情绪波动与货币：金融科技与家庭信贷-英.docx

MoodSwingsandMoney:TheRoleofFinancialTechnologyinHouseholdCreditDemandRanDuchin,PaulFreed,andJohnHackney*December2023AbstractFintechlendingallowsborrowerstoapplyforloansanytimeandfromanywhere,completetheirapplicationswithinminutes,andobtainimmediatecreditdecisions.Assuch,transientmoodswingsthatwouldbemitigatedinatraditionalloansettingcanplayanimportantroleinmodernhouseholdcreditdemand.Usinghourlyfluctuationsinlocalsunshineasaninstrumentforsentiment,wefindthatpositivesentimentleadstohigherloandemandbothattheextensivemargin(moreloanapplications)andtheintensivemargin(higherloanamountsandloan-to-incomeratios).Theeffectsleadtohigherdefaultrates,especiallyforlower-incomeandinexperiencedborrowers.Wealsofindevidenceconsistentwithself-correctiveactionswhereindividualslaterwithdrawIheirapplications,suggestingthatucooling-off,periodscanbeaneffectiveconsumerprotectionmechanism.Overall,weprovidesomeofthecleanestestimatestodatethatsentimentaffectsthedemandforconsumercredit.KeyWords:FinTech,ConsumerCreditDemand,Sentiment,MarketplaceLending,DefaultJELClassifications:D12,D14,G4,G21,G23,033Contact:RanDuchin,CarrollSchoolofManagement,BostonCollege,e-mail:duchinr(5)bc.edu;PaulFreed,DarlaMooreSchoolofBusiness,UniversityofSouthCarolina,e-mail:Paul.Freedgrad.moore.sc.edu：JohnHackney,DarlaMooreSchoolofBusiness,UniversityofSouthCarolina,e-mail:iohn.hackneymoore.sc.edu.WethankseminarparticipantsattheUniversityofWashington,OldDominionUniversity,andtheUniversityofSouthCarolinaforhelpfulcomments.1. IntroductionTheadventoffinancialtechnologyhasfundamentallychangedthelandscapeofhouseholds,financialdecision-making.Borrowersononlinemarketplaceplatformscanapplyforloansfromthecomfortoftheirhomes,dayornight,completetheirloanapplicationswithinminutesusingtheirsmartphoneorcomputer,andneverspeaktoabankeroraloanofficer.Suchdevelopments,inturn,canhaveamaterialeffectonoverallfinancialdecision-making.Attheextensivemargin,lowertransactioncostscanincreasetheconsumptionofcredit.Theunsecuredconsumerloanmarkethasgrowndramaticallyinthelastdecade,from$57.7billionin2009to$156billionin2019,withmarketplacelendersresponsibleforroughly40%ofthemarket.Based on TransUnion data - see:Altheintensivemargin,theycanaffectthequalityofcreditdecisionsandsubjectthemtoinfluencesthatmoretraditionalsettingswouldmitigate.Inthispaper,Weusemicro-leveldatafromanonlinemarketplacelendingplatformtostudytheroleofsentimentandfinancialtechnologyinhouseholds,creditdemand.Theanalysesutilize1.4milliontimestampedloanapplicationsfrom2007-2021tostudytheeffectsoftransitoryemotionalstatesonhouseholds,borrowingdecisions,therealconsequencesofthosedecisions,andtheefficacyoffeaturessuchastcooling-ofP,periodsinmitigatingtheemotionaleffects.Asasourceofexogenousvariationinconsumers,sentimentthatmatchesthehighfrequencyofloanapplications,weexploithourlyvariationinlocalsunshineacross2,482countiesduringtheperiod2007-2021.Thisapproachisgroundedinpriorevidenceontheeffectofsunshineonanagent'smoodfrompsychology(SchwarzandClore,1983),experimentaleconomics(Bassi,Colacito,andFulghieri,2013),andnancialmarkets(HirshleiferandShumway,2003;Goetzmann,Kim,Kumar,andWang,2015).Akeyempiricalchallengeistoseparatetheeffectofsentimentonhouseholds,borrowingdecisions,orcreditdemand,fromitseffectoncreditsupplyandlocaleconomicconditions.Indeed,priorstudieshaveshownthatsunshineaffectsbothcreditsupply(Cortesetal.,2016)andeconomicexpectations(Chhaochhariaetal.,2019).Ourempiricalsettinghasseveralfeaturesthatallowustoovercomethischallenge.First,thedatacontainloanapplicationsirrespectiveoftheireventualoriginationorfundingstatus,thuscapturinghouseholds5creditdemandratherthancreditsupply.Second,thetestspecificationsmatcheachapplication'sgranulartimestampwithhourlyvariationinsunshinewithinacounty-week,thusholdingconstantlocaleconomicconditionsandremovingseasonalvariationinsunshineforagivencounty.Third,bydesign,allcreditdecisionsontheonlinemarketplacelendingplatformarebasedonanalgorithmiccreditmodel,andtheinvestorsarenonlocalandinstitutional.Assuch,thesupplyofcreditontheplatformisunrelatedtovariationinlocalsunshine.Weconfirmthathourlyvariationinsunshinedoesnotaffectcreditsupplybystudyingloanpricing,riskassessment,andfunding.Consistentwithouridentifyingassumption,wefindthatsunshineisUncorrelatedwithloaninterestrates,theplatfrm,sestimatedlossrate,ortheproportionoftheapplicationthatisfunded.Theseresultssuggestthatvariationinlocalsentimentdoesnotaffectloanoriginationorloanterms,norisitaccountedforbytheplatformorinvestors.Ourmainfindingscanbesummarizedasfollows.First,positivesentiment,attributabletohourlyvariationinlocalsunshine,correspondstohighercreditdemandbothattheextensiveandintensivemargins.Attheextensivemargin,wefindthatthenumberofapplicationsis2%higherduringsunnyhourscomparedtocloudyhours.Attheintensivemargin,wefindthatrequestedloanamounts,loan-to-incomeratios,andmonthlypayment-to-incomeratiosincreaseby1.3%,1.3%,and1.1%,respectively,duringsunnyhours.Combined,theseresultssuggestthatsentimentoperatesthroughboththeextensiveandintensivemargins.Theabovefindingsholdaftertheinclusionofcounty-by-weekfixedeffects,whichabsorbweeklyvariationineconomicconditionsspecifictoeachcounty,aswellascreditrating,loanpurpose,hour,andday-of-weekfixedeffects,whichabsorbvariationacrossborrowercreditquality,loantype,time-of-day,andweekday,respectively.Theanalysesalsocontrolforawiderangeofborrowers,characteristics,suchasemploymentduration,incomelevel,priorplatformexperience,andparticipationontheplatformasalender.Assuch,Weprovidenovelcausalestimatesofabehavioralcreditdemandchannel,augmentingrecentstudiesthathavemostlyfocusedontheimplicationsofbehavioralfactorsforcreditsupply,includingpersonalconnections(Engelberg,Parsons,andYao,2012),theperceptionofborrowertrustworthiness(Duarte,Siegel,andYoung,2012),andmostrelatedtoourstudy,sunshine-inducedsentiment(Cortesetal.,2016). A related literature examines the effect of sunshine-induced sentiment on other consumer decisions such as car choice (Busse et al., 2015) housing prices (Hu and Lee, 2020), and credit card spending (Agarwal et al., 2020).Incontrast,wefocusoncreditdemandinasettingthatholdsconstantcreditsupplyandeconomicconditions.Ourfinding,thatsentimenthasconsiderableimplicationsforcreditdemandintheFinTechconsumerloanmarketplace,differsfrompriorevidencethatsentimentdoesnotaffectcreditdemandinmoretraditionalcreditmarkets(e.g.,Cortesetal.,2016).ThesefindingshighlighttheroleOftraditionalloanmarketfeatures,suchasliveinteractionswithloanofficers,inmitigatingtheimpactofsentimentoncreditdemand.Second,wefindthatloanapplicationsinitiatedonsunnyhoursaresignificantlymorelikelytobecharged-offcomparedtothoseinitiatedonovercasthoursduringthesameweekinthesamecounty.Inparticular,loanapplicationsinitiatedduringsunnyhoursare0.39percentagepointsmorelikelytobecharged-off,or1.49%relativetothesamplestandarddeviation,comparedtothoseinitiatedduringcloudyhours.Thesefindingsshowthatsentimenthasrealeffectsonhouseholds'financialoutcomes.Third,wefindconsiderabledemographicdifferencesintheeffectsofsentimentoncreditoutcomesacrossincomegroups.Specifically,wefindthatloansinitiatedbylow-incomeindividualsduringsunnyhoursareroughly1.4percentagepointsmorelikelytobecharged-off,orabout5.3%relativetothesamplestandarddeviation,comparedtothoseinitiatedduringcloudyhours.Incontrast,wedonotfindstrongsentimenteffectsforhigh-incomeborrowers.Theresultssuggestthatlow-incomeborrowersarelesscapableofbearingthefinancialburdenofsentiment-drivenloans,andsubsequentlyexperiencenegativeeconomicconsequences?Fourth,weinvestigatetheroleofpreviousexperienceandcooling-offperiodsinmitigatingtheeffectofsentimentoncreditdemand.Consumerprotectionadvocatespointtobehavioralresearchtojustifyregulationsinfinancialmarkets.Onesuchregulationisatcooling-offperiod,whichallowsborrowerstowithdrawfromfinancialcontractswithinacertaintimewindow.ThalerandSunstein(2008)notethatcooling-offperiods“makebestsense,andtendtobeimposed,whentwoconditionsaremet:(a)peoplemaketherelevantdecisionsinfrequentlyandthereforelackagreatdealofexperienceand(b)emotionsarelikelytoberunninghigh.”ConsistentwiththeThalerandSunstein(2008)view,wefindthattheeffectofsunlightonrequestedloanamounts,loan-to-incomeratios,andcharge-offsisconcentratedinfirst-timeborrowers,anddisappearsforborrowerswithpriorplatformexperience.Inparticular,loanapplicationsinitiatedduringsunnyhoursbyfirst-timeborrowersare1.3%largerandhave1.12-1.4%higherloan-to-incomeratioscomparedtoapplicationsinitiatedduringcloudyhours.Wealsofindsubstantialnegativeoutcomesforinexperiencedborrowers,whoexperiencea0.62percentagepoints,or2.4%,increaseincharge-offratewhentheybegintheirapplicationduringsunnyhours.Incontrast,sunny-hourapplicationsinitiatedbyexperiencedborrowersareindistinguishablefromcloudy-hourapplications.3Theseindividualsmayalsobetheleastfinanciallyliterateandhencemostsusceptibletosentiment(see,i.e.,Campbell,2(X)6;ThalerandSunstcin,2008;LusardiandMitchcll,2014;RuandSchoar,2016).4Wealsofindthatexperiencedapplicantsareconsiderablymorelikelythanfirst-timeapplicantstowithdrawasunny-hourloanlistingbeforeaccessingthefunds.Whileexperiencedborrowersare1.78%morelikelytowithdrawasunny-hourloanapplicationcomparedtoacloudy-hourloanapplication,first-timeborrowersarenoteconomicallyorstatisticallysignificantlymorelikelytowithdrawsuchapplications.Further,experiencedapplicantswhosepreviousapplicationwasinitiatedduringasunnyhouraremorelikelytowithdrawacurrentloanlistingcomparedtothosewhosepreviousapplicationwasinitiatedduringacloudyhour.Assuch,ourresultssuggestthatindividualsIeamfrompreviousexperience,andthatmandatingacooling-offperiodprovidesbenefitsintheconsumercreditmarketbypotentiallyprotectinghouseholdsfromirrational,sentiment-baseddecisions.1.astly,wefindthatsentimentinfluencesthecompositionofloans.Localsunshineincreasesthedemandfordiscretionaryloanssuchasthoseforlargepurchases,vacations,orhomeimprovementby1.1%.Incontrast,itdoesnotaffectbusinessloanapplicationsandreducesthedemandfordebtconsolidationby1.63%.Theseresultsareconsistentwithashifttoward“impulse”creditexpansionforhouseholdswhensentimentishigh.Thiseffect,however,doesnotspillovertobusinesscreditdemand.Thesefindingscomplementresearchinexperimentaleconomicsthatstudiesconsumptioncompositioninconsumermarkets(see,e.g.,Busseetal.,2015).OurpapercontributestothegrowingliteratureontherealeffectsofFinTech.Onthepositiveside,technologymayimprovefinancialeducation(Breza,Kanz,andKIapper,2020),enhanceconsumption-smoothingandrisksharing(JackandSuri,2014;Suri,2017),increasefinancialattention(StangoandZinman,2014;Karlan,Morten,andZinman,2017;Bursztyn,2019;Medina,2021),orrelieveinformationfrictions(Carlin,Olafsson,andPagel,2023).SeveralstudiesspecificallyexaminefutureoutcomesOfFinTechborrowersandfindmixedresults.Balyuk(2023)5findsthatFinTechborrowingprovidesinformationspilloversthatfacilitatefuturecreditaccessfrombanks.Conversely,Chavaetal.(2021)andDiMaggioandYao(2021)findnegativelongtermconsequencesofFinTechborrowingintermsoflowercreditscores,highercostsofcredit,andhigherdefaultrates.WangandOverby(2022)findthatFinTechborrowersoverconsumeloansfrommarketplacelenders,leadingtobankruptcy.Ourresultssuggestthattheeaseofaccessingconsumercreditheightenstheeffectoftransitorysentimentonfinancialdecisions,potentiallyleadingtooverconsumptionofcreditandnegativefutureoutcomes.Ourpaperalsocontributestotheliteratureonhouseholdparticipationindebtmarkets.Priorstudiesrelyprimarilyonsurveys,andprovideevidenceonsocio-demographicvariables,economicvariables,anddeviationsfromoptimalchoice(e.g.,CoxandJappeIli,1993;DucaandRosenthaL1993;Gropp,Scholz,andWhite,1997;LeaandWebley,1995;Leece,2000;GrahamandIsaac,2002;Karlsson,Dellgran,Klingander,andGarlin,2004;Brown,Taylor,andWheatleyPrice,2005;Easterlin,2005;Magri,2007;Siemens,2007;DelRioandYoung,2006,2008;Ranyardetal.,2006;MeierandSprenger,2007,2010;Rohde,2009;Etzioni,2010.).Weaddtothisliteraturebyprovidingevidencefrommicro-levelobservationaldataonhighfrequencycreditdemanddecisionsofhouseholds.Assuch,ourworkisrelatedtoBen-DavidandBos(2021),whouseobservationalSwedishdata,andfindthatanincreaseintheavailabilityofliquorincreasescreditdemand,default,welfaredependence,andcrime.1.astly,wealsoaddtothegrowingliteratureontheroleofsentimentinfinancialmarkets.Earlyworkinthisfieldexaminestheeffectofsunshine-inducedmoodonthestockmarketandontradingbehavior(Suanders,1993;Kamstra,Kramer,andLevi,2003;HirshleiferandShumway,2003;Goetzmannetal.,2015).Weaddtothisliteraturebystudyinghouseholdcreditdemand.62. InstitutionalDetailsandDataThissectiondescribestheinstitutionaldetailsoftheloanapplicationprocessofthemarketplacelender(ProsperMarketplace),aswellasthetheoreticalmotivationforsunshineasamoodprimer.Additionally,thissectiondetailsthedatasourcesandvariablesWeuseinthisstudy.2.1. ProsperMarketplaceTheempiricalanalysesfocusonFinTechconsumerloanapplicationsfromtheuniverseOflistingsprovidedbyProsperMarketplace,thesecondlargestonlineconsumerlenderintheUnitedStates. Prosper has issued over $21 billion in loans to over 1.2 million people since 2005.Toinitiatealoanapplication,individualsstatetheirdesiredloanamount(upto$40,000)andthepurposeoftheloan. Loan purpose is separated into 10 main categories. The most common loan purpose is debt consolidation, whichmakes up roughly 70% of the applications in our data.Theapplicantmustthenprovidehername,address,anddateofbirth,alongwithheroccupation,income,homeownershipstatus,employmentstatus,andothergeneralcredithistoryterms.Theplatformthenperformsasoftcreditinquiry,andgeneratesasetofpotentialloanpackageswitheithera3-or5-yeartermfromwhichtheborrowerchooses.Oncetheapplicantacceptsthedesiredloanterms,theplatformconductsahardcreditinquiry,andusesaproprietarycredit-scoringmodeltoevaluatetheriskinessofeachloan.Theplatformmayalsorequestadditionaldocumentationtoverifycertainitemsontheborrower'screditprofileandprovidedinformation.ThelistingthengoespublicontheProsperplatformandisavailabletobefundedbyinvestors.Theentireapplicationprocesstakesonlyminutes,andtheplatformapprovesmostloansandprovidesfundingwithin1day.Upongoingpublic,alistinghastwoweekstobefunded,andmaybefundedbyindividualsorinstitutionalinvestors.Inpractice,theloansareoverwhelmingfundedbyinstitutionalinvestorsusingpassivemeansbasedonobservablequantitativevariablesfromthecreditbureau(BalyukandDavydenko,2019).Atanytimepriortofunding,aborrowermaywithdrawherapplicationatnocost.Investorsobservetherequestedamount,loanpurpose,andtheborrower,screditcharacteristics,butnotthepersonalcharacteristicsoftheborrower,suchastheborrowers*preciselocation.Investorsthendecidewhethertofundtheloanpartiallyorcompletely.Oncefundingiscomplete,contingentontheborrower'sapproval,theloanisoriginated,andthefundsaredepositedintheaccountoftheborrower.Borrowersincurfeesandnotificationsofpastdueaccountsontheircreditreportsiftheyareunabletomakethenecessaryloanpayments.Iftheborrowermissesfiveconsecutivemonthlypayments,theloanisconsidered“charged-off,“andtheentireloanbalanceisdueimmediately(