《金融经济学杂志》169(2025)104053数据销售与数据稀释

c,d whereusersofdatachoosequantitiesandmarketclearingdeterminestheprice.Thesecondassumption,aclassiccommitmentfriction,isparticu-larlyrelevantindatamarketswhereonecouldeasilytransformdatatomakeitnon-identical,butfunctionallyequivalent.Ourmodelteachesusthat,evenadatamonopolistmayhavelimitedpowertoextractrentfromtheircustomerswhenthedatasellercannotcommittoapriceschedule.Thereasonisthatadatasellercompeteswithitsfutureself.Ifadatasellercannotcommitnottosellthedatatothebuyer’scompetitors,thebuyer’swillingnesstopayforthedatadeclines.Thisforcekeepsdatapriceslow.Inthistypeofenvironment,weshouldworrylessabouttheexcessiveprofitsofdatamonopoliesandworrymoreaboutwhetherdataisunder-provided.Sinceweobservethatmany data producers sell data subscriptions,rather than dataownership,weaddthatfeaturetoourmodel.Wefindthatsubscriptionsfordataallowasellertore-capturemuchofitslostrevenuefromthelackofcommitment. Butif data subscriptions allow sellers to capture more surplusfromconsumers,whywouldanyoneselldataoutright?Weusethemodeltoidentifythreefeaturesofadatasellerthatmakesubscrip-tionslessattractive:financialfrictions,asmallmarket,andhighdatadepreciation.Ourtheorythusprovidesusawaytounderstandtheprevalenceandforceofmonopolypower—itdirectsustoexaminedatasalesmodels.Specifically,weshouldlookfortheprevalenceofdatasalesversusdatasubscriptionsandpatternsinhowthesetradetypesareused.Therefore,howdataissoldbecomesthecenterpieceofourempiricalanalysis.Tomeasureactivityindatamarkets,wehand-collectanoveldatasetfromDatarade,oneofthelargestonlinedatamarketplacesthatconnectsbuyersandsellersofdata.Theevidenceaboutthegeographic,industry,anddatatypecoverageofthismarketplacepaintsanuancedpictureofthewayinwhichdataistraded.Acrossover3,000dataproducts,we find that 64%offer an option to buy the data for aone-timefee.However,over80%offerasubscription-basedpaymentsystem.Thesefractionsdonotsumtoonebecausemanysellersoffermultiplepurchasingoptions.Thisfindingsuggeststhatatleasthalfofalldataprovidershavesignificantabilitiestoextractrents.Totestthepredictionsofourmodel,weneedtomergethedatamarketplaceevidencewithcompany-levelcharacteristicsofthedatasellersandthecharacteristicsoftheirdataproducts.Someofthesedatasellersarepubliclylistedcompanies,butmanyareprivate.Weuseavarietyofdatasources—Crunchbase,Pitchbook,Compustat,andCRSPtocollect information on these companies background informationandfinancinghistory.WeuseEdgar10-Kfilings,combinedwithdataproductdescriptionsonDatarade,tofillinthecharacteristicsofthemarketsinwhichtheyselltheirdata.Themodelpredictsthatdatasellersshouldchooseone-timefeesiftheyarefinanciallyconstrained.Iftheydonoturgentlyneedcash,thesubscriptionmodelofsellingdataistypicallymoreprofitablebecauseitresolvesthecommitmentproblem.However,one-timefeesbringinmorerevenueearlyinthelifeofthedata-sellingfirm.Ourempiricalanalysisconfirms this prediction.We find a significant correlationbetweenthewayinwhichadatasellersellsitsdataanditsage,thenumberofroundsofVCfundingithasreceived,andthetotalamountofthatfunding.Theolder,better-fundeddatasellersaremorelikelytoextractsurplus,throughtheuseofdatasubscriptions.Themodelalsopredictsthatwhenthemarketofdatabuyersissmall,thereislessscopetoerodethevalueofdatawithfuturesales.Therefore,datathatpertainstoamorespecializedgroupofpotentialbuyerscouldbesoldforaone-timefee,withlittleloss.Thedatamarketplaceevidencealsoconfirmsthisprediction.Wedeterminethesizeofthemarketfordatasalesbycomparingthetextualsimilarityofdatadescriptionswiththeuniverseof10-Kreportsandthendeterminingtheindustrieswiththegreatestsimilarities.Then,wecomputethenumberofrelevantindustriestodeterminethesizeofthemarketforthedata.Wefindthatthismarketsizepositivelypredictsdatasubscriptionsandisnegativelycorrelatedwithdatasales.Weacknowledge that it isalso possible that somesettings lendthemselvesto complementarity in the use of data.In settings likespeculativeattacks orprice-setting,the valueof data might riseasothersacquireit.Insuchsettings,thedatasellers’lackofcommitmentwillbelesscostly,becausedatagainsinvaluewhenmorecopiesaresold.Incontrast,dynamiccomplementarity,whereaninvestorwantstolearndatanowthatotherswillacquirelater,stilldecreasesthevalueofdataovertime.Finally,onemightobjectthatdataisnotadurablegoodbecauseitbecomes less relevant over time.We explored the role of datadepreciationinSection1.7.Theseresultsinformongoingdebatesaboutdatapolicy.TheEuro-peanCommission’s‘‘FreeFlowofNon-PersonalData’’initiative(Reg-ulation𝐸𝑈2018∕1807)arguesthatthedevelopmentofefficientdatamarketsthatpromotedatamobilityisessentialtothedevelopmentoftheEUdigitaleconomy.Itspecificallycites‘‘distortionsofcompetition’’ asaproblemtobeaddressed.Ourpaperexplorescompetitivepricingindatamarkets.Traditionalcompetitiontheoryisdesignedforproductionofphysicalproducts,withnon-zerocostsofreplication,thatareun-likelytohavethestrategicsubstitutabilityinherentinmostinformationproducts.Ourresultsteachusthatevenamonopolistdatasellerhaslittleeffectivemarketpowerinthemarketfordatabecauseitcannot