nested苹果5

DeterminantsofFormationandDurationofTerritorialExclusivityAgreements:EmpiricalEvidencefromtheGlobalDistributionoftheAppleiPhoneTimBurnettBangorUniversityt.
burnett@bangor.
ac.
uk15thMarch2015AbstractTheinternationaldistributionoftheAppleiPhone,sinceitslaunchin2007,hasbeencharacterisedbyanirregularpatternofterritorialexclusivityagreements.
Notonlywasthelaunchstaggeredacrossdierentcountries,thedecisiontodistributeexclusivelythroughasingledownstreamnetworkoperator,andthedurationofsuchdecisions,variedbetweencountries.
ThispaperusesaCragg'sTobitAlternativeregressionapproachtodeterminewhetherthisdistributionstrategyfollowedaclearpatternandit'sdeterminants.
Headlineresultsshowthatthestructureandtechnicalcharacteristicsofamarketaresignicantindeterminingthedurationofanyexclusivityagreements.
Otherkeyresultsindicatethatthedeterminantsofdurationofagreementsaredierentfromthosemotivatingitsoriginalimposition.
JEL:L22,L42,L96Keywords:Telecommunications,VerticalRestraintsIwouldliketothankMonicaGiulietti,StephenDavies,BruceLyons,andtheESRCCentreforCompe-titionPolicyfortheirfeedbackandsuggestions.
IwouldalsoliketothankDrRichardCadmanwithoutwhoseassistancewithdatathispaperwouldnothavebeenpossible.
11IntroductionThereexistsawelldevelopedliteraturediscussingthetheoreticalmotivationfortheim-plementationofverticalrestraints,specicallyconcerningexclusiveterritories.
However,therehasthusfarbeenlittleattentionpaidtodynamicapplicationsofexclusivedistri-butionortheoptimaldurationofanyagreements.
Thispaperusestruncated-variableregressionmodelstoempiricalyexaminethereal-worldexampleoftheglobaldistibutionoftheAppleiPhonebetween2007and2010.
Anotablefeatureofthisdistibutionwastheimpositionofanumberofconditionsuponthemobilenetworkoperatorsthatweretosellthedevicetonalconsumers,akeycomponentofwhichwastheallocationofexclusivedistributionagreementsfordistributorsinsomecountries.
Furthermore,wherethesere-straintswereimposed,thesevariedinduration-rangingfrommonthsuptoseveralyears.
Ininvestigatingexclusivitythispaperhasseveralmainobjectives:therstistotesttheunderstandingoftheconditionsunderwhichexclusivityagreementsareenteredinto;thesecondistoinvestigatethedriversofdurationofsuchagreements,specicallythecasewherethesedeviatefromthedriversoftheinitialdecisiontosupplyexclusively.
Inadditiontothis,thepaperalsoexaminesthefactorswhichdeterminedthosecountriesinwhichtheiPhonewassold(whetherexlusively,ornot).
Toachieveitsaimsthispaperusesanumberofeconometricapproachestoanalyseacustomdatasetconstructedbytheauthor;thisdatasetcontainsinformationonthecountriesinwhichtheiPhonelaunched,thenetworkoperatorswhichsoldthedevice,andthedatesthateachoperatorstartedselling.
Thiscollecteddatasetisaugmentedbydatafromanumberofo-the-shelfsourcesincludingtheInternationalTelecommunica-tionsUnion(ITU),theWorldCellularInformationService(WCIS),andtheWorldBank(WB).
Chiefamongsttheeconometricmethodologiesemployedtoanalysethisdatasetistheuseoftruncatedvariablemodels,notablyTobitandCragg'sTobitAlternative.
Thepresentinvestigation,withitseconometricapproach,improvestheunderstand-ingofexclusivityinanumberofways.
Itaddstoaveryslimempiricalliteratureondeterminantsofexclusivedistribution,alsoincludingananalysisofentry.
Morecrucially,thispaperistherstpaper(tothebestoftheauthor'sknowledge)whichquestionstheoptimaldurationofexclusivitycontractsfromeithertheoreticalorempiricalspheres.
Indoingso,theresultsofthepapersuggestthattheexplanationofdurationassomeextensionoftheprocesswhichdeterminesinitialexclusivityisawed.
Thepaperproccedsasfollows:thefollowingsectionprovidesbackgroundinformationonthedistributionoftheiPhone;thisisfollowedbyareviewoftherelevantliterature.
Adetailedlookateconometricmethodlogyandmodelspecicationisfollowedbyade-scriptionofthedatasetemployed.
Thepaperconcludeswiththeeconometricresultsandadiscussionoftheirimplications.
22TheDistributionoftheAppleiPhoneTheiPhone,whichwaslaunchedin2007,waselectronicsmanufacturerApple'srstforayintotheproductionofmobiletelecommunicationhandsets.
Previouslythermhadbeenidentiedwith,amongstothers,homecomputersandportablepersonalmusicplayers.
TheiPhonefeaturedarelativelynoveltouch-screenuserinterface,anaestheticdesignwhichwasverysimilartoApple'sothersuccessfulproducts,andheavilypromotedtherm'sonlineapplicationstore1whichsuppliedsmallinexpensivesoftwareadd-onsforthehandset.
Fromthis2007startingpoint,bythenalquarterof2012theiPhoneaccountedfor9.
2%ofglobalhandsetsalesbyvolume2.
ThedistributionoftheiPhonefollowedapatternofdistributionwhereitwassoldexclusivelybyasinglenetworkoperatorinsometerritories.
Oftheseexclusivityagree-ments,someweretemporaryandcharacterisedbydieringdurationsindierentterri-tories,whileotherswereon-going.
Althoughexclusivedistributionofpremiumhandsetshasbeenselectivelyemployedforseveralyears3,theubiquityoftheiPhonehasenabledthecollectionofabroadinternationaldatasetconcerningthepatternofitsdistribution,thusenablingthepresentanalysisofthechoiceofcountriesinwhichthephonewassold,andthedeterminantsofexclusivityagreementsandtheirsubsequentduration.
Figure1:iPhonedistributionstrategiesSource:Author'sownresearchFigure1indicatesthedistributionofagreementtypeacrossterritorieswheretheiPhonehasbeenmadeavailablesince2007.
Itshowsthatexclusivityagreementswere1Whichdrewheavilyontherm'sprevioussuccessatsellingmusiconline.
2Gartner[2013]3Zhuetal.
[2011]oerashortoverviewofexclusivityagreementsintheUS.
3favouredoverhavingcompetingretailers,andthatinroughlyhalfofthesecasesthisexclusivityhadexpiredbytheendofthestudyperiod.
Inadditiontothedierencesinexclusivity,thelaunchoftheiPhonewasstaggered;itdidnotsimultaneouslylaunchinallcountries.
Figure2illustratesthepatternofdistributionacrossdierenttimeperiods;Exclusiveindicatesthatarminthecountryhadaperiodofexclusivedistribution,whileCompetitiveindicatesthatmorethanonerminthatparticularcountrysimultaneouslysoldtheiPhoneatlaunch.
Figure2:ExclusiveversuscompetitivedistributionofiPhoneatlaunchSource:Author'sownresearchFigure2showsthegeneraltrendinfavourofexclusivityandisdominatedbyaspikeinthethirdquarterof20084;thisquarterrepresentsaperiodofinternationalroll-outoftheiPhone.
Althoughtheearlydistributionwasdominatedbyexclusivityagreements,theystillcontinuedtobeenteredintoafter2008.
Figure3indicatesthat,inadditiontothestaggeredentryandselectiveexclusivity,thedurationoftheexclusivedistributionagreementsalsovariedbetweendierentcoun-tries.
Some,suchasGermanyortheUS,didnothaveasecondvendoroftheiPhoneforaroundthreeyearsfromthedateoftheoriginalexclusiveoperator,whileothershadasecondvendoroftheiPhonemarketafterseveralmonths.
ThisvariabilityispresentevenbetweencountrieswheretheiPhonewaslaunchedinasimilartimewindow.
4Thesecondspikein2011islargelyresultofthesimultaneousintroductionoftheiPhonetoanumberofsmallCaribbeancountries(including,amongstothers,Anguilla,AntiguaandBarbuda,andStLucia).
4Figure3:DateofrstiPhonedistributorandtimeuntilnextentrantSource:Authorsownresearch3ExclusiveDistributionThedecisionbyanupstreamsuppliertosupplyanintermediategoodexclusivelytoasinglerminanotherwisecompetitivedownstreammarketisknownas'exclusivedistribution'(whichcanbeusedinterchangablywith'exclusivesupply'and'territorialexclusivity').
Theeectofexclusivedistributionistosegmentthedownstreammarket(eithergeographically,orbycustomertype),suchthatamarketsegmentisservedbyasinglerm.
Byallocatingamonopolypositioninacertainsegment,territorialexclusiv-itycanhave(amongstotheroutcomes)theeectofreducingdownstreamintra-brandcompetition;thusreducingtheextenttowhichdownstreamrmscompeteagainsteachothertosellproductsproducedusingtheinputoftheupstreamrm.
Exclusivesupplycanalsobeusedasasolutiontohold-upproblems,toensureoptimallevelsofinvest-mentbydownstreamrmsandarerelativelycommonespeciallyamongstfranchisingorganisationswherethefranchiserwillallocatethefranchiseeacertaincatchmentarea;ineectguaranteeingthedownstreampartneracertainlevelofdemand.
Thefollowingreviewfocussesonthebroadliterature,boththeoreticalandempirical,whichinvestigatestheuseofexclusivedistributionagreements.
Thetheoreticallitera-tureisbroadlyfocussedontwomainthemes;exclusivityrestraintstoresolvedivergentincentivesinintra-brandcompetition,andexclusivityasameanstosoftenupstreamcompetitioninthepresenceofintra-brandcompetition.
Theempiricalliteratureismorevariedinscopeanddrawsrelevantconclusionsfromarangeofdierentreal-worldin-vestigations.
53.
1Intra-brandcompetitionThekeyliteratureonterritorialexclusivityandintra-brandcompetitionisfocussed(intermsofmodel)onthecasewhereanupstreamrmfacesnoupstreamrivalsandsellsintermediateproductstodownstreamrms.
Inthiscasetheupstreamrmhastheabil-itytoimposearangeofrestraintstoinuencethedecisionsofdownstreamrmsandbetteraligntheirincentiveswithitsown.
Thecasethatanupstreamrmdoesnotimposerestrictionsonthedownstreammarketdoesnotimplythatthedownstreammarketisotherwisecompetitive.
Onthecontrary,itisdownstreammarketimperfectionswhichmotivatetheupstreamrmtoimplementrestraints.
5Twokeypaperswhichbothexaminetherationalefortheimple-mentationofdierentrestraintsareMathewsonandWinter[1984]andReyandTirole[1986].
Bothseektoexplaintheuseofrestraintsinaverticalsettingwithasingleup-streammonopolistselectinghowtocontractwithoneormoredownstreamagents.
InMathewsonandWinter[1984]theauthorsproposetheminimumsetofverticalrestraints,includingexclusivedistribution,whichcanbeusedtoelicittheequivalentverticallyintegratedmonopolyprotbenchmark(orsimilarlyasingleupstreamrmsellingintoaperfectlycompetitive(freeentry)retailmarket).
Theauthorsspecifyacircularmodelofdownstreamcompetitionwhereretailersmustadvertisetheirproductacrossanendogenouslychosenradiusofinuence.
UnderattempttousesimplelineartarisMathewsonandWinter[1984]highlightthreeexternalities:doublemarginalisa-tion;horizontalpricingandundercutting;andhorizontalinformation(oradvertising)spillovers.
ThedoublemarginalisationresultmatchestheclassicdoublemarginalisationeectofSpengler[1950],whiletheotherhorizontalexternalitiesariseasaresultofadverseselectionbydownstreamrmsasaresultofintra-brandcompetition.
Theadvertisingspillovereectmeansrmsdonotfullyappropriatethefullbenetsoftheiradvertisingeortandconsequentlyadvertisetoolittle,whilethepriceexternalitypromptsrmstopricebelowtheoptimallevelinordertoattractmarginalcustomerswhichwouldotherwisebeservedbyadownstreamrival.
Alltheseeectsresultintheinabilityoftheverticalchaintoappropriatethemaximumavailableprots.
MathewsonandWinter[1984]showthattheuseofamulti-parttari,inconjunc-tionwithclosedterritorialdistribution(retailersarenotcompetingtoattractcustomersfromrivals'territories),eliminatesbothhorizontalandverticalexternalities.
However,ifspilloversinadvertisingarereintroducedintothemodel,despitethelackofhorizontalpriceexternality,theretailerstilldoesnotappropriatethefullbenetoftheiradvertising5Astandardresultinmanyofthefollowingpapers[MathewsonandWinter,1984;ReyandTirole,1986]isthatwhenadownstreammarketisperfectlycompetitive,thiscompetitioneliminatesdownstreamintra-brandprotsandenablestheupstreamrmtosecuretheverticallyintegratedprotbysettingalinearwholesalepriceequaltothemonopolynalprice.
6andthusadvertisessuboptimally.
ImplicitinMathewsonandWinter[1984]istheas-sumptionthat,whilesellingandbuyingactivitycanbesegmentedintoclosedterritories,advertisingeortcannot.
Althoughtheimplementationoftheauthors'interpretationofclosedterritorydistributiondoesnotappropriatethefull'integrated'levelofprotbecauseoftheadvertisinginformationalexternality,itdoessolvethehorizontalpricingexternality.
Thusexclusivedistribution,whencombinedwithasuitabletwoparttari,resultsinasuperioroutcometothatwhichwouldbeelicitedinitsabsence.
TheadvertisingdilemmainMathewsonandWinter[1984]isoneofadverseselection;competingrmsdonotadvertiseoptimallybecausetheybelievethattheywillnotap-propriatethefullbenetsoftheiradvertising.
ThedilemmaissimilartoLafontaineandSlade[2005]intheirapproach.
Theseauthorsciteaseparateformofadverseselectionasthecausefortheimplementationofverticalrestraints;theclassichold-upproblemwhereupstreammanufacturersrequiredownstreamdealerstomakecostlyinvestments.
Theallocationofanexclusiveterritorycanguaranteeadownstreamrmaminimumlevelofdemand,thusremovingtheriskfromtheirinvestmentandaligningtheinterestsofbothup-anddownstreamparties.
AdierentapproachtoMathewsonandWinter[1984]isusedinReyandTirole[1986]whichhighlightstheprincipal-agentnatureofaverticalretailchain.
Ofteninfor-mationisdecentralised–anupstreammanufacturermaynothaveimmediateknowledgeofthedownstreammarketconditions.
Undersuchconditionsattemptstoimposeverticalrestraintsupondownstreamrms,baseduponassumptionsofdemandorsupplycon-ditions,maynotresultinextractionthefullpotentialprotandinappropriateverticalrestraintsimposedbyanupstreamrmmayevenberejectedbydownstreamrms.
ReyandTirole[1986]ndthat,undersimpleconditionsofuncertainty(overlevelofdemandandmarginalcostofdownstreamsupply)decentralisationofnalpricingdecisionsbygrantingofanexclusiveterritorywithatwo-parttarileadstoasuperioroutcometocompetition.
Thisillustratesthestandardresultthatwhenimperfectionsareintroducedintothedownstreammarket,specicallywhenretailersaredierentiated,theeective-nessofcompetitiontoresolveverticalissuesisdiminished.
O'BrienandShaer[1992]reachthesameconclusionasMathewsonandWinter[1984]thatinordertoachievetherst-bestprottheupstreamrmmusteliminatethehorizontalpricingexternality.
Furthermore,theauthorsshowthatthiscanbeachievedeitherthroughresalepricemaintenance(whichtheyshowtoeliminateallverticalandhorizontalexternalities),orotherwisethroughtheuseofexclusivedistribution.
ThisO'BrienandShaer[1992]resultiscontingentuponthenon-violationoftheterritorialexclusivityagreement.
Specicallytheauthorsshowthatitmustnotbeintheupstreamrm'sinteresttotacitlyencouragethisviolation.
Inthearrangementofanexclusiveter-ritorythereexistsamoralhazardprobleminthat,onceatwo-partcontractisagreed,theupstreamrmhastheincentiveto'sellthemarketasecondtime'.
Thisresultsinnegativeprotsforthermwhichagreedtheinitialcontract.
Thecredibilityissue7inherentintheestablishmentofexclusiveterritoriesalsoinformstheanalysisofBoyd[1993],discussedinsection3.
3inrelationtotheempiricalliterature.
Thereisaseparatemoralhazardprobleminthat,onceacontracthasbeenagreed,adownstreamrmmayhavetheincentiveandabilitytoprotablyselloutsidetheiral-lottedterritory.
Thisresultsinrationingofaxedquantityacrosstwoterritoriesratherthanone.
ThegovernanceofexclusiveterritoriesisexaminedbyDuttaetal.
[1994]whoproposeamodelsimilartothatofReyandTirole[1986]wheretheupstreammonopolistsellsaninputtotwodownstreamcompetitorsusingatwo-parttariandallocatestoeachanexclusiveterritory.
SimilarlyDuttaetal.
[1994]alsoincludeuncertaintyinboththecostofsalefordownstreamrmsandthelevelofdemand.
Aspertheearlierpaper,thisoerstheupstreamrmtheincentivetodecentralisedecisionmaking.
Inbroad-eningthescopeofthemodelDuttaetal.
[1994]ndthatinequilibriumtheupstreamrmtoleratesacertaindegreeof'bootlegging'(thesalebyonedownstreamrmintotherival'sterritory).
Asmallamountofbootleggingispreferabletothecaseinwhichnoterritoriesareallocated,orwherethereiszeroenforcementoftheterritories.
Bothoftheseresultinfullintra-brandcompetitionwithzeroinvestmentindemand-inducingservices.
Theauthorsalsoshowthatasdemandbecomesmoresensitivetoresellerser-vices,theincentivetostrictlyenforcethisbootlegginglimitincreasestoensurethatrmsdonot'cheat'ontheagreementandbootlegmore.
Theoverallconclusionfromthisliteratureisthatunfetteredintra-brandretailcom-petitionmaynotelicittherst-bestprotfortheupstreamrm.
Inthisinstancever-ticalrestraints,specicallyexclusivedistribution,canbeusedtoremedyextantissuesofmoralhazardandadverseselectionandalignincentivesofup-anddown-streamrms.
3.
2Inter-brandcompetitionTheliteraturementionedthusfarprimarilyconcernsitselfwiththecaseofintra-brandcompetitionandtherestraintsusedbyanupstreammonopolistinordertomaximiseprotsgivendierentcongurationsofdownstreamcompetition.
Theintroductionofinter-brandcompetitionmeansanupstreamrmsmustconsidernotonlythesalesoftheirownproducts,butthestrategiesoftheirupstreamrivals.
Upstreamrmsfacethechoiceofnotonlywhethertodistributethroughasingledownstreamretailer,butalsothedecisionofwhethertouseacommondownstreamagentwiththeirupstreamrival.
ReyandStiglitz[1995],Lin[1990],andO'BrienandShaer[1993]6allindicatethatcompetingupstreamretailersprefertodealexclusivelywiththeirowndownstreamrms–avoidinghavingacommondownstreamagent.
Byminimisingdownstreaminter-brandcompetitionthissoftensthecompetitiveconstrainttheyimposeononeanother.
Focussingonthesofteningofupstreamcompetition,SaggiandVettas[2002]ndthatupstreamrmschoosetoexclusivelydistribute;wherermsdealwithmultipledown-6O'BrienandShaer[1993]isacritiqueofLin[1990]andutilisesthesamemodel.
8streamagentsthisinducestherivaltobehavemorecompetitively.
Afeaturecommontoalltheliteratureabove(bothintra-andinter-brand)isthatupstreamrmsmaketake-it-or-leave-itoerstodownstreamagents.
Trivedi[1998]andDobsonandWaterson[1996]bothexaminethecasewherepowerismorebalancedandconsidertheincentivesfordownstreamrms.
Bothpapersindicatethatinsomecircum-stancesasingledownstreamrmspreferstocarryproductsfromallupstreamrms.
Subramanianetal.
[2012]isrelevanttothecurrentpaperastheauthorsspecicallyexamineexclusivityagreementswithrespecttotheAppleiPhoneandmobiletelecom-munications.
Liketheabovepapers,Subramanianetal.
[2012]examinethecasewhereanexclusivityagreementisinthemutualinterestsofboththeupstreamtelephoneman-ufacturerandthedownstreamnetworkoperator.
Theauthors,likeReyandStiglitz[1995],utiliseasimpletwo-upstream,two-downstreammodelbutspecicallymodelfeaturesofmobiletelecomssuchasincreasedconsumerhet-erogeneity.
Subramanianetal.
[2012]showthatundercertainscenariosahandsetmanu-facturermaychooseexclusivityinordertoinducearivaltoincreaseitswholesaleprices,thussofteningcompetition.
However,unlikeReyandStiglitz[1995],theauthorsalsofocusonmutualincentives,asopposedtoatake-it-or-leave-itapproach.
Subramanianetal.
[2012]ndthatnetworkoperatorsmaynditadvantageoustoenterexclusivitysincelimitingtheavailablehandsetstotheirrivalsmayraisetheirrival'sinputcosts.
Thendingsaboveoccurunderspecicboundsofrelativequalitybetweenhandsets,thoughSubramanianetal.
[2012]alsondthatlikelihoodofexclusivityagreementsisdecreasinginthedegreeofhandsetdierentiation7.
3.
3EmpiricalliteratureInadditiontothetheoreticalliteraturethereexistsanempiricalliteraturewhichexam-inestheimpositionofexclusiveterritories.
Thisismainlysplitintotwothemes:thewelfareeectsofexclusivityrestraints;andthemotivationforstrategicdelegationofdownstreampricingdecisionsversusverticalintegration.
Thereisaveryslimempiricalliteraturewhichexaminesaspectsofterritorialexclu-sivity.
SassandSaurman[1993]studytherelationshipbetweenbrewersandbeerwhole-salersandtheimpactonpriceofdierentstatelawsconcerningterritorialexclusivity.
Usingacross-sectionoftwentyfourstatesSassandSaurman[1993]ndthatexclusivity7Understandablesinceincreasingthedegreeofdierentiation,reducesthesubstitutabilityofproducts-meaningthathandsetwholesalepricesinabsenceofexclusivityarehigherandlessreactivetothepricesoftherival.
9agreementsservetoincreaseprices,indicatingtheexpectedeectofreducedintra-brandcompetition.
Theauthorsalsoshow,however,thatwhereexclusiveterritoriesareman-dated,theestimateddemandforbeerishigher;suggestingthat,asperLafontaineandSlade[2005],territorialexclusivitycanprovideguaranteestormswhichincentivisesotherwisecostlysaleseort.
InsteadofusingacrosssectionalanalysisBrenkersandVerboven[2006]utiliseanat-uralexperimenttoestimatethewelfareeectsofachangeinlegislationtowardverticalrestraints.
ThecasestudiedinvolvesthesaleofcarsintheEUwherein2002itwasproposedtoabolishrulespermittingterritorialexclusivityandrestrictmanufacturerstopolicieswhichaimedtointensifyintra-brandcompetitionandreducecross-borderpricediscrimination.
TheauthorsappliedtheobserveddatatoanestedLogitmodeltoestimateademandfunction,thenutilisetheobservedcoecientstoestimatewelfarepost-liberalization.
BrenkersandVerboven[2006]ndthattheabolitionofterritorialexclusivitywhichwouldpermitrmstofreelycompetewoulddeliverwelfaregainsofbetween1.
6billionand8billionEuros.
Theauthorsremark,however,thattheseguresexcludethepotentialeciencieswhichcanbeachievedthroughverticalco-ordinationsuchastheincentivisationofsaleseort.
Zhuetal.
[2011]seektomeasurethewelfareconsequencesofthelong-termexclu-sivityagreementbetweenAppleandnetworkoperatorAT&TforthedistributionoftheiPhoneintheUS.
Unlikethepresentinvestigationtheauthorsdonotexaminethemotivesfortheexclusivityagreement,butrathertakeitasagiven.
Theauthorsutiliseresearchdataonmobiletelephoneusefrom2007and2008toestimatearandomcoef-cientLogitmodelforcommonhandsetsandcontractsubscriptionbundlesinordertoestimatethedemandformobileservices.
Theauthorsnextestimatethesupplyfunctionusingutilisingwholesaleandnalconsumerprices.
Bydevelopingacounterfactualanal-ysisbasedupontheestimatedmarginsatoperatorandmanufacturerlevel,theauthorsestimatethatthereductiononintra-brandcompetitioncreatedbytheexclusivityoftheiPhoneresultedinawelfarelossofbetween$210millionand$326million.
Conceptually,themostsimilarpapertothepresentstudyisBoyd[1993];ararepaperwhichtheoreticallyandempiricallyexaminesthemotivationofanupstreamrmtoengageinexclusiveterritorialdistribution.
Inthisinstancetheauthoruseslitigationcasesbasedaroundverticalrestraintoftradetodeterminewhetheranupstreamrmimposedresalepricemaintenance,exclusiveterritoriesorbothontheirdownstreamre-tailers.
Boyd[1993]8observesthat,likeReyandTirole[1986],resalepricemaintenanceandexclusivedistributionmaynotbeperfectlysubstitutable.
Specically,theauthor'sstudyndsthatforsometypesofgood(routinepurchases)upstreamrmshaverou-tinelyembloyedterritorialexclusivity,whileresalepricemaintenanceispreferredfor'experience'goods.
Thoughfocussedonverticalintegration,LafontaineandSlade[2007]warrantsinclu-8AndalsoBoyd[1996].
10sionhereasitsharesmanyfeaturesincommonwiththecurrentstudy,sinceitinvestigateupstreamrms'incentivestoimposeverticalrestraints;inthisinstancetheirinvestiga-tionconsistsprimarilyofareviewofstudies.
3.
4ContributionThereisawealthoftheoreticalliteraturewhichfocussesonincentivesfortheimpositionofvariousverticalrestraints.
Theliteraturewhichconsidersintra-brandcompetitionheavilyfocussesontheuseofrestraintstoresolvedivergentincentivesatdownstreamstagesofproduction.
Theinter-brandliteratureinsteadexaminestheuseofrestraintstosoftencompetitionbetweenupstreamrms.
Alsowithinthescopeofinter-brandliteratureisthatwhichexaminesnotjustrestraintsimposedupondownstreamrms,butalsothecasewheretheyaremutuallybenecial.
Bothbranchesofthetheoreticalliteratureservetoinformvariableswhichcanbeincludedinempiricalinvestigation,asperLafontaineandSlade[2007].
Theempiricalliteraturehastakenseveralapproaches,focussingonbothwelfareeectsandincentives.
Howevertherearefewexampleswheretheincentivesfortheim-positionofexclusivedistributionareempiricallyexamined,perhapsonlyBoyd[1993].
Thepresentinvestigation,withitseconometricapproach,improvestheunderstand-ingofexclusivityinanumberofways.
Itaddstothelimitedempiricalliteratureondeterminantsofexclusivedistribution,alsoincludingananalysisofentry.
Morenotably,asmentionedintheintroduction,thispaperopensadialoguewhichquestionswhetherexclusivitycontractshavesomeoptimalduration,andwhatmightdeterminethis.
Inthisrespectthepaperrepresentsanimportantstageinabroader,emergent,researcharea.
4EconometricApproachThefollowingsectionoutinestheeconometricapproachtobeemployedintheinvesti-gationofthedeterminantsofdecisiontoemployterritorialexclusivity.
Thedecisiontosupplyaproduct(inthiscasetheiPhone)ismodelledasamulti-stagedecisionprocess.
Specically,thedecisiontoenteramarket;thedecisiontosupplyexclusively;and,conditionalonexclusivedistributionbeingchosen,thedecisionoverthedurationofsuchexclusivity.
Thedecisionoverthedurationisnecessarilysubjecttothedecisiontosupplyexclusively.
Fromtheprevailingliterature,thedecisionoverwhethertodealexclusivelymightbedrivenbyanumberoffactorssuchasadverseselectiononthepartofdownstreamrms,11ormarketimperfectionsanduncertainty.
Itmightalsobetoincentivisetheparticipa-tionofdownstreamrms(or,moreaccurately,adownstreamrm);thisislinkedtoLafontaineandSlade[2005]inthatacertainmarketsizemustbeguaranteedtoasingledownstreamrminorderforthemtoparticipate;notablyinthepresenceofdownstreamxedcosts.
Inthiscaseonemightexpectasmall,orparticularlyundeveloped,markettobeunabletosupporttwocompetingdownstreamrms,requiringterritorialexclusivity-inthisrespect(similarlytoReyandTirole[1986])upstreamrmsmightbeconstrainedbythewillingnessofdownstreamrmstoagreetoproposedrestraints.
Thereis,however,littleinthebackgroundmaterialtoinforma-prioriexpectationsconcerningthedurabilityofexclusivityagreements.
InthisrespectBoyd[1993]isrele-vantastheauthorhighlightsthattheoptimalverticalrestraintimposedonadownstreamrmmayvaryovertimeaccordingtotheproductcycle.
Thepreviousstepintheupstreamrm'sdecisionprocess-theentrydecision-canbethoughtofasbeingdeterminedbytheprotabilityofsuchamove;determinedbymarketsize,wealth,speciccostsassociatedwithgivenmarkets,orevenintellectualproperty(IP)issues.
Thedecisionsoftheupstreamsuppliercanbeillustratedusingadecisiontree;splittingdecisionsintotwodistinctstages:ForanycountryorterritoryifatstageBπin=max(πexc,πcomp)>0thenatstageAthedecisionistakentoenterthemarket.
Giventhatentryhasoccurredthemostprotablestrategyispursuedsubjecttoaprotconstraintforthedownstreamrms(s).
9Figure4:Apple'smarketentrydecision9Downstreamrmswillnotenterintoanagreementifitdeliverslowerprotthantheycouldachieveoutsidethedeal.
12Ineconometricallyanalysingthedecision,theanalysisissplitintotwostages:theentrychoiceatA,andtheexclusivitydecisionatB(executedagainstthesubsampleofcountrieswhichchose'Yes'atA).
NotableinthisstudyisthedecisiontomodelthedecisionsatBasasinglerelatedprocess-thisisrelatedtodistributionalassmptionsofusingcensoreddata-discussedatlengthinsection4.
2.
4.
1EntrydecisionThepropensityofAppletosellphonesintoaspecicmarketislikelytobesomefunctionofboththecharacteristicsofthecountryandthemarket:Entrypropensity=E(Countrysize,Countrypopulation,Countryincome,Countryregion,Marketcompetitiveness,Numberofrms,Markettechnology)(1)Whiletheabovefactorsinuencetheentrydecisionitisnotpossibletoobservethetrueentryfunctiononwhichthepropensitytoenterthemarketisbased.
Insteadweobserve,baseduponthecharacteristicsofthecountry,whether(ornot)theiPhonewasactuallylaunched.
Inthisrespecttheobservedentrydecisionisabinaryyes/nobaseduponsomeunderlyinglatentprotfunction.
Thus,inordertoformallyanalysethelaunchoftheiPhoneaProbitmodelisemployedbasedupontherandomutilityframeworkofGreene[2012]].
Ineachcountry'i',wheretherearej=1,2,.
.
.
,Noperators.
ThemaximumpotentialprotwhichcanbeextractedbyApplebycontractingwith'n≤N'networkoperatorsisindicatedbytheexpression'maxnj=1πi,j'.
Atthispointitisnotnecessarytodenetheexactnumbern,onlytounderstandthatineachcountrythereexistssomeoptimallevelofn≥0whichdeliversthehighestlevelofprot.
Thedeterminationofnisthesubjectofthesecondhalfoftheinvestigation.
DeningΠi,outastheprotforAppleifitchoosesnottosellintoacountry,andΠi,inastheprotgaineduponentry:Πi,out=0(2)Πi,inmaxnj=1πi,j=xiα+εi(3)Fromequation(2),Apple'sprotfromnon-entryisequaltozero.
Equation(3)in-13dicatesthattheprotfromentryintoiisdependentuponavectorofcharacteristics(xi).
Thesecharacteristicsareassociatedwithaxedsetofcoecients(α)andanerrortermwhichaccountsforunobservedheterogeneitybetweencountries-εi.
SinceApplewillenteramarketifitmakesnon-negativeprot,theprobabilityofsuchentrycanbeexpressed:Prob[Entryi=1|xi]=Prob[Πi,in>0]=Prob[(xiα+εi)>0|xi](4)Generatingthestandardprobitlatentvariable,Ei=xiα+εi,estimatedagainstthevectorofcharacteristicsofeachcountry(xi).
Oncethedistributionalassumptionsoftheprobitareimposedthiselicitstheprobabilityofentryasbeing:Prob[Enteri=1]=xiα∞φ(Ei)dEi=Φ(xiα)(5)WhereφisthestandardnormaldistributionandΦthecumulativedensityfunctionforthestandardnormaldistribution;whereΦ(Ei>0)>0.
5,creatingapositiveprob-abilityofentering.
TheProbitmodeliscalculatedusingmaximumlikelihoodestimation.
4.
2ExclusivityanddurationStudyingthedurationofexclusivitypresentsonekeyproblem:Becausetherearealargenumberofcountriesinthesampleinwhichexclusivitywasnotpursuedthismeansthatanyvariableexaminingthedurationofexclusivityacrossthesamplewillcontainmanyzeros,moreoverthatinthisinstanceisthatthezerovaluesrepresentacensoringofthedataset.
Theyconstituteanarticialminimumvalueimposedonthedurationvariable.
10Dataisconsideredtobecensoredorlimitedwhenanotherwisecontinuousvari-ablehasaprobabilitymassatoneormorepoints[Wooldridge,2002].
Thisprobabilitymassresultsindistortionstothemeanofthevariableandintuitivelyresultsinbiasedestimatedcoecientswhenusingthevariableinanalysis.
Whenusingatruncatedde-pendentvariableTobin[1958]observesthattheconcentrationofobservationsatthelimitingvaluemakesconventionalOLSmultipleregressioninappropriatesincethiswill10Anothercommonareaofstudywiththisproblemisconsumption-givenanestimatedmodelsomepeoplemaybepredictedtohavenegativeconsumption,butobservationsonpurchasingbehaviourlimitthelowerboundatzero.
14violateanumberoftheregressionassumptionsconcerningthedistributionoferrorterms.
Anumberofmodelswhichhavebeenproposedwhichdealwiththecensoringissuebysplittingtheestimationofthesamplebetweenthoseobservationswhicharecensoredandthosewhicharenot.
TheuseofthecensoredTobitmodel[Tobin,1958]isthemostcommonremedytocensoreddata.
LiketheProbitmodelitassumesthatthecensoredcontinuousvariablewhichisobserved(Duration)isactuallyrelatedtosomeunderlyinglatentvariable,D,suchthat:D=xβ+u,u|xNormal(0,σ2)Duration=max(0,D)Inthisinstancetheunderlyinglatentvariableisonlyobservedsubjecttothepassingofthezerothreshold,butthereisapositiveprobabilitymassatzero.
TheTobitmodelproposesthefollowinglikelihoodfunctionwhich,whenmaximised,yieldsestimatesofthevectorofcoecients,β::f(Duration|x)={1Φ(xβ/σ)}1(D=0)A(2π)12σ1exp(yxβ)2/2σ21(D>0)B(6)In(6)thelikelihoodfunctionhastwocomponents;thelikelihoodthatD=0indi-catedbyA,andthelikelihoodthatD>0shownbyB.
Inthecaseofdatawhichiscensoredatzero,theshortcomingoftheTobitmodelisthatboththeprobabilitytheD>0,andthevalueofDforanyvalueD>0aredeterminedbythesamemechanismandthesamevectorofparameters.
Thisrestrictionislogicalinanexamplewhereconsumptionofagoodissolelysomefunctionofincomeandzeroconsumptionofthegoodtheresultofnancialconstraint.
However,inmanycasestherelationshipbetweenthetwostagesisnotsoclear.
11Inthiscasewearestillinterestedinthevalueofthecontinuousvariable.
Itisstillnecessarytocontrolforthecensoreddistributionofthecontinuousvariable,butthebinaryvariablewhichdeterminesthepassingofthethresholdvalueisdeterminedby11Usingtheexampleofconsumption;supposethatdesiredconsumptionofthegoodisnotonlydeter-minedbyincome,butalsodeterminedbysomeparameterofpersonaltasteunrelatedtoincome(haircolour,forexample).
Inthisinstancetherewouldbeasubsetofindividuals(possiblywithbrownhair)whichwouldneverchoosetoconsumeapositivequantityofthegood,irrespectiveoftheirincome.
Thereisnolinkbetweenpassingthethresholdandthenalvalueofthevariable.
15someseparateprocess.
4.
2.
1Cragg'sTobitAlternativeCragg[1971]proposesatwo-stageor'double-hurdle'alternativetothesimpleTobitmodelinwhichthetwostages(binaryandcontinuous)canbedeterminedbydierentprocesses.
Intherelevantcaseofcensoringatzero(aspertheDurationvariable),inorderforthecontinuousvariabletorecordapositivevaluetwoconditionsmustbemet.
Therstconditiondictatesthattheprocessdeterminingthebinaryprocessmustyieldapositiveresponse.
Thesecondconditionrulesthat,subjecttotherstconditionbeingmet,theprocessdeterminingthecontinuousvariableissucienttodeliverapositiveoutcome.
Thusanobservedvalueofzeromaybetheresultofjustoneoftheseconditionsnotbeingmet,whileapositivevaluerequiresthesatisfactionofbothconditions:f(X,D|x1,x2)={1Φ(x1γ)}1(X=0)AΦ(x1γ)(2π)12σ1exp(yx2η)2/2σ2Φ(x2η/σ)11(D>0)B(7)ThelikelihoodfunctionfortheCragg[1971]doublehurdlemodelindicatedbyequa-tion(7),liketheTobit,hastwocomponents;theprobabilitythatX=0(A),andthelikelihoodthatD>0giventhecensoredcontinuousdurationvariable(indicatedbytheexpressionB).
From(7)themaximumlikelihoodestimateofγisthesameinformastheProbitestimator.
PartBof(7)isreadilyidentiableasaprobability-weightedtruncatednormaldistribution.
ThetermΦ(x2η/σ)1ensuresthatthedensityofthetruncatednormalintegratestoone.
NotableinCragg'smodelisthattheprocessdeterminingthehurdleresultandthepositivevaluearedierent;theydependupondierentvectorsofvariables(x1,x2),eachwithitscorrespondingvectorofcoecients(γ,η).
Wherex1=x2themodelpermitsthesamedependentvariablestoactindierentdirectionsatdierentstagesinthemodelifγ=ησ.
Thisallowsfortheseparabilityofthedeterminantsofthedurabilityofexclusivityagreementstodierfromthosevariablesdeterminingtheimpositionofexclusivity.
Notably,theTobitcanbeviewedasaspecialcaseofCragg'smodelwhereγ=ησ.
ThisformsthebasisoftheLinandSchmidt[1984]LagrangemultipliertestwhichteststhisrestrictiontodeterminetheappropriatenessoftheTobitspecication.
164.
3ModelspecicationsTheuseofCragg'sTobitAlternativemeansthatthesamevectorofcharacteristicsforeachcountrycanaecteachofthethreestagesdierently(eachstageoftheestimationfeaturesitsownvectorofcoecients).
Thissectionoutlinesthespecicationofthemodelswhicharetestedusingtheeconometricmethodologiesoftheprevioussection.
Alsooutlinedistherationalefortestingusingbothafullandrestrictedsample,andalsodescriptionsofsomeoftherobustnesstestingwhichisappliedtotheresults.
4.
3.
1PopulationeectsThisinvestigationusesasampleof187countriesbutthiscontainsanumberofcountrieswithlowpopulations.
Whereacountryhasalowpopulation,orasmallsize,thenitislikelythatoperatorsinthesecountriesarenotfacedwiththesamesetofconstraints;thechoicesetofthermsintermsofinvestment,orthenatureofcompetition,islikelytobedierent.
Assuch,allofthefollowingregressionspecicationsarerunusingafullsampleofcountries,andalsoexcludingthosecountrieswithpopulationlessthan500,000.
Anysignicantchangeintheresultswillindicatethattherelationshipbetweenthedependentandindependentvariablesisdierentforthesubsetofsmallercountries.
TheresultssectionalsofeaturesaMann-Whitneytesttotesttheappropriatenessoftherestriction.
4.
3.
2EntryInanalysingtheentrydecisionsection4.
1explainedthattheapproachbeingtakenistotreattheentrydecisionasbinarywhereavalueof1impliestheiPhonewassoldinaparticularcountry,zerootherwise.
TothiseectaProbitmodelisemployed.
Theanalysisexaminesseveralspecicationsoftheentrymodel;focussingonthecasewhereonlydemographiccontrolvariablesareconsidered,uptoamodelcontainingbothdemographicandtelecommunicationvariables.
DenotingEiastheunderlyinglatentvariabledetermininglaunchincountryi,thespecicationsusedareasfollows:Ineachofthespecications(A)to(D)variablesareiterativelyaddedtothere-gressiontoidentifytheiradditionalinuenceonthedecisiontoselltheiPhone.
Thecoecientsα4toα9representtheinuenceofthedummyvariablesasscociatedwiththeregioninwhichthecountryisbased.
Thesespecicationsarerepeatedwiththerestrictedsample(specications(E)to(H)).
17(A)Ei=α0+α1LogLandAreai+α2LogPopDensityi+α3LogGDPCapitai+εi(B)Ei=α0+α1LogLandAreai+α2LogPopDensityi+α3LogGDPCapitai++α9α4Continenti+εi(C)Ei=α0+α1LogLandAreai+α2LogPopDensityi+α3LogGDPCapitai++α9α4Continenti+α10Num.
Firmsi+α11HHIi+εi(D)Ei=α0+α1LogLandAreai+α2LogPopDensityi+α3LogGDPCapitai++α9α4Continenti+α10Num.
Firmsi+α11HHIi+α12Comp.
Standardsi++α13MobilesPer100i+εi4.
3.
3ExclusivityTworegressiontechniquesareusedintheinvestigationofexclusivity;theTobitmodelandCragg'sTobitalternative.
TheTobitisrestrictiveinanumberofassumptionsbutbytestingtheseassumptionsitallowsinferencestobemadeaboutwhetherthedeter-minantsofthedurabilityofexclusivesupplyarrangementsarethesamefactorswhichinuenceitsinitialimposition.
TheCraggmodelconverselyallowsmuchgreaterexi-bility.
TheTobitmodelTheTobitmodelisusedtodeterminewhetherthesamerelationshipgovernsboththedecisiontosupplyexclusivelyandthedurationoftheexclusivity.
ThespecicationoftheTobitmodelemployedisshownbelow:(I)Durationi=β0+β1LogLandAreai+β2LogPopDensityi+β3LogGDPCapitai++β9β4Continenti+β10Num.
Firmsi+β11HHIi+β12Comp.
Standardsi++β13MobilesPer100i+β14LaunchDelay+εiSection4.
2describesthattheTobitmodelisusedtoovercomeissuesrelatingtothetruncationofacontinuousvariablebyutilisingahurdlewhereDurationi>0.
ThisrestrictstheanalysisinthatthesameprocessmustgovernbothwhetherDurationiexceedszero,anditsultimatemagnitude.
Therelationshipbetweendriversofexclusiv-ityanddurationformoneofthekeystrandsofinvestigation,thusthisassumptionistestedusingtheLagrangeMultipliertestofLinandSchmidt[1984]andresultsdiscussed.
Withthesamerationaleasinthecaseforentry,theTobitmodelisperformedwith18boththefullsampleof84countrieswheretheiPhonewaslaunched,andthepopulation-restrictedsampleof81countries;resultingin,respectively,specications(I)and(J).
Cragg'sTobitAlternativeCragg'sTobitAlternativeallowsseparabilityoftheprocesseswhichgovernexclu-sivity,andthosedeterminingitsduration.
Cragg'smodelutilisesbothaprobitformfortheinitialexclusivitydecision,andatruncatednormalregressionforthecontinuousdurationvariable.
Assuch,eachofthefollowingspecicationshasaregressionrepre-sentingeachofthecomponents;whereXiistheunderlyinglatentvariabledeterminingexclusivityincountryi,andDurationithelengthoftheexclusivityperiod:(K)Xi=γ0+γ1LogLandAreai+γ2LogPopDensityi+γ3LogGDPCapitai++γ9γ4Continenti+εiDurationi=η0+η1LogLandAreai+η2LogPopDensityi+η3LogGDPCapitai++η9η4Continenti+υi(L)Xi=γ0+γ1LogLandAreai+γ2LogPopDensityi+γ3LogGDPCapitai++γ9γ4Continenti+εiDurationi=η0+η1LogLandAreai+η2LogPopDensityi+η3LogGDPCapitai++η9η4Continenti+η10Num.
Firmsi+η11HHIi+η12Comp.
Standardsi++η13MobilesPer100i+η14LaunchDelay+η15VendorPos.
+υi(M)Xi=γ0+γ1LogLandAreai+γ2LogPopDensityi+γ3LogGDPCapitai++γ9γ4Continenti+γ10Num.
Firmsi+γ11HHIi+γ12Comp.
Standardsi++γ13MobilesPer100i+γ14LaunchDelay+εiDurationi=η0+η1LogLandAreai+η2LogPopDensityi+η3LogGDPCapitai++η9η4Continenti+η10Num.
Firmsi+η11HHIi+η12Comp.
Standardsi++η13MobilesPer100i+η14LaunchDelay+η15VendorPos.
+υiTheregressions(K),(L),and(M),baseduponCragg'sTobitalternative,arealsoreplicatedfortherestrictedsample;denoted(N),(O),and(P).
194.
3.
4GoodnessofFitBecausetheabsolutevalueofthelog-likelihoodisnon-increasinginthenumberofvari-ablesthereexistsadangerof'over-tting'themodelsthroughtheinclusionofirrelevantvariables.
BurnhamandAnderson[2004]observethetrade-obetweenparsimonywhichcanresultinbiasedmodels,andthoseover-ttedmodelswheresignicanceislostorotherwisethereisspuriousidenticationofsignicanteects.
InordertoillustratethattheadditionofextravariablesresultsinabetterttingmodelAkaike'sInformationCri-terion(AIC)isemployed.
TheAICimproveswithassociateddecreasesintheabsolutevalueofthelog-likelihood,but(allelsebeingequal)degradeswiththesizeofthemodelasmeasuredbythenumberofincludedvariables.
Inamaximumlikelihoodmodel,suchastheprobit,theAICformulaisexpressedasbelow,whereasmallervalueofAICimpliesabetterttingmodel:AIC=2Df2ln(L)WhereDfrepresentsthedegreesoffreedomofthemodelandln(L)isthelog-likelihoodofthemodel.
Foralloftheregressionspecicationsabove,thespecicationofthemodelsistestedandtheresultsreported.
5DataThisinvestigationusesadatasetgatheredfromanumberofsources;itcontainsvari-ablesfromo-the-shelfdatasetsandalsodatawhichhasbeencollected.
Thedataissummarisedintable1.
Correlationcoecientanalysisindicatesthattherearenoin-stanceswherevariablesareperfectlycorrelated.
5.
1DependentVariablesThedataforalldependentvariablesintheanalysiswascollectedbytheauthoroveraperiodofseveralmonthsinlate-2010andearly-2011.
Afterthisperiodcheckswereregularlymadetomonitorforadditionstothedataor,whereinformationhadbeenmissing,newdata.
Collectionofthedatarequiredamethodicalreviewofpressreleases,industrynews,andlocalnewssources;intotalaround300individualpiecesofevidencewereconsidered.
Datacollectionwasfocussedinanumberofareas;whethertheiPhonewaslaunchedinagivencountry,thedateoflaunchinthatcountry,whetheritwassoldexclusivelythroughasinglenetworkoperator,andthedateatwhichanyexclusivityexpired.
Inadditiontheidentityofthermwhichgainedexclusivitywasrecordedin20Table1:VariablesummaryVariableSourcenMeanStd.
DevMin.
Max.
iPhonelaunched(1=yes)Author'sresearch1870.
4490.
49901Exclusivity(1=yes)Author'sresearch840.
7260.
44901Duration(days)Author'sresearch84695.
619614.
76401836LogLandAreaWB18711.
2652.
6923.
33216.
611LogPop.
DensityWB1874.
231.
535-1.
9869.
805LogGDP/CapitaWB1878.
4411.
5685.
22211.
839AfricaGoogle1870.
2780.
44901NorthAmericaGoogle1870.
1440.
35201SouthAmericaGoogle1870.
0590.
23601AsiaGoogle1870.
2250.
41801Aus&OceaniaGoogle1870.
0590.
23601EuropeGoogle1870.
2350.
42501Num.
FirmsWCIS1873.
5031.
938114HHIWCIS1870.
4770.
1980.
1691CompetingStandardsWCIS1870.
1180.
32301Mobilesper100ITU18782.
15644.
9381.
02232.
068VendorPositionWCIS611.
7380.
85415DelayfromlaunchAuthor'sresearch84461.
167153.
690900ordertobeabletomatchupthecollectediPhonedatawithotherdatasets.
Theelicitedvariablesandtheprocessofcollectionisoutlinedbelow:5.
1.
1IdentityofiPhoneCountriesDataonthecountriesinwhichtheiPhonewasdistributedisthedependentvariableusedwhenexaminingwhatmotivatesAppletosellintoparticularcountries.
Thede-cisiontoselltheiPhoneinacountryisrepresentedasabinaryvariablewithavalueof0or1foreachcountryinthesample.
ToidentifythosecountriesinwhichiPhonewasdistributedtheocialApplewebsitewasconsulted.
ThewebsitefeaturesaspecicsetofpagesforeachcountrywhereAppleproductsarelocallysold.
IftheiPhonewasavailableinacountrythenthesepageswouldeitherallowthepurchaseofthephone(withanaccompanyingairtimecontract)oroeralinktothedistributorpartner'sweb-site(orwebsiteswherethereweremultipledistributingoperators).
IftheiPhonewas21notociallyavailableinacountrythentheoptiontopurchasethephonewouldnotbedisplayed.
Asof2013,intotal109countrieswereidentiedinwhichtheiPhoneislaunchedthroughatleastoneocialnetworkpartner.
Ofthesecountriesasubsetof84wasisolatedwheretheiPhonewasbothlaunchedbefore2010(theperiodcoveredbythesupplementaryo-the-shelfdatasets)andwherethecountriesalsohavecompleteinfor-mationintheo-the-shelfdatasetsused.
5.
1.
2iPhoneExclusivityWhethertheiPhonewaslaunchedexclusivelywithasinglenetworkoperatorformsthebinaryrsthurdleinthetwostagemodelinvestigatingexclusivityanditsduration.
PressreleaseswerefavouredastheprimarysourceofevidenceinordertodeterminewhethertheiPhonewaslaunchedexclusivelyinacountry.
ThesepressreleaseswereprimarilyissuedbyApple,butsometimesfromtherelevantnetworkoperator.
Inaverysmallnumberofcasesitwasnecessarytorelyuponnationalorindustrypress.
Itwasfoundthatin23ofthe84countriesintheusablesampletheiPhonewaslaunchedcom-petitivelywithmorethanoneocialnetworkpartner.
Intheremaining61countriestheiPhonewaslaunchedexclusively.
5.
1.
3DurationofExclusivityThedurationofanyexclusivityisthetruncatedcontinuousvariableusedasthesecondhurdleinthetwo-stageexclusivity-durationmodel.
Inordertoidentifythedurationofanyexclusivityitwasnecessarytodetermine,rst,thelaunchdateoftheiPhoneintherelevantcountries,and,secondly,thedatethatthenextsubsequentvendorstartedsellingthephone.
Again,pressreleaseswereemployed;theinitiallaunchdatewasinalmostallcasesincludedineitherApple'sorthenetworkoperator'spressrelease.
Forthedatesofsecondandlatervendorstherewasagreaterrelianceonpressreleasesfromtheoperatorthemselves.
Finally,asbefore,insomeisolatedcasesitwasnecessarytoconsultlocalpressorindustrynewssourcestolearnlaunchdates.
Usinginformationondates,thedurationvariablewascalculatedasthenumberofdaysbetweentheinitiallaunchbyanexclusiveoperator,andthedateatwhichasubsequentoperatorstartedsellingtheiPhone.
InthecasethattheiPhonewaslaunchedinitiallywithmultiplenetworkoperators(competitively)thenthedurationvariableisequaltozero.
Whereexclusivityisongoing(thereisstillanexclusivenetworkoperator)thedura-tionofexclusivitywastakenasthedaysbetweenlaunchandSeptember2013.
While22thisrepresentedanarticialcensoringofthedataitdidnothaveaprobabilitymass(asatzero)duetotheheterogeneityinlaunchdates.
ToensurerobustnessoftheresultsaseconduppercensoringwastestedwhichspeciedthedurationforongoingexclusivityasthedaysbetweenlaunchandSeptember2014(anextra365days);therewerenosignicantdierencesintheresults.
Fromtheaboveinvestigations,theidentityofthelaunchoperator(s)wasisolated;thisenabledthematchingoftheiPhone-specicdatawiththeo-the-shelfdatasetsper-tainingtorm-leveltelecommunicationsdata.
5.
2IndependentVariablesInordertoanalysethedeterminantsoftheexclusivityanddurationthreeo-the-shelfdatasetswereused:WorldBank(WB)dataonlandcharacteristicsanddemographics;theWorldCellularInformationService(WCIS)datasetonrm-levelmobiletelecom-munications;andInternationalTelecommunicationsUnion(ITU)dataoncountry-leveltelecommunicationsindicators.
WherenecessarytheinternetsearchengineGooglewasalsoconsultedforconsistencyindeterminingthegeographiclocationofsomecountries.
Table1speciesthesourceoftherespectivevariables.
Becausevariablesneededtobecompleteforallcountries,thisinuencedthespecicchoiceofvariablebut,eventakingthemostcompletevariablesavailable,12thenumberofcountriesincludedinthestudywasreducedfromallcountriesintheworldtoawork-ablesampleof187.
5.
2.
1DemographicControlVariablesInordertocontrolfortheunderlyingeconomicandphysicalcharacteristicsofthecoun-triesinthestudyarangeofvariablesareincluded.
CountrySizeandPopulationDensityInordertocontrolforanyeectrelatedtocountrysizeandthepopulation,bothlandareaandpopulationdensityareincluded.
Landareaandpopulationdensityarelikelytoaectthecostandreturnsoninvestmentofimplementingorupgradingamo-bilenetwork.
Boththesestatisticsareincludedinlog-formtocorrectforaleft-sidedbias.
CountryperCapitaGDP12Formostofthethemestobeincludedtherewereseveralpossiblevariableswhichcouldhaveactedasindicator.
23ThepercapitaGDPofacountrycanbeconsideredtobeaproxyfordemandandthusaectsthelikelihoodofarmcarryingoutbusinessinthatcountry.
CountryGDPisincludedintheformofGDPpercapitaincurrentUSdollars.
Per-capitaGDPhasalongright-tail,assuchthenaturallogofper-capitaGDPisincludedintheanalysis.
ContinentDummyVariablesInordertocaptureanyregion-specicvariationinpolicy,ordemandcharacteristics,dummyvariablesareincludedindicatinginwhichcontinentacountryislocated.
Cen-tralAmericancountriesareconsideredaspartofNorthAmerica,whilemanyPacicIslandsareconsideredtobepartofOceaniawithAustralia.
Intheinterestsofconsis-tencywheretherewasambiguitythesearchengineGooglewasconsultedandthemostprevalentanswerfromthesearchresultsincluded.
5.
2.
2MarketstructureInadditiontobroaddemographiceectswhicharelikelytoaectthemutualdesirabilityofAppleofnetworkoperatorstoselltheiPhone,thecharacteristicsandcompetitive-nessoftheindividualmarketsispredictedtodeterminethelikelihoodanddurationofanyexclusivityarrangements.
Market-leveldataisprimarilysourcedfromtheITUandWCISdatasets.
NumberofFirmsandHerndahl–HirschmanIndex(HHI)ForeachcountrydataconcerningthenumberofactivelicensedoperatorsandtheHHIisincludedfortheyearandquarterwhentheiPhonewaslaunched.
IftheiPhonewasnotsoldinacountryduringthestudyperiod,theHHIandrmnumbersinfor-mationisincludedforthenalquarterof2009.
Giventhatmanycountriessincetheintroductionofmobiletelecommunicationshaveexperiencedsignicantconsolidationinthenumberofmobileoperators,anoperatorisconsideredtobeactiveiftheyhaveapositivenumberofsubscriptions.
Foreachcountry(i)theHHIiscalculatedusingeachactiveoperators'shareofthetotalsubscriptionsinthatcountry,denotedsi,jwherejrepresentstheoperator:HHIi=Njsi,jThelevelofcompetitionwithinacountryislikelytoaectthevaluetoarmofsellingtheiPhone.
Ifamarketisconcentratedthenthelackofcompetitionmaymeanthatadominantrmcanearngreaterprotsfromsellingnon-iPhonehandsetsrelative24tormsinhighlycompetitivemarkets.
FornetworkoperatorsincompetitivemarketstheiPhonemayactasakeycompetitivedierentiator-makingitrelativelymorevalu-abletothesermsrelativetormsfacinglittlecompetition.
5.
2.
3TechnologyVariablesareincludedwhichindicateboththetypeoftechnologywhichisbeingusedineachcountryandalsothepenetrationofmobiletechnology.
CompetingStandardsInsomecountriesthereareanumberoftechnologicalstandardsthatareusedtotransmitmobiletelecommunicationsignals(themaintwobeingGSMandCDMA).
Thesestandardsrendersomemobilehandsetsincompatiblewithsomenetworks.
Thepresenceofcompetingstandardshasbeensuggestedtobeasourceofswitchingcostsforsubscribers(inadditiontothosewhichwouldnormallyexistwhenswitchingnetworksbaseduponthesametechnology).
TheiPhonewasinitiallyincompatiblewithCDMAnetworksandwasonlylaunchedwithCDMAcompatibilityinFebruary2011.
GiventhetimelimitationsofthedatasetthismeansthatthestudyislimitedonlytothelaunchoftheGSMmodel.
Adummyindicatorisincludedequalto1(zerootherwise)ifthereismorethanonecompatiblestandardinoperationwithinacountry.
Becauseinmanycountriesthereareanumberoflegacystandardsfromtheearlystagesofmobilephonedevelopmentwithaverysmallnumberofsubscribers,thecompetingstandardsdummyconsidersasinglestandardcountrytobeonewherethedominantstandardhasover90%ofsubscriptions.
Whenconsideringasinglestandardthecompetingstandardsvariableconsiderstheen-tirefamilyofthatstandard;forexample,whenconsideringGSMsubscriptions,bothGSMandthesubsequentW-CDMAstandardareincluded.
MobilePenetrationInordertocapturetheuseofmobilephonesineachcountryavariableisincludedwhichindicatesthenumberofmobilesubscriptionsper100inhabitants.
Thisvariableindicateshowwidespreadtheuseofmobilephonesisand,thus,thepotentialsizeofthemarket.
5.
2.
4OtherVariablesTwovariablesareincludedwhichareusedonlyinthetwo-stagemodelinvestigatingexclusivityanddurationtoprovidefurtherinformationabouttheiPhonelaunchand25thermwhichisallocatedexclusivity.
VendorPositionVendorpositionisusedtoinvestigatewhetherthemarketpositionofthermwhichisallocatedexclusivitydeterminesthedurationoftheexclusivityagreement.
Necessarilyitisonlyobservedinthecasethatanexclusivitystrategywaspursued.
TimeThevariable'Delayfromlaunch'isincludedwhichrepresentsthenumberofdaysbetweentheinitiallaunchoftheiPhoneintheUSin2007,andthelaunchoftheiPhoneineachcountry.
Thistimeaspectisdesignedtocapturetheeectofdemanduncertainty(commontomanyofthetheoreticalmodelsofsection3.
1)diminishingovertime.
ItisnecessarilyincludedonlyforthosecountrieswheretheiPhonewaslaunched.
6ResultsandDiscussionTheresultsfromtheinvestigationindicatethat,assuggested,therearedistinctionsbe-tweenthemechanismswhichdetermineimpositionanddurationofexclusivityrestraints.
Furthermore,inexaminingthedurationoftheagreementsthereissignicantevidencethatboththelevelofcompetitionandthepresenceofcompetingstandardshaveapos-itiverelationshipwiththedurationofexclusivity.
6.
1PopulationEectsThedecisiontoinvestigatetwosamples-separatedbypopulation-isdrivenbythepredictionthattheoperationofmobiletelecommunicationsissubjecttodierentcon-straintsinsmallercountriesvis-a-vislargerones.
Theseconstraintsarelikelytobedierentdistributionsofincome,dierentchallengesintermsofnetworkinfrastructure,anddierencesinthenatureorstrengthofcompetition.
Inordertotestthispredictionthesampleofcountriesissplitintosubsetsof'small'(accordingtoanupperpopulationlimitof499,999forasmallcountry)andothercountries.
Foreachofthesubsetsofcountriesthesetofcontinuousvariablesanalysedinthemodels(whichisassumedtobeaproxyofthedierentconstraintsrmsface)iscompared.
ThisissubjectedtoaMann-Whitneytesttodeterminewhethertheyaredrawnfromthesamedistribution.
Table2indicatesthatthemeanvaluesforthevariablesaredierentbetweenthesmallerandothercountries'continuousvariables.
TheMann-WhitneyU-valueforeachofthevariablesisalsosignicantata1%levelinthecaseoflandsize(smallerpopula-tioncountriesarealsosmaller),GDPpercapita(smallercountriesarericheronaper26Table2:Mann-WhitneytestofpopulationrestrictionVariableMeanSmallMeanOthersMWUP>|U|LogLandSize7.
03812.
017.
4420.
000***LogPop.
Density4.
6564.
155-2.
4120.
016**LogGDPperCapita9.
2698.
295-2.
9120.
004***Mobilesper10094.
09980.
052-1.
730.
084*Num.
Firms2.
0363.
7615.
5560.
000***hlinecapitabasis),andthenumberofcompetingnetworkoperators(lessinsmallercountries).
Theseresultssupporttheadditionalanalysisofarestrictedsample.
Forrobustnessdif-ferentpopulationthresholdswerealsoconsidered,butultimatelyrejectedbaseduponabalanceofMann-Whitneyscoringandintuitiononwhatthepopulationcensoringisdesignedtoachieve.
6.
2EntryDecisionAProbitmodelisutilisedtoexaminethebinarynatureofApple'sdecisiontoselltheiriPhoneinparticularcountriesandistestedusingafullandarestrictedsample.
Theresultsareshownintable3whilethemarginaleectsforspecications(D)and(H)areshownintable4.
Table3indicatesthatunderboththefullandrestrictedsample,irrespectiveofmodelspecication,thecoecientsassociatedwithrawdemographicvariables(α1,α2,andα3)aresignicantatthe1%level.
TheprobabilityoftheiPhonebeingsoldwashigherinthoseterritorieswhichwerebigger,moredenselypopulated,andricherintermsofpercapitaGDP.
GDPwasincludedasaproxyfortheoveralllevelofdemandwithinacountryandhastheexpectedsign;countrieswithahigherlevelofdemandarelikelytobemoreprotable,bothdownstreamandconsequentlyupstream.
Higherpopulationdensitymeansthatinfrastructurecostperuserislower,howeverthelinkbetweenthisandtheeconomicrationalefortheintroductionoftheiPhoneisnotclear.
Wherethexedcostofinfrastructureperuserislowerthiswouldreducethecostassociatedwithnetworkinfrastructureupgrades(ifneeded).
Howeverthelikelyexplanationisthatthesenetworksaresimplybetterdeveloped.
Landsize,havingcontrolledforpopulationdensityandGDP,doesnotimmediatelylenditselftoatheorysupportingthesaleoftheiPhone.
Landsizeisnegativelycorre-latedwithbothGDPpercapitaandpopulationdensity27Table3:ProbitdeterminantsofentryAvailabilityofiPhoneIndependentFullsampleRestrictedsamplevariablesCoecient(A)(B)(C)(D)(E)(F)(G)(H)LogLandAreaα10.
426***0.
509***0.
537***0.
559***0.
337***0.
428***0.
434***0.
455***(0.
068)(0.
081)(0.
096)(0.
1)(0.
085)(0.
094)(0.
113)(0.
119)LogPop.
Densityα20.
464***0.
645***0.
672***0.
692***0.
369***0.
549***0.
558***0.
57***(0.
1)(0.
123)(0.
142)(0.
149)(0.
11)(0.
132)(0.
153)(0.
16)LogGDP/Capitaα30.
603***0.
611***0.
601***0.
631***0.
593***0.
612***0.
605***0.
583***(0.
089)(0.
113)(0.
116)(0.
156)(0.
091)(0.
116)(0.
119)(0.
167)Africaα4-0.
263-0.
164-0.
018-0.
1140.
0290.
18(0.
392)(0.
407)(0.
423)(0.
403)(0.
424)(0.
438)NorthAmericaα5-0.
130.
0440.
2790.
4570.
6740.
932(0.
41)(0.
43)(0.
459)(0.
495)(0.
527)(0.
56)SouthAmericaα60.
5780.
5170.
7120.
6390.
6240.
817(0.
603)(0.
603)(0.
635)(0.
594)(0.
596)(0.
626)Asiaα7-1.
188***-1.
144***-1.
059***-0.
974**-0.
92**-0.
816**(0.
376)(0.
385)(0.
397)(0.
382)(0.
391)(0.
404)Aus&Oceaniaα8-0.
1830.
1020.
423-0.
0760.
3020.
631(0.
879)(1.
084)(1.
04)(0.
943)(1.
199)(1.
151)Europeα9(omitted)(omitted)(omitted)(omitted)(omitted)(omitted)Num.
Firmsα10-0.
168-0.
144-0.
137-0.
115(0.
11)(0.
114)(0.
112)(0.
116)HHIα11-2.
408**-2.
4*-2.
509**-2.
387*(1.
204)(1.
245)(1.
263)(1.
315)CompetingStandardsα12-0.
634-0.
571(0.
486)(0.
499)Mobilesper100α130.
0010.
003(0.
005)(0.
005)Constantα0-12.
045***-13.
456***-12.
192***-12.
983***-10.
46***-12.
232***-10.
749***-11.
267***(1.
586)(2.
023)(2.
195)(2.
391)(1.
741)(2.
139)(2.
339)(2.
497)n187187187187159159159159LogLikelihood-83.
055-75.
125-72.
888-70.
729-76.
662-69.
215-67.
04-64.
679AIC174.
111168.
25167.
776167.
458161.
323156.
43156.
081155.
357Standarderrorsinparentheses*(p1),**(p1)28Intermsofgeographiccontrols,countriesinAsiawerelesslikelytoselltheiPhone,relativetoEurope–thegeographicbasevariable.
Thisconrmedthepatternswhichwereobservedinthedata.
Intermsoftelecom-specicvariables,α11-thecoecientassociatedwithHHIissignicantateither5or10%levelin,respectively,specicationsCandD(andalsoGandH).
TheinclusionofavariablerepresentingthenumberofrmsmeansthatHHIisinterpretedasameasurepurelyoftheinequalityofrmmarketsshares.
Persistentinequalityinrmmarketsharescanbeperceivedaswheremarketsarelesscompetitive.
Thislinksthenotionofmarketcompetitivenesswithrms'likelihood(orlackof)ofsellingtheiPhone.
TherationaleforthisoutcomemaylieintheiPhoneasacompetitivetool;networkshopetoattractcustomersbysellingtheiPhone.
IfthisisnotthecasethennetworksmayfeellittleneedtoselltheiPhone.
OnthebasisthateverysaleofaniPhonehasanopportunitycostintermsofanothervarietyhandsetthatcouldbesold.
Ifcompetitionisnotstrong,andtheunitprotfromaniPhoneislessthanotheravailablehandsets,thentheincentivestoselltheiPhonewillbeweak.
Noneoftheothertelecommunications-speciccoecientsaresignicantat10%orhigherinanyofthespecications.
Giventheinterpretationofthevariable'Mobilesper100'asanindicatorofthedevelopmentandpenetrationofthemobilemarket,thissuggeststhatthishadlittlebearingonthesaleoftheiPhoneinagivencountry.
Result1ThedeterminantsofthesaleoftheiPhoneinagivencountryisdominatedbybasicdemographicfactorsoflandsize,populationdensity,andGDPpercapita.
Ofthetelecomspecicvariablesincluded,onlythemeasureofinequalityofrmsizesissignicant;indicatingthattheiPhonewaslesslikelytobesoldinconcentratedmarkets.
6.
2.
1EntryProbitGoodnessofFitIncomparingthemodelspecications,becausethelog-likelihoodisnon-decreasinginthenumberofvariablesadded,Akaike'sInformationCriterion(AIC)isused.
Thismod-eratesthelog-likelihoodbaseduponthenumberofvariableswhichareaddedtoavoidspuriousconclusionsfromproliferationofexplanatoryitems.
TheAICstatistic,wheresmallerisbetter,indicatesthatthemodelimproveswiththeadditionoftheextravari-ables,evenwhereindividuallythevariablesareinsignicant.
ThesmallerAICresultsindicatethatunderbothrestrictedandunrestrictedsam-plesthefullmodels((D)and(H))providethebestexplanationofthedata.
Althoughthetelecom-specicvariablesareoflimitedsignicance,themodelperformsbetterwiththeirinclusion.
296.
2.
2EntryProbitMarginalEectsThemarginaleectsforaunitchangeintheincludedvariablesontheprobabilityareincludedintable4.
Table4:EntryProbitMarginalEectsFullsampleRestrictedsampleMarginaleect(D)(H)LogLandArea0.
117***0.
103***(0.
015)(0.
023)LogPop.
Density0.
145***0.
129***(0.
025)(0.
032)LogGDP/Capita0.
132***0.
132***(0.
028)(0.
033)Africa-0.
0040.
041(0.
089)(0.
099)NorthAmerica0.
0580.
21*(0.
096)(0.
124)SouthAmerica0.
1490.
184(0.
131)(0.
139)Asia-0.
221***-0.
184**(0.
078)(0.
088)Aus&Oceania0.
0880.
142(0.
217)(0.
259)Europe(omitted)(omitted)Num.
Firms-0.
03-0.
026(0.
023)(0.
026)HHI-0.
502**-0.
539*(0.
251)(0.
287)CompetingStandards-0.
133-0.
129(0.
1)(0.
111)Mobilesper10000.
001(0.
001)(0.
001)Standarderrorsinparentheses*(p1),**(p1)Thesignicanceofthemarginaleectsnaturallymirrorstheresultsinthetable3.
Theyshowthatanincreaseof20%intheHHIofacountry(i.
e.
from0.
2to0.
4)wouldreducetheprobabilityoftheiPhonebeingsoldbyaround10%.
UnitincreasesinthelogdemographicvariablesindicateanincreasedlikelihoodofthesaleoftheiPhonebybetween(approximately)10and15%.
Onemustbewaryofinterpretingtheseresultstooliterally.
Thedemographicvariablesaretransformedandassuchaunitincrease30doesnotimplyalinearincreaseintheirrespectiveunderlyingvalue.
6.
3ExclusivityanddurationThestudyofexclusivityisformedoftwoparts:evidencethatthedeterminantsoftheimpositionofexclusivedistributionagreementsdierfromthosedeterminingitsdura-tion;andthestudyofthedriversofbothstagesexaminedwithrespecttothetheoreticalpredictions.
Modellingexclusivityanddurationasatwodistinctcomponentsinthesameprocessrequiresadierentapproachtostandardregressionmodels.
Thus,inthisinvestigationtwo-stagemodelsareused;theTobitmodel,andCragg'sTobitalternative.
Twostagemodelsallowtheanalysisoftruncatedorcensoredvariables.
Specically,whereavari-ableiscensoredatzeroatwo-stagemodelallowsanalysisoftheconditionsrequiredbreachthezerothreshold,andthesubsequentanalysisofdeterminantsofthenalmag-nitudeofthevariable.
6.
3.
1TheTobitapproachOneofthecoreaimsofthepaperistodeterminewhetherthedurationofanexclusivityagreementisdeterminedbythesamevariableswhichdetermineitsimposition.
ThecoreassumptionoftheTobitmodelisthatasingleprocessgovernsboththebreachofthethresholdandthenalmagnitudeofthedependentvariable.
BytestingtheTobitassumptionsusingtheLinandSchmidt[1984]LagrangeMultipliertestitcanbedeter-minedwhetherseparateprocessesareinactioninexclusivity.
TheresultsfromtheTobitestimationareshownintable5wheremodel(I)representstheunrestrictedsampleofcountriesand(J)thelimitedsample.
TheLikelihoodRatiotestscoresarealsoincludedforeachspecication,measuredagainstthenullthatthedependentvariableshavenoexplanatorypower.
ThendingsfromtheTobitmodelcanonlybeacceptedinthecasethattheTobitisthecorrectspecicationandthusthatasingleprocessgovernsbothstagesofthemodel.
Inthepresentinvestigationthiswouldimplythatexclusivityanditsdurationaregovernedbythesamemechanism.
InordertotestthisassumptiontheLagrangeMultiplier(LM)testofLinandSchmidt[1984]isused.
Theresultsareshowninintable6whichcomparestheLMtestscoreagainstthenullthattheTobitiscorrectlyspecied.
TheLMtestscoresforbothrestrictedandunrestrictedmodelaresignicantatgreaterthan1%level.
ThismeansthatwerejectthenullhypothesisthattheTobitiscorrectlyspeciedandinsteadassumethattheresultsofthetwohurdles(exclusivityandduration)andgovernedbyseparateprocesses.
31Table5:TobitexclusivitymodelDurationofexclusivityIndependentFullsampleRestrictedsamplevariablesCoecient(I)(J)LogLandAreaβ1-145.
435***-219.
833***(50.
962)(56.
823)LogPop.
Densityβ2-206.
696**-278.
421***(81.
301)(82.
879)LogGDP/Capitaβ3-166.
968**-103.
387(81.
621)(81.
813)Africaβ4478.
504425.
414(289.
752)(283.
316)NorthAmericaβ5-460.
381*-551.
255**(264.
564)(261.
742)SouthAmericaβ6-1127.
745***-1103.
342***(311.
925)(300.
845)Asiaβ773.
9129.
327(249.
364)(243.
808)Aus&Oceaniaβ8-553.
277-645.
667(525.
507)(509.
215)Europeβ9(omitted)(omitted)Num.
Firmsβ1035.
11555.
039(53.
93)(52.
687)HHIβ111623.
16*1631.
735*(870.
271)(851.
644)CompetingStandardsβ12595.
601**578.
71**(272.
922)(263.
654)Mobilesper100β130.
791-1.
942(2.
776)(2.
901)LaunchDelayβ14-0.
397-0.
338(0.
575)(0.
555)Constantβ04113.
576***4947.
083***(1403.
69)(1396.
699)σ573.
745***552.
837***(55.
418)(54.
2)n8481LogLikelihood-493.
347-474.
317LR46.
3650.
55P(LR)>χ2(13)00Standarderrorsinparentheses*(p1),**(p1)Result2Theimpositionandthedurationofexclusivedistributionagreementsaregov-ernedbydierentprocesses.
Wheretheconditionsforimpositionarestrongthisdoesnotimplythattheresultingexclusivedistributionagreementwillbedurable,andviceversa.
BecausetheTobitspecicationisrejectedtheanalysisoftheresultsissuitablybrief.
32Table6:TobitLagrangeMultiplierTestSampleLMScore10%5%1%Full(I)21.
4744.
416.
11810.
189Restricted(J)22.
494.
2686.
37411.
681Intable5thereareanumberofvariableswhicharefoundtobesignicant.
Theco-ecientsassociatedwiththekeydemographiccontrolvariablesareallnegative,thoughGDPpercapitaisonlysignicantinthenon-restrictedsampleofcountries(ata10%level).
Thenon-separabilityofexclusivityanddurationdecisionintheTobitimpliesthatexpecteddurationisshortergivenlargeranddenselypopulatedcountries.
ThesameconclusioncanbedrawnofbothcountriesinNorthandSouthAmericarelativetoEurope(thebasecategoryforgeographicallocation).
Bothcoecientsformarketconcentration(HHI)andthepresenceofcompetingstan-dardsarepositiveandsignicant(at10%and5%levelsrespectively).
Theseresultsarerepresentativeofthemarketstructuresinthepresenceofeitherheavyconcentrationorwheretherearemultipletechnologicalstandardsinoperation.
Onthebasisofbothloglikelihoodandlikelihoodratiotestscorestherestrictedsam-plemodel,asintherestoftheanalyses,isabettertrelativetothatforthefullsample.
6.
3.
2Cragg'sTobitalternativeTherejectionoftheTobitspecicationinResult2motivatestheuseofCragg'sTobitalternativewhichallowsdierentprocessestogoverneachstageoftheanalysis.
LiketheTobit,Cragg'smodelstillcontrolsforthetruncatednatureofthedurationvariable.
Table7showstheresultsforCragg'sTobitalternativeappliedtotheexclusivityanddurationdata.
ThetwohurdlesareestimatedseparatelyusingrstaProbitapproach,andsecondalinearregressionofatruncatednormalvariable,calculatedusingmaximumlikelihood.
Forthefullsampleandtherestrictedsampleofcountriesthreespecicationsaretested:ademographic-onlymodel;afullmodelusingallvariables;andahybridwithtelecommunicationvariablesinthesecondstage.
ForeachofthespecicationsAkaike'sInformationCriterionisreportedinadditiontotheoverallloglikelihood(LL).
Thelikelihoodratiotestscoresfortherst-stageProbitarealsoreported.
Thevariableσistheestimateofthevarianceofthesecond-stagecensoredregression.
33Table7:ResultsfromCragg'sTobitalternativeExclusivityanddurationhurdleIndependentFullsampleRestrictedsamplevariablesCoecient(I)(J)(K)(L)(M)(N)Tier1=ExclusivityLogLandAreaγ1-0.
172*-0.
172*-0.
15-0.
311**-0.
311**-0.
421**(0.
103)(0.
103)(0.
224)(0.
132)(0.
132)(0.
182)LogPop.
Densityγ2-0.
438**-0.
438**-0.
422*-0.
664**-0.
664**-0.
777**(0.
211)(0.
211)(0.
127)(0.
26)(0.
26)(0.
31)LogGDP/Capitaγ30.
0520.
0520.
110.
1530.
1530.
329(0.
176)(0.
176)(0.
236)(0.
194)(0.
194)(0.
26)Africaγ40.
8740.
8740.
8721.
1341.
1340.
731(0.
778)(0.
778)(0.
909)(0.
833)(0.
833)(0.
927)NorthAmericaγ50.
0340.
034-0.
3630.
0710.
071-0.
583(0.
529)(0.
529)(0.
658)(0.
553)(0.
553)(0.
721)SouthAmericaγ6-1.
613***-1.
613***-2.
009***-1.
653***-1.
653***-2.
096***(0.
6)(0.
6)(0.
68)(0.
622)(0.
622)(0.
716)Asiaγ70.
8840.
8840.
8491.
233*1.
233*1.
21*(0.
597)(0.
597)(0.
687)(0.
677)(0.
677)(0.
734)Aus&Oceaniaγ8-1.
408-1.
408-2.
069*-1.
827*-1.
827*-2.
612**(1.
007)(1.
007)(1.
232)(1.
066)(1.
066)(1.
283)Europeγ9(omitted)(omitted)(omitted)(omitted)(omitted)(omitted)Num.
Firmsγ10-0.
0230.
022(0.
129)(0.
138)HHIγ113.
4992.
495(2.
642)(2.
796)Comp.
Standardsγ120.
8580.
755(0.
659)(0.
649)Mobilesper100γ130.
001-0.
011(0.
008)(0.
009)LaunchDelayγ1400(0.
002)(0.
002)Constant1γ04.
1374.
1371.
9165.
88*5.
88*6.
148(2.
811)(2.
811)(4.
08)(3.
098)(3.
098)(4.
711)Tier2=DurationLogLandAreaη1-76.
987**-119.
593***-119.
593***-92.
806**-141.
301***-141.
301***(36.
946)(40.
616)(40.
616)(37.
575)(41.
359)(41.
359)LogPop.
Densityη2-72.
171-101.
311*-101.
311*-83.
284-115.
459**-115.
459**(60.
294)(60.
82)(60.
82)(58.
909)(58.
963)(58.
963)LogGDP/Capitaη3-164.
834***-251.
131***-251.
131***-149.
987***-225.
75***-225.
75***(51.
201)(62.
828)(62.
828)(50.
402)(60.
67)(60.
67)Africaη4374.
605*362.
06*362.
06*388.
781*331.
024*331.
024*(203.
497)(193.
699)(193.
699)(199.
185)(190.
646)(190.
646)NorthAmericaη56.
195-393.
143*-393.
143*-12.
165-464.
744**-464.
744**(177.
525)(219.
451)(219.
451)(173.
141)(215.
427)(215.
427)SouthAmericaη632.
453-132.
501-132.
50127.
818-171.
443-171.
443(309.
551)(305.
363)(305.
363)(301.
408)(295.
451)(295.
451)Asiaη7-72.
642-195.
048-195.
048-84.
968-215.
177-215.
177(168.
507)(181.
683)(181.
683)(164.
431)(176.
102)(176.
102)Aus&Oceaniaη8478.
191-21.
862-21.
862431.
281-112.
918-112.
918(410.
114)(440.
621)(440.
621)(400.
611)(428.
996)(428.
996)Europeη9(omitted)(omitted)(omitted)(omitted)(omitted)(omitted)Num.
Firmsη1094.
47**94.
47**98.
852**98.
852**(42.
03)(42.
03)(40.
711)(40.
711)HHIη111077.
373*1077.
373*1281.
305**1281.
305**(637.
755)(637.
755)(626.
579)(626.
579)Comp.
Standardsη12508.
144**508.
144**530.
46**530.
46**(231.
921)(231.
921)(223.
771)(223.
771)Mobilesper100η132.
4432.
4431.
6841.
684(2.
15)(2.
15)(2.
144)(2.
144)Delayfromlaunchη14-0.
239-0.
239-0.
269-0.
269(0.
42)(0.
42)(0.
406)(0.
406)VendorPositionη15-88.
503-88.
503-65.
115-65.
115(64.
626)(64.
626)(64.
134)(64.
134)Constant2η03602.
241***4281.
804***4281.
804***3724.
546***4358.
935***4358.
935***(851.
357)(1016.
889)(1016.
889)(842.
981)(988.
622)(988.
622)σ391.
029***348.
274***348.
274***382.
296***337.
668***337.
668***(41.
749)(35.
981)(35.
981)(40.
813)(34.
884)(34.
884)n848484818181LL-484.
995-478.
225-475.
627-465.
661-458.
762-455.
986LR19.
3619.
3624.
5523.
5123.
5129.
06P(LR)>χ20.
0130.
0130.
0260.
0030.
0030.
006AIC1007.
9911006.
451011.
253969.
322967.
52971.
972Standarderrorsinparentheses*(p1),**(p1)34BeforeanalysingtheresultsofCragg'sTobitalternativeindepth,thereareanum-berofobservationswhichcanbemadeconcerningthemodelspecication.
First,therestrictedsamplemodelsareabettertofthedatathantheunrestricted(asmeasuredbytheabsolutevalueoftheloglikelihoodandAIC)-thisagainvindicatesrestrictingthesample.
Second,theLikelihoodRatioscoresindicatethatthedemographic-onlyspec-icationsfortherststagearemoreeectivethanthefullmodelunderbothfullandrestrictedsamples.
Third,associatedwiththedierencesintheLRtestscoresfortherst-stage;theAICstatisticindicatesthatspecications(J)and(M)provideabettertthantheircounterpartsforthetwosamples.
GiventhatAkaike'sInformationCriterionpenalisestheinclusionofirrelevantvariables,itcanbeconcludedthattheadditionoftelecom-specicvariablesintherststagedonotaddtotheexplanatorypowerofthemodel.
Thefollowinganalysisofthevariablesfocussesontherestrictedsampleresults,thoughconsiderationisgiventothefullsamplewhereappropriate.
Intherststagethedeterminationofanexclusivedistributionagreementthevari-ablesassociatedwithbothlandarea(γ1)andpopulationdensity(γ2)arenegativeandsignicantinallspecications.
Thistswiththestylisedfactsfromtheobserveddatawherelargecountriesoftenhaveregionaltelecommunicationnetworks-suchthatexclu-sivitydealsmaybemadeonaregionalbasis.
Secondly,thistswiththenotionthatthereislessscopeforfree-ridingonpromotionaleort,asperMathewsonandWin-ter[1984],inlargesparselypopulatedcountries.
Thisisareasonableassumptionsincespilloversininformationwillbemorepronouncedwherepopulationdensityishighest.
ThepersistenceofthesignicanceofsomegeographicalidentiersacrossspecicationsconrmstrendswhichwerepresentinthedatasuchasthetendencyofnotenteringintoexclusivedistributioninSouthAmerica.
Notelecommunications-specicvariablesprovedtohaveastatisticallysignicantim-pactontheprobabilityoftheintroductionofanexclusivedistributionagreement.
Inthetheoreticalliterature[ReyandTirole,1986]suggeststhatdemandorcostuncertaintywillinuencethechoiceofexclusivity.
Giventheuseof'LaunchDelay'asaproxyforthepassageoftime,overwhichuncertaintywouldberesolved,itwasexpectedthatthiswouldhavesignicance.
Thenear-zeroresultonthetimefromlaunch(γ14)emphaticallydeniesthisprediction.
Theresultspertainingtothedurationofanexclusivityagreementpresentanumberofvariableswhicharesignicant;bothdemographicandtelecom-related.
Thepaucityofliteratureconcerningthedeterminantsofdurationinexclusivityagreementslimitsthescopefora-prioriexpectationsregardingtheandmagnitudeorsignofthecoecients.
Theresultsarediscussedbelow.
Evenforcontrollingforregionaldierences,whicharelikelytobehighlycorrelated35withsomedemographicmeasures,allkeydemographicindicatorsarenegativeandsig-nicantat5%orbetter.
Comparativelylarger,morepopulous,andwealthiercountriesarelikelytohaveshorterdurationexclusivityrestraints.
OneissuethatApplemayhavefacedinitsexclusivecontractingarrangementsisthatrelatingtoswitchingcostsorbarrierstoswitching-issuesthatareparticularlyprevalentasaresultofthexed-termnatureofmobilecontracts.
BycontractingwithasinglenetworkoperatorApplemayhavelimiteditspotentialmarkettoonlythosecurrentlysubscribedtothatoperatorandthosewillingtoswitch.
Whereincomesarehighertheopportunitycostofsuchexclusivityislikelytobehigher.
Similarlyinlargercountrieswithregionalpatternsofprovisionthebarrierstoswitchingwillbeveryhighforindividualsinareasnotintensivelycoveredbytheexclusivesupplier.
AsecondinterpretationforthesignicanceofGDPliesintakinganalternativeview.
WhereGDPpercapitaisverylowthepotentialmarketmaybesmallandnetworkin-frastructureunderdeveloped.
Inwhichcase,iftherearexedcostsforanoperatorinprovidinginfrastructuretosupportiPhone,themarketmayonlybeabletofeasiblysus-tainasingleprovidertoguaranteeareturnonnecessaryinvestment,asperLafontaineandSlade[2005].
Geographicindicatorsarealsosignicanttovaryingdegrees;indicatingthat,relativetoEurope,NorthAmerican(includingCaribbeanandCentralAmerica)countrieswerelikelytohaveshorterduration,whileAfricancountrieslonger,thoughtheirinclusionismotivatedbyaneedtocontrolforregionalpolicyratherthandrawconclusionsbasedupontheirresult.
6.
3.
3TelecomVariablesandDurationAnumberoftelecom-specicvariablesaresignicantinthesecondstageofCragg'smodel.
SimilarlytotheTobitintable5,coecientsassociatedwithmarketconcentra-tionandcompetingstandards(η11andη12,respectively)arepersistentlysignicantata5%level.
Inadditionthenumberofrmsactiveinthemarketisalsosignicantata5%level-indicatedbythecoecientη10.
CompetingStandardsWhereamarkethascompetingstandards,thedierenttechnologicalstandardsareoftenusedbydierentrms.
Theearlierdiscussionfocussedoncompetingstandardsasasourceofswitchingcostsinmobiletelecommunications;competingstandardsexagger-atetheswitchingcoststhatalreadyexistconcerningchangingprovider.
TheuseoftheiPhoneasacompetitivelevertodierentiatearmreliesupontheabilityofthatrmtoattractcustomersfromrivalproviders.
Whereswitchingcosts36existthiswilldelayorinhibitconsumersfromswitchingtothatoperatorwhichoerstheiPhone.
Thenatureofswitchingcostsintelecommunicationsisalsocloselylinkedtothenatureofxed-termsubscriptions.
Theyarenotconstantovertimeandwillbeloweratperiodsneartheendofthexed-term.
Itnaturallyfollowsthat,inthepresenceofsuchswitchingcosts,anexclusiveprovidermayrequirealongerperiodofexclusivityinordertofullyexploitthecompetitiveadvantageinherentinoeringtheiPhone.
ConcentrationandFirmNumbersBycontrollingforrmnumberinthemarketinalltheapproaches(throughthein-clusionoftherelevantvariable)HHIcanbeinterpretedasbeingprimarilythemeasureofinequalityofrmsizes.
Thus,fromtheentryinvestigationintable3,weobservedthatmarketswithunequalmarketshareswerelesslikelytoselltheiPhone.
Fromtable7HHIhasnobearingontheactuallikelihoodofexclusivitybut,conditionalonthecasethatexclusivityisimposed,thisexclusivityislikelytobesignicantlylongerinmarketscharacterisedbyunevenmarketshares.
TheinsignicanceofthevariableindicatingthemarketpositionofthermthatsellstheiPhone(η15meansthatthereisnopatternthateitherthedominantrminthemarket,norsmallerrmsaresellingtheiPhone).
Thismakesthedrawingofconclusionsoverthecompetitiveaspectsoftheobservedlongduration.
Inaddition,whereamarketisunevenlydistributed,long-termcontractingwithadominantprovidermayprovideaccesstothemajorityofthemarket,reducingtheneedtocontractwithfurthernetworksafteratime.
Astrongresultisthatthecoecientassociatedwiththenumberofrms(η10)ispositiveandpersistentlysignicantat5%acrossspecications.
Wheretherearemanyrms,thisissignicantlylikelytoincreasethedurationofexclusivityrestraints.
ThisresultonrmnumbershasmorebasisintheinformationspillovertheoriesfromtheliteraturesuchasMathewsonandWinter[1984].
Thendingsupportstheviewthatwheretherearemorermscompeting,spilloversinadvertisingarehighest.
InthiscaseApplehastheincentivetomaintainexclusivitytoensurethatrmsexpendappropriateeortinadvertising.
Noneoftheremainingindustry-specicvariablesaresignicant,thispromptsthenalresultoftheinvestigation:Result3Thedecisiontodealexclusivelyisdominatedbydemographicvariables,whereasthedurationofexclusivityisdeterminedbyarangeofvariablesrelatingtotechnicalandcompetitiveaspects.
Specically,wherethenumberofnetworkscompetinginthemarketishighestweobservelongerexclusivityperiods.
Longerperiodsofexclusivityarealsoassociatedwithmarketconcentrationandcompetingstandards.
377ConclusionThispartofthethesissetoutwithseveralobjectives:tounderstandthedeterminantsofApple'sdecisiontoselltheiPhoneincertaincountries;tounderstandthedeterminantsofthedecisionanddurationofanyexclusivedistributionagreements;andtoshowthatthedecisiontodealexclusivelyanddecisionsoverdurationarenotnecessarilyon-and-the-samedecision.
Intheprocessoftheachievingtheaboveobjectivestheinvestigationhasreturnedthreekeyresults:Result1ThedeterminantsofthesaleoftheiPhoneinagivencountryisdominatedbybasicdemographicfactorsoflandsize,populationdensity,andGDPpercapita.
Onlythemeasureofinequalityofrmsizesissignicantofthetelecomspecicvariables,in-dicatingthattheiPhonewaslesslikelytosoldinconcentratedmarkets.
Result2Theimpositionandthedurationofexclusivedistributionagreementsaregov-ernedbydierentprocesses.
Wheretheconditionsforimpositionarestrongthisdoesnotimplythattheresultingexclusivedistributionagreementwillbedurable,andviceversa.
Result3Thedecisiontodealexclusivelyisdominatedbydemographicvariables,whereasthedurationofexclusivityisdeterminedbyarangeofvariablesrelatingtotechnicalandcompetitiveaspects.
Specically,wherethenumberofnetworkscompetinginthemarketishighestweobservelongerexclusivityperiods.
Longerperiodsofexclusivityarealsoassociatedwithmarketconcentrationandcompetingstandards.
Oftheseresultsthemoststrikingisnumbertwo.
Thisquestionstheprevailingas-sumptionofdurationofexclusivedistributionassomefunctionoftheoriginalexclusivity.
Thesignicanceofmarket-specicvariablesisstrongestwhenexaminingthedurationofexclusivity.
Inthiscasetherewasevidencethatswitchingcosts-proxiedbycompet-ingstandards-resultinlongerexclusivitywhichisanaturalresultwhenconsideringcustomerinertiainswitchingproviders.
SimilarlythevariablesconcerningcompetitionandmarketinequalitiesbothsuggesttheuseoftheiPhoneasacompetitivelever,andalsosupportthepredictionsoftheliteratureconcerningadverseselectioninadvertising.
Theremainingresultsaregenerallyinsignicantandsubjecttointerpretation.
Inconclusion,thispaperhasempiricallydemonstratedtheexistenceofmultiplepro-cessesintheimpositionofexclusivityrestraints.
Thisresulthighlightsthescarcityoftheliteraturewhichexaminesoptimaldurationofverticalrestraintsandrepresentsaspringboardtofurtherresearchinthearea.
38ReferencesBoyd,D.
W.
(1993).
Thechoicebetweenresalepricemaintenanceandexclusiveterri-tories:Evidencefromlitigation.
ReviewofIndustrialOrganization,8:755–763.
Boyd,D.
W.
(1996).
ResalepricemaintenanceordealerexclusiveterritoriesTowardatheoryofproductdistribution.
TheAmericanEconomist,40(2):86–94.
Brenkers,R.
andVerboven,F.
(2006).
LiberalizingaDistributionSystem:TheEuro-peanCarMarket.
JournaloftheEuropeanEconomicAssociation,4(1):216–251.
Burnham,K.
P.
andAnderson,D.
R.
(2004).
MultimodelInference:UnderstandingAICandBICinModelSelection.
SociologicalMethods&Research,33(2):261–304.
Cragg,J.
G.
(1971).
SomeStatisticalModelsforLimitedDependentVariableswithApplicationtotheDemandforDurableGoods.
Econometrica:JournaloftheEcono-metricSociety.
Dobson,P.
W.
andWaterson,M.
(1996).
Exclusivetradingcontractsinsuccessivedierentiatedduopoly.
SouthernEconomicJournal,63(2):361–377.
Dutta,S.
,Bergen,M.
,andJohn,G.
(1994).
Thegovernanceofexclusiveterritorieswhendealerscanbootleg.
MarketingScience,13(1):83–99.
Gartner(2013).
GartnerSaysWorldwideMobilePhoneSalesDeclined1.
7Percentin2012.
http://www.
gartner.
com/newsroom/id/2335616,((accessed13thOctober2013)).
Greene,W.
H.
(2012).
EconometricAnalysis.
PearsonEducation,NewJersey,US,8edition.
Lafontaine,F.
andSlade,M.
(2005).
Exclusivecontractsandverticalrestraints:Em-piricalevidenceandpublicpolicy.
InBuccirossi,P.
,editor,HandbookofAntitrustEconomics,numberJune.
MITPress.
Lafontaine,F.
andSlade,M.
(2007).
VerticalIntegrationandFirmBoundaries:TheEvidence.
JournalofEconomicLiterature,45(3):629–685.
Lin,T.
-F.
andSchmidt,P.
(1984).
AtestoftheTobitspecicationagainstanalternativesuggestedbyCragg.
TheReviewofEconomicsandStatistics,66(1):174–177.
Lin,Y.
(1990).
Thedampening-of-competitioneectofexclusivedealing.
TheJournalofIndustrialEconomics,39(2):209–223.
Mathewson,G.
F.
andWinter,R.
A.
(1984).
AnEconomicTheoryofVerticalRestraints.
TheRANDJournalofEconomics,15(1):27.
O'Brien,D.
P.
andShaer,G.
(1992).
Verticalcontrolwithbilateralcontracts.
TheRANDJournalofEconomics,23(3):299–308.
39O'Brien,D.
P.
andShaer,G.
(1993).
OntheDampening-of-CompetitionEectofExclusiveDealing.
TheJournalofIndustrialEconomics,41(2):215–221.
Rey,P.
andStiglitz,J.
(1995).
TheRoleofExclusiveTerritoriesinProducers'Compe-tition.
RandJournalofEconomics,26:431–451.
Rey,P.
andTirole,J.
(1986).
TheLogicofVerticalRestraints.
TheAmericanEconomicReview,76(5):921–939.
Saggi,K.
andVettas,N.
(2002).
Onintrabrandandinterbrandcompetition:Thestrate-gicroleoffeesandroyalties.
EuropeanEconomicReview,46:189–200.
Sass,T.
R.
andSaurman,D.
S.
(1993).
Mandatedexclusiveterritoriesandeconomiceciency:Anempiricalanalysisofthemalt-beverageindustry.
JournalofLawandEconomics,36(1):153–177.
Spengler,J.
(1950).
Verticalintegrationandantitrustpolicy.
TheJournalofPoliticalEconomy,58(4):347–352.
Subramanian,U.
,Raju,J.
S.
,andZhang,Z.
J.
(2012).
ExclusiveTie-upForaHandsetintheWirelessIndustry:ACompetitiveAnalysis.
ForthcominginMarketingScience.
Tobin,J.
(1958).
Estimationofrelationshipsforlimiteddependentvariables.
Econo-metrica:JournaloftheEconometricSociety.
Trivedi,M.
(1998).
DistributionChannels:AnExtensionofExclusiveRetailership.
ManagementScience,44(7):896–909.
Wooldridge,J.
M.
(2002).
EconometricAnalysisofCrossSectionandPanelData.
MITPress,London,UK.
Zhu,T.
,Liu,H.
,andChintagunta,P.
(2011).
WirelessCarriers'ExclusiveHandsetArrangements:AnEmpiricalLookattheiPhone.
ChicagoBoothPaperNo.
13-09.
40