunderinvestmentduetohold-upproblems.
Whatweobserveinrealitywhenanewpartistobedevelopedandprocured
isthatinmostcasesmultiplesuppliersareinvitedtodevelopablueprintto
matchtheOEM’sspecifications.Dependingonthedevelopmentstage,these
canrangefromextremelyvague(pre-developmentupto3yearsinadvanceof
seriesproduction)toratherspecific(detaileddevelopmentabout6-9months
priortoseriesproduction).Alargepartofthecompensationofthesupplier
iscomprisedoftheOEM’ssubsequentbusinessregardingseriesproduction
ormoredetaileddevelopment—butintheselaterphasesonlyasubset(often
onlyonefirm)ofthepreviouslyemployedsupplierswillbeawardedthesecond
ct.atrcon

4.2.1.StructureandTimingoftheProcurementGame

Wemodelthisinthefollowingway:AssumethatamonopolisticOEMneeds
theinputofasupplierinordertoproduceoutputwithavalueofv(θi,Ii),
wheretheargumentsofvdesignatetheintrinsicqualityandeffortchoiceofthe
suppliertobedefinedexactlybelow.
First,int=0,theOEMirreversibly9choosesnex-anteidenticalsuppliers,
invitingthemtodevelopandsubmitablueprintforthepartitwantstopro-
cure.Foreachsupplier,itincurscostsk,e.g.administrationandcoordination
efforts,sothatthetotalcostsarenk.Thesupplierstheneachindependently

9Imaginethattheinterfacesbetweenthepartinquestionandtheotherpartshavebeendesignedbythe
timetheOEMcouldreconsider,makingthisoptioneconomicallyimpossible.

87

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

drawanintrinsicqualityparameterθifromtheknowndistributionQ(θ),which
iscontinuouslyincreasingwithinthedomain[0,1].Therearemanywaysto
interpretthisparameter-forexampleitcouldmeasurehowwellthesupplier’s
employeesinvolvedintheprojectarecompatiblewiththeOEM’sengineers,
orhowwelltheresearchcapacitiesofthetwofirmscomplementeachother
withregardtotheperiodoftimeinquestion.Clearlythereisacertainlevel
ofrandomnessinvolvedhere,evenifthefirmshavecooperatedalreadyinthe
past,andwewouldarguethatinanestablishedindustryliketheautomotive
industry,thisrandomnessshouldbeofsimilarorthesamedegreeovercompa-
rablesuppliers.Aftereachsupplierprivatelyobservesthevalueofitsdraw,it
chosesaneffortlevelIi,giventherevelationofθiandthenchosenbytheOEM.
Exertingeffortinducesthecostsc(Ii),withbothcandcstrictlypositive.
Then,int=1,thesuppliers’qualitiesandinvestmentchoicesbecomecom-
monknowledge.Welimitthesubsetofsuppliersthatreceiveasecondcontract
toasingleton,whichisselectedinthefollowingfashion.Ifn=1,thenthis
supplierisawardedthecontract,otherwise,thesuppliersengageinasecond-
priceauction(orequivalentlyBertrandcompetition)underperfectlysymmetric
information.Intuitively,thisensuresthatthesuppliercontributingthehighest
netvalueisawardedthecontract.
Finally,int=2,thesupplierwiththewinningbidandtheOEMbargain
abouthowtosharethejointsurplusfromproductionv(θi,Ii),whichisstrictly
increasingbothinθiandinIi.Inthisbargainingprocess,wedenotetheOEM’s
outsideoptionasg,whichisexogenouslygiven.Thesupplier’soutsideoption
isendogenouslydeterminedbyitseffortchoiceandwillbedenotedf(θi,Ii).10
Wemakethefollowingsimplifyingassumptionswithregardtotherelationship
ofthejointsurplusandoutsideoptionf(θ,I)ofthesuppliers:

A1:f(θ,I)<v(θ,I)and∂f∂(Iθ,I)<∂v∂(θI,I)foranygiven(θ,I).
Thisassumptionimpliesthatthesurplusgeneratedfromthesupplier’sin-

10Inanexcursionbelow,wewillproposeanalternativesettinginspiredbydiscussionswithindustryrep-
resentatives,inwhichtheoutsideoptionisreplacedbytheOEMcoveringashareofthesuppliers’
costs.

88

orkramewFAnalytical4.2.

vestmentwithintherelationshipisalwaysgreaterthanthesurplusobtained
fromthatinvestmentoutsidetherelationship.Further,anincreaseininvest-
mentbythesupplierleadstoagreatercreationofvaluewithinthanoutside
relationship.the

A2:v(θ,I)−f(θ,I)>v(θ,I)−f(θ,I)ifθ>θ.
Higherqualitysuppliersgeneratealargersurplusabovetheiroutsideoption
thanlowerqualitysuppliersforagivenlevelofinvestment.

A3:v(θ,I)>v(θ,I)⇔v(θ,I)−f(θ,I))>v(θ,I)−f(θ,I)
Thehigherthetotalsurpluswithinarelationship,thehigheristheefficiency
losssufferedifbargainingbreaksdown.Thisimpliesthatasupplierwithalower
intrinsicqualityθcanovercomethisdisadvantagetobecomemoreefficient
throughhigherinitialinvestmentI.A1throughA3wouldbesatisfiedforan
exogenousoutsideoptionfˆofthesupplier.Giventheseassumptions,clearly
thefirstbestinvestmentlevelisdeterminedbythefirstorderconditionv2=c.
Wewillcallanylevelofinvestmentbelowthefirstbestunderinvestment.
Wedistinguishtwopolarsubcaseswithrespecttothebargainingsituation
att=2.Insubcasea),whichwewillrefertoasthedefactoscenario,the
OEMhastheentirebargainingpowerandisinthepositiontomakeatake
itorleaveitoffertothesupplier.Insubcaseb),whichwecallthedejure
scenario,itisthesupplierwhoisendowedwiththesuperiorbargainingposition
andwhomakesthetakeitorleaveitoffer.Onepossibleinterpretationisthat
thesupplierobtainsthepropertyrightsoftheblueprintsithasdevelopedat
t=1.InthedejurescenariotheOEMhonorsthesepropertyrights,whilein
thedefactoscenarioheisabletoextractrentsfromthesupplierbeyondhis
formallegalrightswithouthavingtofearrepercussions.

Results4.2.2.

Wesolveforasubgame-perfectequilibriumbybackwardinductionforeachof
thetwobargainingsettings.

89

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

Casea)defacto-OEMmakestakeitorleaveit-offerinperiodt=2

Atthebargainingstageint=2,theOEMoffersthesupplierafixedpay-
mentT2,maximizingv(θi,Ii)−T2.Hehastotakethesupplier’sparticipation
constraintintoaccount,thereforehemustofferatleasttheoutsideoptionsuch
thatT2≥f(θi,Ii).Thisistheonlyconstraintoftheproblematthisstage,
thereforetheOEMwillchooseexactlyT2=f(θi,Ii).
Nowconsiderthecompetitivesituationamongsuppliersint=1.Ifasupplier
isawardedthesubsequentcontract,wehavejustseenthathewillobtainan
offerfromtheOEMthatisequivalenttohisoutsideoption,i.e.theOEM
extractstheentirerentfromcooperationandthesupplierisindifferentbetween
winningorlosingthebid.11Nevertheless,thereisasubstantialeffectofn
fromthepointofviewoftheOEM.Assuppliersareindifferentandthereis
symmetricinformationatthispoint,heisabletoselectthehighest-quality
supplieroutofn.Mathematically,thisisasimpleorderstatisticproblem.
Weknowthatthemaximumorderstatisticwithndrawsfollowsadistribution
Q1n=[Q(θ)]n.Fornon-degenerateQ,thisstochasticallydominatestheoriginal
distribution,therefore∂Eθ(n)/∂n≥0andincreasingthenumberofsuppliers
nleadstoahigherexpectedqualityθ1ofthehighestqualitysupplieramong
thecompetitors,withoutaffectingtheirinvestmentincentives.
Asaresult,eachsupplierfacesthefollowingmaximizationproblemint=0:
maxIf(θi,Ii)−c(Ii),leadingtothefirstorderconditionf2=c.ByA1the
suppliers’investmentsaremorevaluablewithintherelationshipthanoutsideit.
Inthissettingthereforethesuppliersunderinvest.Ifwedenotethesupplier’s
optimalinvestmentlevelinthiscaseasIa(θi),theexanteexpectedprofitof
is:OEMthe

(4.1)EΠaOEM=E[v(θ1,Ia(θ1))−f(θ1,Ia(θ1))|n]−nk

wherethesubscript1indicatesthatitistheexpectedqualityandinvestment

11Thereceivewinningitbecausebiddertheyreceivhavesetohisfalloutsidebackonoptionit.astheoutcomeoft=2bargaining,thelosingsuppliers

90

orkramewFAnalytical4.2.

ofthehighestqualitysupplierthatisrelevant.TheOEMbenefitsfromlargern
onlythroughthehigherexpectedqualityofthebestsupplierandextractsthe
entirevalueaddedthroughcooperation.Thereforetheoptimalna∗ischosenat
thepointwherethedifferencebetweenmarginalbenefitsfromexpectedprofit
gainsduetohigherexpectedqualityandadministrativecostsbecomesnegative
forna∗+1.

Caseb)dejure,suppliermakestakeitorleaveit-offerinperiod
2=t

Toclearlydemonstratetheeffectsweareinterestedin,wefirstconsiderthe
casen=1,inwhichasinglesupplierhasbeenchosenoriginally.Then,int=2,
thesupplierofferstheOEMhisoutsideoption,whichwedenotebyg,according
tothesamerationaleasinthepreviouscase.Wemaketheassumptionthatthe
OEM’soutsideoptiondoesnotdependupontheeffortprovidedbythechosen
supplier,whichcaneasilybejustified.12Theninperiodt=0,thesupplier’s
maximizationproblemismaxIv(θi,Ii)−g−c(Ii),leadingtothefirstorder
conditionv2=c.Obviously,thiscontractinducesfirstbesteffortchoiceby
thesupplier,whointhiscaseownstheentireproject.
Thecaseinwhichmorethanonesuppliercompeteamongsteachotherint=
1differssubstantiallyfromthis.Again,firstconsiderthesupplier’sbargaining
problemint=2whoprevailedinthepreviouscompetitionwiththewinning
bid,whichwecallb1.Asthesupplierownsthepropertyrightstotheblueprints,
thisbidcanbeinterpretedastherentgrantedtotheOEMbythewinning
supplier.Howhighwillthesuppliersbewillingtobid?Thewinner’spayoff
–ignoringtheinvestmentcoststhataresunkatthisstage–willbethetotal
surplusoftheprojectminushisbid,i.e.v(θi,Ii(θi))−bi,whilethelosing
biddersstillobtaintheiroutsideoptionworthfj(θj,Ij(θj)).ByA1,therefore,
thesuppliersderivehigherrentsfromwinningthecompetition,aslongasthe
amounttheyhavetopayissmallerthanbˆi=v(θi,Ii(θi))−fi(θi,Ii(θi)).

12Theresultsfromcontactinganothersupplierdonotdirectlydependontheeffortthefirstsupplierhas
example.forvided,pro

91

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

Att=2,thesupplierscompeteàlaBertrandorinasecondpriceauc-
tion.Forsimplicity,weassumedthatatthisstage,theindividualqualities
andinvestmentlevelsarecommonknowledge.13Weobtainthewellknownre-
sultforthiscompetitionthatinequilibriumthemostefficientsupplieri,with
i=argmaxi∈Nv(θi,Ii(θi))−fi(θi,Ii(θi)),submitsanoffertotheOEM,at
whichthesecondmostefficientsupplierisindifferentbetweenmatchingthe
bidandheroutsideoption.14
Tounderstandtheeconomicmechanismsatplayhere,itishelpfultoconsider
theOEM’spayoffsatgivenlevelsofinvestmentandgiventhechoiceofsuppliers
n.Ifwedenotethemostefficientsupplierassupplier1andthesecondmost
efficientassupplier2,theOEM’spayoffatgiveninvestmentlevelsissimply:
v(θ2,I2(θ2))−f2(θ2,I2(θ2))−nk,whichisequivalenttothesurplusthesecond
mostefficientsupplierwouldreceiveifrunningtheproject,minusthecosts
ofsolicitingthenoffers.Ignoringtheinvestmentincentivesforthemoment,
twoeffectsareatplay:asincasea)withincreasingntheexpectedquality
ofthesecondbestsupplierimproves–thoughitfollowsthesecondinstead
ofthemaximumorderstatisticdistribution.Weconsiderthisaleveleffect–
theOEMprofitsfromtheoverallexpectedqualityincreasing.Forthesecond
effectinquestion,itisusefultofirstderivethesurplusofthemostefficient
suppliernetofhisoutsideoptionandignoringthesunkeffortcostsincurredin
d:erioppreviousthe

(4.2)[v(θ1,I1(θ1))−f(θ1,I1(θ1))]−[v(θ2,I2(θ2))−f2(θ2,I2(θ2))]

Whileincasea)theOEMbyconstructionabsorbstheentiresurplusfrom
cooperationnomatterhowmanysupplierscompete,nowthemostefficient
supplieralsoreceivesashare,whichisdeterminedbythedifferenceinsurplus

13Thisassumptionisnotnecessaryinthissetting,asbiddingtheirrealvaluationsisaweaklydominant
strategyinthesecondpriceauction.
14Wemostobtainefficienthetonesamepaysoutcomethesecondwithamostsecondefficienpricetsupplauctionier’s–herebid.eachsupplierbidshervaluationandthe

92

orkramewFAnalytical4.2.

betweenhisandthesecondbestdesign.Asnbecomeslargerthisdifference
shrinks,sothattheOEMreceivesagreatershareofthesurplus,whichisthe
secondeffectwealludedtoabove.

Upuntilnow,wehavetakentheinvestmentlevelofsuppliersasgivento
examinetheeffectsofcompetitiononthebargainingoutcome.Nowweturnour
attentiontotheinvestmentdecisionofsuppliersint=0.Firstletusconsider
theexpectedpayoffsgiventhatthesupplierisawardedtheproductioncontract
lateron,conditionalonthenumberofsuppliersinvolvedinthecompetitionand
therealizationofhisownqualityθi.Againdenotingthemostefficientsupplier
assupplier1andthesecondmostefficientsupplierassupplier2,theseare:

E[(v(θ1,I1(θ1))−v(θ2,I2(θ2))+f2(θ2,I2(θ2))|n,θ1=θi]−ci(Ii))

Fornotationalsimplicity,weabbreviatethisexpressiontoE(Si|n,θi)−ci(Ii).
Wecantakethesurplusgeneratedbysupplier1outoftheexpectationoper-
ator,whichmeansthathisprofitsgiventhathereceivesthecontractarethe
valuehegenerates,reducedbyhisinvestmentcostsandtheexpectedsurplus
generatedbythesecondbestsupplier.Asnotedabove,thelatterincreases
inthenumberofsuppliersinvolvedforgivenlevelsofinvestment.Butsup-
plieridoesnotreceivethispayoffwithcertainty,butinsteadheanticipates
thathewillbeawardedthesubsequentcontractonlyifheisthemostef-
ficientsupplier.Asatthispointtheintrinsicqualityoftheothersuppliers
isunknown,thesupplierexpectsthiseventtotakeplacewithprobability
p[v(θi,Ii)−f(θi,Ii)≥supjv(θj,Ij)−f(θj,Ij)]withi=j.ApplyingA3,
thiscanbereducedtotheshorterexpressionp[v(θi,Ii)≥supjv(θj,Ij)].

Now,spellingoutthemaximizationproblemofsupplieriint=0,weget:

(4.3)Imaxp(θi,Ii,n)E(Si|n,θi)+[1−p(θi,Ii,n)]fi(θi,Ii)−c(Ii)

93

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

Thisleadstothefirstorderconditionwithrespecttooptimalinvestment:
(4.4)c(Ii)=∂p(θi,Ii,n)v(θi,Ii(θi))−maxv(θj,Ij(θj),n)−fi(θi,Ii)
∂Iij=i
+p(θi,Ii,n)∂(v(θi,Ii)−fi(θi,Ii))+∂fi(θi,Ii)
∂Ii∂Ii

Weobserveanumberofcountervailingeffects.Ontheonehand,investment
incentivesaresomewhatdiluted,asthesupplieronlyprofitsfromtheadded
valuewithsomeprobability.Ontheotherhand,hecanincreasetheprobability
thatthecontractwillbeawardedtohimbychoosinghigherlevelsofinvestment.
Notethatirrespectiveofthenumberofsuppliersn,theinvestmentincentives
arestrictlyhigherinthiscasethanincasea)–intuitively,inthedefacto
scenario,thesuppliersneverparticipateinthesurplustheygeneratewithin
therelationship,thereforeeventheincentivesdilutedbycompetitioninducea
highereffortlevel.Onlyasngetsverylarge,thetwoeffortlevelsconvergeat
f2=c.
Againdenotingasupplier’sinvestmentchoiceinthissubcaseanalogouslyto
aboveasIb(θi)andthemostandsecondmostefficientsuppliersassupplier1
andsupplier2,respectively,theexpectedprofitoftheOEMinthiscaseis:

(4.5)EΠbOEM=E[v(θ2,Ib(θ2,n))−f2(θ2,Ib(θ2,n))|n]−nk

Comparingequations(1)and(5)mayleadtotheimpressionthattheOEM’s
profitsarenecessarilyhigherinthedefactocase,asoneiscomparingthemax-
imumandthesecondorderstatistic.Butunlikecasea),inwhichthesupplier’s
investmentlevelwasnotaffectedbythechoiceofn,herethesupplier’sinvest-
mentchoicedoesdependonthelevelofcompetitionandis,atthesametime,
strictlylargerthanabove.Asaresult,wecannotingeneralrankthetwoprofit
.elslevWeareabletosayabitmoreabouttheoptimalnumberofsuppliersin
thetwoscenarios.Ina),increasingthenumberonlyhastheeffectthatthe

94

orkramewFAnalytical4.2.

expectedqualityofthebestsupplierincreases.Inb),wehavethreeeffects:1)
TheOEMexpectsahigherquality,2)hereceivesagreatershareofthesurplus
asnincreases,and3)thesuppliersinvestlessintoquality.Itisextremely
simpletoshowthateffects1)and2)dominatethequalityeffectinscenario
a),i.e.wouldleadtoahigherchoiceofn,wereitnotfortheinvestment
effect.Then,wereallyareonlycomparingthemaximumwiththesecondorder
statistic–weknowthatasn→∞,thesecondorderstatisticapproachesthe
maximumorderstatisticfrombelow.Asbotharemonotoneincreasing,the
slopeofthesecondorderstatisticmustbestrictlygreater.Asaresult,the
OEMwillchooseahighernincaseb)thanincasea)unlesstheinvestment-
deterringeffect3)is“toolarge”inrelationto1)and2).Toenhancethe
intuitionforthisresult,considerthefollowingsimpleexample:Letsupplier
qualitybeuniformlydistributedsothatQ(θ)=θ.Thesuppliers’investment
costfunctionisdeterminedbyc(I)=I22.Thesurplusderivedfromcooperation
isv(θ,I)=θIandeachsupplier’soutsideoptionisworthf(θ,I)=av(θ,I)
witha∈(0,1).Finally,theOEM’soutsideoptionisg=0.Hereweareable
toderiveclosed-formsolutionswhichwesummarizeinthefollowingtable:

I(θ)E(v)EΠOEM
a)n=1aθ3aa(13−a)−k
a)n>1aθnan+2a(1n−+2a)n−nk
b)n=1θ31−k
b)n>1(1−a)θn+aθ(12n−a+1)n+nan+2a((1n−+1)(a)nn(n+2)−1)+(n−2(2n1)(1+1)−a)2−nk
Table4.1.:Investmentdecisions,expectedsurplusandexpectedprofitsofthe
0=tinOEM(forQ(θ)=θ,c(I)=I22,v(θ,I)=θI,f(θ,I)=av(θ,I)).

Inthedejurecase,thesuppliers’investmentishigherby(1−a)θnthanin
thecaseinwhichtheOEMextractsallsurplusexpost.Withincreasingn,
theseinvestmentincentivesdecrease,butduetothefunctionalform,forθ→1,
thepaceofthisprocessgoestowardszero.Asaresult,a“‘wedge”betweenthe
expectedvaluevofthetwostrategiesremainsevenasthenumberofsuppliers

95

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

grolarge.eryvws

Figure4.1.:OEM’sexpectedprofitsdependingonn
(Blue:Πa(lowk);red:Πa(highk);green:Πb(lowk);brown:Πb(highk)).

Whatwecareaboutatthispoint,though,istheoptimalchoiceofn.For
this,wecancomparetheexpectedprofitsinthethirdcolumn.First,compare
onlythefirsttermsofEΠaandEΠb.Theterminthedejurecasestarts
outat0andapproachesthelevelofthefirstterminthedefactocaseas
n→∞.Forthegivenfunctionalform,thisisolatestheeffectofthesecond
orderstatisticapproachingthemaximumorderstatisticatgiveninvestment
levelsalludedtoabove.ThesecondtermofEΠbisalsostrictlyincreasinginn,
whichreflectsthehighershareofthe(larger)totalsurplusbeingabsorbedby
theOEM.Therefore∂∂EnΠb>∂∂EnΠaandtheOEM’soptimalchoiceofnislarger
inscenariob)thaninscenarioa).Figure4.1depictstheOEM’sprofitsin
thesetwoscenariosandgivesanexampleforwhichtheyarelargerineachofthe
twosettings,respectively,dependingonlyontherelativesizeofadministrative
costsk.∗Note∗thatwhileitisnotclearwhichstrategyismoreprofitablein
general,na<nbholds.

96

4.2.3.ASimpleNotionofTrust

TrustandUnderinvestment

4.2.orkramewFAnalytical

Upuntilthispoint,trustdoesnotexplicitlyplayaroleinourmodel.We
attempttointegratetheconceptinthefollowingway.Assumethatpriorto
theirinvestmentchoice,itisnotcertainwhichsubcase,a)orb),willbeplayed
lateroninthegame.Instead,withprobabilityλitisthesupplierwhomakes
thetakeitorleaveitofferinperiod2,whilewithprobability1−λitisthe
OEMwhomakestheoffer.Correspondingly,thesupplierisabletogeneratea
rentabovehisoutsideoptionwithprobabilityλ.
Howdoesthisrelatetotherealityintheindustry?Inourin-depthinter-
views,theclearpictureemergedthatOEMsareatgreatlibertyindesigning
andenforcingcontractualdetailsinrelationshipswithsuppliers,almostirre-
spectiveofthesizeormarketpoweroftheircounterparts.15Ben-Shaharand
White(2006)reportequivalentorevenmorepronouncedfindingsfortheNorth
Americanautomobileindustry.Atfirstglancethereforeitwouldappearasif
onlythea)-subcase(allottingtheentirebargainingpowertotheOEM)de-
scribedabovearerelevantwithrespecttoreality.Thisconfrontationalsetting,
inwhichtherelationshipisdefinedmainlythroughthepureholdup-problem,
isoftensubsumedunderthetermofan“American”procurementstrategy.Yet
theglobalsuccessofJapanesecarmakersbeginninginthe80shasprompted
muchinterestinalternativewaysofsupply-chainmanagement,perhapsmost
famouslyincarnatedintheMIT’sInternationalMotorVehicleProgram.Asa
result,researchersstartedtostresstheimportanceofcooperativeandmutually
beneficialrelationshipsbetweenOEMsandsuppliersintheindustryinachiev-
ingthegoalsofleanproduction.Inthiscontext,TaylorandWiggins(1997)
showthatgrantingthesupplierspositiveeconomicrentscanbeasubstitute
forcontrol(i.e.monitoringqualitylevelsandeventualpunishment).16Tra-
ditionally,cooperativerelationshipsareprevalentintheJapaneseautomotive

15SeeMülleretal.(2008)fordetails.
16SeealsoAghionetal.(2002)forqualitativelysimilarresultsinamoregeneralsetting.

97

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

industryandwerecommonpracticeinGermanyatleastupuntilthemid-90s.
Fora“traditional”micro-economicperspective,considertheentiregamefrom
theprevioussub-sectionstobeonlythestagegameofaninfinitelyrepeated
Markov-game.λthenisastatevariablethatdependsonthe(inourreduced-
formunobserved)previoushistoryofthegame.Atthepointintimeinwhich
thestagegamethatweobserveisplayed,λisgivenandresemblesthesubjec-
tiveprobabilitywithwhichsubcaseb)willbeplayedinthisperiod,i.e.the
probabilitywithwhichtheOEMrespectsthesupplier’spropertyrights.This
ispreciselyourdefinitionoftrustinthefollowing:Highervaluesofλdenotea
highersubjectiveprobabilitywithwhichtheOEMwillgrantthesupplierthe
extractionofashareoftherentgeneratedwithintherelationship–inother
words,thehigherλ,themorethesuppliertrustsintheOEM’swillingnessnot
toexploithissuperiormarketpower.
Werefrainfromspellingouttherepeatedgameforthefollowingreasons:
First,thedataonwhichwebaseouraccompanyingempiricalanalysisarecross-
sectionalwithoutapanel-dimension.Therefore,whatweobservecanbecon-
sideredexactlyoneperiodofthestagegame,inwhichthecurrentrelationship-
specificlevelsoftrustareexogenouslygiven,andthisallowsustoapplythe
modelmoredirectly.Second,inthissetupofthemodel,whatmattersisthe
supplier’sleveloftrusttowardtheOEM.Webelievethatthemaindriving
forceintheserelationshipsisthehold-upproblemandthereforeourobjective
istoseewhetherandhowthiscanbemitigatedbythebeliefofsuppliersthat
abuseisless(ormore)likely.Finally,lookingatthisreducedformallowsus
toremainagnosticaboutthereasonsfortheOEMwantingtograntthesup-
plierthisrent.17Tostatejusttwoexamples,asinTaylorandWiggins(1997)
therecouldbeatradeoffbetweenrentextractionandcontrolcosts,orrelat-
edly,itmaybeworthwhiletomakethesupplierfeartheconsequencesofthe
relationshipbeingterminatedasinAkerlofandYellen(1990).
Ourmodelallowsustostateourcentralpredictionverystraightforwardly:

17Forthegametobeontheequilibriumpath,thesubjectiveandtheobjectiveprobabilityofsubcaseb)
beingplayedwouldhavetobethesame.

98

orkramewFAnalytical4.2.

Hypothesis1:HigherlevelsofsuppliertrustintheOEMareassociated
withhigherrelationshipspecificinvestmentsbysuppliers.

Asshownabove,thisholdsforanygivenprocurementstrategychosenbythe
OEM–buteventhisreducedformmodelconveyssomeofthecomplexityofthe
procurementdecision,whichalsodependsonthelevelofqualityuncertaintyre-
gardingtheproduct.Clearly,allelsegiven,higherlevelsofqualityuncertainty
willinduceprocurerstoinducecompetitionbetweenmoresuppliers.

TrustandtheOptimalInducedLevelofCompetition

Whilewegetaclearcutandveryintuitiveresultwithrespecttotherelation-
shipbetweentrustandupstreaminvestment,theanalysiswithregardtothe
levelofcompetitioninducedbytheOEMissomewhatmorecomplicated–and
surprising.Asdescribedabove,alargepartoftherelationalcontractingliter-
ature,seee.g.recentlyCalzolariandSpagnolo(2009),arguesthatcompetitive
arms-lengthagreementsontheonehandandrelationalcontractsontheother
willbeusedinmutuallyexclusivesettings,depending,forexample,ontheen-
forceabilityofcomplexclauses.Supportedbythesefindingsonemightexpect
inanaïvefirstapproachthatsupplier-OEMrelationshipsgovernedbytrust,
i.e.withrelativelyhighλinourmodel,shouldbeassociatedwithlessinduced
competition,i.e.lowern.Ouranalysisabovehasshownthatgenericallythe
oppositeisgoingtobethecase,asλsimplygeneratesaconvexcombination
betweenscenariosa)andb)andinthelatter,ahigherlevelofcompetitionis
optimallyinduced.Figure4.2displayshowtheoptimalnincreasesforthe
simpleexamplestatedaboveasλincreasesfrom0to1.
Theintuitionbehindthisresultisassimpleasitisstriking.Inthede
factosetting,theOEMextractsthesupplier’sentirerentthroughhissuperior
bargainingposition,exploitingtheexistinghold-upsituation.Inthedejure
setting,heforgoesthispossibility–inducingcompetition,i.e.achoiceofa
highernishisalternativemechanismtoextractrentfromthecooperation.In
thissense,trustandcompetitionarecomplementaryinthissetting.Wecan

99

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

Figure4.2.:(Blue:OEM’s0.0;redexp:0.2;ectedgreen:profits0.4;brodepwn:ending0.6;onorange:nfor0.8,cydifferenan:t1.0).valuesofλ

directlyderivethesecondhypothesisfromtheformalmodel:

Hypothesis2:HigherlevelsofsuppliertrustintheOEMareassociated
withmoreintensesuppliercompetitionintheprocurementprocess.

Whilethisresultsratherdirectlyandclearcutfromtheformalmodel,it
willprovetobeverydifficulttodistinguishourconceptoftrustfrom,e.g.,
theconceptofrelationalcontractingora“naïve”understandingoftheterm.
Ineachofthesecases,thepredictionswouldbetheoppositeofoursecond
hypothesis,andthereforeweapproachthetaskathandasoneofcompeting
hypothesesandwillletourdatadecide.

100

4.3.AnalysisEmpirical

DataofSource4.3.1.

AnalysisEmpirical4.3.

Ourdataresultsfromanonlinequestionnairestudythatwascarriedoutforthe
Germanautomotiveindustryassociation18(VDA)fromFall2007untilSpring
2008.Thequestionnairewasdesignedonthebasisoftheresultsofacase
studyperformedinSpring2007,inthecourseofwhichinterviewswithhigh
rankingexecutivesintheautomobileindustrywereconducted.19Weobtained
auniqueviewoftherelationshipsbetweenoriginalequipmentmanufacturers
andtheirtier1-suppliers,withatwofoldapproach:First,eachof13partic-
ipatingsupplierswasaskedtoevaluatetheirrelationshipwitheachofupto
11OEMsactiveintheGermanmarketfordifferentrepresentativeproductsin
theirportfolioinclinicaldetail–morethan300questionswereaskedcovering
allcentralfunctionswithinthefirms.Inaddition,theparticipatingOEMswere
askedtoevaluatetheirsourcingrelationshipsingeneral–i.e.notspecifically
forindividualsuppliers–foreachofthefourdifferentproductclassesaccording
classification:industryestablishedtheto

•Commodities:physicallysmallandtechnologicallyunsophisticated
(e.g.shockabsorbers)

•High-techcomponents:physicallysmallbuttechnologicallysophis-
ticated(e.g.electronicbrakecomponent),inthefollowingreferredto
t.onencompassimply

•Modules:physicallylargebuttechnologicallyunsophisticated(e.gfront
end)

•Systems:physicallylargeandtechnologicallysophisticated(e.g.break
system)

18VerbandDeutscherAutomobilunternehmene.V.
19Forthequalitativeresultsofthiscasestudy,seeMülleretal.(2008).

101

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

AsOEMsansweredasetofquestionsalmostidenticaltothesupplierques-
tionnaire,wearetherebyabletocomparetheirviewoftheirgeneralpolicies
withthesuppliers’view.Intotal,morethan1,500questionnaireswerefilled
inbycompetentengineers,procurement-andsalesofficerswiththefollowing
methodology.Aparticipantfirstwouldhavetoindicatehisfunctionwithinthe
companyoutofthefollowing20:

•Pre-development:“Basic”technologicalresearch,notmodel-specific.

•VehicleDevelopment:Car-modelspecific(technologyadaptation).

ctionduProSeries•

trolConyQualit•

Sales•

Logistics•

Aftermark•et

Thenshewouldchooseaproductforwhichshehadthenecessaryknow-how
aswellasthecustomerssheworkedwith,thelatterfromalist.Foreach
productandcustomer,shewouldthenanswerasetofquestionssuitedtoher
functionwithinthecompany.
Oneobservationinourdataiscomposedoftheanswersoftheentiresupplier
questionnaireforagivenproductandagivencustomer.Thisexceedsthe
extentofanygivenfunction-questionnairelistedabove–therefore,inorder
toobtainascompleteobservationsaspossible,wemergetheanswersfroma
givensupplier,productandcustomeroverallfunctionstocoverallaspects
oftherelationship.Whateachobservationthereforedescribes,isoneview
(potentiallyofmanypeople)oftherelationshipbetweenthesupplieranda

20Foradetaileddescriptionoftheindividualfunctionsandtheautomobiledevelopmentandproduction
process,werefertoMülleretal.(2008)

102

Empirical4.3.Analysis

givenOEMforoneproduct.WemergethiswiththeresultsfromtheOEM-
questionnaireinordertobeabletocontrolforthenonsupplier-specificbehavior
OEMs.ofOnpaper,wehave792observations,butfortworeasonsthesearenotneces-
sarilycomplete:First,noteachfunctionwithinacompanyfilledoutaquestion-
naireforeachproductstudied.Inthiscase,wholesectionsofthequestionnaire
aremissingfortheobservationcoveringthegivenproduct.Second,participants
couldskipindividualquestionsandmadeampleuseofthisoption.Therefore
thenumbersofobservationsovertheindividualquestionsdiffersubstantially,
asseeninthedescriptivedatabelow.

StatisticseDescriptiv4.3.2.

Theunderlyingquestionnairesoughttodepictcomplexrelationshipsinhith-
ertounmatcheddetail.Inthefollowingsubsections,thereforewewillexert
effort–perhapsmorethanusual–tointroducethevariablesofthestudy
andshinesomelightonthebasicforcesandtensionsthatareatplaybetween
manufacturersandsuppliers.

ParticipatingCompaniesandCharacteristicsofParts

OntheOEMside,10ofthelargestplayersintheGermanmarketparticipated
activelyinthesurvey,7producersofpassengercarsand3truckmakers.Onthe
otherhand,13suppliersactiveintheGermanmarketprovidedtheirinputon
11manufacturers-i.e.the10participatingplusonefurthercarmanufacturer.
Thesuppliersampleisverystronglybiasedtowardslargeparticipants,with
average2007revenuesof9.4billionEuro(standarddeviation(std)12.4)and
eventhesmallestparticipantpostingrevenuesabove700millionEuro.Thisis
emphasizedbytheself-reportedEuropeanmarketsharesfortheproductinour
sample:For161observationsthiswasprovidedonafivepointscalewithan
averageof3.76(std.90),whichtranslatesintoashareofmorethan25%.Not
surprisingly,thecorrelationofmarketsharewithintensityofsuppliercompe-

103

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

tition–alsoona5-pointscale–isnegativewithavalueof-0.20(significantat
5%-level).Further,weobserveanegativecorrelationofsuppliersize(measured
bythe2007-revenuesinbillionEuros)andtheintensityofsuppliercompeti-
tion(-.144,p-Value:.072)andapositivecorrelationofsuppliersizeandmarket
shareintheobservedpart(.124,p-Value:.083).

Thiscouldraisetheworrythatthelargersuppliersmaybeabletoexert
monopolypoweroverOEMsforsomeofthepartswestudy–tocounterthis,
wemadesurethatthereareatleasttwosuppliersactiveinGermanyforeach
partinoursample.Nevertheless,wewillhavetotrytocontrolforrelative
marketpowerinourregressions,asitmayclearlyaffectbargainingstrength
option.outsideOEM’stheand

Apartfromtheirtype,productsarefurtherspecifiedbytheR&D-shareof
totalcostsaswellastheassessmentbythesupplierhowimportantthedegree
ofinnovationisfortheparticularpart.Bothweremeasuredona5-point
scale–theimportancerangingfrom1–verylittle–to5–veryhigh,while
thecost-shareswereprovidedin2%increments,thereforerangingfrom<2%
to>8%.Asonewouldexpect,theanswerstothequestionsarestrongly
correlated(0.27,significantat0.1%-level).Moreinterestingly,though,they
allowustorevisitthemeritsoftheunderlyingtype-classification.Thefollowing
Table4.2displaysthedescriptivestatisticsforthesequestionsbyunderlying
e:yptductpro

Performingpairwiset-testsshowsthatthemeansforbothvariablesaresig-
nificantlylowerforcommoditiesandmodulesthanforsystemsandcomponents,
whileamongthesetwogroupsthehypothesisofequalmeanscannotberejected,
whichisexactlyinlinewiththeindustryspecificationdiscussedabove.This
allowsustointroduceanadditionaldummymeasureoftheinnovativenessof
apart(dummy_soph)whichtakesthevalue1ifthetypeisasystemorcom-
ponent.Further,toaccountforpotentialpricedifferencesduetothesheersize
ofapart,weintroducedummy_bigwhichtakesthevalue1forsystemsand
les.dumo

104

ariableVSystemsR&DShareCostImportanceR&D
MoCostduleSharesR&D
ImportanceR&D
tsonenCompR&DShareCostImportanceR&D
tiesdiCommoR&DShareCostImportanceR&D

4.3.AnalysisEmpirical

Mean(Std.Dev.)MinMaxObs

(1.43)4.19(.74)4.16

(1.76)2.06(.76)2.89

(1.40)3.48(.87)3.42

(1.35)2.35(.69)2.91

11

11

12

12

375505

185184

215545

591934

Table4.2.:ImportanceofInnovationandCostshareR&Dbyproducttype.

onshipRelatitheofCharacteristics

Inthissection,wepursuetwodifferentgoals.Whileintroducingthevariables
describingtherelationshipswealsoattempttoshedadditionallightonthedif-
ferencespertainingtoproduct-type.Thereisafurtherdimension,inwhichwe
canexploitexistingvariation:Asnotedabove,wegatherinformationonthe
relationshipbetweenOEMsandsuppliersovertheentirecar-modellife-cycle.
Inthefollowing,thedistinctionbetweenthreeofthesephasesisespecially
important:pre-development,developmentandseriesproduction.Thelastof
thesephases,seriesproduction,istheleastcomplicatedcase–supplierswork
withexistingblueprintsandcompletelydesigned(orexisting)toolstoproduce
givenquantitiesofthepartinquestion.Theproductandservicescanclearlybe
specifiedthroughcontractswithoutmuchroomformisunderstanding,forex-
amplespecifyingacceptablefailureratesanddeliveryconditionsindetail.The
(model-specific)developmentphaseisinmanywayslessclearcut.Whilethe
generalrequirementsthataparthastomeetaredefinedbyitsfunctionwithin
theautomobile(abrakehasarelativelyspecificfunctionandplace,giventhe
projectedweightandtop-speedofthemodelinplanning),aplethoraofother
partswithwhichithasinterfacesarebeingdesignedinparallel.Blue-prints

105

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

forthepartdonotyetexistatthebeginningofthedesignphase.Clearly,the
objectivescannotbedrawnuppreciselyexanteincontracts,butaresubject
toacontinuouscooperativeprocess.Finally,theseuncertaintiesbecomeover-
whelmingwhenconsideringthe(notmodel-specific)pre-developmentphase:
Hereforexample,thesupplierisresearchingbrake-technologieswithoutknow-
inghowfastorheavythemodelinwhichitwillbeusedisgoingtobe.In
general,morefundamentalresearchisinvolvedhere–and,asshouldbeclear
fromthenatureoftheendeavor,itisevenhardertowritespecificandpre-
ciseenforceablecontractsregardingtheoutcomes.Asourrespondentswere
involvedinthedifferentstagesoftheproductlife-cycle,inawaylikeBrown
etal.(2004)thisallowsustoexogenouslychangethelevelofexternalenforce-
abilitywhilekeepingproductandrelationshipcharacteristicsconstantfora
numberofquestions.

Asacaseinpoint,werequestedsupplierstoevaluatetheOEM’ssupplier
choicecriteriaonasix-pointscalefrom1–norelevanceto6–veryimportant,
foreachofthephasesintheproductlife-cycle.Fromthediscussioninthe
literaturewewouldexpect“relational”choicecriteriatogrowrelativelyless
importantasopposedto“hard”criteria–suchasprice–asoneprogresses
frompre-developmentonward.Ourempiricalresultsstronglysupportthishy-
pothesis.Theimportanceofpricestrictlyincreasesfromanaverageof5.10
(seeTableB.1),to5.37(seeTableB.2)to5.70(seeTableB.3)(t-testsfor
differenceofmeanareeachsignificantatthe0.1%level).Ontheotherhand,
theimportanceoftrustisrespectively4.89(seeTableB.1),4.90(seeTable
B.2)and4.73(seeTableB.3)forpre-development,developmentandseriespro-
duction.Thereforetrustisonlysignificantlylessimportantwhenchoosinga
seriessupplier(p<0.1%),whilethereisnodifferencebetweenpre-development
anddevelopment.Butwiththiskindofquestion,itiseasiertointerpretrel-
ativemagnitudeofanswers:Whenwelookatthedifferencesindifferences
betweentheimportanceofthechoicecriteria,thereisamonotonerelation-
ship,withpricebecomingrelativelymoreimportantforeachstep(p<5%for
pre-developmenttodevelopmentandp<0.1%fromdevelopmenttoseriespro-

106

duction).

Empirical4.3.Analysis

[TableB.1,B.2andB.3abouthere]

Next,wesuggesttwoproxiesforthevalueoftheOEM’soutsideoption(de-
notedginthetheoreticalframework).Thefirstistheshareinthevolumeof
thepartprovidedbythesuppliertotheOEM(measuredona5-pointscale
whereeachpointresemblesa20%difference,with1–<20and5–>80%).Pre-
sumably,itismoredifficulttoshiftalargershareofproductionawayfromone
suppliertoanotherthanasmallerone,thereforegshouldbenegativelyrelated
tothismeasure.Lookingathowthismeasurebehavesforthedifferentprod-
ucttypesshowsthattheshareprovidedhasstatisticallynon-distinguishable
meansformodules,componentsandcommodities(thesub-samplemeansare,
respectively,3.08,3.26and3.23,i.e.attheupperendoftherange40-60%),
whileforthesystems,thisvalueissignificantlyhigherat4.07(orattheupper
endoftherange60-80%).Therelianceonanindividualsupplierthereforeis
significantlystrongerinthecaseofsystemsthanfortheothertypesofpart.
Asthesecondpotentialproxy,therespondentswereaskedtoassesshow
oftentheOEMchoosestoproduceagivenparthimselfona6-pointscale
from1–neverto6–veryfrequently,with4–about50%ofcasesasafurther
anchor.Thisalsoallowsustocreateadummyvariablewhichtakesthevalue
1whenevertheanswerisdifferentfromnever.Theabilitytoproduceapart
himselfisperhapsthemostintuitiveoutsideoption–anditisoneofwhich
OEMsaremakingampleuse,astherecentwaveof“in-sourcing”demonstrates.
Inday-to-daybusiness,therecanbedifferentreasonsforthis,mostcommonly
capacityutilizationsmoothing21orworriesaboutsuppliers’abilitytoprovidea
partasagreed.Thecomparisonofsub-samplemeansshowsthatsystemsand
components,i.e.thetechnologicallysophisticatedparts,aresignificantlyless
likelytobealsoproducedbytheOEMhimselfthancommoditiesandmodules,
withthelatterbeingthemostlikelytobein-sourced.Thismaybedueto
thefactthatphysicallylargerpartsaremorecostlytostockpileandutilize
21Laborlawsmakeshort-termadjustmentstotheworkforceallbutimpossible.

107

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

morecapacity.Clearly,higherlevelsofthisvariableshouldbeassociatedwith
ahigheroutsideoptionofmanufacturers,i.e.ahigherg.
Finally,weaskedtherespondentstoevaluatethelevelofspecificityofthe
contractualrequirementsatdifferentdevelopmentstages,aswellasthe“degree
offreedom”inrelationshipinordertobeabletogeneratesomeinsightsinto
theinteractionsbetweentrustandcontractspecificity.

ProcurementDecisionsbytheOEM

WehavetwosetsofvariablesthatmeasuretheOEM’sprocurementdecisionsat
differentpointsintheproductlife-cycle,onequalitativeandonequantitative.
Forthequalitativemeasure,weaskedtherespondentstoevaluatehowoften
differentprocurementstrategieshavebeenemployedbytheOEMforeach
ofthedifferentstages.Thismayappearslightlyparadox,asforeachpart
amanufacturershouldapplyonestrategy,butpartsareprocuredanewfor
eachnewseriesofagivenmodel,i.e.thereisanewprocurementprocess
every1.5to2yearsandclearlydifferentstrategiescouldbeusedatdifferent
pointsoftimeinthepast.Forpre-development,theoptionswerepreselection
ofaspecificsupplierandprocurementamongalimitednumberofsuppliers,
eachona6-pointscalefrom1–neverto6–veryfrequently.Fordevelopment
andseriesproduction,afurtheroptionwasadded,openprocurement,which
playsnoroleinpre-development.Eventhepurelydescriptiveresultsoffer
someinterestinginsights.Forpre-development,OEMsareactuallysignificantly
morelikelytocontractwithspecificsuppliers(mean4.43),thantogothrough
alimitedcompetitiveprocurementprocess(mean3.95,t-testfordifference
ofmeanssignificantat1%level;seeTableB.4).Incontrasttothis,pre-
selectionofsuppliersissignificantlylesslikelybothfordevelopment(mean
3.06)andseriesproduction(2.98),seeTablesB.5andB.6,respectively.On
theotherhand,fordevelopmentOEMsaresignificantlymorelikelytoprocure
amongalimitednumberofsuppliers(mean5.18,seeTableB.5),therefore
thereisaclearshifttomoremarket-basedinteractionsfrompre-development
todevelopment.Thesamekindofshifttakesplaceagainfromdevelopment

108

AnalysisEmpirical4.3.

toseriesproduction,whereprocurementamongalimitednumberofsuppliers
growslessimportant(mean4.55,seeTableB.6),butthereisasignificant
increaseintheuseofopenprocurement(2.44insteadof1.97,seeTablesB.5
andB.6).Clearly,thepicturethathasbeguntoemergeabove,i.e.ashift
tomorearms-lengthinteractionsastheproductreachesthedevelopmentand
seriesproductionphaseissupportedbythesedata.Webelievethistobedriven
mainlybyincreasingcontractibility,whenviewedtogetherwiththeresultsof
etitativquanthemeasure.

[TableB.4,B.5andB.6abouthere]

Forthis,weaskedhowmanysuppliersprovidedthegivenserviceorpro-
ducethepartinparallel,differentiatedforadditionalphaseswithineachof
the(bynow)familiarstages.Thedevelopmentstagewassubdividedintothe
phasesproductplanning,productspecification,conceptdevelopmentandde-
taileddevelopment(startingfromtheearliest).Forseriesproduction,weasked
forthenumberofsuppliersatseriesstart,after1-2yearsandaftermorethan
2years.Theresultsfromthisappeartobesomewhatcounter-intuitive.For
pre-developmentthereareonaveragemorethantwo(2.16,TableB.4)suppliers
competinginparallel.Thisnumberstaysaboutconstantinthefirststagesof
development,beforeitsignificantlydecreasesforthelastdevelopmentphase
downto1.51(seeTableB.5).Itreachesitsnadiratthebeginningofseries
productionwith1.20,beforeitincreasesagainto1.59twoyearsintoproduction
(seeTableB.6).Howdoesthismeshwithourpriorresults?Thepreviousques-
tionsonlyaimedatthechoiceprocedure,insteadofathowmanysuppliersare
selected.Forpre-development,duetothelackofspecificityconcerningtheob-
jectives,openprocurementisnotfeasible–preciselyforthisreason,thereisthe
greatestuncertaintyregardingtheoutcomeoftheprocess.ThewaythatOEMs
dealwiththis–alsosuggestedbyourmodel–istohavemultiplesupplierswork
onthedesigns.Asseenabove,thesearefrequentlyhand-picked.Onthebasis
ofthemostpromisingapproachtheOEMthenentersintothedevelopment
process.Thereisastrongincentiveforsuppliersfortheirpreliminarydesign

109

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

tobechosen,asthecontractualreimbursementforpre-developmentworkison
averagebelow60%oftheactualcosts,whetherornotthecompanyisawarded
asubsequentdevelopmentcontract.Ananalogousprocessisrepeatedagain
forthedevelopmentprocess,whichresultsinaspecificblueprint.Withthis
blueprint,thequalityuncertaintyispracticallyeliminated,giventhatsuppliers
aregenerallycertifiedthroughstringentqualityassuranceprocesses,therefore
thiscomponentiseliminatedfromthedecisionproblem.Inproduction,fewer
supplierswithhighervolumespromisethehighesteconomiesofscalesandthe
steepestlearningcurves,thereforethenumberofsuppliersdropssignificantly
atproductionstart.Oncetheseeffectshavebeenrealized,theOEMcanstart
in.suppliersadditionalbringtoWebelievethatthisbackgroundisextremelyvaluableandshouldbeborn
inmindforthefollowinganalysisofwhichroletrustplaysintheinteractions
setoutabove.

4.3.3.MeasuresofTrust:WhoTrustsWhom-andWhy?

Trustisasensitiveconceptwhichhasprovedtosomedegreeelusivetoat-
temptsatexplanationbyeconomists.Whileexistingstudieshaveemployed
eitherexperimental/behavioralevidenceorsubjects’answerstovariationson
thequestion“Towhichdegreecanotherpeoplebetrusted?”,ourdatahasthe
hugeadvantagethatitisrelationship-specific:Weaskrepresentativesofcom-
panyAabouttheirstanceandmisgivingstowardcompanyBwithregardto
theinteractionsconcerningaspecificproduct.Clearly,therearedrawbacksto
thisapproachaswellthatneedtobeaddressed.Wedevotethefollowingthree
subsectionstodeterminehowrobusttheindividualmeasuresareandwhether
andhowtheycanbeapplied.Firstweintroducethequestionsthatwebelieve
toberelatedtotheconcept“trust”.Thenweuseamethodwell-established
insociology,exploratoryfactoranalysis,totrytoshedadditionallightonthe
dimensionalityoftheconstructweareobserving–i.e.,isthereonlyonekind
of“trust”,ordothequestionsweobservereallydepictaconstructcomposedof

110

AnalysisEmpirical4.3.

variousdifferent“factors”.Finally,wetrytoprytheblackboxfromSection
4.2.3open(ifonlyaslit),bytakingaglanceatwhichpastoutcomesand
behavioraffectthesuppliers’evaluationoftrustintherelationship.

HowDoWeMeasureTrust?

Weattempttocapturetrustinthevarioussupplier-manufacturerrelationships
intwo,notnecessarilymutuallyexclusiveways.Oneapproachisrelatively
direct,askingtoevaluatemutualtrustorinquiringwhichroletrusthasplayed
forimportantdecisions.Thesecondapproach,whichwecallindirect,istolook
atreportedbehaviorforwhichtrustcanbeconsideredaprerequisite.
Animportantcontextfortrustintheserelationshipsistheareaofintellectual
property.Especiallybasic,nonmodel-specificresearchresemblesanimportant
shareofsuppliers’capitalandembodiestheirabilitytodifferentiatethemselves
–thisabilityallbutdisappears,forexample,ifanOEMweretotakeasupplier’s
blueprintsforapartandmakethemaccessibletocompetitors.Muchofthis
know-howisinvolvedintheearlieststages,thepre-developmentofproducts,
wheresuppliersshowcasetheiradvancedknow-how.Inourinterviewsinthe
preparationofthestudy,therewasamentionofthepracticetodenythemost
advancedtechnologytoOEMswhowereexpectednottotreatitwiththe
necessarycare,i.e.whowerenottrusted.
Thereforeasafirstdirectmeasureoftrust,weinquiredaftertheimpor-
tanceofthetrustrelationshipwiththeOEMinafirm’sdecisiontoinitiate
apre-developmentprojectonasix-pointscalerangingfrom1–norelevanceto
6–veryhighrelevance(fromnowonTrust1).Tobeabletorelatethistoother
criteria,weaskedthesamequestionsfortheimportanceofthefactorssales
potential,productpositioningandlong-termcooperation,sothatwecanuse
boththeabsolutevalueoftheansweraswellastherelativerankasmeasures
trust.ofAstheseconddirectmeasure,weaskedthequestion:“Howdoyouevaluate
mutualtrustbetweenOEMandsupplierwithrespecttohonoringeachother’s
intellectualpropertyrights(IPR)?”onafive-pointscalerangingfrom1–very

111

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

littleto5–veryhigh(fromnowonTrust2).Whilethefirstquestiononly
involvestheleveloftrustofthesuppliertowardstheOEM,thesecondquestion
isphrasedtocoverbi-directionaltrust.Clearly,adisadvantagetothissecond
questionisthatthesuppliermustalsogiveanestimateoftheotherparty’s
t.assessmenWealreadyencounteredthethirddirectmeasureoftrustabove.Foreach
phaseoftheproductlifecycle,thesupplierswereaskedfortheirviewofthe
OEM’schoicecriteriaforchoosinghissupplier(pre-development:Trust3,
development:Trust4,seriesproduction:Trust5).Again,wehaveboththe
absolutevalueoftheimportanceoftrustaswellastherelativerankcompared
tocost,personalcontact,durationofcooperationandcertification.Thisisthe
supplier’sassessmentoftheOEM’spreferences,only,soweclearlyneedto
evaluatethereliabilityofthismeasure.Second,analogoustoSapienzaetal.
(2007),wewillhavetocheckwhetherrespondentsperhapsrelatedtheirown
leveloftrustintheOEMinthesequestions,instead.
Ourpotentialindirectmeasuresoftrustareassociatedwithbehaviorthat
isrelatedtoIPRprotectionandtothesecrecyofthecost-structureofthe
supplier.22Suppliersstatebothhowoftentheyprovideoriginalresearchdata
totheOEMonafive-pointscale(1–veryrarelyto5–veryfrequently)aswellin
aseparatequestionhowoftentheOEMprovidesaccesstohisoriginalresearch
dataonthesamescale.Clearly,boththelevelsandthedifferencebetweenthe
twovaluesmaybeofinterest.Furtherweinquireonthesamescalehowoften
thesupplier’scostsaremadetransparenttotheOEM.Aninterestingissue
withthesemeasureswhichwewillhavetoattempttodisentangleiswhether
suppliersareforcedintorevealingthesedataduetotheOEM’ssuperiormarket
powerorwhetherthisistrulyaresultoftrust.Inordertodeterminethis,
therelationbetweenthefrequencieswithwhichasupplierandtheOEMreveal
originalresearchdatawillbeofinterestandweintroducethedifferencebetween
22Thesuppliers’costsareanextremelycontentiousissueinnegotiations.Cost-cuttingmanufacturers(have
to)Therefore,accepttheythatasuppliertraditionallyprotryducintogbnegotiateelowcostpriceswillthathavearetoasgocloseouttoofthebusinescostssassoponerossibleratherandbthanegrudgelater.
thesuppliersanypositivemarginthattheyobtain.Foranexcellentandcomprehensivediscussion,see
theclassicalWomacketal.(1991).

112

4.3.AnalysisEmpirical

thetwoasanadditionalvariable.
Asafirststeptowardsbetterunderstandingthesemeasures,TableB.7dis-
playspairwisecorrelationsbetweeneachofthem.Asonewouldhaveexpected,
thereisasignificantpositivecorrelationbetweenallofthedirectmeasuresof
trust.Fortheindirectmeasures,thepictureismoreinteresting:Thefrequency
withwhichthesuppliermakeshiscoststransparentisnegativelycorrelatedto
twoofthedirectmeasures–theimportanceoftrust(fromtheOEM’sview)
inchoosinghisdevelopmentandseriessupplier.Thesemeasuresreflectthe
attitudeofOEMsintheselectionandnegotiationprocessofdevelopmentand
seriessuppliers–itappearsplausiblethatitistheinsistenceoftheOEM,
therefore,andnotnecessarilytrustthatcausessupplierstobaretheircosts
morefrequently,whichmakesthismeasurenon-satisfactory.

[TableB.7abouthere]

Thesupplier’sprovisionoforiginalknow-how,ontheotherhand,isnot
correlatedwithanyofthedirecttrustmeasures,whilethereisarelatively
strongsignificantpositivecorrelation(.443)withtheprovisionofknow-how
bytheOEM.Thelatterisalsopositivelycorrelatedwiththemutualtrust
regardingthetreatmentofIPR.Whatwouldweexpecta“trusting”asopposed
toa“forced”relationshiptolooklike?IftheOEMforcesthesuppliertoreveal
intellectualpropertysecrets,thisshouldnegativelyaffectthelevelofmutual
trustwithregardtoIPR.Further,wewouldexpectthat–inthesekindsof
relationships–theOEMprovidesrelativelylittleintellectualpropertyintothe
relationshiphimself.Finally,therelativemarketandbargainingpowermay
playaroleinthiskindofrelationship.Todetermine,whetherthiseffecttruly
existsinthedata,weregressthedifferenceofIP-secretsprovidedontothe
levelofmutualtrustwithregardtoIPR,adummywhethertheproductis
technologicallysophisticatedandthesupplierrevenuesasaproxyforrelative
marketpower.TheresultsoftheOLS-regressionareprovidedinTable4.3
23w.elob23Wealsoperformedorderedlogitregressions,whicharemoresuitedtothestructureofthedata.Theresults

113

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

VariableCoefficient(Std.Err.)
Trust2-0.233**(0.113)
(0.201)0.249y_sophdummSupplierRevenue-0.015**(0.007)
3)(0.412.009const.

Table(N4.3.:=129,(OLS):*significanDifferencetatin10%,**frequencysignificanoftatrev5%,ealing***originalsignificantatresearc1%)hdata

Theregressionresultsshowasignificantnegativeassociationofthediffer-
enceinIP-provisionandmutualtrustaswellassupplierrevenue,which,aswe
showedabove,canbeusedasaproxyforrelativemarketpower.24Therefore,
thelowertherelativemarketpowerofasupplier,themorelikelyitisthathe
providesmoreoriginalresearchthanhisopposite,whichwetakeasanindica-
tionthatintheseasymmetricsetting,enforcedrevelation(whichisnegatively
associatedwithtrust)doestakeplace.Inthesimplepairwisecorrelationsthe
overalleffectofsupplierprovisionofIPonourmeasuresoftrustisneutral
(notsignificantlydifferentfromzero).Thereforetheremustbeinformationin
thismeasurethatcountervailstheeffectoftheonaverageincreasingdifference
inprovisionofIP.Intuitively,onecouldimaginetheretobethreecoexisting
IP-regimes:Onesymmetriconecharacterizedbydistrust–here,bothparties
providelittleornoresearchfindingstoeach-other.Oneasymmetriconechar-
acterizedbyforce,theexistenceofwhichissuggestedbytheregressionsabove.
Andfinally,onesymmetriconecharacterizedbytrust–here,bothpartiespro-
videresearchtoeachotherrelativelyoftenandinsimilaramounts.Wetryto
usethefollowingmeasuretobeabletoaccountforthedifferencesinthethree
regimes:Firstwecreateadummythattakesthevalue1ifthedifferencebe-
tweentheprovidedIPisnottoolarge,i.e.nolargerthan1.25Next,weinteract

arequalitativelyidentical(signsandp-values),wereporttheOLSregressionfortheeasierinterpretability
theofts.efficienco24Thecorrespondingregressionwithmutualtrustasadependentandthedifferenceasanindependent
variableshowsasignificantnegativeeffectofthedifferenceontrust.
25Takingthisasanottoolargedifferenceissomewhatarbitrary.Therearetworeasonswhywefindit
sensible.Peopletendtooverestimatetheirowncontributioncomparedtoothers.Further,thevalueof
1leadstothehighestcorrelationofthefinalmeasurewithourdirecttrustmeasures.

114

AnalysisEmpirical4.3.

thisdummywiththefrequencyofIP-provisionbythesupplier.26Theresulting
measurehasasignificant,positivecorrelationwiththereportedmutualtrust
withregardtoIPR(.163,p-value.05).
Asaresultoftheseconsiderations,weareleftwiththe5“direct”trust
measures.Ourknowledgeoftherelationshipbetweentheobservedmeasures
upuntilthispointisbasedonpairwisecorrelationsalone.Factoranalysisis
amethoddesignedtomakebetteruseofthese“within”correlationsbetween
asetofvariablesinordertoextendwhatcanbelearnedfromthem.Using
alatentvariableapproach,itmaximizesandrecordstheshareofvariation
intheobservedvariablesthatcanbeexplainedbyoneunobservedfactor(or
more),whilereproducingthecorrelationsbetweenvariables.27Themethodhas
beencriticizedinthepastforproducingresultsthatarenotunique,butwe
finditperfectlysuitedtoproduceakindof“upperbound”inourexploratory
setting,i.e.toexplainhowmuchofthevariationinourmeasurescanatmost
beexplainedthroughtheunobservedunderlyingfactor,whichweassumeto
be(atleastassociatedwith)trust.Oneremainingdifficultythatwefaceis
thatweonlyhave59observationsinwhichall5variablesareincluded,but
eventhislownumberofobservationscanbesufficientina1-factor,5-variable
modelasMacCallumetal.(1999)argue,andweperformanumberoftestsfor
robustness.Table4.4displaysthefactorloadingsanduniquenessoftheindividualvari-
ablesusingtheprincipal-factormethodandlimitingtheadmissiblenumberof
factorsto1.Theresultingpatternisrobusttousingthemaximum-likelihood
estimationapproach,toallowingasecondandthirdunderlyingfactorandto
recursivelyeliminatingindividualfactors(therebyobtainingsignificantlymore
observations).Inallspecifications,theuniquenessforthevariablesthatmea-
suretheimportanceoftrustinprocurementnegotiationsatdifferentstages
(trust3-5)isclosetoorbelowthelevelof.5,whichisseenastherelevant
thresholdintheliterature.Amongthese,theexplanatorypowerregardingpre-

26Thisinteractiontermhasthelowestvalueforthe“force”regime(0),lowvaluesforsymmetricmistrust
regimesandthehighestvaluesforsymmetrictrustregimes.
27ForanintroductiontoFactorAnalysis,werefertoHarman(1976).

115

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

variableFactorloadingUniqueness
.648.594rust1T(1)(2)(3)TTrust2rust3.473.679.776.539
.288.844rust4T(4).406.771rust5T(5)

Table4.4.:FactorAnalysisResults
Factorloadingsanduniquenessreported,principal-factormethod(N=59).

developmentnegotiation(trust3)seemstobesmallest.Thegeneralrulesof
thumbwouldsuggesttoremoveallvariablesexceptfor(trust3-5)fromthe
el.dmoForus,thisentailsthefollowingresult:Itappearsthatthe5measures
donotcapturetheexactsamething,i.e.“trust”,orequivalently,itseems
thatthecommonperceptionoftherebeingonehomogenouskindoftrustis
inadequateinourcontext.Inthefollowingsectionwewillperformacloser
analysisofthepotentialdeterminantsoftrusttoachieveanunderstandingof
causalrelationships,andthereforetheindividualmeaningsofourmeasures.

DeterminantsofSuppliers’Trust–IPRHold-up,PayandFairness

Bothourmodelandourindustrysurveysuggestthattheinherenthold-up
problemisatthecenteroftrustformation:suppliers“sink”effortintoresearch
anddesignforpartswhichresultinblueprints.Afterobtainingtheseblueprints,
thesuperiorbargainingpositionoftheOEMenableshimtoextractadditional
rents.Thereforeourfavoredinterpretationoftrustframesitasthebeliefof
thesupplierregardingtheprobabilitythattheOEMwillrefrainfromsuch
vior.ehabundesiredTotestthis,weturnedtothesuppliersevaluationofsuchbehaviorbythe
OEMinourquestionnaire,specificallythefrequencyofconflictsregardingthe
treatmentofpatentsandtrade-secretsaswellasthefrequencywithwhichthe
OEMpassesontechnologicalsecretsofthesuppliertothirdpartieswithout
ermission.p

116

AnalysisEmpirical4.3.

Clearly,therearemoredirectwaystoextractrentsfromapositionof
power,especiallybyexertingpressureinpricenegotiations.Thereforeforpre-
developmentwelookedatthedegreetowhichtheOEMsharesthe(consid-
erable)riskofhigherthanexpectedcosts.Fordevelopment,wecanusethe
evaluationoftheadequacyoflicensefeesinthecasethattheOEMmakes
useofprotectedknow-howofthesupplier.Andforseriesproduction,bothan
evaluationofthefrequencywithwhichtheOEMdemandslump-sumpricere-
ductionsinrenegotiationsaswellastheextenttowhichheattemptstoextract
costinformationbyemployingsub-suppliermanagementwereavailabletous.
WeperformedOLS-regressionswiththeindividualtrust-measuresasthe
dependentandthemeasuresintroducedaboveasexplanatoryvariables,while
controllingforthesizeandtechnologicalsophisticationofthepart.Table
4.5belowpresentsthecoefficientsandp-values,neglectingtheeffectsofthe
product-typedummies.

Trust1Trust2Trust3Trust4Trust5
telopmenPre-DevFrequencyIPRconflicts-.637(.05)-.323(.01)-.632(.00)-.149(.277)-.416(.02)
HowoftendoesOEM-.147(.05)-.521(.00)-.291(.00)-.153(.00)-.225(.00)
IPRsupplier’sleakOEMhigherdevshareselopmenriskoftcosts.180(.06)-.156(.31).040(.82).242(.00).301(.00)
telopmenDevFHowrequencyoftendIPRoesOEMconflicts-.302-.170(.08)(.17)-.463-.392(.00)(.00)-.450-.333(.00)(.00)-.134-.118(.01)(.11)-.116-.089(.09)(.39)
IPRsupplier’sleakAdequacyoflicensefees-.035(.74).100(.32).398(.00).134(.04).256(.00)
ductionProSeriesFnegotiationrequencyprice(lumpre-sum).007(.91)-.124(.13)-.079(.45)-.180(.00)-.265(.00)
extractEffortsofcostOEMinformtoation-.026(.65)-.042(.41)-.038(.54)-.151(.00)-.166(.00)

Table4.5.:DeterminantsofTrustmeasures.
CoefficientsofOLSregressionscontrollingforproducttypeand(p-values)re-
orted.p

Thepatternthatemergeslendsitselftointerpretation:Thefirsttrust-

117

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

measure(importanceoftrustforthesuppliertoinitiatepre-developmentco-
operationwithOEM)isnotsignificantlyinfluencedbyanyoftheanswersto
thequestionsaimingatcompensationandpay.Instead,thereisaverystrong
negativecorrelation(-.637)withthereportedfrequencyofIPRconflictsduring
pre-development,i.e.themorefrequentIPRconflicts,thelowerthismea-
sureoftrust.Amuchweaker,butstillsignificanteffectwiththeexpected
sign(-.147)resultsfromtheOEMleakingsensitiveIP-relateddatamoreoften.
IPR-conflictsduringthedevelopmentphasehavearelativelystrongnegative
effectaswell(-.302),thep-valueof.08isinpartexplainedbythesmaller
numberofcommonobservations.
Thesecondtrust-measure(mutualtrustwithrespecttoIPR)followsthe
samegeneralpattern,thoughwithdifferentindividualweightings.Again,the
compensationmeasuresshownosignificantinfluenceonthetrust-measure.But
here,theleakingofsensitiveinformationshowsafarstrongereffect(-.521in
pre-development,-.392indevelopment)thanintheformercase.Further,in
thismeasuretheimportanceofIPR-conflictsduringthedevelopmentphase(-
.463)ishigherthanduringthepre-developmentphase(-.323),whichprobably
reflectsthefactthatahighershareofrespondentstothisquestionwereinvolved
inlaterdevelopmentstages.
Thepicturechangesfortheremainingtrustmeasuresthreethroughfive
(ImportanceoftrustfortheOEM’ssupplierchoiceforpre-development,de-
velopmentandseriesproduction,respectively).Formeasuresfourandfive,
theimportanceofIPR-relatedbehaviordecreasesmarkedly,whiletheeffectsof
adequatecompensationandprice-cuttingbecomesignificantwiththeexpected
signsacrosstheboard.Interestingly,thethirdtrustmeasure,relatedtopre-
developmentpresentsitselfasahybridcase,inwhichtheIPR-relatedfactors
arestillpredominant,butneverthelessalsotheadequacyoflicensefeesplaysan
importantrole(.398),whiletheothercompensationrelatedmeasuresdonot.
Theresultsfurtherprovideajudgmentonthe–asofyetunanswered–issue
ofwhosetrustthesequestionstrulymeasure:Asperceivedmisbehaviorbythe
OEMaffectstheminasignificantlynegativemanner,wefeelcomfortableusing

118

AnalysisEmpirical4.3.

themasmeasuresforthesupplier’strustintheOEM.
Tocrudelysummarizethesefindings:Trust1andTrust2arenegatively
associatedwithattemptsatrentextractionbytheOEMintheareaofIPR
withslightlydifferentfocuses,whiletheyarenotaffectedbydirectattempts
atprice-reductions.ThisbalanceshiftstowardthelatterforTrust3andeven
moresoforthemeasuresTrust4andTrust5.Aftergoingtogreatlengths
toestablishthesemeasures,wenowtrytoshowthattheunderlyingconstruct
significantlyaffectsrelevanteconomicbehaviorandoutcomes.

4.3.4.MistrustandUnderinvestment

Themainhypothesisderivedfromourmodelstatesthathigherlevelsoftrust
shouldbeassociatedwithmorerelationshipspecificinvestmentbysuppliers.
Duetothecross-sectionalstructureofourdata-set,determiningthedirection
ofcausalityisanissue.Thecaseforhigherinvestmentbysuppliersleadingto
higherlevelsoftrustofsuppliersintheOEMcanbemade:Lessinvestment
mayleadtomoreconflictsbetweentheparties,whichletstrustdeteriorate.
Whileacknowledgingthis,wefindtheoppositeargumentmoreconvincing,but
leavetheissueopenandprefertorefertoassociationinsteadofcausality.
Measuringinvestmentofsuppliersposesachallenge.Aswedonotobservea
directmeasure,weproposetwoproxiesrelatedtothequalityofpartsinstead.
Onestandardinterpretationofqualityrelatedeffortintheliteratureisthat
itaffectsfailureratesofparts(see,forexample,TaylorandWiggins(1997)).
Alongtheselines,weaskedrespondentstwoquestions:Withrespecttothe
partconsidered,howoftendoqualityproblemsoccur?and...howoftendo
recallactionsoccurduringseriesproduction?Bothquestionsaremeasuredon
a5-pointscalewith1resemblingthelowestand5thehighestfrequency.The
correlationbetweenthemis(only)0.41,whichcanbeexplainedbythefact
thatnoteveryqualityproblemleadstoarecallandthelatterareextremelyin-
frequent.Asaconsequence,89%ofrespondentsreporteda1–veryinfrequently

119

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

fortherecallquestion,whichposesseriousempiricallimitations.28Specifically,
in50%ofthecasesinwhichthelowestpossiblevalueforrecallsisreported,
thereissomehigherlevelofqualityproblems.Ontheotherhand,whenquality
problemsbecomemorefrequent/severe(levelsof3orhigher),whichinitself
isratherrare(17cases),in54%ofcasesrespondentsreportlevelsofrecall
frequencyabove1.
Ingeneral,difficultiesarisewhentryingtoassessunderinvestment-related
qualityissuesempirically,asa)theobservedfailureratesofcarscannotnec-
essarilybelinkedtoindividualparts,b)thediligenceofthemanufacturerin
assemblyalsoaffectsqualityandc)ifqualityproblemsarediagnosedbefore
thepartsareinstalled,thisisgenerallynotobservable.Thehugeadvantageof
ourquestionnaireisthattheresponsesarepart-specificwhichaddressissuea).
Thephrasingofthequestionaddressissuec).Byincludingcustomer-orOEM-
effectsintheregressions,wehopetoalleviateissueb).Apotentialdrawback
isthefactthatthefrequenciesareself-reported,sothatrespondentsmaybe
temptedtounder-reportproblems.Tocounterthis,completeanonymitywas
guaranteedattheoutsetofthestudy.
Forourempiricalstrategy,wechoosethefollowingapproachwithydenoting
thefrequencyofproblemsarising,κdenotingaconstant,idenotingthepart
inquestionandjdenotingthecustomer:

(4.6)yij=κ+αj+β∗trustij+γ∗dummy_sophi+δ∗dummy_bigi+ij

ThisModel1spellsoutthesetofOLS-regressionsincludingcustomerfixed-
effects(α).29Toaddresstheissueofsubjectivedifferencesintheunderstanding
ofthequestionsathand,wealsospecifieddummieswhichtakethevalue1
whenevertheanswertothequestionisthelowestpossiblefrequency(i.e.1)and

28Themodelquality,asrepresentedbytheF-statistic,ingeneralisverylowasmeaningfulvariationis
t.absen29Wealsoperform(moresuitabletothedataathand)tobitandordered-logitregressions,whichdelivered
qualitativelyidenticalresults.Wedonotreporttheresultsinthispaper,buttheycanbeprovidedby
request.onupauthorsthe

120

AnalysisEmpirical4.3.

takethevalue0wheneverthereportedfrequencyislargerthanthis.Thisallows
ustoestimateaprobit-model,inwhichyissimplyreplacedbytheprobability
thatnoproblemsoccur/problemsoccurasrarelyaspossible(Model2).This
specificationhasthefurtheradvantagethattheresultscanbemorereadily
interpreted.WewouldexpectnegativecoefficientsforβinModel1andpositive
coefficientsforβinModel2fromourhypothesis.30
Weestimatebothmodelswithandwithoutcustomerfixed-effects.Wein-
cludethesetocapture,forexample,potentialcomplementaryeffortexerted
bythecustomers(OEMs),whichalsomayinfluencetheprobabilityofquality
issuesarising.Asthiseffortmayaffectsuppliers’incentivestoprovidebetter
quality,itisnotclearhowtheseeffectsaredirected.Asaconsequenceofthe
verylimitedvariationinthefrequencyofrecalls,weonlypresentthesetof
OLSregressionsforthisquestiontoshowthatthetrustmeasurescontribute
explanatorypowerinthecasesthatrecallsarereported.Theresultsofthe
threesetsofregressionscanbefoundinTablesB.8,B.9andB.10.

[TableB.8,B.9andB.10abouthere]

Firstnotethatthecoefficientsinallestimationshavetheexpectedsigns,
thoughtheirlevelsofsignificancevary.Thelattercan,toalargedegree,
beexplainedbythefactthatdifferentnumbersofobservationsareavailable
fortheindividualregressions.InthoseincludingTrust4,whichincludethe
highestnumberofobservations(122),βissignificantlydifferentfromzeroat
the5%levelforallspecifications.Asonewouldexpect,bothtechnological
sophisticationandsize(orinterfaceswithin)raisetheprobabilityofquality
issuessignificantlyacrosstheboard.Thesizeofthecoefficientsisnoteworthy
–morecomplexpartsthatarebothtechnologicallysophisticatedandlargeare
morethan50%morelikelytohavequalityissues.Theothersurprisingresultis
theextenttowhichourmeasureTrust4influencesthequalitylevelasreported
bysuppliers.Anincreaseinthemeasureby1(i.e.1.1standarddeviations)

30AsModel1makesuseofmorevariationwithintheoutcomevariable,itusestheexistinginformation
.tlyefficienmore

121

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

decreasestheprobabilityofqualityissuesarisingby12.9%(excludingcostomer
fixedeffects)oreven16.7%(includingcustomerfixedeffects),seeTableB.9.
Wetakethisasevidenceinfavorofourfirstcentralhypothesis.

4.3.5.TrustandSourcingDecisions

Inthiscontext,theunderlyingcausalmechanismiscloselyrelatedtothehy-
pothesisonewouldexpect:Ifthelevelofcompetitionaffectstrust,inthesense
thatthenatureofarms-lengthmarketagreementsislessamenabletofostering
trustrelationships,thenoneshouldbemoreinclinedtosidewiththeexisting
literatureinexpectinganegativeassociation.Ontheotherhand,wecouldalso
findourselvesinasetupsimilartothemodelwepropose.Then,thesupplier
facesattemptsatrentextractionbytheOEMeitherthroughexploitationofthe
hold-upsituationorthroughcompetition.Inthiscasemorecompetitionmay
beasignalofhigherlevelsofcompliancewithpropertyrightsbythepartywith
thesuperiorbargainingpower,andthereforehigherlevelsoftrust.Therefore,
atbestwecandeterminecorrelationwithourdata.
Ourempiricalapproachistoanalyzehowthreedifferentmeasuresofsup-
pliercompetition,i.e.thenumberofparallelsupplierscontractedtoworkon
pre-development,developmentandseriesproductionprojects,respectively,is
associatedwithourmeasuresoftrustderivedabove,whilecontrollingforthe
sizeandtechnologicalcomplexityoftheproductinquestion.Asasecondtest,
wealsodirectlycheckfortherelationshipbetweenthenumberofcompetitors
andthemethodsofrentextractionthatweidentifiedinSection4.3.2asaf-
fectingtherelevantmeasuresofsuppliertrustmost.Forpre-development,this
isthefrequencyofIPRconflictsandthefrequencywithwhichtheOEMhas
discontinueddevelopmentprojectsinthepast5years.31Fordevelopment,this
isthefrequencyofIPRconflictsandtheadequacyoflicensefees,whilefor

31Weobservemulti-collinearitybetweenIPRconflictsandtheOEMleakingsupplierIPR(thesameholds
formentdevcosts,elopmenwet),looseallthereforebutwe50omitobservtheations.latterfromThereforetheweregression.replaceIfthiswevuseariabletheriskwithofthehigherfrequencydevelop-of
discontinuationofprojects,whichcapturestheoppositeoftheeffectofrisk-sharing.

122

AnalysisEmpirical4.3.

seriesproduction,thisisthefrequencyofpricere-negotiationandtheOEM’s
costextractingateffortsinformation.WeagainusesimpleOLS,aswearemainlyinterestedincorrelationbetween
thevariables.Inthefollowingapproach,yisthenumberofsuppliersemployed
bytheOEM,iisthepartinquestionandjisthecustomer,κagaindenotes
aconstant.xresembleseitheratrustmeasureorthevectorofthetworent-
extractionmechanismdescribedabove,dependingonthespecification.

(4.7)yij=κ+β∗xij+γ∗dummy_sophi+δ∗dummy_bigi+ij

Regardingoursecondhypothesis,wewouldexpectapositivesignforβ
inthespecificationswiththetrustmeasures.Higherlevelsoftrustwould
beassociatedwithahigherincentivetoextractrentsthroughinducingmore
competition,i.e.employingmoreparallelsuppliers.Forthespecificationswith
thedirectmeasuresofrent-extraction,wewouldexpecttheopposite:Themore
anOEM,forexample,exertspricepressuredirectlyonthesupplier,theless
heneedstoinducecompetitionthroughmultipleparallelsuppliers.

[TableB.11,B.12,B.13andB.14abouthere]

Theresultsoftheregressiononthetrustmeasuresforthenumberofsuppliers
involvedinpre-developmentareinterestinginthateventhesignificanceofthe
modelitselfisrejected(TableB.11).Thereappearstobenosignificant
correlationbetweenourtrustmeasuresandthenumberofsuppliersinpre-
development,withthesamealsoholdingforthecomplexityandsizeofthe
product.The“direct”approachontheotherhandyieldsinterestinginsights
intotheposedquestion(seecolumn1inTableB.14):Wefindahighly
negativecorrelationbetweenthefrequencyofconflictsregardingIPRandthe
numberofsuppliersinvolvedinthepre-developmentproject.Thispointsinthe
directionofourhypothesis,inthatthesupplier’sintellectualpropertyplaysa
pivotalroleinthecourseofpre-development,andabuseofhisrightsinthis

123

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

realmisastrongindicationofattemptsatrentextraction.Thefactthatthere
isnosignificantcorrelationwiththesizeandthecomplexityofthepartmay
beattributabletothefactthatpre-developmentisbydefinitionnotmodel
specific–e.g.researchontheapplicationofanewalloycanbeconnectedto
acar-modelandpartexpost,butatthetimethebasicresearchtakesplace,
clear.fromfaristhisWithrespecttothedevelopmentstage,themodelsincludingthetrust-
measuresdirectlyrelatedtothedevelopmentprocess(Trust2andTrust3)
havethestrongestexplanatorypower(TableB.12).Whilethecoefficients
herearenotsignificant,theyarepositiveandinthecaseofTrust2veryclose
tosignificancewithap-valueof.16,whichpointsinthedirectionofourhypoth-
esis.Forthedevelopmentphase,wealsofindoveralnegativeandsignificant
effectsofthesizeandcomplexityoftheproductinquestiononthenumberof
suppliersinvolvedintheprocess.Wemaycapturetheeffectoflesscompetition
inthesemoredifferentiatedgoodmarkets,here.Finally,thedirectmethodsof
rentextractioninthecaseofdevelopmentshowneitherasignificantpositive
nornegativesign(seecolumn2inTableB.14).
Inseriesproduction,theevidenceismixed(TableB.13).Interestingly,
thecoefficientforTrust5,themeasuredirectlyrelatedtoseriesproduction,
isinsignificant(p=.758).Amongsttheothers,Trust3showsastrongpositive
correlationwiththenumberofsuppliersinvolvedinthefirststageofproduc-
tion,whileTrust2alsohasapositive(evenlarger)coefficientbutisbarely
insignificantatthe10%level(p=.132).OnlythecoefficientofTrust1isout
ofline,whichresemblestheimportanceoftrustforthesupplierapproaching
theOEMforapre-developmentproject.The“direct”approachyieldsavery
clearpicture(seecolumn3inTableB.14).Thefrequencyofpricerenegotia-
tionswithlumpsumreductionsdemandedbytheOEMishighlysignificantly
andnegativelyassociatedwiththenumberofsuppliersinvolvedinthebegin-
ningofseriesproduction.Inourviewthisstronglyindicatesthatthesetwo
instrumentsofrentextractionaresubstitutive.
Overall,wefindsomewhatmixedevidence,withmostofitinsupportofour

124

okOutloandConclusion4.4.

secondhypothesis.Weactuallyconsiderourfindingstoberelativelystraight-
forward,ifoneallowsforthefactthatweusedinterviewdata,inwhicheach
observationsiscompiledfrommultipleparticipantsandthequestionsincluded
psychologicallyloadedtermssuchas“trust”.Thestructureofourdatadoes
notallowustocompletelydisentangletheseideas,butwestronglybelievethat
moreresearchintothisiswarrantedandimportant.

4.4.ConclusionandOutlook

Trustisanimportantingredientinalmostallmeaningfulsocialandeconomic
interactions.Whilemostrecenteconomicresearch,toalargeextentdueto
availabilityofdata,hasfocusedonthewillingnessofindividualstotrustothers
ingeneral,wewereabletoshedlightontheroleoftrustthathasdeveloped
–i.e.beenfosteredorsquandered–inspecificeconomicrelationships.Inthis
context,weproposeanintuitivedefinitionfortrustthatwedemonstratein
aformalmodel:Inthepresenceofahold-upsituation,wedefinetrustasthe
subjectiveprobabilitythatthepartywiththesuperiorbargainingsituationwill
e.exploitativebWethenshowintheempiricalpartofourstudythatbehaviorthatcanbe
construedasappropriatingrentsinexcessoftheformalpropertyrightsofthe
OEMdoesleadtolowerlevelsofsuppliertrust,usingvariousdifferentmeasures
forthis.Wefurthershowthat“trust”isnotasinglehomogenousconstruct,
butinsteaddifferentpeopleinthesamefirmwillfocusondifferentfactors
whenassertingtheirleveloftrustintheotherparty.Someofourmeasuresare
morecloselyrelatedtotheadherencetoandrespectforintellectualproperty
rights,othermeasuresareassociatedwithfaircompensationvs.frequentprice
renegotiations.WefindthatanOEM’sinvestmentinhissupplier’strustbyforgoingthese
(oftenshort-term)opportunitiesatappropriatingrentcanpayoff:Inrelation-
shipssignifiedbyhigherlevelsoftrustwefindindicationforsignificantlyhigher
investmentbythesupplierresultinginfewerfailuresandcallbacks.

125

4.TrustandInvestment-ATheoreticalandEmpiricalAssessment

Ourmodelmakesasecondpredictionwhichmayappearmoresurprisingat

firstglance:Weshowthatinducingupstreamcompetitionisasubstitutefor

theOEMexploitinganexistinghold-upsituation–ourmodelpredictsthat

higherlevelsoftrustofthesupplierintheOEMshouldbeassociatedwitha

largernumberofupstreamsupplierscompeting.Incontrasttoresultsfrom

theliteratureonrelationalcontractsourempiricalfindingssupportthistoa

certaindegree,thoughnotunequivocally.Whileemployingalargernumber

ofsuppliersisrelatedwithsignificantlylowerlevelsofrentextractionthrough

otherchannels,suchaspricerenegotiationsexpost,theeffectsofourtrust

measuresonthenumberofsuppliersismixed.Inthefaceofourfindingthat

thereisnoone-size-fits-allconceptof“trust”,thiswastobeexpected–and

showsthatmorespecificresearchinthisdirectioncouldbeveryuseful.

Finally,webelievethatourunderstandingoftrustasbeingconnectedto

hold-upsituationsorsimilarsettingsisextremelyuseful,butclearlydoesnot

coverclosetoallpotentialapplications.Inparticular,webelievethatanap-

plicationtosettingsofincompletecontracts,potentiallyalsoinprocurement

environmentssuchasBajariandTadelis(2001),couldgreatlyenhanceour

field.thisinunderstanding

126

A.AppendixtoChapter3-Regression
Results

Dep.Variable:Prob(Severance)Model1a∗∗∗Model2aModel3a∗∗∗
Ownmusicofsevereratt=0.0000139-.0000169
(2.65e-06)(2.72e-06)Ownmusicofseveredatt=0-.0000182∗--.0000161∗∗
(7.10e-06)(7.71e-06)Play-listsofsevereratt=00.00017∗∗∗-.0000592
(.000043)(0.0000301)Play-listsofseveredatt=0.0000641-.000207
(.0000197)(.000183)Musicuploadedbysevererperday-.00483∗∗∗.00136∗∗∗
(.000382)(.00053)Musicuploadedbyseveredperday--.0120∗-.00966∗
(.00555)(.00634)Play-listscreatedbysevererperday--.0449∗∗∗-.0186∗∗∗
(.00465)(.00652)Play-listscreatedbyseveredperday-.0775.0754
(.0467)(.055)1,5211,5211,521ationsobservpseudoR2.429.312.469
2379.64252.54347.12ChiTableA.1.:FreeRidingandNegativeReciprocitya):Logisticregressionson
theprobabilitythatagivenfriendshipissevered.
Marginaleffectsandsignificanstandardtaterrors5%,at***meansignificanreptatorted:1%*significantat10%,**

127

A.AppendixtoChapter3-RegressionResults

SongsDep.Vlistenedariable:perProdaby(Sevbyseverance)ererMo-.130del∗1bMo-.0450del2b
(.0455)(.009)Songslistenedperdaybysevered.00952∗∗∗.00519∗∗∗
(.00154)(.00244)Ownmusicofsevereratt=0-.0000145∗∗∗
(2.38e-06)Ownmusicofseveredatt=0--.0000215∗
(.0000115)Play-listsofsevereratt=0-.0000623
(.0000451)Play-listsofseveredatt=0-.0003717
(.000234)Musicuploadedbysevererperday-.000746∗∗∗
(.000286)Musicuploadedbyseveredperday--.00862∗∗
(.00366)Play-listscreatedbysevererperday--.0279∗∗∗
(.00707)Play-listscreatedbyseveredperday-.0901∗∗
(.00707)1,4771,477ationsobserv2pseudoChi2R.047326.64201.58.358
TableA.2.:FreeRidingandNegativeReciprocityb):Logisticregressionson
theprobabilitythatagivenfriendshipissevered.
Marginaleffectsandsignificanstandardtaterrors5%,at***meansignificanreptatorted:1%*significantat10%,**

128

Dep.Variable:musicfilesadded
friendsusicmusicmwnofriendsusicmΔexperienceddummy
timetscommenmessagesdsfriensy-listpla

Dep.Variable:musicfilesaddedModel1Model2Model3
musicfriends-5.44e-06-.0000134-.0000292
(.0000199)(.0000138)(9.66e-06)ownmusic-.0602665-.0666011-.0920428
(.0928703)(.0827902)(.0585873)Δmusicfriends4.24e-06∗5.56e-06∗∗6.75e-06∗∗
(2.91e-06)(2.49e-06)(2.29e-06)experienceddummy--.0423019.4954036
time-.0234988(.0632985).0301578(.5668963)
(.016886)(.0155074)comments-(4.517221)1.478684(4.346862)1.352071
messages-(1.362083).0867128(1.33175)-.1812929
play-listsfriends--.0073593
(.0072321)ownplaylists--.4954036
constant3.3649623.9069915.863737(.5668963)
(2.266137)(3.610217)(2.804955)observgroupsations177,82825,03125,031177,82825,031177,828
overallR2.0294.0241.0208
9.155.522.96F.962.933.924rhoTableA.3.:FreeregressionsRidingconandtProllingositiveforpReciprootentialcityforuser-clusterInformationheteroskProvisioedasticitn,FE-y,
sampleireten*significantat10%,**significantat5%,***significantat1%

plawnoylists

-

-

129

A.AppendixtoChapter3-RegressionResults
Dep.Variable:musicfilesaddedModel1Model2Model3∗
musicfriends-.0000124-.0000293-.0000572
(.0000345)(.0000197)(.0000198)ownmusic-.0593032-.0661795-.1171941
(.1071663)(.0836462)(.0590668)Δmusicfriends3.87e-069.61e-06∗∗.0000105∗∗
(4.06e-06)(3.71e-06e-06)(2.81e-06)experienceddummy--1.636376-1.314448
(2.485865)(2.713547).2793232.3370503-timecomments-(.17308)1.48062(.1722218)1.352071
(4.346862)(4.511961)-.1812929.0533741-messages(1.33175)(1.376997)play-listsfriends--.0126578
(.0123971)ownplaylists--1.011472
constant11.126515.5182314.21924(1.176967)
(13.35873)(9.963463)(7.765481)observgroupsations9,42824,4139,42824,4139,42824,413
overallR2.0472.0409.0317
Frho2.27.7255.93.7576.67.896
TableA.4.:FreeRidingandPositiveReciprocityforInformationProvision,ac-
sub-sampleuseretiv*significantat10%,**significantat5%,***significantat1%
130

Dep.Variable:play-listsaddedModel1Model2Model3
play-listsfriends-.0002687-.0007544-.0013166∗∗
(.0006179)(.0007107)(.0005835)ownplay-lists-.1671713∗∗∗-.2086505∗∗∗-.2901718∗∗∗
(.0489084)(.0453584)(.0393094)Δplay-listsfriends.0009297∗∗.0010383∗∗.0010542∗∗∗
(.0003998)(.0004011)(.000391)experienceddummy-.0046399.0072075
time--.0046172(.0084659)∗∗∗.0019167(.0085653)
(.002046).0010961comments-(.4122087)-.0715414(.3545851)-.0329556
messages-(.1184353).1556359(.1029976).0831986
dummyactive--.0000785.0040834
(.0073418)(.0079719)musicfriends---5.21e-06∗∗∗
(1.97e-06)ownmusic--.0123697∗∗
(.0056172)constant(.0710501).3138434(.0549475).4337222(.2757704).5619726
177,828177,828177,828ationsobserv25,03125,03125,031groupsoverallR2.0435.0000.0005
22.2716.519.43F.975.923.797rhoTableA.5.:FreeRidingandPositiveReciprocityforInformationOrganization,
sampleireten*significantat10%,**significantat5%,***significantat1%
131

A.AppendixtoChapter3-RegressionResults
Dep.Variable:play-listsaddedModel1Model2Model3
play-listsfriends-.0003837-.0006297-.0012703
(.0009274)(.0009335)(.0007793)ownplay-lists-.123411∗∗∗-.1525017∗∗∗-.2550607∗∗∗
(.0808259)(.0836462)(.029512)Δplay-listsfriends.0012131∗∗.0012077∗∗.0012261∗∗
(.0005189)(.0005203)(.0005418)experienceddummy-.0961424-.0053575
(.1435919)(.1195154).0036746-.0199061-timecomments-(.0116495)-.0874014(.011876)-.0446645
(.3362418)(.3545696).0663375.0924774-messages(.1116464)(.1224277)musicfriends---.0000168∗∗∗
(3.25e-06)ownmusic--.0108558∗
constant.5901143.979095(.0064666)1.46338
(.9223387)(.353222)(.1505223)observgroupsations24,4139,4289,42824,4139,42824,413
overallR2.0892.0026.0008
rhoF.52412.18.65111.53.91424.35
TableA.6.:FreeRidingandPositiveReciprocityforInformationOrganization,
sub-sampleusereactiv*significantat10%,**significantat5%,***significantat1%
132

B.AppendixtoChapter4:Descriptive

ResultsRegressionandStatistics

ariableVWhenissupplieraskedtoparticipate?
Howoftenisprogresscoordinated?
CostShareofreimeffortsbursemenabsorbtifedbysubsequensuppliertcontract
Costreimbursementifnosubsequentcontract
Specificitydevelopmentobjectiveswrt...
......contentime-framet
...OEM’sfinancialsupplierengacgemenhoicetcriteria:
...importanceofsupplierprice
...importanceofdurationcooperation
trustofortanceimp...

TRelationshipB.1.:able

view)

Characteristics:

Mean(Std.Dev.)MinMaxObs.
14461(1.37)2.773.502.98(1.33)(.57)1155151142
24651(1.52)2.3123251(1.59)2.391.852.33(.96)(.97)1155350350
34351(1.14)2.22

(1.16)5.10(.99)4.70(.98)4.89

111

telopmenPre-Dev

666

158160159

(Suppliers’

133

B.AppendixtoChapter4:DescriptiveStatisticsandRegressionResults
VariableMean(Std.Dev.)MinMaxObs.
Howspecificanddetailedarespecifications?2.39(1.02)15231
DesiredSupplier’sdegreedegreeofoffreedomfreedom3.622.91(.77)(.86)1155229231
FOEM’srequencyconoftributionIPRtoconflictsdevelopment2.372.24(1.10)(.87)1155200194
criteria:hoicecsupplierOEM’s...importanceofsupplierprice5.37(.72)2.56387
......impimportanceortanceofofpdurationersonalcoconoptacteration4.524.52(.98)(1.00)1166387387
......impimportanceortanceofoftrustcertification4.394.90(.93)(1.14)1166384377
TableB.2.:RelationshipCharacteristics:Development(Suppliers’view)
VariableMean(Std.Dev.)MinMaxObs.
HowoftendoesOEMproduceparthimself?1.69(1.31)16210
criteria:hoicecsupplierOEM’s......impimportanceortanceofofsupplierdurationcopriceoperation4.385.70(1.07)(.52)1366253253
...importanceofpersonalcontact4.44(1.10)16253
......impimportanceortanceofoftrustcertification4.284.73(.98)(1.19)1166252250
TableB.3.:RelationshipCharacteristics:SeriesProduction(Suppliers’
view)VNumariableberofcompetingsuppliersMean2.16(Std.(.84)Dev.)Min1Max5Obs.137
FHowrequencyoftenofwerepsubsequenrojectstdevdiscontinelopmenuedtinprolastject5syrs.3.232.00(.88)(1.11)1155139322
Howoftenwerethefollowingemployed...
...preselectionofaspecificsupplier4.43(1.26)16351
...procurementamongaltd.numberofsuppliers3.95(1.44)16338
TableB.4.:ProcurementDecisions:Pre-Development(Suppliers’view)
134

VariableMean(Std.Dev.)MinMaxObs.
FNumrequencyberofjointprosupplierscuremenemplotydev.edanddurinprog...duction3.76(1.24)15363
...productplanning2.22(1.13)15167
...productspecification2.03(1.02)15177
...conceptdevelopment2.12(1.07)15208
...detaileddevelopment1.51(0.90)15210
...Howpreselectionoftenwoerefathespecificfollowingsupplieremployed...3.06(1.52)16259
...procurementamongaltd.numberofsuppliers5.18(1.10)16264
...openprocurement1.97(1.41)16255
TableB.5.:ProcurementDecisions:Development(Suppliers’view)
VariableMean(Std.Dev.)MinMaxObs.
...Numatbproerofductionsuppliersstartemployed...1.20(.58)15251
...after1-2years1.47(.78)15249
...aftermorethan2years1.59(.81)15246
Howoftenwerethefollowingemployed...
......propreselectioncurementofaamongspaecificltd.nsupplierumberofsuppliers2.984.55(1.63)(1.52)1166248248
...openprocurement2.44(1.66)16243
TableB.6.:ProcurementDecisions:SeriesProduction(Suppliers’view)
135

B.AppendixtoChapter4:DescriptiveStatisticsandRegressionResults

(1)(2)(3)(4)(5)(6)(7)(8)(9)
1.001rustT(1)(2)Trust2.2501.00
(.048)(3)ProvisionIPR-.0296-.0371.00
(.663)(.764)supplieryb(4)ProvisionIPR-.036.202.4431.00
byOEM(.718)(.016)(.000)
(5)Differencein-.009-.185.756-.2541.00
theProvisionIPR(.930)(.029)(.000)(.000)
(6)ProvisionCosts.037.005.120.032.1211.00
bysupplier(.557)(.968)(.229)(.748)(.225)
(7)Trust3.535.432-.046.136-.147-.0181.00
(.000)(.000)(.569)(.092)(.071)(.889)
(8)Trust4.339.385-.030.052-.078-.193.5091.00
(.000)(.000)(.663)(.447)(.260)(.002)(.000)
(9)Trust5.320.382-.090-.011-.104-.143.408.7001.00
(.000)(.002)(.365)(.912)(.296)(.025)(.001)(.000)

TableB.7.:PairwiseCorrelationsofTrustMeasures
alues)(p-vTrust1:importanceofthetrustrelationshipwiththeOEMinafirm’sdecision
toinitiateapre-developmentproject(six-pointscalerangingfrom1–norelevance
to6–veryhighrelevance).Trust2:Howdoyouevaluatemutualtrustbetween
OEMandsupplierwithrespecttohonoringeachotherÕsintellectualpropertyrights
(IPR)?(five-pointscalerangingfrom1–verylittleto5–veryhigh)Trust3to5:How
importantistrustintheOEM’sselectionofasupplierinpre-develpoment(Trust
3),development(Trust4),seriesproduction(Trust5)?(six-pointscalerangingfrom
1–norelevanceto6–veryhighrelevance)

136

(10)

(9)

(8)

(7)(6)(5)(4)(3)

(2)(.42)-.054(.395)-.052(1)

1trusT(.51)-.069(.396)-.072

2trusT-.099(.192)-.120*(.087)

3trusT(.022)-.159**(.03)-.133**

4trusT(.006)(.23)es.18790y-.080-.066(.440)(.437).186.183
(.001)(.193).12690no.401***.354***.271**.244**.215*.218*
15)(.076)es0.162122.(y(.024)(.058).118122no(.625)es(.146).084.277.23870y(.766)(.152).049.251.09270noes(.418).144(.364).135.17167y04(.614).085(.26).0567no.2.327**

5trusTesy95.327**.298**(.020)(.026).223.189(.138)(.171)no95.162.102
d_sophd_bigOEM-FEobs.#2R

1%attsignifican***5%,attsignifican**10%,attsignifican*
problemsyqualitoffrequencyfor1)del(MoresultsOLS-regressionB.8.:ableT
tsefficiencoorted:repalues)(p-vand

(10)

(9)

(8)

(7)(6)

(5)(4)(3)

(2).040(.487)(.427).041(1)

1trusT(.262).101(.212).085

2trusT(.128).111

.096(.111)

3trusT(.016).167**(.026).129**

4trusT9***(.065)(.003)9es90.16y-.34.092.086(.308)(.249)-.238*-.219*.
(.064)(.008)-.298***.12790no(.015)(.016)-.237**-.258**es.138122y(.015)(.029)-.246**-.207**.107122no(.007)-.478***(.881)es.024.24170y.030(.015)-.344**(.833).10370no-es(.162)-.242.(.574)-.084.18667y(.051)-.278*(.836)-.028no.07467-.278**-.283**

5trusT(.014)(.027)esy95.149-.240**(.028)-.224**(.050)no95.096
d_sophd_bigOEM-FEobs.#2Ps-R

1%attsignifican***5%,attsignifican**10%,tatsignifican*
problemsyqualitservingbonotofyprobabilit,)2del(MoresultsProbit-regressionB.9.:ableT
seffectmarginaleragevaorted:repalues)(p-vand

(10)(9)(8)(7)(6)(5)(4)(3)(2)(1)-.090**-.071**(.011)(.033)

1struT-.087(.168)-.052(.290)

2struT(.028)-.090**(.027)-.081**

3struT(.049)-.073**(.142)-.049

4struT

8)(.65)(.055)-.116*es(.39.153.040.07087y9)(.360)(.204)-.063.053.075(.31.08087no(.347)(.317)es.118.061.062115yo(.168)(.389).048.087.052115nes(.560)(.821).210.060-021.66y2)(.84).018.089.08066no(.39-es(.659)(.471).15564-.054.078y(.898)(.508).03364no.014.067es.195(.260)92.089(.523).047y

5struT

.041(.568).109(.148)no92.077
d_sophd_bigOEM-FEobs.#2R

d_soph

1%significan*attsignifican***5%,attsignifican**10%,att
recallsduct-relatedprooffrequencyfor1)del(MoresultsOLS-regressionB.10.:ableTorted:repalues)(p-vandtsefficienco

B.AppendixtoChapter4:DescriptiveStatisticsandRegressionResults

-.098(.214)

(5)(4)(3)(2)(1)-.0061trusT(.947)-.1322trusT(.123)-.0983trusT(.214)-.0844trusT(.322).2085trusT(.157)d_soph-.006.059.099.062-.035
(.982)(.695)(.522)(.686)(.891)
d_big-.023.133.075.119.001
(.934)(.446)(.68)(.504)(.997)
#obs.6212613412761
R2.000.028.020.014.036
TableB.11.:OLS-regressionresultsfornumberofsupplierspre-development
coefficientsand(p-values)reported:*significantat10%,**significantat5%,***
%1attsignifican

140

1trusT2trusT3trusT4trusT5trusT

(1)-.018(.820)

(2)

.125(.161)

(3)

.079(.333)

(4)

.101(.188)

(5)

Trust5-.139(.276)
d_soph-.382*-.239-.299*-.238*-.380*
d_big-.161(.079)-.458**(.126)-.488**(.068)-.208(.060)-.162(.079)
(.494)(.126)(.014)(.015)(.498)#R2obs..05297.116113.117119.036206.06696

TableB.12.:OLS-regressionresultsfornumberofsuppliersdevelopment
coefficientsand(p-values)reported:*significantat10%,**significantat5%,***
1%attsignifican

141

B.AppendixtoChapter4:DescriptiveStatisticsandRegressionResults

.200(.132)

(2)(1)-.061**1trusT(.049).2002trusT(.132)3trusT4trusT5trusT

(3)

.188**(.047)

(4)

.041(.260)

(5)

-.0315trusT(.758)d_soph-.187**-.448*-.462*-.196***-.186**
(.014)(.077)(.057)(.010)(.014)
d_big-.153*-.256-.229-.154**-.167**
(.034)(.049)(.364)(.323(.053)#obs.2445963249248
R2.066.170.195.057.053
TableB.13.:OLS-regressionresultsfornumberofsuppliersseriesproduction
coefficientsand(p-values)reported:*significantat10%,**significantat5%,***
%1attsignifican

142

(1)FrequencyIPRconflictspre-development-.389***
(.003)Frequencydiscont.ofprojectsbyOEM.079
(.350)FrequencyIPRconflictsdevelopment
Adequatelicensefees
renegotiationpricerequencyF

Effortsatextractionofcostinfo

d_sophd_bigobs.#2R

(2)

.035(.710)-.014(.879)

(3)

-.125***(.000).005(.801)-.090.265*-.028(.192)(.091)(.139)-.090-.079.261(.200)(.643)(.179)195147111.133.024.102

TableB.14.:OLS-regressionresults:Relationshipofdifferentrent-extraction
devicesandnumberofsuppliersduring(1)predevelopment,
(2)developmentand(3)seriesproduction
coefficientsand(p-values)reported:*significantat10%,**significantat5%,***
1%attsignifican

143

yBibliograph

Aghion,P.,Y.Algan,P.Cahuc,andA.Shleifer(2008).Regulationanddistrust.
HarvardUniversity,mimeo..

Aghion,P.,M.Dewatripont,andP.Rey
ropeanEconomicReview46,745–753.

(2002).Onpartialtracting.conEu-

Akerlof,G.A.andJ.L.Yellen(1990).Thefairwage-efforthypothesisand
unemployment.TheQuartelyJournalofEconomics105,255–283.

Andreoni,J.(1988).Whyfreeride?JournalofPublicEconomics37,291–304.

Anton,J.J.andD.A.Yao(2007).Findingllostpprofits:Anequilibrium
analysisofpatentinfringementdamages.TheJournalofLaw,Economics&
Organization23,1,186–207.

Asvanund,A.,K.Clay,R.Krishnan,andM.D.Smith(2004,June).Anempir-
icalanalysisofnetworkexternalitiesinpeer-to-peermusic-sharingnetworks.
InformationSystemsResearch15(2),155–174.

Bajari,P.andS.Tadelis(2001).Incentivesversustransactioncosts:atheory
ofprocurementcontracts.RANDJournalofEconomics32,387–407.

Baker,G.,R.Gibbons,andK.J.Murphy(1994).Subjectiveperformance
measuresinoptimalincentivecontracts.TheQuarterlyJournalodEco-
1125–1156.(4),109nomics

Baker,G.,R.Gibbons,andK.J.Murphy(1999).Informalauthorityinorga-
nizations.TheJournalofLaw,Economics,&Organization15,56–73.

144

yBibliograph

Bala,V.andS.Goyal(2000,September).Anoncooperativemodelofnetwork
formation.Econometrica68(5),1181–1229.

Ben-Shahar,O.andJ.J.White(2006).Boilerplateandeconomicpowerin
automanufacturingcontracts.MichiganLawReview104,953–82.

Bessen,J.E.andM.J.Meurer(2006).Patentlitigationwithendogenous
disputes.TheAmericanEconomicReview96,77–81.

Bessen,J.E.andM.J.Meurer(2008a).PatentFailure.PrincetonUniversity
Press.

Bessen,J.E.andM.J.Meurer(2008b,February).Theprivatecostsofpatent
.omimelitigation.

Birke,D.(2008).Theeconomicsofnetworks-asurveyoftheempiricallitera-
ture.NottinghamUniversityOccasionalPaperSeries,1–33.

Bloch,F.andB.Dutta(2009).Communicationnetworkswithendogenouslink
strength.GamesandEconomicBehavior66,39–56.

Boldrin,M.andD.K.Levine(2002).Thecaseagainstintellectualproperty.
AmericanEconomicReview92,209–212.

Boldrin,M.andD.K.Levine(2008).AgainstIntellectualMonopoly.Cambridge
Press.yersitUniv

Bottazzi,L.,M.DaRin,andT.Hellmann(2007).Theimportanceoftrustfor
investment:Evidencefromventurecapital.mimeo.

Bottazzi,L.,M.DaRin,andT.Hellmann(2009).Whatistheroleoflegalsys-
temsinfinancialintermediation?theoryandevidence.JournalofFinancial
559–598.,18diationInterme

Bramoullé,Y.andR.Kranton(2007,
ofEconomicTheory135,478–494.

July).Publicgoodsinwnetorks.Journal

145

yBibliograph

Brown,M.,A.Falk,andE.Fehr(2004).Relationalcontractsandthenature
ofmarketinteractions.Econometrica72(3),747–780.

Bull,C.(1987).Theexistenceofself-enforcing
JournalofEconomics102,147–159.

Butler,WorkingJ.,P.PaperGiuliano,No.and15344L..

Guiso(2009).implicitThetrightracts.contamounofQuartelyThe

trust.NBER

Cagno,D.D.andE.Sciubba(2007).AdvancesinDecisionMakingUnder
RiskandUncertainty,ChapterTheDeterminantsofIndividualBehaviourin
NetworkFormation:SomeExperimentalEvidence,pp.219–241.Springer-
erg.HeidelbBerlinerlagV

Cagno,D.D.andE.Sciubba(2008,January).Socialnetworksandtrust:not
theexperimentalevidenceyoumayexpect.workingpaper.

Caillaud,B.andA.Duchene(2006).Patentofficeininnovationpolicy:No-
body’sperfect.mimeo.

Calzolari,G.andG.Spagnolo(2009).Relationalcontractsandcompetitive
screening.CEPRDiscussionPaperNo.7434.

Chiou,J.-Y.(2008).Thepatentqualitycontrolprocess:Canweaffordan
(rationally)ignorantpatentoffice?mimeo.

Cho,M.(2006,December).Endogenousformationofnetworksforlocalpublic
goods.workingpaper.

Choi,J.P.(2006).Howreasonableisthe’reasonable’royaltyrate?damage
rulesandprobabilisticintellectualpropertyrights.CESifoWorkingPaper
.1778No.

Coleman,J.,E.Katz,andH.Menzel(1957).Thediffusionofaninnovation
amongphysicians.Sociometry20,253–270.

146

yBibliograph

Cook,J.P.(2007).Onunderstandingtheincreaseinu.s.patentlitigation.
AmericanLawandEconomicsReview9(1),48–71.

Corbae,D.andJ.Duffy(2008).Experimentswithnetworkformation.Games
andEconomicBehavior64,81–120.

Crampes,C.andC.Langinier(2002).Litigationandsettlementinpatent
infringementcases.TheRANDJournalofEconomics33(2),258–74.

Falk,A.andM.Kosfeld(2003,May).It’sallaboutconnections:Evidenceon
networkformation.IZADiscussionPaper(777).

Farrell,J.andC.Shapiro(2008).Howstrongareweakpatents?American
EconomicReview98,1347–1369.

Fehr,E.(2009).Ontheeconomicsandbiologyoftrust.JournaloftheEuropean
EconomicAssociation7,235–266.

Fehr,E.andS.Gächter(2000).Fairnessandretaliation:Theeconomicsof
reciprocity.JournalofEconomicPerspectives14,159–181.

Goeree,J.K.,A.Riedl,andA.Ule(2008).Insearchofstars:Networkforma-
tionamongheterogeneousagents.GamesandEconomicBehaviorarticlein
.esspr

Gowrisankaran,G.andJ.Stavins(2004).Networkexternalitiesandtechnol-
ogyadopotion:Lessonsfromelectronicpayments.TheRANDJournalof
260–276.,35onomicsEc

Goyal,S.(2005).Strongandweaklinks.JournaloftheEuropeanEconomic
608–606.,3ciationAsso

Goyal,S.,M.J.v.d.Leij,andJ.L.Moraga-González(2006).Economics:An
emergingsmallworld.JournalofPoliticalEconomy114,403–412.

147

yBibliograph

Granovetter,M.S.(1974).GettingaJob:AStudyofContactsandCareers.
HarvardUniversityPress.

Granovetter,M.S.(1976).Network
sampling:JournalofSociology81,1287–1303.

Somefirststeps.anmericAThe

Grossman,S.J.andO.D.Hart(1986).Thecostsandbenefitsofownership:
Atheoryofverticalandlateralintegration.TheJournalofPoliticalEcon-
691–719.,94omy

Gu,B.,Y.Huang,W.Duan,andA.B.Whinston(2007,October).Onlinepeer-
to-peercommunities:Anempiricalinvestigationofamusicsharingnetwork
asadynamictwo-sidednetwork.workingpaper.

Gu,B.,Y.Huang,W.Duan,andA.B.Whinston(2009).Indirectreciprocity
inonlinesocialnetworks-alongitudinalanalysisofindividualcontributions
andpeerenforcementinapeer-to-peermusicsharingnetwork.McCombs
ResearchPaperSeriesNo.IROM-06-09,1–40.

Guiso,L.,P.Sapienza,andL.Zingales(2004).Theroleofsocialcapitalin
financialdevelopment.AmericanEconomicReview94,526–556.

Guiso,L.,P.Sapienza,andL.Zingales(2005).Trustingthestockmarket.
NBERWorkingPaper11648.

Guiso,L.,P.Sapienza,andL.Zingales(2006).Doescultureaffecteconomic
outcomes?JournalofEconomicPerspectives20,23–48.

Guiso,L.,P.Sapienza,andL.Zingales(2009).Culturalbiasesineconomic
exchange?TheQuarterlyJournalofEconomics124,1095–1131.

Hall,B.H.andD.Harhoff(2004).Post-grantreviewsintheu.s.patentsystem-
designchoicesandexpectedimpact.BerkeleyTechnologyLawJournal19(3),
989–1015.

148

yBibliograph

Harman,H.H.(1976).Modernfactoranalysis(3rded.).TheUniversityof
Press.Chicago

Hart,O.(2008).Hold-up,assetownership,andreferencepoints.mimeo.

Hart,O.andJ.Moore(1988).Incompletecontractsandrenegotiation.Econo-
755–785.,56ametric

Hart,O.andJ.Moore(1990).Propertyrightsandthenatureofthefirm.
JournalofPoliticalEconomy98,1119–1158.

Hart,O.andJ.Moore(2007).Incompletecontractsandownership:Somenew
thoughts.AmericanEconomicReview97,182–186.

Häussler,C.,D.Harhoff,andE.Müller(2009).Tobefinancedornot...-the
roleofpatentsforventurecapitalfinancing.CEPRDiscussionPaper(7115).

Jackson,M.O.(2004).GroupFormationinEconomics;Networks,Clubsand
Coalitions,ChapterASurveyofModelsofNetworkFormation:Stabilityand
Efficiency.CambridgeUniversityPress.

Jackson,M.O.andA.Watts(2002).Theevolutionofsocialandeconomic
networks.JournalofEconomicTheory106,265–295.

Jackson,M.O.andA.Wolinsky(1996).Astrategicmodelofsocialandeco-
nomicnetworks.JournalofEconomicTheory71,44–74.

Kosfeld,M.(2003).Networkexperiments.IEREWorkingPaper(152).

Kossinets,G.andD.J.Watts(2006).Empiricalanalysisofanevolvingsocial
network.Science311,88–90.

Kumar,R.,J.Novak,andA.Tomkins(2006).Structureandevolutionofonline
socialnetworks.InProceedingsofthe12thACMSIGKDDinternational
conferenceonKnowledgediscoveryanddatamining,Philadelphia2006,pp.
611–617.

149

yBibliograph

LaPorta,R.,F.L.deSilanes,A.Shleifer,andR.W.Vishny(1997).Trustin
largeorganizations.AmericanEconomicReview87(2),333–338.

Lemley,M.A.(2001).Rationalignorance
UniversityLawReview95,1495–1529.

Lemley,M.A.andC.Shapiro(2005).
EconomicPerspectives19,75–98.

atthepatenProbabilistictoffice.ts.patenNorthwestern

ofJournalThe

Levin,J.(2003).Relationalincentivecontracts.AmericanEconomicRe-
835–857.(3),93view

MacCallum,R.C.,K.F.Widaman,S.Zhang,andS.Hong(1999).Sample
sizenfactoranalysis.PsychologicalMethods4,84–99.

MacLeod,W.B.(2007).Reputations,relationshipsandcontractenforcement.
JournalofEconomicLiterature45,595–628.

Meurer,M.J.(1989,Spring).Thesettlemen
JouralofEconomics20(1),77–91.

toftpatenlitigatio.nRANDThe

Müller,F.,K.O.Stahl,andF.Wachtler(2008).Upstreamrelationshipsinthe
automotiveindustry:Acontractualperspective.mimeo.

Newman,M.(2001).Thestructureofscientificcollaborationnetworks.Proceed-
ingsofthenationalAcademyofSciencesoftheUnitedStatesofAmerica98,
404–409.

O’Dea,D.(2008,November).Heterogeneityinsocialnetworksandtheclassical
theoryofpublicgoods.workingpaper.

Prady,D.(2008).ïsthataguninyourpocket...?ädvertisingandevaluation.
.omime

Reitzig,M.,J.Henkel,andC.Heath(2007).Onsharks,trolls,andtheirpatent
prey-unrealisticdamageawardsandfirms’strategiesof“beinginfringed”.
ResearchPolicy36,134–54.

150

yBibliograph

Rosenblat,T.S.andM.M.Mobius(2004).Gettingcloserordriftingapart?
QuarterlyJournalofEconomics119,971–1009.

Saloner,G.andA.Shepard(1995).Adoptionoftechnologieswithnetwork
effects:Anempiricalexaminationoftheadoptionofautomatedtellerma-
chines.TheRANDJournalofEconomics26,479–501.

A..,PSapienza,.13387WP

oldra,TandL.Zingales(2007).Understandingtrust.NBER

Schankerman,M.andS.Scotchmer(2001).Damagesandinjunctionsinpro-
tectingintellectualproperty.TheRANDJournalofEconomics32(1),119–
220.

Schuett,F.(2009).Inventorsandimpostors:Aneconomicanalysisofpatent
.omimeexamination.

Shapiro,C.(2003).Antitrustlimitstopatentsettlements.TheRANDJournal
ofEconomics34,391–411.

Straus,J.andN.-S.Klunker(2007).Harmonisationofinternationalpatent
law.InternationalReviewofIntellectualPropertyandCompetitionLaw8,
907–36.

Taylor,C.R.andS.N.Wiggins(1997).Competitionorcompensation:Supplier
incentivesundertheamericanandjapanesesubcontractingsystems.The
AmericanEconomicReview87,598–618.

Womack,J.P.,D.T.Jones,andD.Roos(1991).TheMachineThatChanged
theWorld:TheStoryofLeanProduction.HarperPaperbacks.

Xia,M.,Y.Huang,W.Duan,andA.B.Whinston(2007).Toshareornotto
share?-anempiricalanalysisonuserdecisionsinonlinesharingcommuni-
er.paporkingwties.

151

yBibliograph

152

Erklngäru

heEidesstattlic

und

angefertigt

selbständig

Dissertationorliegendev

dieh,icerkläreHiermit

derinihrangegebenenHilfsmittelbedientzuhaben.

michkeineranderenals

h-ennzeicgek

solche

als

Quellen

enander

InsbesonderesindsämtlicheZitateaus

ersehen.v

enQuellenangabmitundnet

12.05.2010.Mannheim,

Johannes

enenKo

yBibliograph

154

VitaeCurriculum

1981

1998

2000

2000-2006

2003-2006

2006-2010

2007-2010

2009

BorninFrankfurtamMain,Germany

USHigh-SchoolDiploma,PhillipsAcademyAndover,
Andover,Massachusetts

GraduationfromSecondarySchool,Kaiserin-Friedrich-Gymnasium,
yGermanburg,HomBad

UndergraduateStudiesinEconomics,UniversityofMannheim,Germany

UndergraduateStudiesinLaw,UniversityofMannheim,Germany

inGraduateEconomicsStudies(CDSE),inUnivEconomicsersityatoftheCenMannheim,terforGerDomctoralanyStudies

ResearcProfessorhKAssistanonradtatStahl,thePhD,ChairUnivforersitAppliedyofMicroMannheim,economics,Germany

VisitingScholaratDartmouthCollege,Hanover,NewHampshire

Un pour Un
Permettre à tous d'accéder à la lecture
Pour chaque accès à la bibliothèque, YouScribe donne un accès à une personne dans le besoin