2025金融科技借貸與債務(wù)螺旋（英）

FintechLendingandDebtSpiralLiliDai,1JianleiHan,2JingShi,3andBohuiZhang4March20251UNSWBusinessSchool,UniversityofNewSouthWales,Sydney,Australia,Email:lili.dai@.au2MacquarieBusinessSchool,MacquarieUniversity,Sydney,Australia,email:jianlei.han@.au3MacquarieBusinessSchool,MacquarieUniversity,Sydney,Australia,email:jing.shi@.au;4SchoolofManagementandEconomicsandShenzhenFinanceInstitute,ChineseUniversityofHongKongShenzhen(CUHK-Shenzhen),email:bohuizhang@WeappreciatethecommentsfromseminarparticipantsattheHunanUniversity,HuazhongNormalUniversity,NanjingUniversityofAeronauticsandAstronautics,XianJiaotongUniversity,ZhongnanUniversityofEconomicsandLaw,aswellasconferenceparticipantsatthe2023InternationalForumofFintechandFinancialMarketHigh-qualityDevelopment,the2023FinTech,AI,andDataAnalytics,FAIDAWorkshop,the2023BusinessFinancingandBankingResearchGroupWorkshop,the2023ChinaFinancialMarket,Innovation,ESGAcademicConference,andthe18th2023ChinaFinanceScholarsAssociationMeeting.WethankShanghaiQiaopanTechnologyforprovidinguswiththefintechlendingdataandsupportforourempiricalanalysis.FintechLendingandDebtSpiralAbstractWeexaminewhetherandhowconsumerscanbetrappedindebtspiralsinfintechlendingmarkets.Withalternativedatasources,fintechlendersmaytargetandprovideeasy-to-accesscreditproductstonewcustomers.Wefindthatafterobtainingthefocalfintechlender’sloanapproval,borrowersreceivemorepromotionalmessagesfromotherfintechlendingplatforms,causinggreatertendenciesofsubsequentnewborrowing,personalexpenditure,loandelinquency,andexperiencingadversesocialoutcomes.Thiseffectisstrongerforborrowerswithlowfinancialliteracyandlimitedcreditaccess.Takentogether,wedocumenttheeconomicandsocialconsequencesoffintechconsumers’overborrowingbehavior.JELclassification:D14;G23;G51Keywords:Debtspiral;Fintech;Loandelinquency;SocialoutcomePAGEPAGE1“KeycompetitiveadvantagesofFinTechlenders…allowformoreelasticloansupplybutalsohavethepotentialtoinduceoverborrowingbyna?veconsumers.”–Berg,Fuster,andPuri(2022)IntroductionPriorfinancialtechnologystudiessuggestthatfintechlenderscanrespondmoreelasticallytodemandshocks(Fusteretal.2019;2021)andhelpexpandcreditaccesstounderservedborrowers(Bergetal.2020;DiMaggioetal.2022),whileregulatorsaroundtheglobeareincreasinglyconcernedaboutthebusinessmodelsoffintechlenders,whichmayinduceoverborrowingandharmconsumers’financialwell-being(OECD2021;Treasury2022).Forexample,theU.S.ConsumerFinancialProtectionBureau(CFPB)cautionsthatborrowerswhotakefintechpersonalloansbasedonthe“BuyNow,PayLater”modelexhibithighlevelsoffinancialdistressandfaceoverextensionproblems(CFPB2022;2023).1However,researchontheoverextensionofcreditinthefintechlendingmarketisscarce(Bergetal.2022).Thisstudyprovidesdirectevidenceofhowconsumerscanbetrappedintoadebtspiralthroughthereceiptofmobilemessagesfromfintechlenders,aspecificmechanismofenticingnewborrowers.2Theexistingresearchonconsumerfinanceshowsthatindividualborrowersmayencounteroverextensionandeconomichardshipproblemsarisingfromvariouscreditmarkets,including,forinstance,mortgagelending(Bondetal.2009),creditcards(Bentonetal.2007;HeidhuesandKoszegi2010),andpaydayloans(Gathergoodetal.2019;Allcottetal.2022).Thiseffectisespeciallysignificantforlow-incomeconsumersandhouseholds(Melzer2011;Melzer2018)andmayspillovertoconsumers’othersocialoutcomes(CarrellandZinman2014).Inthisstudy,wecomplementthisstreamofresearchbyinvestigatinganddocumentingtheeconomicandsocialconsequencesofconsumers’engagementinoverborrowingbehaviorinthedigitalerathroughthefintechlendingmarket.1Overborrowing/overextensionforconsumersreferstotwoforms:1)loanstacking,theriskthataborrowertakesoutconcurrentloansatdifferentlendersandisunabletorepaysomeorallofthem,and2)sustainedusage,theriskthatfrequentloanusagemaythreatenborrowers’abilitytomeetotherfinancialobligations,suchasrentandutilities(CFPB2022).2Relatedly,DiMaggioandYao(2021)findsthatfintechborrowersaremorelikelytopurchaseacarafterloanoriginationanddefaultthanthoseborrowingfromtraditionalfinancialinstitutions.Ourstudycomplementstheirfindingsbydocumentingaparticulartoolforfintechlenderstoenticenewcustomersthroughthedisseminationofloanadvertisingmessagestopotentialborrowers.Withaccesstoalternativedatasources,fintechlendersmaystrategicallytargetandenticenewcustomers,potentiallyleadingtoborrowers’unsustainableindebtedness.Forinstance,inChina,consumersoftencomplainaboutmobileadvertisementsfromfintechplatforms,andmanyfintechlendershavebeenaccusedofmisleadingadvertising,suchastrickingborrowersintoloanserviceswithlimitedinformationonprocessingcosts.Asaresult,theChinaBankingandInsuranceRegulatoryCommission(CBIRC)hashighlightedoverborrowingriskfacedbyfintechborrowersrelatingtoseveralfintechlendingissues,includingambiguousloanterms,inducedexcessiveexpenditures,andpersonalinformationleakage(CBIRC2020).Fintechborrowersarelikelytobetrappedindebtspiralsthroughthreepotentialchannels.First,thoseborrowerswithalowerleveloffinancialliteracy(Gathergood2012)aremorelikelytoengageinself-controlproblemsandimpulse-drivenexpenditures,leadingtooverborrowingproblems(Bentonetal.2007;HeidhuesandKoszegi2010).Inthefintechlendingmarkets,thecompetitivefeaturesoflendingproductsandplatforms,suchasthestreamlinedapplicationprocess,acceleratedprocessingtime,andlesseffortofhumaninputdemandedfromconsumers,willamplifyborrowers’self-controlproblems,thusinducingtheiroverborrowingbehavior(Bergetal.2022).WeconsiderthismechanismastheFinancialLiteracyChannel.Second,inthefintechlendingmarkets,thoseconsumerswithastrongandurgentdemandforaccessingcreditservicesaretypicallyassociatedwithahighlevelofpovertyandareunderservedbytraditionalfinancialinstitutions(Hauetal.2019).Prioreconomicstudiessuggestthatpovertycanperpetuateitselfbyfurtherunderminingindividuals’self-controlcapacity(e.g.,Bernheimetal.2015),thereforecausingoverborrowingbehavior.Especiallywhenfintechlendershavetheabilitytoexploitalternativedatasources,theycaneasilytargetthosepotentialborrowerswithlimitedaccesstotraditionalfinancialservicesandexecutepredatorylendingstrategiesonthoseborrowers(DiMaggioetal.2022),whowillconsequentlybeinducedintooverborrowing.WetreatthismechanismastheCreditAccessChannel.Third,theregulatoryoversightandenforcementmechanismsagainstfintechcompaniestendtobelessrestrictivecomparedtotraditionalfinancialinstitutions(Buchaketal.2018;Thakor2020).TheconcernsaboutthelackofregulationleadingtofraudwithfintechlendershavemaderegulatorsinChinaconsiderstrengtheningregulationsinthefintechlendingmarkets(Liaoetal.2023).Forexample,theCBIRChascautionedconsumersagainst“theriskofoverborrowingviaonlineplatformswhilethecountryissteppingupitsregulationoffintechcompanies”(Jiang2020).Thus,fintechborrowersmaybetrickedintodebtspiralsbecauseofthelackofregulationinthefintechlendingmarkets.WelabelthismechanismastheLendingRegulationChannel.WeinvestigatetheprocedureofhowconsumersmayengageinadebtspiralusingthedataonpersonalloancontractsfromaleadingfintechcompanyinChinaandthemobilephonelogsofconsumersthatcanhelpidentifytheirfinancialandsocialconsequences.3Thefintechcompanyasksaborrowertoprovidehermobilenumberandnationalidentitynumber.Withsuchinformation,thelendercanobtainotherinformationabouttheborrowerfromthird-partydataproviders,including,forexample,thebalanceofheronlinepaymentaccounts,thehistoricalrecordofpastloanapplications,andthedataofmobilephonelogs.Thelenderthenevaluatestheborrower’screditworthinessandcomputestheinternalcreditscore.Ouranalysesstartbyexaminingwhethertheborrowerreceivesmoremobilemessagesfromotherfintechlendersafterobtainingtheloanapprovalfromthefocalfintechlender.Thefullsampleinthebaselineanalysesconsistsof2,787,505borrower-dayobservationsfromJuly2017toNovember2019.Theresultsfromthedifference-in-differences(DID)analysesindicatethatborrowerswhoseloanapplicationsareapprovedbythefocalfintechlender(inthetreatmentgroup)aremorelikelytoreceivepromotionalmessagesthatencouragethemtoapplyfornewloansatotherfintechlendingplatformsthanthosewhoseapplicationsarerejectedbythefocallender(inthecontrolgroup).Thisfindingsuggeststhatotherlendingplatformscanusealternativedatasourcestodeliberatelytargetborrowerswithapprovedloanapplicationsbecausetheseborrowersmayhaverelativelyhighcreditworthinessandwillhaveastrongcashdemandtoapplyfornewloanswhentheyfacerepaymentpressureinfutureperiods.Thislendingstrategybyfintechlenders(i.e.,acquiringborrowers’informationfromalternativedatasourcesandtargetingthosewithapprovedloanapplications)doesnotapplytobanksandothertraditionalfinancialinstitutions,whicharesubjecttorigorouslendingregulationsrestrictingtheuseofalternativedatatoenticepotentialborrowers.Economically,wefindthataftertheloanapproval,thedailylikelihoodofreceivingpromotionalmessagesfortreatedborrowersincreasesby6.42percentrelativetothoseinthecontrolgroupwithinawindow[-180,180]aroundtheloanapprovaldate,comparedtothesamplemean.3WethankShanghaiQiaopanTechnologyforprovidinguswiththefintechlendingdataandsupportforourempiricalanalysis.PAGEPAGE10Next,toaddresspotentialendogeneityconcerns,weconducttheregressiondiscontinuitydesign(RDD)analysesbasedonthefocalfintechlenders’internalcreditworthinessscore,whichisusedtoevaluateborrowers’creditriskanddeterminetheloanapplicationoutcome.Specifically,thehigherthescorevalue,thelowerthecreditriskofaloanapplicant,andthefocallenderwillapproveanapplicationwhenthestandardizedscorevalueisgreaterthanzero.OurRDDanalysesareperformedinsub-sampleswithcreditworthinessscoreswithinasmallrangearoundthecutoffvalueofzero.Bydoingthis,weassumethatborrowers’characteristicsaresimilarbetweentreatmentandcontrolgroupsinthesesub-samples,exceptthattheircreditscoreshappentobeslightlygreaterorlowerthanthevalueofzero(e.g.,creditscoreswithintherange[-0.05,0.05]).OurfindingsfromtheRDDanalysesareconsistentwiththeresultsfromourbaselinefindings.Wefurtherperformseveraltwo-stageRDDanalysestoinvestigatewhethertreatedborrowersindeedengageinadebtspiralinducedbyotherfintechlenders’loanpromotionmessagesthroughaseriesofactionsandconsequences,i.e.,registeringloanswithotherlenders,spendingmoreexpenditures,defaultingonnewloans,andexperiencingadversesocialoutcomes.Inthefirststage,overtheperiodafteraloanapplicationfromthefocalfintechlender,weregressthenumberofpromotionalmessagesreceivedbyaborrowerwithinoneweekpriortoaparticulardayontheindicatorvariableoftheborrower’sloanapprovaldecision.Inthesecondstage,weexaminetherelationbetweentheborrower’sactionorconsequenceonthatdayandthepredictedvalueofthepromotionalmessagenumberinstrumentedfromthefirst-stageregression.TheseRDDanalysesareperformedinreducedsamplesthatrequireborrowers’creditworthinessscoreswithintherange[-0.05,0.05].Wediscusstheresultsofdebtspiralengagementasfollows.First,weshowthattheborrowers’tendencytoregisternewloanswithotherfintechlendingplatformsinawindow[91,180](i.e.,threemonthsaftertheloanapprovaldecisionmadebythefocallender)ispositivelyassociatedwiththepredictedvalueoftheloanpromotionmessagenumber.Weperformtheanalysesinapost-approvalwindow[91,180]toalleviatethepotentialimpactsofloaninitiationwiththefocalfintechlenderonborrowers’actionsandconsequences.4Infurtheranalyses,wedifferentiatetheloanregistrationsenticedandnotenticedbylenderswhosendpromotionalmessagesandfindthepositiveassociationwiththeloanpromotionmessagenumberholdsforbothtypesofregistrations.Theseresultssuggestthattreatedborrowersarenot4Ourresultsarerobusttoalternativewindowssuchas[61,180]twomonthsor[121,180]fourmonthsaftertheloanapprovaldecision.onlyenticedtoregisternewloanswithpromotinglendersbutalso,interestingly,theyareencouragedtoborrowmorefromnon-promotinglendingplatforms.Second,wefindapositiverelationbetweenborrowers’personalexpendituresandthepredictedloanpromotionmessagenumber.Inparticular,forborrowershavingnewloanregistrationswithotherfintechlenders,thedailylikelihoodofincurringexpendituresandthedailyamountofexpendituresarepositivelyassociatedwiththepredictednumberofloanpromotionmessagesinawindow[91,180]threemonthsaftertheloanapprovalbythefocallender.Incontrast,thepredictedpromotionalmessagenumberisinsignificantlyrelatedtotheexpenditurebehaviorofnon-registeredborrowers.Thesefindingssuggestthatafterobtainingcashfromotherlendingplatforms,treatedborrowers’personaldailyconsumptionincreases.Thisincreasedexpenditurebehaviormaypotentiallyreduceborrowers’futurerepaymentcapabilities.Third,theresultsfromtheloandelinquencytestsfurthersuggestapositiveassociationbetweenthedailylikelihoodofreceivingadebtcollectionmessage(orthedailynumberofdebtcollectionmessages)fordelinquentloanpaymentinawindow[121,180]fourmonthsaftertheloanapprovalandthepredictednumberofpromotionalmessages.Comparedtoourprevioustestsbasedonthewindow[91,180],weperformtheloandelinquencyanalysesfromDay121toallowthefirstmonthlypaymenttobedueonemonthafterDay91.Ourfindingsfromthesedelinquencytestsimplythatafterregisteringnewloansandincurringmorepersonalconsumptions,treatedborrowersaremorelikelytofailtorepaytheirloanand,therefore,receivedebtcollectionmessagesfromlendingplatforms,trappedintoadebtspiral.Fourth,weinvestigatethepotentialadversesocialoutcomesofengagingindebtspiralsbyfocusingonthemobilemessagesthatborrowersreceivewithnegativeChinesephrasessuchas“divorce”and“breakup”asasignalsuggestingthattheyarelikelyengagedintheinitiationofadversesocialevents.Wedocumentthatthelikelihoodofreceivingmobilemessageswithnegativephrasesandthenumberofsuchmessagesarepositivelyrelatedtothepredictednumberofpromotionalmessagesinawindow[91,180]threemonthsaftertheloanapprovalbythefocalfintechlender.Theseresultsshowthattreatedborrowersnotonlytendtosufferfinanciallossesbutalsomorepossiblyexperienceadverseconsequencesintheirpersonallives(CarrellandZinman2014),suggestinganegativesocietalexternalityeffectoftheinducementoffintechlendingtodebtoverextension.Takentogether,wefindsupportiveevidencefortheadverseeconomicandsocialconsequencesforborrowersbeingenticedintodebtspirals.However,beforehand,wemaynotnecessarilyexpecttoobservethesenegativeoutcomesforthefollowingreasons.First,approvedloanapplicationswillenableborrowerstosatisfytheircashdemandforconsumptionpurposes,thuslikelyreducingtheirneedforfurtherborrowing.Second,moreloanapprovalscanpotentiallylowerthelikelihoodofreceivingfurtherloans,especiallyforthoseconsumerswhohaveshortcredithistoriesandfailtomakeloanpaymentsontime.Lastly,weperformafewcross-sectionalanalysestoexaminethethreeeconomicchannelsthroughwhichborrowerscanbetrappedinadebtspiral.First,wedocumentthattheeffectoffintechlendingpromotionalmessagesonborrowers’involvementinoverextensionproblemsisstrongerforborrowerswithalowerleveloffinancialliteracy.Thissuggeststhatborrowerswithlimitedknowledgeandexperiencefortheirpersonalfinancialmanagement,budgeting,andinvestmentaremorevulnerabletopredatorylendingstrategiesbyfintechlendingplatforms(LusardiandScheresberg2013;Allcottetal.2022),consistentwiththeFinancialLiteracyChannel.Second,wefindthattheeffectofbeingenticedtooverborrowingismorepronouncedforthosewhohavelimitedcreditaccesstotraditionalfinancialservices.Thisisconsistentwiththenotionthatborrowerswithlimitedfinancingaccessfromtraditionallendingchannelsrelymoreontheaccessprovidedbyconsumercreditmarkets.Therefore,consistentwiththeCreditAccessChannel,theseborrowersaremorelikelytobetargetedandenticedintooverborrowingbyotherfintechlenders,ultimatelyleadingtotheirunsustainableindebtedness.Third,theresultsfromthecross-sectionalanalysesoflendingregressionsuggestaninsignificantmoderatingeffectofthelowregulationindicator,whichisdefinedbasedontheprovince-levelfintechlendingregulationscore,ontherelationbetweenlendingpromotionalmessagesandborrowers’overextensionoutcomes.ThesefindingsdonotprovidesupportfortheLendingRegulationChannel,suggestingthatthevariationinfintechlendingregulationdoesnotaltertheeconomicandsocialconsequencesofconsumersforbeingtrappedindebtspirals.Ourpapermakescontributionstotwostreamsofliterature.First,priorstudiessuggestthatpersonalloanscanhelpconsumersmeetfinancialemergenciesandsmoothconsumptions(Morse2011;Maggioetal.2023)butalsolikelyleadtoagreatlikelihoodoffinancialdistress(e.g.,Campbelletal.2012;Gathergoodetal.2019;SkibaandTobacman2019),andthatindividuals’borrowingdecisionsmaybeinfluencedbycognitivebiasesandlimitations(BertrandandMorse2011).Wecomplementthislineofliteraturebydocumentingthat,inaneraofdigitalfinance,howindividualconsumerscanbeengagedinoverborrowingbehaviorwhentheyfaceinducementfromfintechlendingplatformswhichenticenewcustomersbyprovidingstreamlinedloanapplicationprocessesthatamplifyborrowers’self-controlproblemsandbyexploitingborrowers’personalinformationthroughalternativedatasources.Second,theresearchonfintechlendingsuggeststhatadvancesintechnologyallowfintechlenderstobetterassessthecreditworthinessofcustomersusingalternativepersonaldata(Bergetal.2020;Agarwaletal.2021;DiMaggioetal.2022)androbo-advisingtools(D’Acuntoetal.2024).Fromtheborrowers’perspective,theyfaceatradeoffbetweentheirpersonaldataprivacyandaccesstocreditfinancing(Chenetal.2021;Tang2021).Alignedwiththislineofliterature,recentstudiesondataprivacyregulationalsoindicatetechnologyfirms’abilitytocollectandshareconsumers’personaldata(e.g.,Peukertetal.2022).Weaddtothisstreamofliteraturebydocumentingthepotentialcostsforfintechborrowerswhenlendersaggressivelyadoptandapplyfinancialtechnologiesinpredatorylendingstrategiesintheconsumercreditmarketsforbeingtrappedindebtspiralsandimpulse-drivenconsumptions.Therestofthepaperisorganizedasfollows.Section2describestheinstitutionalbackground,variableconstruction,andsummarystatisticsofthevariablesusedinourstudy.Section3presentsthemainresultsandSection4reportstheadditionaltests.Section5concludesthepaper.Institutionalbackground,variabledefinitions,anddescriptivestatisticsInthissection,wedescribetheinstitutionalbackgroundofourresearchsettingandpresentthedefinitionsandsummarystatisticsofthevariablesusedinourmainanalyses.InstitutionalbackgroundWecollectthedataonpersonalloancontractsfromShanghaiQiaopanTechnology,aleadingfintechcompanyinChinathatcommenceditslendingoperationsinJuly2017.Tofacilitateloanapplicationsfromprospectiveborrowers,thecompanyestablishedapplicationterminalsinretailstoreslocatedinmajorcitiesthroughoutChina.Theseterminalsserveaspointsofaccessforindividualsseekingloans.Duringtheapplicationprocess,applicantsarerequiredtoprovidetheirnationalidentitycardinformationandmobilenumbers.Followingthesubmissionoftheloanapplication,thelenderutilizestheapplicants’mobilenumbersandnationalidentitynumberstoretrievetheirpersonaldatafromthird-partydataproviders,includingtheinformationoftheironlinepaymentaccount,onlineloanrecord,andmobilephonelogs.Byanalyzingtheinformationobtainedfromdataproviders,thefintechlendercreatestheinternalcreditscoreandthenevaluateswhethertheloanapplicationsshouldbeapprovedorrejected(Bergetal.2020).Aftertheevaluationprocess,thelenderpromptlycommunicatesthedecisiontoapplicants,normallywithintenminutes.Oncetheloanapplicationisapproved,borrowerswillobtaincashfromthefintechlenderanduseittopurchasegoodsorkeepthecashforfutureuse.Itisworthnotingthatallthesuccessfulapplicantsinoursampleopttotakeuptheirloans.Thiscouldpotentiallybeattributedtofactorssuchaslimitedaccesstotraditionalbankingservices,leadingtostrongdemandforpersonalborrowing.Additionally,theborrowersinoursamplearemotivatedtotakeloansduetothestandardizedloantermsandinterestrates,whichareunrelatedtotheirindividualcharacteristics.Subjecttoloanmaturity,inthenextsixortwelvemonths,theborrowerneedstopaythelenderthemonthlyprincipalandinterest.Whentheborrowerfailstomakerepayment,thelenderwillsendremindermessagestotheborrowertocollectrepayment.Figure1illustratestheprocessofloanapplication,approval,repayment,andcollectionbythefintechlender.[InsertFigure1Here]Furthermore,weobtainthedataofborrowers’personalmobilephonelogsaftertheinitiationoftheloansfromthefocalfintechcompany.Themobilelogdataisalsoobtainedfromthird-partydataproviders,whocontinuouslyupdateinformationsourcedfromborrowers’mobilecarrierswiththefocallender.Thisdatasetenablesustocreatemeasuresrelatedtoloanpromotionalmobilemessagesfromotherfintechlendingplatformsinthepost-loan-initiationperiodandallowsustoconstructmetricsthatcapturetheeconomicandsocialconsequencesofcreditoverextensions,whicharereflectedinborrowers’subsequentmobilemessagesafterloaninitiations.VariableconstructionToconstructthemeasuresofthereceiptofloanpromotionalmessages,weidentifythepromotionalmobilemessagesreceivedbyeachborrowerfromotherfintechlendingplatforms.Forexample,onemessagecanbe:“[AntCreditPay]Congratulations!Youaresuccessfullyincludedinourwhitelistforacreditlineof15,000yuan!Applywithinonehourtogetthemoney.ReplyTtounsubscribe.”InaccordancewiththeregulationssetbytheMinistryofIndustryandInformationTechnologyofChina,commercialfirmsarerequiredtoincludetheircompanynameswithinbracketseitheratthebeginningorendofmobilemessagessenttoconsumers.Thisregulatoryprovisionenablesustoextractthenamesofpotentialfintechlendersfromthesemessages.Subsequently,wemanuallyverifywhethertheextractedtextsindeedpertaintoothernon-focalfintechlenders.Forinstance,inthisexample,thefintechlendingfirmAntCreditPayisdenotedwithinthebracketinthemessagesenttoborrowers.Employingthismethodology,weidentify4,274uniquelendingcompaniesfromthemobilemessagesofborrowers.AppendixApresentsthetoptenlendersreleasingthelargestnumberofmobilemessageswithinoursample.Next,weproceedwithclassifyingmobilemessagesassociatedwithfintechlendersintothecategoryofloanpromotionalmessages.Thisclassificationisbasedonloanpromotion-relatedphrases,suchas“applyandreceivealoan,”“l(fā)oaninterestwaived,”and“activateyourcreditquota.”AppendixBpresentsthesephrasesinbothChineseandEnglish,andFigure2presentstheexampleofafullpromotionmessage.Further,werequireacompulsorykeyword,“unsubscribe,”tobeincludedinpromotionalmessages,whichismandatedbytheMinistryofIndustryandInformationTechnologyinmessagesformarketingpurposes.Byapplyingthesecriteria,weidentifyloanpromotionalmessagesfromthemobilemessagedataanddefineMessagePassive,asthedailynumberofpersonalloanpromotionmessagesthataborrowerpassivelyreceivesfromothernon-focallendingplatforms.[InsertFigure2Here]Wealsoidentifyanothercategoryofmobilemessagesthataretriggeredwhenborrowersproactivelyapproachnon-focalfintechlenders.Whenaborrowervisitsandregisterswithalendingplatform,theborrowerwillreceivemessagescontainingphrasessuchas“l(fā)oanaccountregistered”and“completepersonalprofile”(seeAppendixB).WespecificallyidentifythesemessagesanddefineMessageProactiveasthedailycountofsuchmessagesreceivedbyaborrowerrelatingtotheproactivebehaviorofsearchingandapproachingotherfintechlenders.InourmainDIDanalyses,weconstructanindicatorvariable,Approval,whichequalsoneifaborrower’sloanapplicationsubmittedtothefocalfintechlenderisapprovedandzerootherwise,anddefinetheindicatorvariable,Post,whichequalsone(zero)ifaborrower-dayobservationisdatedafter(before)thesubmissionoftheborrower’sloanapplicationtothefocallender.Furthermore,weincludeafewcontrolvariablesforaborrower’spersonalcharacteristics,includingtheborrower’sage(Age),theindicatorvariableforthefemaleborrower(Female),andthecreditscoreprovidedbythefocalfintechlender,withahighvaluesuggestinglow-riskprofile(Score).SeeAppendixCforvariabledefinitions.Weexpectborrowerstobemorelikelytoreceiveloanpromotionalmobilemessagesfromotherlendingplatforms(MessagePassive)aftertheyreceivetheloanapprovalfromthefocalfintechlender(Approval×Post).Thisisbecausebyidentifyingborrowerswithapprovedloanapplications,theselenderscanleveragethisinformationtotailortheirmarketingeffortstargetingtheseindividuals,whopossessrelativelyfavorablecreditprofilesandaremorelikelytoseekadditionalloanfinancingwhenconfrontedwithrepaymentobligationsandconsumptiondemandsinthefuture.Wefurtherconstructaseriesofmetricstoexaminetheeconomicandsocialconsequencesofbeingengagedinoverborrowingbehavior.First,weidentifyborrowers’r