FinTech and Banks: Strategic Partnerships That Circumvent State Usury Laws

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) FinTech and Banks: Strategic Partnerships That Circumvent State Usury Laws Gregory Elliehausen and Simona M. Hannon 2023-056 Please cite this paper as: Elliehausen, Gregory, and Simona M. Hannon (2023). “FinTech and Banks: Strategic Partnerships That Circumvent State Usury Laws,” Finance and Economics Discussion Series 2023-056r1. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2023.056r1. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

FinTech and Banks: Strategic Partnerships That Circumvent State Usury Laws* GregoryElliehausen‡ SimonaM.Hannon§ Retired FederalReserveBoard September22,2023 Abstract Previousresearchhasfoundevidencesuggestingthatfinancialtechnology(FinTech)lenders seek out opportunities in markets that have been underserved by mainstream banks. The research focuses primarily on the effect of bank market structure, limited income, and economichardshipinattractingFinTechcompaniestounderservedmarkets.Thispaperexpands thescopeofFinTechresearchbyinvestigatingtheroleofinterestrateregulationofconsumer creditandinstitutionalrisksegmentationinFinTechlenders’effortstosolicitnewcustomers inthepersonalloanmarket. WefindthatstrategicpartnershipsbetweenFinTechcompanies and specialist banks target marginal-risk, near-prime, and low-prime consumers for credit card and other debt consolidation loans. These FinTech-bank partnerships especially target marginal consumers in states with low interest rate ceilings. Mainstream banks largely avoidhigher-riskconsumers, andlowrateceilingsinhibitconsumerfinancecompanylending,whichhistoricallyhasbeenthemajorsourceofpersonalloansforhigherriskconsumers andmaycompetewithbanksatthemargin. Inpartneringwiththespecialistbanks,theFin- Techlendersareabletotakeadvantageoffederalpreemptionsfromstaterateceilingstolend profitablytohigher-riskconsumersinstateswithlowrateceilingstocompeteinthesemarkets. Keywords: Consumer Credit, Access to Credit, Interest Rate Cap, Financial Regulation, Fin- Tech. JELclassification:G21,G23,G4 *WethankHannahCase,JessicaFlagg,andLucasNatheforsuperbresearchassistance,MollyGraebnerandAnuj ShahaniforgeneroushelpwithMinteldata,RobertAdams,FabioBraggion,TimDore,JohnDriscoll,HarryP.Huizinga, ThomasLambert,GengLi,MichaelPalumbo,DamjanPfajfar,andR.BurakUrasforhelpfulcommentsandsuggestions, andDavidJenkinsandChristopherKarlstenforoutstandingediting. Theviewsinthispaperarethoseoftheauthors anddonotnecessarilyreflectthoseoftheBoardofGovernorsoftheFederalReserveSystemoritsstaff. ‡ThefirstdraftofthispaperwascompletedwhileGregoryElliehausenwasaPrincipalEconomistattheBoardof GovernorsoftheFederalReserveSystem. §Correspondingauthor. Address: BoardofGovernorsoftheFederalReserveSystem,20thStreetandConstitution AvenueNW,Washington,DC20551,USAEmail:simona.m.hannon@frb.gov.

1. Introduction Financial technology (FinTech) company partnerships with specialist banks commonly market personalloansforcreditcardorotherdebtconsolidation. TheseFinTech-bankpartnershipsuse onlineplatformstofacilitateoriginationofloansthroughapartnerbankanduseproprietaryunderwriting algorithms to supplement conventional underwriting criteria. Over the past decade, thenumberofFinTechcompanieshasincreasedalongsideloanvolume. EvidencefrompreviousresearchsuggeststhatFinTechlendersseekopportunitiesinmarkets that have been underserved by mainstream banks. These studies focus primarily on the effect of bank market structure, low income, and economic adversity in attracting FinTech companies tounderservedmarkets. Forexample, Buchaketal.(2017)findthatFinTechpenetrationinresidentialrealestatelendingispositivelyassociatedwithlargerminoritypopulations, lowerunemployment, and higher market concentration. Havrylchyk et al. (2020) find that FinTech lenders expanded in areas with lower-density branch networks. Jagtiani and Lemieux (2018) find that a largeFinTechfirm’s(LendingClub)originationsofpersonalloansfordebtconsolidationwereassociated with greater concentration in credit card lending, fewer bank branches per capita, and weakerlocaleconomies. Cornaggia,Wolfe,andYoo(2018)findthatFinTechpresencewasassociatedwithlowerpersonalloanvolumeatbanksinlesscompetitivemarkets,whilevolumeatbanks inmorecompetitivemarketsappearedlargelyunaffected. Tang(2019)findsthatconsumerswho areconsideredmarginalbybanksandconsumersseekingsmallamountsofcreditaremostlikely to benefit from credit expansion opportunities offered by FinTech firms. Examining data from a largeFinTechfirm, Balyuk(2019)arguesthatFinTechfirmsreduceinformationimperfectionsin consumercreditmarkets,leadingtobetterconsumeroutcomes,includingreducedbankrents,increasedspeedandconvenienceinobtainingcredit,andbroadenedfinancialinclusion.DiMaggio and Yao (2020) find that FinTech borrowers seeking a personal loan tended to have lower credit scoresthanbankborrowersbutgenerallywerenotsubprime. FinTechlenderssoughtoutriskier consumers than banks at first but later expanded their market share by extending credit to less riskyconsumers.1 Theunderservedmarketinvestigatedbythisstudyisthepersonalloanmarketforriskyconsumers.MostpreviousstudiesfocusonFinTechfirms’andbanks’provisionofpersonalloansused for debt consolidation and other purposes. They have not considered personal loans from consumer finance companies, which have traditionally been the major source of personal loans for higher-risk consumers (National Commission on Consumer Finance, 1972; Durkin et al., 2014). Mainstreambanksgenerallyhaveavoidedhigherriskconsumers. Thissegmentationispartlythe consequenceofhistoricalrisktolerance,butregulationhasalsoplayedarole.Insomestates,consumer finance companies commonly had higher interest rate ceilings and lower loan size limits thanbanks. Thisdifferencemeantthatsmallpersonalloanstohigher-riskconsumerstendedto 1SimilarresultshavebeenfoundforFinTechparticipationinthesmallbusinesscreditmarket.Forexample,seeErel andLiebersohn(2020)orBalyuketal.(2020). 1

bealmosttheexclusiveprovinceoffinancecompanies,andbanksmadelargerloanstolower-risk consumers (Rogers, 1975; Durkin et al., 2014; Durkin, Elliehausen, and Hwang, 2016).2 In other states,wherelowrateceilingsmadehigh-riskandsmalldollarloansunprofitable,riskyconsumers experiencedrationing. FinTechfirmsoperatebothaloneasfinancecompaniesorinpartnership withabank. Wheretheyfitintheinstitutionalstructureofthepersonalloanmarkethasnotbeen fullyexamined. Despite the tendency to segment the market based on risk, the risk profiles of bank and finance company customers, to some extent, overlapped (Boczar, 1978; Durkin and Elliehausen, 2000), suggesting that these sources competed at the margin. Federal regulation permitting national banks and banks insured by the Federal Deposit Insurance Corporation (FDIC) to charge interest rates allowed by their home states facilitated interinstitutional competition but did not eliminaterisksegmentationinthemarketforpersonalloans. Manybankscontinuedtobereluctanttolendtohigher-riskconsumers. Thefederalregulationsthatallowedabanktochargeratespermittedinitshomestateprovided anopportunityforsomebankslocatedinhigh-orno-rate-ceilingstatestopartnerwithFinTech companies. Partnerbankscouldlegallychargeratesthatreflecttheriskofloans, whichenabled themtooriginateloanstoriskierconsumersinlow-ratestates. Whenpartneredwithabank,Fin- Techcompaniesareabletoaccesslow-rateceilingmarketsthatotherfinancecompaniesfindunprofitable. Thispaperinvestigatestheinfluenceofinterestrateregulationoncreditavailabilityformarginal consumers in personal loan markets. Using data on solicitations (an indicator of credit supply), wefindthatFinTech-bankpartnershipsheavilytargetednear-andlow-primeconsumersinstates with restrictive interest rate ceilings. The partnerships did not heavily solicit high-prime consumers, regardless of rate-ceiling regulation. They also appeared to have little interest in subprime consumers in high-rate states. However, FinTech-bank partnerships moderately solicited subprimeconsumersinstateswithlowrateceilings,likelybecausetheyfacedrelativelylittlecompetitioninthesestates. Asfinancecompanieswithoutbankpartnerscannotoperateprofitablyin low-ratestates,ourresultsshowthatfinancecompaniesheavilysolicitedsubprimeconsumersin high-ratestates.Incontrast,banksshowedlittleinterestinconsumersinanypartoftheriskspectrum other than the high-prime part. Notably, our data also allow us to confirm payday lenders little-known,butsignificantparticipationwithintheinstallmentloanspace. The remainder of this paper proceeds as follows. Section 2 discusses how state interest rate ceilingsaffectthestructureofconsumercreditmarkets.Section3providesdetailsonourresearch designanddatausedinthisstudy. Section4discussesourempiricalanalysisandresults. Section 5concludes. 2Smallerloansarealsorelativelymorecostlytoproducethanlargerloansbecausemuchofthecostofproducinga loanisfixed.Costsareincurredlargelybecausealoanismadeandarenotespeciallysensitivetotheamountoftheloan. Consequently,breakeveninterestratesarehigherforsmallerloansthanforlargerloans.Thisfactpreventsbanksfrom makingsmallerloansizes,whichriskyconsumersarebetterabletoaffordthanlower-riskbankborrowers. SeeChen andElliehausen(2020). 2

2. RateCeilings,CreditAvailability,andtheEmergenceofFinTech-Bank Partnerships Interest rate ceilings have existed long before the emergence of FinTech-bank partnerships.3 Whenarateceilingislessthanthemarketrate,creditisoftenrationed. Lendersmayconsolidate officesorexitthemarketcompletely. Themarketforpersonalloanshasbeenespeciallyexposed torisksofrateregulation(NationalCommissiononConsumerFinance,1972;Durkinetal.,2014). Theemergenceofnon-banklender-bankpartnershipsisamarketresponsetolimitationsresultingfromregulatoryrestrictions.Thismodelwasusedinthe1990sbypaydaylender-bankpartnershipstocircumventrestrictivestateregulations(Stegman,2007),anditcontinuestodaywiththe FinTech-bankpartnershipsexploredinthispaper.4 Foranoverviewof"rent-a-bank"partnerships throughthelegallensseeLevitin(2021). 2.1. ABriefHistoryoftheStateRegulationofConsumerCredit Historically,consumercreditintheUnitedStateshasbeenregulatedbythestates,whoseprimary concerns were thepriceofcredit, nonprice terms, andcreditorconduct.5 States regulated creditpricesthroughinterestrateceilings. Statecreditprice-ceilinglawsgenerallyincludedusury laws and a variety of special laws allowing higher rates than those allowed under usury laws for specifictypesofcreditfromcertainclassesoflenders.Statesalsohopedtoinfluenceothersignificantcharacteristicsofcreditofferingsinthemarketplacebyrestrictingmarketentryto“legitimate” lendersthroughlicensingandotherrequirements. Statelawsfurtherprovidedforlimitsoncreditors’rightsinthecaseofdefault(creditorremedies). Stateregulationofinterestrates—inparticular,theestablishmentofceilingsforinterestrates— hasunquestionablyexertedatremendousinfluenceonthedevelopmentofconsumercreditinstitutionsandmarketsintheUnitedStates.Restrictionsonentryandcreditors’remedieswerealways closelyconnectedandjustifiedinthesamewayasrateceilings,aspartofaregimeofcontrolling thepriceandcharacterofcreditservice.Theunderlyingrationalefortheregulationsincluded,first andforemost,attemptingtoprotectunsophisticatedborrowersfromunneededcredit,uninformed 3Changesinstaterateceilingsareinfrequent.Politicalmeasuresandeconomicconditionshavelittleornoeffecton ceilingrates(seeGlaeserandScheinkman,1998). 4Otherknownmodelsofstrategicpartnershipsincludethosebetweenspecializedfinancecompaniesandsmall bankswiththepurposeofcircumventingstatelaws(Bhattacharyya,2021),thosebetweentaxpreparationfirmsand banksfortheissuanceoftaxrefundanticipationloans(theloansarerepaidbyacustomer’staxrefundandareissued throughthetaxpreparationfirminpartnershipwithabank,withtheconsumerbeingrequiredtofilethetaxreturns electronically(Duffy,2004;Millerbernd,2021)),thosebetweencryptolendersandbanks(Crosman,2018),andthose betweencreditunionsandeitherFinTech(Gargano,2021)orbuynow,paylater(BNPL)companies(Adams,2021).Aside fromthe"financialinstitutionwithfinancialinstitution"model,embeddedfinanceevolvedtoaddressvariousmarket needs.Forexample,TabBankinUtahwasfounded23yearsagoasanaffiliateofachainoftruckstopswiththepurpose ofprovidingfinancialsupporttotruckersontheroad(Crosman,2021). 5UntilpassageofthefederalTruthinLendingAct(TILA)in1968,thestatesweretheprimaryregulatorsofconsumer creditintheUnitedStates. TILA’sconcernwasmostlydisclosureratherthanregulationofcreditpricesorterms. With thepassageoftheoriginalEqualCreditOpportunityActin1974,thefederalgovernmentsoughttoensurefairnessin creditgrantingandpromotecreditavailability,especiallyforhistoricallyunderservedconsumers. 3

useofcredit,andharshcreditterms.Theyalsowereintendedtoredressunequalbargainingpower arisingfromborrowers’urgentneedorlenders’marketpowerandtodiscourageprodigalspending andexcessiveindebtedness. Despite relaxation of many state interest rate ceilings for consumer credit in the 1970s and 1980s, many states still retained interest rate restrictions on consumer credit. Significant in relaxingtheinfluenceofinterestrateceilingsonconsumercreditduringtheseyearswastheriseof interstatecompetitionandSupremeCourtdecisionsonpermissibleratesforinterstatebanks. In1978,theU.S.SupremeCourtinMarquetteNationalBankv. FirstOmahaServiceCorporation ruled that national banks could charge interest rates permitted by the lending bank’s home stateregardlessoftheratepermittedbytheborrower’sstateofresidence. Untilrecently,thissignificantcourtrulinghaditsgreatesteffectonthecreditcardmarket,asitenabledcreditcardcompanies to expand their offerings geographically to consumers located across states with various interest rate ceilings. As a result of the Marquette ruling, credit card companies moved to states withhighornocreditcardrateceilings,andthecreditoperationsofmanylargeretailstoresand consumerfinancecompanieswereacquiredbyorotherwisebecameaffiliatedwithnationalbanks and their subsidiaries. We observe a similar phenomenon with FinTech companies. In the currentenvironment,astherestrictiveinterestrateceilingsinsomestateslimittheiraccess,FinTech companiesstrategicallypartnerwithbanksinordertoaccessthosemarkets. Theeffectsofinterestrateceilingsinpersonalloanmarketshavebeenlargelydiscussed,and a significant amount of evidence developed over many years suggests several conclusions about them(Durkinetal.,2014): •Stateinterestrateceilingsrestrictedcreditavailabilitywhensetatlevelsthatwerelowerthan equilibrium market rates for higher-risk borrowers. When rate ceilings were set at higher levels, higher-riskborrowerswerelesslikelytoexperiencereductionsincreditavailability. •Differentrateceilingsfordifferentinstitutionaltypesoflenderstendedtosegmentconsumer credit markets (Rogers, 1975), with lenders that had higher ceilings (until recently, finance companies) lending to higher-risk borrowers more frequently than institutional types that had lower ceilings(banks). Suchrestrictionstendedtoreinforcebanks’preferencetospecializeinthelowrisksegmentofthemarket. •Despitemarketsegmentation,empiricalriskdistributionsofbankandfinancecompanyborrowerspartlyoverlapped(Boczar,1978). Thatmanyfinancecompanyborrowershadriskcharacteristicssimilartothoseofbankborrowerssuggeststhatthesefinancecompanycustomersmay havebeenabletoobtainloansfrombanksandthatbanksandfinancecompaniesmayhavecompetedover(atleastsomepartof)theriskdistributionofconsumers. Asresearchersdidnothave allvariablesrelatedtorisk,thisresultisonlysuggestive. •Observedinterestrateswerenotgenerallyatthestaterateceilingsunlessceilingrateswere already low (National Commission on Consumer Finance, 1972). Average interest rates for unsecured installment loans charged by banks in high-ceiling states were not much different from 4

averageratesinlow-ceilingstates, andinterestratesforconsiderablesharesoffinancecompany personalinstallmentloanswerebelowrateceilings. 2.2. TheSpecialCasesofArkansasandIowa Arkansasisapopularplacetostudyeffectsofrateceilingsbecauseofitsverylowrateceiling. Unlikemoststates,Arkansas’susurylimitisconstitutional,notstatutory,andiscurrentlysetat17 percent. In1957,theArkansasSupremeCourtaffirmedthatallformsofcreditinthestateweresubject tothe10percentusuryceilinginthestateconstitution,regardlessofactionsthelegislaturemight take(Sloanv. Sears,228Arkansas464,308S.W.,2d8021957). Morerecently,Arkansashasrevised its state constitution and usury law, but the revised law remains restrictive compared with most other states. Today, the ability of Arkansas residents to obtain revolving credit from out-of-state creditcardcompaniesheadquarteredinstateswithahighrateceilingornoceilingvitiatesinlarge partthestate’srestrictiveusurylaw. However, higher-riskconsumersmayhavedifficultyaccessingrevolvingcredit. Therate-ceilinglimitstheseconsumers’accesstopersonalinstallmentcredit fromfinancecompanies.LukongoandMillerJr.(2018)showthatthenumberofconsumerfinance company personal loans in Arkansas is lower than the number of such loans in the neighboring states. As no consumer finance company offices are located in Arkansas, the prevalence of consumerfinancepersonalloansinbordercountiesofArkansasandavirtualabsenceofsuchloans ininteriorcountiessuggestthatArkansasconsumersinbordercountiestraveloutofstatetoget theloans,and,asaresult,thereisa“creditdesert”inthemiddleofthestate. Still,therestrictive effectisclearlyevident.RecentworkbyElliehausen,Hannon,andMillerJr.(2021)usingadifferent datasource,theFederalReserveBankofNewYork(FRBNY)ConsumerCreditPanel(CCP)/Equifax creditbureaudata,supportsthepreviouslydiscussedfindingsoncreditavailabilitywhenrateceilingsarelow. Iowaprovidesanotherusefulcomparisongroupforstudyingthegeographicalconsequences oftheMarquetteruling. IowaistheonlyU.S.statethatoptedoutoftheinterestrateexportation regimeprovidedbytheMarquetteruling,byinvokingtheexceptiontothefederalpreemptionunderSection525oftheDepositoryInstitutionsDeregulationandMonetaryControlActof1980(see BrennanandUdell,2018).6 AsIowaisalowinterestratestatelocatedbetweenthreehighinterest ratestates(Wisconsin,Illinois,andMissouri)andthreelowinterestratestates(Minnesota,South 6PuertoRicoalsooptedoutofthecoverage,andseveralstates—Colorado,Maine,Massachusetts,Nebraska,North Carolina,andWisconsin—previouslyoptedoutofthecoverageaswellbuthaveeitherrescindedtheiropt-outorletitexpire.(Seefootnote18onpage44148oftheOfficeoftheFederalRegister,NationalArchivesandRecordsAdministration, 2020.) 5

Dakota,andNebraska),IowaprovidesacounterfactualforanydevelopmentsenabledbytheMarquetteruling.7 2.3. EffectsofFinTechPresenceinConsumerCreditMarkets TheemergenceoftheFinTechsectoraddedanotherlayerofcomplexitytothepersonalloan market. Ascreditissuedbytraditionalfinancialinstitutionssuchasbanks, thrifts, creditunions, and finance companies decelerated in the aftermath of the financial crisis, the FinTech sector emergedtoprovidecreditorrefinancedebtsofborrowerswhomightotherwisebepricedoutof themarket. FinTechcompaniestargetlow-primeandnear-primeborrowerswhoarelessattractive for banks and who may be able to obtain better loan terms from companies outside of the financecompanyuniverse. Giventhenascentnatureofthefield,researchresultsontheeffectsof FinTechpresenceinconsumercreditmarketsaremixed. Adams(2018)findsthatconsumerloans issuedbyFinTechlendersareprimarilyusedforrefinancingvarioustypesofconsumerdebt.Dore and Mach (2019) show that at Prosper, borrowers’ credit scores initially increase after they take out loans, and their credit card utilization rates fall relative to nonborrowers. In the longer run, however, total debt levels for Prosper borrowers are higher than those of otherwise similar nonborrowers,butdelinquencyratesarelowerforborrowers. DanisewiczandElard(2018)showthat restrictingFinTechpresenceinaconsumercreditmarketisassociatedwithasignificantincrease inpersonalbankruptcy. ThisresultfollowedtheMaddenv. MidlandFunding courtverdict,which ruledthatabove-usuryloansissuedbybankstoresidentsofConnecticutandNewYorkwerenull andvoidiftheloanswereimmediatelysoldtonon-bankentities. Theirfindingssuggestthatrestrictingmarketplacelendingdelaysfilingforbankruptcyandconsistentlyhelpssomehouseholds avoidbankruptcy. Asinterestrateceilingslimitnon-partneredFinTechpresenceinsomestates,FinTechcompaniescancircumventtheusuryratelegislationandpartnerwithspecialistbanksinordertoavoid these states’ low(er) interest rate ceilings. As a result of this arrangement, the FinTech sector is segregated into (1) companies that function more like finance companies and end up being restrictedinlow-ratestatesand(2)companiesthatplacethemselvesinapositiontotakeadvantage ofweakercompetitionfromfinancecompaniesandotherFinTechsinsuchstates. 2.4. FinTech-BankPartnershipMechanism The FinTech-bank partnership opportunity begins when state usury rate ceiling restrictions andtheMarquetterulingconverge. AftertheMarquetteruling,banksbecameabletoexporttheir homestateinterestratestomoststates,irrespectiveofthestate’sinterestrateceiling.8 Incontrast, 7Iowalawallowsamaximuminterestrateof5percentunlessadifferentrateisagreeduponinwriting.Inthatcase, theinterestratecannotexceedthestate’sofficialusuryratesetbythesuperintendentofbankinginaccordancewith theprovisionsofIowaCodeSection535.2(3)(a),consistingofthemonthlyaverageofthe10-yearTreasuryrateplusa2 percentpremium. 8Iowaisanexceptiondiscussedearlier. 6

FinTech companies operating alone remained limited by the interest rate ceiling and found that doingbusinessinstateswithlowinterestrateceilingswaslessprofitablethaninhigh-orno-rate ceiling states. As result, some FinTech companies have partnered with banks to circumvent the usury rate legislation by offering joint personal loan offers. When these partnerships occur, the bankbecomesthe“truelender”bymakingtheloanonbehalfoftheFinTechcompany.9 Thebank holds onto the loan for a number of days or even months and may retain a certain percentage of the loan production. For example, LendingClub reveals in its 2019 third quarter 10-Q report thatWebBankholdsontoitsloansfortwobusinessdays, afterwhichtheloansarepurchasedby themarketplacelenderatparplusaccruedinterest.10 CrossRiverBankretains5to10percentof themonthlyloanproductionandtendstoholdontotheloansforsixtoninemonths. Loansare then sold back to the FinTech company, which in turn sells them in either the private or public securitizationmarket(seeScully,2015). Importantly, in order for these partnerships to be profitable, the banks that are typically involvedinjointofferings,thespecialistbanks,needtobeeitherexemptfromrateceilingsorlocated instateswithhighornoconsumerfinancerateceilings.InthenextsectionwenotehowWebBank, oneofthemajorspecialistbanksengagedinpartnerships,hasaUtahindustrialloanbankcharter, thatexemptsitfromstatelenderlicensinglaws,interestrateceilings,andmoneyservicebusiness laws,andCrossRiverBank,theothermajorspecialistbankengagedinpartnerships,islocatedin NewJersey, astatethatallowsitscharteredbankstochargeuptoamaximuminterestrateof30 percent. Thesearrangementscanbemutuallybeneficial.FinTechcompaniesareabletoaccessmarkets previouslyconsideredinsufficientlyprofitableandearntransactionfeesfortheirroleinprocessing theloanapplications.LendingClubdisclosesinits2019third-quarter10-Qreportthattheamount offeeschargedisbasedonthetermsoftheloan, includinggrade, rate, term, andchannel—and thatasofSeptember30,2019,thesefeesrangedfrom0to6percentoftheinitialprincipalamount of a loan.11 In addition, FinTech lenders benefit from their partner banks having the regulatory, compliance, and licensing frameworks in place in the respective states in which they choose to operate. Banks’ revenues in such a partnership mainly come from loan issuance fees. For example, CrossRiverBankchargesaprocessingfeebetween30and100basispointsperloan,conditional onthevolumeofloansprocessed;ontheloansitretains,thebankisgenerallyopentoarevenuesharestructurewiththeplatformsuchthatiftheplatformelectstoworkwithanotherbank,ithas topayafeeof10to20basispointsperloanuntilthecontractends(seeScully,2015). Inaddition, banksmaybenefitfromthetechnologicalnoveltyFinTechcompaniesbringtothesepartnerships. 9Theofferdetailstypicallyincludeinformationonthetruelender. SeefiguresA1throughA3inAppendixAfor sampleoffers. 10Seepage49ofForm10-Qfor2019:Q3. 11Seepage68ofForm10-Qfor2019:Q3. 7

BothFinTechandbankpartnersaresubjecttoapprobationandregulatoryscrutiny.Regulatory scrutinyisespeciallylikelyforbankpartnersinFinTech-bankpartnerships. Bankregulatorsused supervisoryguidancetodiscouragebanksfrompartnershipsofferingpaydayloansandrefundanticipation loans—loans issued through tax preparation firms and repaid with tax refunds.12 For FinTechcompanies,lossofbankpartnershipsmayjeopardizetheviabilityoftheirbusinessmodels. 2.5. PartnerBanks But what do we know about these banks that partner with FinTech companies? Although in ourdata(discussedbelow)weobserveanumberofbanksthatpartnerwithmarketplacelenders, we will focus the discussion around WebBank and Cross River Bank, the predominant banks in FinTech-bankpartnerships. WebBank is a $961 million FDIC-insured state-chartered industrial bank located in Salt Lake City,Utah,whichwasfoundedin1997.13 Asoftheendof2019,personalloansrepresentcloseto 50percentofitstotalassets. WebBankisownedbySteelPartnersHoldingsLP,adiversifiedinternationalholdingcompany.SteelPartnersHoldingsL.P.revealsinitsQ310-QreportthatWebBank representsitsfinancialservicesbusiness:14 “WebBankengagesinafullrangeofbankingactivitiesincludingoriginatingloans,issuingcreditcardsandtakingdepositsthatarefederallyinsured. WebBankoriginates andfundsconsumerandsmallbusinessloansthroughlendingprogramswithunaffiliatedcompaniesthatmarketandservicetheprograms(“MarketingPartners”),where theMarketingPartnerssubsequentlypurchasetheloans(orinterestsintheloans)that are originated by WebBank. WebBank retains a portion of the loans it originates for its Marketing Partners. WebBank also has private-label financing programs that are branded for a specific retailer, manufacturer, dealer channel, proprietary network or bankcardprogram.WebBankparticipatesinsyndicatedcommercialandindustrialas well as asset-based credit facilities and asset-based securitizations through relationshipswithotherfinancialinstitutions.” 12Forexample,bankregulatorsdidsoviaOperationChokePoint,aDepartmentofJusticeinitiativethatinvestigated banksandbusinessesiftheyengagedindealingswithcompaniesconsideredtobeatahighriskforfraudandmoney laundering,whichincludedpaydaylenders. 13Industrialbanksevolvedfrom“MorrisPlan”banks,namedafterArthurJ.Morris,whodevisedin1910amethod toprovidesmallloansthatdidnotviolateusurylaws. Theircustomerswereprimarilyindustrialworkers. Underthe MorrisPlan,lenderswouldofferaloanatthelegalrateallowedundertheusurylaw.Interestwasquotedonadiscount basis,whichcollectsinterestinadvanceoutoftheloanprincipal. Thispracticereducedthelender’soutlay,thereby increasingtherevenuerelativetoothermethodsforquotinginterest(seeMors,1965). TheMorrisPlanalsorequireda simultaneouspurchaseoninstallmentsbytheborrowerofanon-interest-bearingcertificateofdepositfromthebank inthesameamountastheloanprincipal. Whenthecertificatewasfullypaidfor,itwouldbeusedtopayofftheloan (Michelman,1966;Oeltjen,1975). Thisarrangementincreasedtheyieldtothebankbybringingitsfundsbackbefore theactualmaturityoftheloancontract. 14Seepage35ofForm10-Qfor2019:Q3. 8

According to the Utah Department of Financial Institutions, as a Utah industrial loan bank charter,WebBankissubjecttothesameregulatoryframeworkascommercialbanksandisauthorizedtomakeallkindsofconsumerandcommercialloans. Thebankcanacceptfederallyinsured deposits, but it cannot issue demand deposits if its assets are greater than $100 million.15 However,banksthatfallundertheUtahindustrialbankcharterarenotconsideredsubjecttotheBank HoldingCompanyAct. Industrialbanksareexemptfromstatelenderlicensinglaws,interestrate ceilings,andmoneyservicebusinesslaws. Industrialbankshavetheadvantageofreducedcostof fundinginpartduetoaccesstotheFederalReservediscountwindow.ThesetypesofbanksaresupervisedonlybytheUtahDepartmentofFinancialInstitutionsandtheFDIC,thuscircumventing Fedoversight. Importantly,giventhatWebBankischarteredinUtah,ithasnosmall-loaninterest rate cap because Utah has no state interest rate limits. As a Utah industrial bank, WebBank can legally charge interest rates that exceed other states’ rate ceilings—a key feature for the FinTech partnership. OtherbanksengagedinpartnershipsthatarelocatedinUtahareFinWiseBankand FirstElectronicBank. CrossRiverBankisa$2billionFDIC-insuredstate-charteredbanklocatedinNewJerseyand created in 2008 with the specific purpose of facilitating FinTech partnerships. As of the end of 2019, nearly half of Cross River Bank’s total assets consist of personal loans. In March 2019, the bankwasawardedGrowNewJerseycreditsbytheNewJerseyEconomicDevelopmentAuthority, thusenablingittoexpandwithinthestateandencouragingitnottoleavethestate.16 Thebankis venturefunded,withinvestorssuchasventurecompaniesAndreessenHorowitz,BatteryVentures, RibbitCapital, andKKR& Co. Accordingto the state’sgovernor, “InnovativeFinTechcompanies likeCrossRiverchoosetolocateinNewJerseybecauseofourunrivaledpoolofdiverse,tech-savvy talent, andtheuniqueadvantagesofourlocation.”CrossRiverBankissubjecttoregulationand supervisionbythestateofNewJerseyandtheFDICandisabletoacceptdeposits.AsaNewJersey charteredbank,CrossRivercanchargeuptoamaximuminterestrateof30percent. 2.6. LegislativeChallenges:Maddenv.MidlandRuling In2015,FinTechlenderswereaffectedbytheMaddenv.MidlandFunding ruling,discussedat lengthbyDanisewiczandElard(2018),whentheU.S.SecondCircuitCourtofAppeals—presiding overConnecticut,NewYork,andVermont—overturnedalowercourtrulingthatloansoriginated to residents of those states with an interest rate exceeding the existing usury limits are null and voidiftheloansareheldbynonbankfirms. 15AccordingtoinstructionsintheFederalReserveBoard’sFR2900,ReportofDepositsandVaultCashforbanks,savingsandloanassociations,andsavingsbanksinstructions,demanddepositsaredepositsthatarepayableimmediately ondemand,orthatareissuedwithanoriginalmaturityorrequirednoticeperiodoflessthansevendays,orthatrepresentfundsforwhichthereportinginstitutiondoesnotreservetherighttorequireatleastsevendays’writtennoticeof anintendedwithdrawal.Demanddepositsaretransactionaccounts. 16GrowNewJerseyisajob-creationandincentiveprogramcreatedtostrengthenNewJersey’seconomicposition. Accordingtotheprogram’swebsite,businessesthatarecreatingorretainingjobsinNewJerseymaybeeligiblefortax creditsandbonuscredits. 9

SalihaMadden,aNewYorkresident,becamedelinquentoncreditcarddebt. Thecreditcardissuing bank deemed the debt to be uncollectable and sold it to Midland LLC, a debt collection firm. Midlandtriedtocollectthedebtata27percentrate,whichexceededNewYork’s25percent rate ceiling. Madden sued, alleging that the loanwas usurious because Midland was not a bank andthereforewasnotentitledtopreemptionoftheNewYorkrateceiling. ThepenaltyforviolationoftheusuryceilinginNewYorkandConnecticutisforfeitureofinterest and principal. The penalty in Vermont is forfeiture of only the interest above the ceiling. AlthoughnotdirectlyrelatedtoFinTechlending, thisrulingaffectedthespecialistbank-FinTech partnershipmodeldescribedinsection2.4thatallowsthebanktoissuetheloansandsubsequently selltheloanstoanonbankpartner. TheSecondCircuitrulingheldthattheexemptionfromstate rateceilingsnolongerappliesonceloansaresoldtononbankfirms. TheMaddenrulingappliesonlytoloansoriginatedtoresidentsofConnecticut,NewYork,and Vermont,butitmaymorebroadlyraiseconcernsaboutriskinloansalesandsolicitations. 3. ResearchDesignandData We use a quasi-experimental design. Our dependent variable is the solicitation mail volume senttohouseholdsbydifferenttypesoflenders: FinTechcompany-bankpartnerships(hereafter, FinTech-bankpartnerships), financecompanieswithoutabankpartner(hereafter, financecompanies), payday lenders, banks without a FinTech company partnership (hereafter, mainstream banks),FinTechlenderswithouttheparticipationofbanks(hereafter,FinTech),andbanksthatare typicallyinvolvedinpartnershipsmakingindependentoffers(hereafter,specialistbanks). Our treatment groups consist of consumers in different credit risk classes residing in states with low or high rate ceilings for personal loans (subprime/low rate, subprime/high rate, near prime/low rate, near prime/high rate, low prime/low rate, low prime/high rate, high prime/low rate,andhighprime/highrate).Ourcomparisongroupconsistsofhigh-primeconsumersresiding instateswithhighrateceilingsornoceilingforpersonalloans. High-primeconsumersinhighratestateswouldbeleastconstrainedbytheircreditriskorlenders’abilitytooffersmallpersonal loansprofitably. Thesubstantiverestrictionsimposedonnonbanklenderswhooperateinstates withlowinterestrateceilingsstimulatesomeofthemtopartnerwithbanksinordertocircumvent thelegislationthatmakeslendingtoriskierconsumersunprofitableandexpandtheirgeographicalpresencewithintheirtargetmarkets. Sincesomeoftheselenderssuccessfullycircumventthe restrictivestaterateceilings,theyendupbeingabletochargeinterestrateshigherthanmaximum ratesspecifiedinthestatelegislation(seeTable1). Stateswithhighornointerestrateceilings— SouthCarolina,Georgia,Texas,Oklahoma,Louisiana,Tennessee,Missouri,Illinois,NewMexico, Kentucky,Alabama,Wisconsin,Indiana,Mississippi,andIdaho—arestatesinwhichhigh-rateconsumerfinancecompaniesoperate. In addition, to illustrate the effects of the Madden v. Midland ruling on credit supply for all lendercategoriesobserved, inlinewithotherstudiesexaminingthesupplyofFinTechloans, we 10

use a differences-in-differences estimation, but we expand the treatment group to include Vermont,inadditiontoNewYorkandConnecticut. Thecomparisongroupconsistsofpersonalloan offersolicitationssenttoconsumerslocatedinallotherstates. Weusedatafromthreesources. OurprimarydatasourceisMintelComperemedia(hereafter, Mintel),adatasetconsistingofmonthlyunsecuredpersonalloanacquisition,cashadvanceproduct,andvehicletitleloanoffers(solicitations). Asthesolicitationsarecreditoffers,theyareoften usedasameasureofcreditsupply.17 Thedatarepresentmonthlycampaign-levelmailvolumesent toconsumersoveraperiodoftenyears,startingwith2010.Mintelrandomlyselectsroughly4,000 consumersfromapoolof1millionconsumersthatMintelpurchasedfromalargesurveyservice provider. The Mintel panel is balanced on four major demographic characteristics: region, age, income, and household size. Each month, about 2,500 consumers participate in the Mintel surveybymailingbacktoMinteloffersfromacrossthesectorsmonitoredbythecompany.18 Mintel motivatesparticipationwithrafflesofferingprizes,suchasgiftcards.Thecompanyrecordsallthe offerdetailsinitsdatabases,thusprovidinginsightsintotherichlandscapeofsupplyoffers.Every month, post-collection, the data are sent to TransUnion alongside the name and address of the panelist.TransUnionthenappendstheVantageScorecreditscoreforeverypanelistanddepersonalizestheinformationbeforesendingitbacktotheFederalReserveBoard.19 Mintelconductsan additionalsurveyofparticipatingconsumerstocollecthousehold-leveldemographicandsocioeconomic information. This additional information is merged with the mail offer information.20 Finally,Mintelappliesweightstotheapproximately2,500consumersparticipatinginthesurveyto representtheentireU.S.adultpopulation. Cruciallyforouranalysis,thedataenableustoobserveclearly,byname,thecompanysending theofferanditsexactbrandpartnercompanies,ifapplicable. (WeincludeinAppendixAseveral examples of various types of offers made to consumers.) Knowledge of the name of each companyinthedatasetalsoenablesustocategorizetheoffersbylendertype. Minteldatashowthat in 2019, there were approximately 1.8 billion acquisition offers for unsecured personal loans.21 Thepersonalloanscategoryisrelativelybroad. Itincludesnon-vehicleloansforanypersonalor household-relatedexpenditureorforconsolidatinganytypeofdebt. Themajorityoftheobservationsinthedatasetincludedetailedmailinglocationinformation.22 Morethanone-thirdofthese offersareissuedbyFinTechlendersinpartnershipwithspecialistbanks(33percent),withanother one-thirdissuedbyfinancecompanies(29percent). Theremainderofoffersareissuedbybanks otherthanthoseinvolvedinpartnerships(15percent),banksthataretypicallyinvolvedinpartner- 17DettlingandHsu(2021)andHan,Keys,andLi(2018)discusstheuseofsolicitationsasameasureofsupply. 18Iftheconsumerschoosetorespondtotherespectiveoffer,theysendMintelonlytheremainderoftheoffermaterials,exclusiveoftheresponseportion. 19Personallyidentifiableinformation(PII)isnotincludedinthedatasetavailabletoresearchers. 20ThedemographicandsocioeconomicinformationcollectedbyMintelappliestothehouseholdheadandisrepresentativeatthehouseholdlevel,whiletheVantageScorecreditscoreisthatofthepanelist. 21AccordingtoMintel,solicitationsforpersonalloansrepresentthelargestmailvolumecategory,supersedingthat formortgageloansorcreditcards. 22Wedroptheobservationswithinsufficientgeographicalinformationtoavoiderrors. 11

shipsmakingindependentoffers(10percent),FinTechlenderswithouttheparticipationofbanks (6percent),paydaylendersmakinginstallmentloanoffers(4percent),otherfinancialinstitutions (3percent),andcreditunions(lessthan1percent).2324 Figure1showstheannualunsecuredpersonalloanoffermailvolumebylendertype. Minteldataallowustoidentifythepartnerbanks. Weobservethenotableparticipationofthe market leaders: WebBank and Cross River Bank, alongside other participants such as First Bank ofDelaware,FirstElectronicBank,FarmersMerchantBank,MidAmericaBank&TrustCompany, County Bank of Rehoboth Beach, Republic Bank, The Brand Banking Company, FinWise Bank, Goldman Sachs Bank USA, First Bank & Trust, and Capital Community Bank. When examining thepartnershipstructure,wenotethedominanceofWebBank(61percentofofferings)andCross RiverBank(35percentofofferings). TheFinTechcompanypresenceinpartnershipstructuresis less concentrated than that of partner banks. LendingClub has the largest share of offerings (41 percent), followed by Best Egg (24 percent). Prosper, Upgrade, and Upstart have much smaller sharesofofferings(11percent,10percent,and6percent,respectively). Figure2showsadiagram ofFinTech-bankpartnershiprelationships. Wecomplementthesupply-sideinformationwederivefromMinteldatawithloan-levelinformationfromthetwomainFinTechlenders:ProsperandLendingClub.25 Prosperdataareavailable atamonthlylevelstartingwith2014,whileLendingClubdataareavailableonlybetweenJanuary 2014 and December 2017. Both lender data sets allow us to observe the geographic location of theborrower,theloanamounttaken,theloaninterestrate(Prosperdataalsoincludetheloanannualpercentagerate,orAPR),andtheborrower’sriskscorebothgeneratedbyatraditionalcredit bureauandsuppliedbyaninternalproprietarymodelusedbythistypeoflender,amongothernumerousdetails.In2017,themostrecentyearforwhichwehaveavailabledata,LendingClubmade 443,579newloans,amountingtonearly$6.6billion. Thatsameyear,Prospermade201,906loans, amounting to $2.7 billion.26 Examining data for Arkansas, the state with the lowest interest rate ceiling,between2014and2017forbothlenders,werevealthat36percentofProsper’sloansand 19percentofLendingClub’sloansissuedovertheperiodhaveinterestratesexceedingthestate’s ceilingrate(Table1).Thisinformationservesasanadditionalmotivatingfactorforourstudy. Finally,inordertocomparethemailvolumeofsolicitationstoaggregatesofindividualswith creditscores,weusetheFRBNY’squarterlyCCP,adatabaseonconsumers’credituseandpayment performancedrawnfromEquifaxcreditbureaurecords.Theanonymizedrandomsampleisrepre- 23Percentagesmaynotaddto100becauseofrounding. Vehicletitlelendersmakeasmallnumberofoffersthat appeartobeinstallmentloanoffers,butthedatadonotallowthedifferentiationbetweentraditionalinstallmentloan offersandvehicletitleloansthatarerequiredtoberepaidininstallments. 24Althoughpaydaylendersareknowntomakeprimarilycashadvanceorlumpsumoffers,ourdataallowustoobserveinstallmentloanoffersmadebytheselenders,whichareanewertypeofproductforthem(seeMillerJr.,2019). 25LendingClub’sLoanStatsdatawasavailabletoresearchersviahttps://www.lendingclub.com/untilDecember 2017. 26In2019,Prospermade156,671loans,amountingto$2.2billion. Theloancharacteristicsforthetwolendersare similar,withbothreportingloanamountsof$1,000or$2,000to$40,000—LendingClubwithamedianof$12,900anda meanof$14,926andProsperwithamedianof$12,000andameanof$13,709.Astableshareoftheseloans—about70 percentforbothlenders—isusedfordebtrefinancing. 12

sentativeofthepopulationofcreditusersineachquarter.27 Thedatabasecontainsindividual-level dataonvirtuallyeverydebtowedbyeachconsumerand,importantlyforouranalysis,thecredit bureauscore,whichweusetocreateaggregatesforeachcreditriskgroupofborrowers.28 Weusea 1percentsamplefromthetotalavailable5percentsample,coveringtheperiodbetween2010:Q1 and2019:Q4. Attheendofthefourthquarterof2019,theCCPtotaledabout252millionindividuals. 4. EmpiricalAnalysisandResults Toidentifytheroleofinterestrateregulationofconsumercreditandinstitutionalsegmentation inFinTechlenders’effortstosolicitnewcustomersinthepersonalloanmarket,wefirstcompare thestate-levelshareofmailvolumeintotalmailvolumeovertheperiodforeachlendertype.Then, wecomparethecumulativemailvolumesolicitationssentbyFinTech-bankpartnershipsforeach credit score category (subprime, near prime, low prime, and high prime) and level of the state consumerfinancerateceiling(withhighandlow).Wealsoexaminethedifferencesintheaggregate mailvolumeissuedbyFinTech-bankpartnershipsbeforeandaftertheMaddenrulinginthestates representingourtreatmentgroupversusallotherstates.Inaddition,weuseregressionanalysisto estimatedifferencesinlenders’unsecuredpersonalloanmailoffers. 4.1. PersonalLoanSolicitationsbyTypeofLender,InterestRateCeilingCategory,and CreditRiskGroup Weexaminetheaggregatenumberofpersonalloansolicitationsovertheanalyzedperiod(2010– 19). Solicitations are an indicator of credit supply, but it is important to note that they do not directly and uniformly translate into loans. The take-up rate of mail offers is relatively low—for example,usingthedataavailabletous,weestimateittobeapproximately9.63loansper100LendingClubsolicitationsand12.76loansper100Prospersolicitations. Thatsaid,theseestimatesare likelybiasedasborrowerscanobtainloansbycontactingeachlenderdirectly, withoutbeingsolicitedviamail.Thetake-uprateislikelyinfluencedbythelocaldemandandsupplycharacteristics which are not uniform across states, within a state, or among borrower risk profiles. For example,asFinTechlendersinpartnershipwithbanksandalsoactingindependentlytypicallycompete withfinancecompanies,thetake-uprateforFinTech-issuedloanscouldbehigherinstateswith lowconsumerfinanceinterestrateceilings,wherefinancecompaniescannotoperateprofitably.29 27Thesamplingprocedureensuresthatthesameindividualsremaininthesampleineachquarterandallowsfor entryintoandexitfromthesample,sothatthesampleisrepresentativeofthetargetpopulationineachquarter. See LeeandderKlaauw(2010)foradescriptionofthedesignandcontentoftheCCP.Seealsohttps://www.newyorkfed. org/medialibrary/interactives/householdcredit/data/pdf/data_dictionary_HHDC.pdf. 28Thevariablesincludetypeofcredit, typeoflender, originationdate, accountbalance, scheduledmonthlypayments,delinquency,andadverseeventsassociatedwithcreditaccounts.Variablesalsoincludeyearofbirth. 29BothLendingClubandProspershowhighertake-uprates(20.1and25.63loansper100solicitations,respectively) inArkansas,forexample,thestatewiththelowestconsumerfinancerateceilingwherenofinancecompaniesareoperating.Theseestimateshoweverarebiased,aspreviouslynoted. 13

Moreover,asLukongoandMillerJr.(2022)andElliehausenetal.(2021)showthatinArkansas,for example, there is a credit desert in the middle of the state and that there is a higher concentrationofconsumerloansinthecountiesborderingtheneighboringstates,thetake-uprateformail volumesolicitationsfromvariouslenderscouldbedifferentintheinteriorcountiesversustheexteriorcountiesofthestate. Finally, thetake-upratecouldalsodifferamongborrowercreditrisk groups,withcertaincreditconstrainedgroupsbeingmorelikelytopursueanofferreceivedinthe mailthanothers. 4.1.1. State-LevelLenderShareinTotalMailVolume We first examine the cumulative mail volume distribution at the geographic level looking at a state-level share for each category of lender identified earlier. Our share is constructed as the ratioofallmailvolumefromeachlendertypetototalmailvolumesentbetween2010and2019to borrowersineachstate.Figure3showstheshare. Focusing on the share for the mail volume sent by FinTech-bank partnerships (Figure 3a), wenotethatstateswithlowerinterestrateceilings(nothatched)appeartohavehigherlevelsof FinTech-bank partnership mail volume than states with high interest rates (hatched). Notably, Arkansas,thestatewiththemostrestrictivelegislation,hasbyfarthehighestFinTech-bankmail volumepresenceofallstatesfortheobservedperiod. Asthemostprominentofthelowinterest rate states, Arkansas has a very low, constitutional interest rate ceiling that does not allow consumerfinancecompaniestooperateprofitablyinthestate. Itislikelythat,ofthestateswithlow interestrateceilings,ArkansasisattractivetoFinTechlendersbecauseofthelackoflocalfinance companycompetitorsinthehigher-riskmarketsegment. Incontrast, anotableexceptiontothe ruleisIowa. WebelievethisexceptionisduetoIowa’soptingoutoftheinterestrateexportation regimediscussedpreviously. AnotherexceptionisWestVirginia,whichislikelyduetolegislation prohibitingProsperandLendingClubfromdoingbusinessinthestate.30 Incontrast,mailofferssentbyfinancecompanies(Figure3b)areconcentratedinstateswith highinterestrateceilings,astheabilityoffinancecompaniestolendprofitablyinlow-ratestates islimitedbythelowinterestrateceilings. FinTech lenders sending out independent mail offers without bank partners (Figure 3c) and paydaylendersmakinginstallmentloanoffers(Figure3d)appearmostlyindifferenttotheinterest rateceilingregime,but,notably,theyavoidArkansas.31 Banksofbothtypes,mainstreamandspecialist,alsoappeartobeindifferenttostateinterest rateceilingregimes(Figures3eand3f). 30Forexample, Prosperexplicitlynotesonpage55ofits2021:Q1Form10-Qthat"noloanshavebeenoriginated throughtheProsperplatformtoWestVirginianssinceJune2016."Thislimitationisreferencedasthe"WestVirginia Matter"andisattributedtoanunresolveddiscussionbetweenthestate’sattorneygeneralandthelenderinreferenceto thepotentialviolationofWestVirginia’sConsumerCreditandProtectionAct. 31AlthoughpaydaylendersareprohibitedinArkansas, thestate’sconsumersreceiveinstallmentloanoffersfrom onlinepaydaylenders. 14

4.1.2. AggregateNumberofFinTech-BankPartnershipOffersbyRiskCategoryforIndividuals withCreditScores Next,weexaminethenumberofFinTech-bankpartnershipoffersreceivedovertheperiodof analysisbyborrowersforeachofthefourriskcategoriesweconsideredbasedonMinteldata:subprime (VantageScore credit score lower than or equal to 660), near prime (VantageScore credit scoregreaterthan660andlowerthanorequalto719),lowprime(VantageScorecreditscoregreater than719andlowerthanorequalto792),highprime(VantageScorecreditscorehigherthan792). In order to do so, we obtained the credit score distribution as of 2019:Q4 from the FRBNY CCP/ Equifaxdata. AlthoughMinteldataandtheFRBNYCCPusedifferentcreditscores—Minteluses VantageScorecreditscoresandFRBNYCCPusesEquifaxRiskScores—theircreditscoredistributionscanbeconsideredcomparable. Assuch,wesumthemailofferinformationovertheperiod acrossthefourcreditriskcategoriesanddivideoffersbythenumberofborrowersineachcredit scoregroupintheFRBNYCCPdata.Figure4showstheresults. FinTech-bank partnerships targeted borrowers with near-prime and low-prime credit scores morethantheirsubprimeorhigh-primepeers,irrespectiveoftheirstateofresidence(Figure4a). Forlow-riskborrowers,thestateofresidencemakeslittledifference,astheyreceiveaboutthesame numberofpersonalloanmailoffers—anaverageof10inlow-ratestatesand9inhigh-ratestates, respectively, overtheobservedperiod. Incontrast, thestateofresidencemakesanotabledifferenceforsubprimeborrowers,asthoseresidinginlow-ratestatesreceivedabout5moreoffersthan theirpeersresidinginhigh-ratestates,whoreceivedanaverageof8offersovertheobservedperiod. InArkansas,FinTech-bankpartnershipsheavilytargetednear-primeborrowersresidinginthe state. Near-primeArkansasconsumersreceivednearlyfourtimesasmanyoffersthannear-prime consumers residing in other low-rate states—81 offers relative to the 21 received by their peers (Figure4b). Wealsonotethatlow-primeborrowersarealsomuchmoretargetedbytheFinTechbankpartnershipsthantheircounterpartsinlow-ratestates,asnear-primeborrowersresidingin Arkansasreceived9moremailoffersthanthoseresidinginotherlow-ratestates.Thestateofresidence,makeslittledifferenceforsubprimeborrowersandhigher-risk-scoreprimeborrowers. ParticularlyinterestingaboutArkansasisthefocusonnear-primeborrowers.Inlow-ratestates (excludingArkansas),theFinTech-bankpartnershipstargetsimilarproportionsofnear-primeand low-primeborrowers. InArkansas,FinTech-bankpartnershipsfocusmoreheavilyonnear-prime borrowers. 4.1.3. AggregateNumberofFinTech-BankPartnershipOffersinLightoftheMaddenv. MidlandRuling In this subsection, we examine the changes in the aggregate number of FinTech-bank partnershipoffersfollowingtheMaddenv. Midland ruling. AlthoughHornandHall(2017)highlight severalfar-reachingeffectsoftheMaddenv. Midland ruling, ourdataenableustocomplement 15

theirkeyfindingthattherulingledtoadecreaseinmarketingandlendingprogramsinthestates affected, and we show that the Madden v. Midland ruling caused uncertainty that had a ripple effectonthesupplyofpersonalloansacrossothergeographicareasaswell.32 InFigure5,wecontrast the supply trends in New York, Connecticut, and Vermont with those observed across the unaffectedstates. BeforeMadden,NewYork,Connecticut,andVermontsolicitationswerefollowinganascendingtrend,similartothoseinotherstates.However,solicitationsinthosethreestates fellsharplyfollowingtheruling, whilesolicitationsinotherstatesremainedflat. Althoughsolicitations later increased to about pre-Madden levels, more recently, subsequent declines for New York,Connecticut,andVermontweresteeperthandeclinesinotherstates. Uncertaintyfollowing theMaddendecisionlikelyreducedsolicitationvolumeintheaffectedstates,buttheuncertainty appearstohavehadspillovereffectsacrossothergeographicareasaswell. 4.2. RegressionAnalysisandResults Onaverage,lenderssent202monthlyunsecuredpersonalloansolicitationspercampaignper 10,000 individuals with credit scores. The number of personal loan solicitations varied widely amonglendertypes.FinTech-bankpartnershipshadthelargestaveragenumberofsolicitations,72 per10,000individualswithcreditscores(Table2),followedcloselybyfinancecompanies,with68.7 per 10,000 individuals. Mainstream banks had an average of 30.1 solicitations per 10,000, about half the personal loan solicitations as the previous two categories. Specialist banks soliciting on theirownwithoutaFinTechpartnermade, onaverage, 14.9solicitations. FinTechfirmsoffering personalloanswithoutapartnerbankshowedanaverageof9.21solicitationsper10,000individuals,whilepaydaylendersmade7.17per10,000.Creditunionsandtheotherlendercategorymade fewsolicitationsandthereforewillnotbeconsideredinthesubsequentdiscussion. 4.2.1. PersonalLoanSolicitationsbyTypeofLenderandInterestRateCeilingCategory We look first at the effect of rate ceilings on the volume of solicitations for personal loans of differentlendertypes.Thedependentvariableisthenumberofsolicitationsper10,000individuals withcreditscoresintherespectivestateofresidenceforthereceiveroftheofferintotalandforeach ofthesixcategoriesoflenders.Ourregressionmodelis Y =β +b ·Rate +Σ β ·X +ε , (1) ijt 0 1 j rit j rit ijt whereY representsolicitationsbylendercategoryiinstatejandmonthandyeart. Rate isan ijt j indicatorvariablethatequals1ifthestatehasahighrateceilingforloansfromlicensedlendersand 32Theothereffectsincludethecappingofinterestratesaccordingtotheusurylimits;theexclusionofloanstoNew York,Connecticut,andVermontresidentsfromsecuritizationpoolsor,viceversa,havingtheseloansspecificallyacquiredbyspecialpurposevehicles; andtherestructuringoftherelationshipwiththebankpartnersothatthebank retainsaninterestonalltheloans. 16

0otherwise.TheX arervariablesaccountingforeconomic,creditrisk,andlife-cyclecharacterrjt isticsthatlendersmighttargetforsolicitationsandtimefixedeffectstoaccountforseasonalityand macroeconomicconditions. WiththeexceptionofthestateGDPinformation,whichcomesfrom theBureauofEconomicAnalysisandweincludeasaproxyforconsumerwealthandemployment opportunitiesinthestate,alloftheX variablesarefromMinteldata,includingtheriskscore.33 rjt ε istheerrorterm.Standarderrorsareclusteredatthestatelevel. ijt Table3presentsestimationresultsinpanelA.Thecoefficientofparticularinterestisthehighrate-ceilingindicatorvariable.Thiscoefficientisthemeandifferenceinthenumberofsolicitations inhigh-ratestates(thetreatmentgroup)relativetolow-ratestates(thereferencegroup).FinTechbank partnerships (column II) and banks (columns VI and VII) solicited less in high-rate states thanlow-ratestates(partnerships,44.4per10,000fewersolicitations;mainstreamandspecialized banks,24and12.1fewer,respectively).Financecompanies(columnIII)solicitedmoreinhigh-rate statesthanlow-ratestates(9.46solicitationsper10,000,ceterisparibus). Insolicitinginhigh-rate states, finance companies would be more likely to be able to charge rates that enabled them to coverthecostsofrelativelysmalldollarandriskyloans. Thatmainstreambankssolicitedlessin high-ratestatesisconsistentwithbanksseekingtomakelargerloansandtolessriskycustomers andavoidingsmall-sizedloansandriskyborrowers. Specialistbankssolicitingloansontheirown alsoshowedlessinterestinsolicitingloansinhigh-ratestatesthaninlow-ratestates. Significanceofborrowercharacteristics(includedinthecontrols)inthisandsubsequentestimationssuggeststhatsolicitationstargetareasinwhichtake-upratesandpaymentperformance arefavorable. Thesizeandsignificanceofthesevariablesoftenvarybytypeoflender, reflecting differencesinbusinesspracticesofdifferentlendertypes. Usingconsumersaged40to54—who stillmaybefinancingacquisitionofdurableswithdebt,haveaccumulatedsubstantialdebts,and benefitfromrefinancing—asthereferenceagegroup,wefindthathouseholdsheadedbyyounger consumers(agedunder25andaged25to39)andolderconsumers(55andolder)arelesslikely than the reference group to be solicited by finance companies, FinTech lenders acting independently,andpaydayinstallmentlenders.34 Differencesinsolicitationsbyincomeareconsistentwith institutionalrisksegmentation.Financecompaniessolicitmostconsumerswithincomeslessthan $25,000. Aslower-incomeconsumerstendtohavelittlediscretionaryincomeabovetheirnormal livingexpenses,theyposegreaterdelinquencyanddefaultriskthanhigher-incomeconsumers.In contrast,FinTech-bankpartnershipssolicitleastconsumersinthelowestincomebracket.Finance companies specialize in lending to higher-risk consumers, while FinTech-bank partnerships do notseekouthigher-riskconsumersforsolicitations. 33Intheregressions,weusehouseholdheads’VantageScorecreditscoreinformationfromTransUnionincludedin Minteldatainadepersonalizedfashion.Weusethepopulationwithcreditscoresintherespectivesolicitationreceiver’s stateofresidencefromEquifaxtoscaleourresults,asourdependentvariablesrepresentsolicitationsper10,000individualswithcreditscoresforeachlendercategory. 34RecallthatinMinteldata,thedemographicandsocioeconomicinformationappliestothehouseholdhead,while theVantageScorecreditscoreisthatofthepanelist. 17

TheVantageScorecreditscorepointstorisksegmentation. HigherVantageScorecreditscores areassociatedwithgreatersolicitationsfrombanks.Inaddition,higherVantageScorecreditscores areassociatedwithfewersolicitationsfromfinancecompanies,FinTechfirmsoperatingwithouta bankpartner,andpaydayinstallmentlenders. Ingeneral,rentersaresolicitedmoreheavilythan owners across lender types. Higher state-level GDP is associated with fewer solicitations by all lendersirrespectiveoftype. Variablesforthechildren,housingtype,race,andeducationaregenerallynotstatisticallysignificant. As previously mentioned, banks historically have avoided high-risk consumers, and finance companies provided credit to high-risk consumers, when the rate ceiling allowed them to do so profitably. Institutionaldifferencesinsolicitationsbytheheightofthesmall-loanrateceilingmay reflect continuing differences in risk tolerances of established lenders. Institutional differences mayalsorevealrisk-relatedperceptionsoflendingopportunitiesforrelativelyrecentFinTechentrants. Toexplorethispossibilityfurther, weexaminesolicitationsbyrate-ceilingandconsumer riskcategories,asdescribedintheprevioussection.Asbefore,thedependentvariableisthenumberofsolicitationsper10,000consumerswithcreditscores.Theregressionmodelis Y =β +d1·SL +d2·SH +d3·NL +d4·NH +d5·LL +d6·LH +d7·HL ijt 0 it it it it it it it +Σ β ·X +ε , (2) rjt j rjt ijt Variablesindicatingtherateceilingcategory(L,low; andH,high)andtheconsumerriskcategory(S,subprime;N,nearprime;L,lowprime;andH,highprime)replacethehigh-rateindicatorvariablefromthepreviousregression. Thehigh-prime,high-ratestate(HH )isthereference it group. EstimationresultspresentedinTable3inpanelBsuggestthathistoricalrisktolerancesoffinancecompaniesandbankspersistandthatrateceilings,whenbinding,influencethesupplyof personalloans. Financecompanies(columnIII)solicitedquiteheavilyhigher-riskconsumersespeciallyinhigh-ratestates,consistentwithrisksegmentationofthepersonalloanmarketbyinstitutionalclass. Subprimeconsumersinhigh-ratestatesreceived58.6solicitationsper10,000more thanhigh-primeconsumersinhigh-ratestates, whilesubprimeconsumersinlow-ratestatesreceived48per10,000moresolicitationsfromfinancecompanies.Financecompaniesalsosolicited near-primeconsumersinhigh-ratestatesmorethanhigh-primeconsumersinhigh-ratestates.As awhole,thesefindingssuggestthatfinancecompaniesspecializeinthehigh-risksegmentofthe marketandthattheyespeciallyfocustheirattentiononhigh-ratestatesinwhichhigh-risklending isprofitable. Mainstreambanksolicitations(columnVII)targetedhigh-primeconsumers. Subprime,nearprime,andlow-primeconsumersreceivedfarfewersolicitationsfrombanksthanhigh-primecon- 18

sumers,regardlessofrateceiling.35 High-primeconsumersinlow-ratestatesreceivedsubstantially more solicitations than high-prime consumers in high-rate states. Fewer solicitations of all but high-primeconsumerspointtocontinuedriskaversionbybanks. FinTech-specialistbankpartnerships(columnII)heavilysolicitednear-primeandlow-prime consumers,whichappearnottointerestbanksandfinancecompaniesverystrongly. Near-prime consumersinlow-ratestatesreceived79.1per10,000moresolicitationsfromFinTech-bankpartnershipsthanhigh-primeconsumersinhigh-ratestates,whilelow-primeconsumersinlow-rate statesreceived72.3per10,000morethanthereferencegroup. FinTech-bankpartnershipsolicitationsofnear-primeandlow-primeconsumersinhigh-ratestateswerealsonotable(35.2and38.2 per10,000,respectively). Notablealsoaresolicitationsofsubprimeconsumersinlow-ratestates (47.8per10,000).Thelattergroup(subprimeconsumersinlow-ratestates)likelyarenotprofitable forfinancecompanies,becausetheyaresubjecttolocalstaterateceilings.FinTech-bankpartnershipscanusethebankpreemptionofstaterateceilingstolendtothisgroup. Specialist banks (column VI) originating loans on their own solicited fewer subprime consumers and near-prime consumers in high-rate states than high-prime consumers in high-rate states. Thesebankssolicitedlow-primeandhigh-primeconsumersinlow-ratestatesatasomewhathigherratethanthereferencegroup,perhapsbecausetheirexperienceinFinTechpartnershipsmadethemcomfortableusingtheirpreemptionfromstaterateceilingstolendinlow-rate states.Inavoidinghigher-riskconsumers,specialistbanksoperatingontheirownactedmuchlike theirmainstreambankbrethren. FinTechfirmsoperatingontheirown(columnIV)targetedriskierconsumers. Subprimeand near-primeconsumersinlow-andhigh-ratestatesweresolicitedatnotablyhigherrates(6.97,13.7, and7.95per10,000consumers, respectively)thanthereferencegroup. Whenoperatingontheir own,FinTechfirms’riskacceptancemorecloselyresembledthatoffinancecompaniesthanbanks. Paydaylenders(columnV)focusedontheriskiestcustomers,solicitingsubprimeconsumers morethananyothercreditriskcategory. Subprimeconsumerslocatedinlow-ratestatesreceived 10.5 solicitations per 10,000, while subprime consumers located in high-rate states received the highestnumberofsolicitations—18.9per10,000.Similartofinancecompanies,whenmakingpersonalloans,paydaylendersfocusonhigh-ratestatesinwhichhigh-risklendingisprofitable. InTableB1inAppendixB,weshowthatourresultsremainrobustevenaftertheeliminationof outlierstates—ArkansasandIowa. 4.2.2. MailVolumeaftertheMaddenv.MidlandDecision ToexaminetheimplicationsoftheMaddenv. Midlandrulingonthesupplyofpersonalloans, weuseasimilardesignemployedbyotherstudies(seeDanisewiczandElard,2018),butweexpand the treatment group to include Vermont. As a result, our treatment group consists of all three 35Insomestates,rateceilingsforbanksmaydifferfromrateceilingsinsmall-loanlaws. Banksmaychargerates allowedintheirhomestate,regardlessofwheretheborrowerislocated. 19

statesinwhichtheU.S.SecondCourtofAppealshasjurisdiction—NewYork, Connecticut, and Vermont.36 Otherstatesmakeupourcomparisongroup.Ourregressionmodelis Y =β +d ·NYCTVT +d ·PostMadden +d ·NYCTVT·PostMadden ijt 0 1 i 2 t 3 it +Σ β ·X +ε , (3) rjt j rjt ijt where Y represent solicitations by lender category i in state j and month and year t in aggreijt gate and by credit risk group. NYCTVT represents an indicator variable that equals 1 if the i state is New York, Connecticut, or Vermont, and 0 otherwise. It carries the coefficient d , which 1 showsthemeandifferenceinthenumberofsolicitationsintreatmentstates(NewYork,Connecticut, and Vermont) relative to control states (all others). PostMadden represents an indicator t variablethatequals1fortheperiodaftertheMaddenrulingand0otherwise. Thecoefficientd 2 captures the change in the number of solicitations after the ruling, relative to the previous period. NYCTVT ·PostMadden istheinteractiontermunderpinningthecoefficientofinterest, it d , whichrepresentsthechangeinthenumberofsolicitationsintreatmentstatesrelativetothe 3 changeinthenumberofsolicitationsincontrolstates. Thiscoefficientisexpectedtobenegative andstatisticallysignificantforbank-FinTechpartnerships. X arervariablesaccountingforecorjt nomic, creditrisk, andlife-cyclecharacteristicsthatlendersmighttargetforsolicitations. ε is ijt theerrorterm.Standarderrorsareclusteredatthestatelevel. Aftercontrollingforconsumercharacteristics,consumersinNewYork,Connecticut,andVermont received more solicitations from FinTech-bank partnerships (Table 4, panel B, column II) andbanks(panelsFandG)thanconsumersinthecomparisongroupstates. Financecompanies (panelC)madefewersolicitationsintreatmentstates. NewYork, Connecticut, andVermontare subjecttorelativelylowrateceilings. Theseresultsareconsistentwithlowrateceilingsrestricting nonbanklending. Lowrateceilingsmakeriskyloansandsmall-sizedloansunprofitable, sothey are not offered in the market. Banks generally are not so constrained. Banks and FinTech-bank partnershipsmaytakeadvantageofregulationsthatallowbankstochargeratesthatarelegalin theirhomestateregardlessofthestateinwhichaconsumerresides. Overall,FinTech-bankpartnerships(panelB,columnII)andspecialistbanks(panelF,column II) made more solicitations, while nearly all other lender types made fewer solicitations, in the periodaftertheMaddendecisionthanintheperiodbeforethedecision.Aftercontrollingforconsumer characteristics, the net effect of the Madden decision in treatment states was 15.9 fewer solicitations per 10,000 individuals (41.2 per 10,000 post-Madden plus negative 56.1 per 10,000 post-Madden in New York, Connecticut, and Vermont). Finance companies made notably more 36Althoughweareawareoftheheterogeneityissuewithinourgroup,asVermontdiffersinitstreatmentofusurious loans,giventhatweexaminethesupplyeffectsoftheruling,weincludeallthreestatesinourtreatmentgroup. Our resultsareconsistentwhenexcludingVermontfromourestimation.(SeeHonigsberg,JacksonJr.,andSquire,2017;and DanisewiczandElard,2018.) 20

(47.4per10,000)solicitationsintreatmentstatesaftertheMaddendecisionthantheymadeinthe periodbeforethedecision.37 Danisewicz and Elard (2018) found a decline in FinTech lending following the Madden decision, especially among low-income households. Low income is not the same as high risk, but low income does tend to make consumers vulnerable to financial difficulties that adversely affecttheircreditscores. Post-Madden,wefindreductionsinFinTech-bankpartnershipoffersinall riskgroupsrelativetoconsumersincomparisonstates(Table4,panelB),withthemostsignificant reductionsinsolicitationsoccurringfornear-prime(76.2per10,000),low-prime(50.7per10,000), and high-prime (37.4 per 10,000) consumers in treatment states. With the exception of finance companies, solicitationsfromtheotherlendertypeswerelittlechangedintreatmentstatesrelative to consumers in comparison states in the post-Madden period. The most notable increases shownbyfinancecompaniesarefornear-,low-,andhigh-primeconsumers(60.8,48.4,and56.5, respectively). TheMaddendecisionappearstohaveaffectednonbanklendingmorebroadlythanFinTechbankpartnershiplending. Maddendidnotdisputethatthecreditcardcompanychargedalawful rate. Rather, the decision invalidated the long-standing valid-when-made doctrine, which holds that on a contract that is legal at inception, a transferee has the right to enforce the contract on the same terms as those that had been available to the transferor when the contract was made. This doctrine is a core and generally accepted principle of contracts and is central to the stabilityofcreditmarkets. InalegalanalysisoftheprinciplesinvolvedintheMaddendecision, Horn andHall(2017)pointedtotheuncertaintythatarisesfromthedecisionandthatmayhavebroad implicationsforlenders’willingnesstolend: "Byeffectivelyinvalidatingthecollectionofpost-defaultinterestonalawfulloanagreementbyreasonofitstransfer,Maddenhascreatedsubstantialuncertaintyinthereliability of the valid-when-made doctrine in the Second Circuit. The outcome of the Madden decision involves a somewhat specific interaction of usury and federal preemption principles ... Madden, however, has cast at least a temporary pall on loan salesandtradingactivity,andhasforcedbankandnonbankbuyersandsellersofloans toreviewcriticallytheirloansalesandtradingpoliciesandprocedures,andinmany casesrevisetheirbusinesspractices. TheultimateissuethatMaddenraises,however, ishowfar-reachingareitsholdingsandtheirramifications,andwhetherthedecision willmateriallyaffecttheinterpretationandapplicationoflongstandingprinciplesof usuryandthevalidityofloanagreements. [T]helegalandcommerciallandscapefor 37Theregressionresultsforfinancecompaniesandpaydaylendersneedtobeinterpretedwithcaution,asthepost- MaddenrulingperiodoverlapswiththeproposalandimplementationoftheConsumerFinancialProtectionBureau’s transientPaydayLoans,VehicleTitleLoans,andCertainHigh-CostInstallmentLoansrulewhichlimitedthesupplyof loansissuedbythesetypesofinstitutions,amongothers.TheproposedrulewasissuedonJune2,2016,whilethefinal rulewasissuedonOctober5,2017,andrescindedonFebruary6,2019. 21

loan origination and sales activities would become, at the very least, materially less predictable(pp.1–2)." OurfindingssuggestthattheMaddendecisionindeedhadnotableimplications.FinTechcompaniespartneringwithbanksconsiderablyreducedsolicitationsforpersonalloansinSecondCircuitstatesfollowingtheMaddendecision,especiallyforhigher-riskconsumers. Atthesametime, financecompaniesincreasedtheirsolicitationsinthesestates. 5. Conclusion Our paper looks at FinTech-bank partnerships’ solicitations for personal loans for debt consolidationandotherpurposes, anindicatorofsupplyforsuchcredit. Weexaminepersonalloan solicitationsbyFinTechfirms,financecompanies,banks,andpaydaylenders.Weinvestigatehow institutionalarrangementsandinterestrateregulationinfluencethesupplyofpersonalloansfor differentriskclassesofconsumers. PreviousresearchfocusesonFinTechfirms’andbanks’provisionofpersonalloansusedfordebtconsolidationandotherpurposes. Theyhavenotconsidered personalloansfromconsumerfinancecompanies,whichhavetraditionallybeenthemajorsource ofpersonalloansforhigher-riskconsumerswithinthebroadermarket. WealsoexaminetheeffectofacourtdecisionthatinvalidatestheFinTech-bankpartnership businessmodelspecificallyinseveralstatesandmayhavebroadimplicationsforthefunctioning ofconsumercreditmarketsingeneral. Thesolicitationdataprovideevidenceofrisksegmentationbyinstitutionalclassoflender.Our findings show that finance companies, FinTech companies without a bank partner, and payday lendersconcentrateonlendingtohigh-risksubprimeconsumers.Banksfocusonhigh-primeconsumers. We find that FinTech-bank partnerships focused on solicitations for personal loans to marginalconsumers,nearprimeandlowprime.Theyshowsomeinterestinsubprimeconsumers inlowratestates. Ourfindingssuggestthatstateinterestrateceilingsinfluencedthesupplyofpersonalloansby institutionalclass.FinTech-bankpartnershipsheavilytargetednear-andlow-primeconsumersin stateswithrestrictiveinterestrateceilings.Thepartnershipsdidnotheavilysolicithigh-primeconsumersregardlessofrate-ceilingregulation.Theyalsohadlittleinterestinsubprimeconsumersin high-ratestates. However,FinTech-bankpartnershipsmoderatelysolicitedsubprimeconsumers instateswithlowrateceilings,likelybecausetheyfacedrelativelylittlecompetitioninthesestates. Financecompanieswithoutbankpartnerscouldnotoperateprofitablyinlow-ratestates. Butfinance companies heavily solicited subprime consumers in high-rate states, and payday lenders andFinTechcompanieswithoutabankpartnerfollowedsuit. Banksshowedlittleinterestinconsumersinanypartoftheriskspectrumotherthanthehigh-primepart. TheSecondCircuit’sMaddenv. Midland decisionappearstohavehadasizablenegativeeffectonFinTech-bankpartnershiplending. Thedecisionalsohadsizableeffectsonsolicitationsof 22

otherlendersthatspecializeinpersonalloanstohigher-riskconsumers—financecompanies,that steppeduptheirsolicitationsintreatmentstatesafterthedecision. Our Madden exercise using a supply indicator complements other investigations examining other outcome variables (Danisewicz and Elard, 2018; Honigsberg, Jackson Jr., and Squire, 2017; HornandHall,2017). 23

Figure1.UnsecuredPersonalLoanOfferMailVolume 2.5 2.0 1.5 1.0 0.5 2010 2013 2016 2019 Billions Other Lenders Credit Unions Mainstream Banks Specialist Banks Payday Lenders FinTech Companies Finance Companies FinTech−Bank Partnerships Note: Thisfigureshowstheannualunsecuredpersonalloanoffermailvolumeforeachlendercategory. FinTech-bankpartnershipsandfinancecompaniesdominatethisspace. Source:MintelComperemedia. 24

Figure2.FinTech-BankRelationshipDiagram LendingClub Circleback Lending Marlette Best Egg WebBank Funding Borrowers First Prosper CrossRiver Upstart LendingPoint Bank Upgrade BetterCash Inc. Freedom Avant FinWiseBank Financial NewCredit Network America FreedomPlus Personify FirstElectronic Financial Barclays RISE Bank Rocket RocketLoans Holdings Payoff,Inc. CountyBank of Rehoboth Beach Fingerhut LendingTree LoanDepot Note:ThisfigureshowsthefootprintofFinTech-bankrelationshipsobservedinourdata.Inblue,weshowa listofthespecialistbanksengagedinpartnershipswithFinTechcompanies.Inorange,weshowtheFinTech companiespartneringwiththespecialistbanksweobserve. Ingray,weshowtheparentsoftheFinTech companiesengagedinthepartnerships. Wedecreasethepigmentfromstrongtoweaktoindicatemarket shares,andweusethicklinestodesignatestrongerpartnershiprelationships(typicallydefinedasmorethan 100partneredcampaignsduringtheperiodofanalysis)andthindashedlinesforweakerrelationships.The relationshipsandtheparticipantshighlightedarenon-exhaustive. Source:MintelComperemedia. 25

Figure3.ShareofMailVolumeinTotalMailVolume Share 60 50 40 30 20 10 (a) FinTech-BankPartnerships Share 50 40 30 20 10 (b) FinanceCompanies Share 7.5 5 2.5 (c) FinTechCompanies Continuedonnextpage 26

Figure3.ShareofMailVolumeinTotalMailVolume(continuedfrompreviouspage) Share 7.5 5.0 2.5 (d) PaydayLenders Share 20 15 10 5 (e) SpecialistBanks Share 40 30 20 10 (f) MainstreamBanks Note: Thisfigurecontraststheshareofmailvolumeintotalmailvolumeforthesixlendercategoriesanalyzed. Panel(a)showstheshareforFinTech-bankpartnerships,panel(b)forfinancecompanies,panel(c) forFinTechcompanies,panel(d)forpaydaylenders,panel(e)forspecialistbanks,andpanel(f)formainstreambanks. Source:MintelComperemedia. 27

Figure4.FinTech-BankPartnershipMailOffersperIndividualwithCreditScore Low−Rate States 21 21 High−Rate States 20 19 13 10 9 8 Subprime Near Prime Low Prime High Prime (a) Low-RateStates(showningray)vs.High-RateStates(showninyellow) 81 Low−Rate States Arkansas 28 21 19 15 13 10 6 Subprime Near Prime Low Prime High Prime (b) Arkansas(showningreen)vs.AllOtherLow-RateStates(showningray) Note:ThisfigurecontraststheFinTech-bankmailoffersperindividualwithcreditscoreacrossstateinterest rateceilingregimesandwithinlowinterestratestatesbycreditriskgroup.Panel(a)contraststhemailoffers receivedinlow-ratestates(showningray)withthosereceivedinhigh-ratestates(showninyellow). Panel (b)contraststhemailoffersreceivedinArkansas(showningreen)withthosereceivedintheremainderof low-ratestates(showningray).Dataaresummedovertheobservedperiod(2010–19).Thepopulationwith creditscoreswasobtainedfromFRBNYCCP/Equifax(asof2019:Q4). Source: Mintel Comperemedia; Federal Reserve Bank of New York Consumer Credit Panel (FRBNY CCP)/Equifax. 28

Figure5.MailVolumeaftertheMaddenv.MidlandRuling 800 60 600 40 400 20 200 0 0 2010 2013 2016 2019 Year snoilliM Millions All Other States (left scale) New York, Connecticut, and Vermont (right scale) Note:ThesolidlineindicatestheaggregateFinTech-bankpartnershipmailvolumeforourtreatmentgroup, thestatesundertheSecondCircuit(NewYork,Connecticut,andVermont),whilethedottedlineshowsthe aggregateFinTech-bankpartnershipmailvolumeforallotherstates. AfterMay2015,whentheMaddenv. Midlandrulingoccurred,wenotethatthesolicitationmailvolumeinthestatesprimarilyaffecteddropped considerably. Theuncertaintycausedbytherulinghadspillovereffectsintheotherstatesaswell,buttoa lesserextentandforashorterperiod. Source:MintelComperemedia. 29

Table1:ShareofFinTechLoansExceedingTheInterestRateCeilingInArkansas State Ceiling ShareofLoansExceedingIt Prosper LendingClub Arkansas 17 36 19 Source:Authors’calculationsbasedonProsperandLendingClubdatafortheperiodbetween2014and2017. 30

Table2:SummaryStatisticsandVariableDescriptions Description Number Mean Standard Source ofObservations Deviation Dependentvariables Totalunsecuredsolicitationmailvolume 105,420 202 275 Mintel FinTech-bankpartnershipunsecuredsolicitationmailvolume 105,420 72 198 Mintel Financecompanyunsecuredsolicitationmailvolume 105,420 68.7 178 Mintel FinTechlender(nonpartnered)unsecuredsolicitationmailvolume 105,420 9.21 68.2 Mintel Paydaylenderunsecuredsolicitationmailvolume 105,420 7.17 62.7 Mintel Specialistbank(nonpartnered)unsecuredsolicitationmailvolume 105,420 14.9 96.3 Mintel Mainstreambankunsecuredsolicitationmailvolume 105,420 30.1 130 Mintel Explanatoryvariables VantageScorecreditscoreofthereceiveroftheoffer 105,420 720 133 Mintel State-levelGDP(logarithm) 105,420 13.2 0.947 BEA Indicatorvariables Stateinterestratelevel(Rate) 105,420 0.372 0.483 Mintel Householdhead’sageislessthan25years 105,420 0.0226 0.149 Mintel Householdhead’sageis25to39years 105,420 0.272 0.445 Mintel Householdhead’sageis40to54years(omitted) 105,420 0.364 0.481 Mintel Householdhead’sageis55yearsorolder 105,420 0.34 0.474 Mintel HouseholdheadidentifiesasAsian 105,420 0.0114 0.106 Mintel HouseholdheadidentifiesasBlack 105,420 0.0501 0.218 Mintel Householdheadidentifiesaswhite 105,420 0.387 0.487 Mintel Householdheadidentifiesasanotherrace(omitted) 105,420 0.0192 0.137 Mintel Householdhead’sraceisnotspecified 105,420 0.532 0.499 Mintel Householdincomelessthan$24,999 105,420 0.144 0.351 Mintel Householdincome$25,000–$59,999 105,420 0.374 0.484 Mintel Householdincome$60,000–$99,999 105,420 0.306 0.461 Mintel Householdincomeover$100,000(omitted) 105,420 0.157 0.364 Mintel Presenceofchildrenisunknown(omitted) 105,420 0.504 0.5 Mintel Presenceofchildreninthehousehold 105,420 0.123 0.329 Mintel Nochildreninthehousehold 105,420 0.373 0.484 Mintel Householdhead’seducationislessthanhighschool 105,420 0.185 0.388 Mintel Householdheadiseducatedattheundergraduatelevel 105,420 0.214 0.41 Mintel Householdheadiseducatedatthegraduatelevel 105,420 0.0401 0.196 Mintel Householdhead’seducationisunknown(omitted) 105,420 0.561 0.496 Mintel Single-familyhome 105,420 0.377 0.485 Mintel Multi-familyhome 105,420 0.0841 0.277 Mintel Trailerhome 105,420 0.031 0.173 Mintel Residencetypeisunknown(omitted) 105,420 0.508 0.5 Mintel Homeisrented 105,420 0.267 0.443 Mintel Homeisowned 105,420 0.705 0.456 Mintel Thehomestatustypeisunknown(omitted) 105,420 0.0276 0.164 Mintel Subprimeconsumersinlow-ratestates(SL) 105,420 0.142 0.349 Mintel Subprimeconsumersinhigh-ratestates(SH) 105,420 0.11 0.313 Mintel Near-primeconsumersinlow-ratestates(NL) 105,420 0.156 0.363 Mintel Near-primeconsumersinhigh-ratestates(NH) 105,420 0.0938 0.291 Mintel Low-primeconsumersinlow-ratestates(LL) 105,420 0.163 0.369 Mintel Low-primeconsumersinhigh-ratestates(LH) 105,420 0.0873 0.282 Mintel High-primeconsumersinlow-ratestates(HL) 105,420 0.167 0.373 Mintel High-primeconsumersinhigh-ratestates(HH)(omitted) 105,420 0.0811 0.273 Mintel ConsumersinNewYork,Connecticut,orVermont(NYCTVT) 105,420 0.0559 0.23 Mintel PeriodafterMay2015(PostMadden) 105,420 0.786 0.41 Mintel Source:MintelComperemedia,BureauofEconomicAnalysisfortheGDP,andFederalReserveBankofNewYorkConsumerCredit Panel/Equifaxforthecreditscoredistribution. 31

Table3:RegressionResultsforMailVolumebyLenderType All FinTech-Bank Finance FinTech Payday Specialist Mainstream Lenders Partnerships Companies Companies Lenders Banks Banks VARIABLES I II III IV V VI VII A. Rate -67.1 -44.4** 9.46 1.58 2.51 -12.1** -24** (45.677) (21.139) (10.657) (2.237) (2.011) (5.712) (10.330) R-squared 0.527 0.173 0.193 0.024 0.026 0.048 0.097 B. SL 73.9 47.8** 48*** 6.97** 10.5*** 1.86 -41.2*** (47.039) (23.603) (15.115) (3.191) (3.166) (6.019) (9.220) SH 1.77 -19.8** 58.6*** 13.7*** 18.9*** -12*** -57.5*** (7.917) (8.704) (12.439) (3.282) (3.938) (3.225) (11.647) NL 61.1 79.1*** 6.58 3.51 .746 6.22 -35*** (42.729) (22.596) (12.579) (2.544) (1.962) (4.596) (7.777) NH 3.91 35.2*** 16.5*** 7.95*** 3.15* -4.09* -54.9*** (6.284) (8.279) (4.248) (2.357) (1.578) (2.121) (10.611) LL 72.4 72.3*** -4.38 4.28* -.0283 13*** -12.8* (45.338) (22.742) (13.292) (2.413) (1.869) (4.609) (7.021) LH 7.45 38.2*** -1.06 1.98** -1.47* 2.53** -32.7*** (4.531) (8.316) (3.601) (0.784) (0.819) (1.181) (7.485) HL 74 27.9 -10.9 3.99* .932 12.7** 39.4*** (45.417) (22.885) (13.553) (2.298) (1.992) (5.907) (14.313) R-squared 0.527 0.185 0.199 0.026 0.032 0.049 0.119 Controls YES YES YES YES YES YES YES TimeFE YES YES YES YES YES YES YES Obs. 105,420 105,420 105,420 105,420 105,420 105,420 105,420 Note:*,**,and***denotesignificanceatthe10%,5%,and1%levels,respectively.Thistablereportsthecoefficientsandstandard errorsclusteredatthestatelevelobtainedusingequations1(panelA)and2(panelB).Thedependentvariablesrepresentthemail volumeper10,000individualswithcreditscores,inaggregateandforeachofthesixlendercategoriesanalyzed:FinTech-bankpartnerships,financecompanies,FinTechlenders,paydaylenders,specialistbanks,andmainstreambanks.Rateisthekeyexplanatory variableinpanelAandequals1forhigh-ratestatesand0otherwise.ThekeyexplanatoryvariablesforpanelBareindicatorvariablesforthecreditscorecategory(subprime,VantageScorecreditscorelowerthanorequalto660;nearprime,VantageScorecredit scorehigherthan660andlowerthanorequalto719;lowprime,VantageScorecreditscorehigherthan719andlowerthanorequal to792;andhighprime,VantageScorecreditscorehigherthan792)andthelevelofthestateconsumerfinancerateceiling(withhigh andlow).Controlvariablesincludefinancialanddemographiccharacteristicssuchasage,incomelevel,thepresenceofchildren inthehousehold,whetherthehomeisownedorrented,ethnicity,residencetype,householdhead’slevelofeducation,andthe state-levelGDP.TheVantageScorecreditscoreisincludedonlyamongcontrolsusedforregressionresultspresentedinpanelA. Source: Mintel Comperemedia, Bureau of Economic Analysis, and Federal Reserve Bank of New York Consumer Credit Panel/Equifax. 32

Table4:RegressionResultsforMailVolumebyCreditRiskGroupinLightoftheMaddenRuling AllConsumers Subprime NearPrime LowPrime HighPrime VARIABLES I II III IV V VI A.AllLenders NYCTVT -69.8 87.5 116 64.1 78.1 95.4 (99.095) (52.451) (76.203) (41.607) (50.166) (61.125) PostMadden -29.4** -11.4 -9.61 -13* -16.6 -7.91 (12.339) (7.168) (11.964) (6.871) (10.100) (8.952) NYCTVT·PostMadden 15.9 -14 -51.3 2.37 -1.17 -23.5 (24.176) (27.264) (68.049) (13.735) (21.689) (40.041) B.FinTech-BankPartnerships NYCTVT 17.3 72.8*** 106* 98.9*** 68.1** 45.1** (35.683) (22.496) (63.066) (31.348) (26.362) (17.405) PostMadden 38.4*** 41.2*** 37.2*** 47.2*** 40.9*** 30.5*** (7.516) (8.310) (10.833) (8.736) (11.412) (6.040) NYCTVT·PostMadden -44.4*** -56.1*** -66.8 -76.2*** -50.7** -37.4*** (11.285) (17.235) (51.842) (27.368) (20.673) (12.984) C.FinanceCompanies NYCTVT -101*** -43.1** -4.72 -51.5*** -48.5*** -56.8*** (22.650) (19.196) (16.884) (14.373) (15.967) (20.233) PostMadden -69.2*** -60.8*** -43.4*** -75.4*** -57.2*** -53.6*** (12.943) (9.792) (8.505) (12.001) (14.721) (10.658) NYCTVT·PostMadden 55.7*** 47.4*** 16.6 60.8*** 48.4** 56.5*** (14.754) (13.685) (27.166) (17.875) (19.638) (12.438) D.FinTechCompanies NYCTVT -6.52*** .375 -6.2 .149 2.44 2.5 (1.690) (3.344) (5.409) (2.666) (3.184) (3.151) PostMadden 1.94* 2.52* .695 .24 1.78 6.44*** (1.142) (1.348) (3.323) (1.379) (1.509) (1.459) NYCTVT·PostMadden 3.02 1.83 6.95 .105 3.23 -.236 (3.582) (2.818) (5.189) (1.986) (4.973) (3.895) E.PaydayLenders NYCTVT -10.7*** -4.12 -10.2 -2.42 -3.1*** -4.21*** (2.040) (2.511) (8.089) (1.787) (1.137) (1.435) PostMadden -4.62*** -4.96*** -10.8*** -1.37 -4.35*** -3.86*** (1.229) (1.183) (3.791) (1.413) (1.491) (0.849) NYCTVT·PostMadden 4.17*** 3.38* 8.44 1.09 2.97 3.79*** (1.273) (1.718) (6.241) (1.470) (1.942) (1.121) F.SpecialistBanks NYCTVT 2.84 15.3** 28.3 11.8** 14.6* 19.4** (5.674) (6.448) (23.700) (5.024) (7.960) (8.811) PostMadden 12.7*** 12.9*** 5.26* 10*** 15*** 20.7*** (2.416) (2.652) (3.019) (2.406) (3.363) (4.683) NYCTVT·PostMadden 5.65 3.03 -17.5 -.387 15.7 -7.26 (12.587) (11.505) (21.457) (8.685) (26.524) (8.830) G.MainstreamBanks NYCTVT 28.3 46.3 3.01 7.14 46 89.4 (46.268) (36.856) (2.375) (5.065) (33.087) (54.945) PostMadden -8.52*** -2.24 1.32* 6.38*** 2.93 -8.05 (3.120) (2.210) (0.714) (1.490) (3.376) (5.632) NYCTVT·PostMadden -8.15 -13.5 .942 16.9 -22.1 -38.9 (17.556) (18.210) (3.084) (23.437) (24.922) (39.066) Controls NO YES YES YES YES YES Observations 105,420 105,420 26,531 26,359 26,337 26,193 Note:*,**,and***denotesignificanceatthe10%,5%,and1%levels,respectively.Standarderrorsclusteredatthestatelevel. Source:MintelComperemedia,BEA,andFederalReserveBankofNewYorkConsumerCreditPanel/Equifax. 33

References Adams,J.(2021). "CreditUnionsPlayCatch-UpinBuyNow/PayLater". AmericanBanker,August 30. Adams,R.M.(2018). "DoMarketplaceLendingPlatformsOfferLowerRatestoConsumers?"FEDS Notes. Washington: Board of Governors of the Federal Reserve System, October 22, https:// doi.org/10.17016/2380-7172.2268. Balyuk, T. (2019). "Financial Innovation and Borrowers: Evidence from Peer-to-Peer Lending," working paper (May), https://www.fdic.gov/bank/analytical/fintech/papers/ balyuk-paper.pdf. Balyuk, T., A. N. Berger, and J. Hackney (2020). "What Is Fueling FinTech Lending? The Role of Banking Market Structure," working paper, June (revised July), https://papers.ssrn.com/ sol3/papers.cfm?abstract_id=3633907. Bhattacharyya, S. (2021). "Behind OppFi’s Approach to Reaching the Credit Underserved". Fin- Ledger,April30. Boczar,G.E.(1978). “CompetitionbetweenBanksandFinanceCompanies:ACrossSectionStudy ofPersonalLoanDebtors”. JournalofFinance,vol.33(March),pp.245–258. Brennan, C. M. and N. R. Udell (2018). "What’s Old Is New Again: The Future of Bank PartnershipProgramsfromSmallDollarInstallmentLoanstoMortgagestoEverything". Conferenceon ConsumerFinanceLawQuarterlyReport,vol.72(4). Buchak, G., G. Matvos, T. Piskorski, and A. Seru (2017). "Fintech, Regulatory Arbitrage, and the RiseofShadowBanks,"NBERWorkingPaperSeries23288.Cambridge,Mass.: NationalBureau of Economic Research, March (revised September 2018), https://www.nber.org/papers/ w23288. Chen, L. and G. Elliehausen (2020). "The Cost Structure of Consumer Finance Companies and ItsImplicationsforInterestRates: EvidencefromtheFederalReserveBoard’s2015SurveyofFinanceCompanies,"FEDSNotes.Washington:BoardofGovernorsoftheFederalReserveSystem, August12,https://doi.org/10.17016/2380-7172.2610. Cornaggia,J.,B.Wolfe,andW.Yoo(2018). "CrowdingOutBanks: CreditSubstitutionbyPeer-To- PeerLending",workingpaper,March(revisedSeptember2019),https://papers.ssrn.com/ sol3/papers.cfm?abstract_id=3000593. Crosman,P.(2018)."CryptoLendingMayBeRisky,butTheseFirmsSayThey’veSolvedtheRiddle". AmericanBanker,April30. 34

Crosman, P. (2021). "U.S. Banks Give Embedded Finance a Whirl. Will It Catch On?". American Banker,August30. Danisewicz,P.andI.Elard(2018). "TheRealEffectsofFinancialTechnology:MarketplaceLending and Personal Bankruptcy", working paper, July (revised April 2022), https://papers.ssrn. com/sol3/papers.cfm?abstract_id=3208908. Dettling,L.J.andJ.W.Hsu(2021). "MinimumWagesandConsumerCredit:EffectsonAccessand Borrowing". ReviewofFinancialStudies,vol.34(May),pp.2549–79. DiMaggio,M.andV.Yao(2020). "Fintechborrowers:Lax-ScreeningorCream-Skimming?",NBER WorkingPaperSeries28021.Cambridge,Mass.:NationalBureauofEconomicResearch,October, https://www.nber.org/papers/w28021. Dore, T. and T. Mach (2019). "Marketplace Lending and Consumer Credit Outcomes: Evidence fromProsper,"FinanceandEconomicsDiscussionSeries2019-022.Washington: BoardofGovernorsoftheFederalReserveSystem,April,https://doi.org/10.17016/FEDS.2019.022. Duffy,D.(2004). "RefundAnticipationLoans,"OLRResearchReport,OLRResearchReport2004- R-0346.Hartford,Conn.:ConnecticutGeneralAssembly,March24. Durkin,T.andG.Elliehausen(2000). "InterinstitutionalCompetitionforConsumerCreditatthe EndoftheTwentiethCentury,"unpublishedpaper. Durkin, T.A., G.Elliehausen, andM.Hwang(2016). "RateCeilingsandtheDistributionofSmall DollarLoansfromConsumerFinanceCompanies:ResultsofaNewSurveyofSmallDollarCash Lenders,"mimeo. Durkin,T.A.,G.Elliehausen,M.E.Staten,andT.J.Zywicki(2014). ConsumerCreditandtheAmericanEconomy. NewYork:OxfordUniversityPress. Elliehausen,G.,S.M.Hannon,andT.W.MillerJr.(2021). "ANewLookattheEffectsoftheInterestRateCeilinginArkansas,"FinanceandEconomicsDiscussionSeries2021-045.Washington: Board of Governors of the Federal Reserve System, July, https://doi.org/10.17016/FEDS. 2021.045. Erel,I.andJ.Liebersohn(2020). "DoesFinTechSubstituteforBanks? EvidencefromthePaycheck ProtectionProgram,"NBERWorkingPaperSeries27659.Cambridge,Mass.: NationalBureauof EconomicResearch,August(revisedDecember),https://www.nber.org/papers/w27659. Gargano, F. (2021). "Credit Unions Enlist Fintechs to Vie with Commercial, Challenger Banks". AmericanBanker,August27. 35

Glaeser,E.L.andJ.Scheinkman(1998). "NeitheraBorrowerNoraLenderBe:AnEconomicAnalysisofInterestRestrictionsandUsuryLaws". TheJournalofLawandEconomics,vol.41(April), pp.1–36. Han, S., B. J. Keys, and G. Li (2018). "Unsecured Credit Supply, Credit Cycles, and Regulation". ReviewofFinancialStudies,vol.31(March),pp.1184–1217. Havrylchyk,O.,M.Carlotta,R.Talal,andV.Marianne(2020)."TheExpansionofPeer-to-PeerLending". ReviewofNetworkEconomics,vol.19(September),pp.145–187. Honigsberg,C.,R.J.JacksonJr.,andR.Squire(2017). "HowDoesLegalEnforceabilityAffectConsumerLending? EvidencefromaNaturalExperiment". JournalofLawandEconomics,vol.60 (November),pp.673–712. Horn, C. M. and M. R. Hall (2017). "The Curious Case of Madden v. Midland Funding and the SurvivaloftheValid-When-MadeDoctrine". NorthCarolinaBankingInstitute(vol.21). Jagtiani,J.andC.Lemieux(2018). "DoFintechLendersPenetrateAreasThatAreUnderservedby Traditional Banks?". Journal of Economics and Business, vol. 100 (November–December), pp. 43–54. Lee, D. and W. V. der Klaauw (2010). "An introduction to the FRBNY Consumer Credit Panel". Technical report, New York: Federal Reserve Bank of New York, November 10, https://www. newyorkfed.org/research/staff_reports/sr479.html. Levitin, A.(2021). "Rent-a-Bank: BankPartnershipsandtheEvasionofUsuryLaws". DukeLaw Journal,vol.71,pp.329–417. Lukongo,O.E.B.andT.W.MillerJr.(2018). "EvaluatingtheSpatialConsequenceofInterestRate CeilingUsingaSpatialRegimeChangeApproach". AmericanEconomist,vol.63(October),pp. 166–86. Lukongo,O.E.B.andT.W.MillerJr.(2022). Measuringtheconsequencesofabindinginterestrate caponsmall-dollarinstallmentloans. JournalofFinancialResearch,forthcoming. Michelman,I.S.(1966). ConsumerFinance: ACaseStudyinAmericanBusiness.NewYork: FrederickFell. Miller Jr., T. W. (2019). How Do Small-Dollar, Nonbank Loans Work? Arlington, Va.: Mercatus CenteratGeorgeMasonUniversity. Millerbernd,A.(2021). "TaxRefundLoans:GetanAdvancein2021". NerdWallet,December7. Mors, W. P. (1965). Consumer Credit Finance Charges. New York: National Bureau of Economic Research. 36

NationalCommissiononConsumerFinance(1972). ConsumerCreditintheUnitedStates.Washington:GovernmentPrintingOffice. Oeltjen, J. C. (1975). Usury: Utilitarian or Useless? Florida State University Law Review, vol. 3 (Spring),pp.167–235. OfficeoftheFederalRegister,NationalArchivesandRecordsAdministration(2020). FederalRegister,vol.85,issue141(July22),pp.44145–44450. Rogers,D.(1975). ConsumerBankinginNewYork. NewYork:ColumbiaUniversityPress. Scully,M.(2015). "HowBankPartnershipswithMarketplaceLoanPlatformsReallyWork". AmericanBanker,February24. Stegman,M.A.(2007). "PaydayLending". JournalofEconomicPerspectives,vol.21(Winter),pp. 169–190. Tang, H. (2019). "Peer-to-Peer Lenders Versus Banks: Substitutes or Complements?". Review of FinancialStudies,vol.32(May),pp.1900–1938. 37

Appendix A. AdditionalFigures FiguresA1throughA3showafewexamplesofmailofferssentbyFinTechlendersinpartnership withspecialistbanks.Thehighlightedportionindicatesthebankpartnerissuingtheloan. 38

FigureA1.Prosper/WebBankPersonalLoanOfferforCreditCardLoanConsolidation Source:MintelComperemediacampaigndatabase. 39

FigureA2.LendingClub/WebBankPersonalLoanOffer Source:MintelComperemediacampaigndatabase. 40

FigureA3.BestEgg/CrossRiverBankPersonalLoanOffer Source:MintelComperemediacampaigndatabase. 41

B. AdditionalTable Table B1 presents robustness checks consisting of the main results of the estimation using the firsttwoequationscontrastedwiththoseobtainedwhenexcludingfromtheestimationtheoutlierstates,ArkansasandIowa. 42

TableB1:RegressionResultsforMailVolumebyLenderType—RobustnessCheck All FinTech-Bank Finance FinTech Payday Specialist Mainstream Lenders Partnerships Companies Companies Lenders Banks Banks VARIABLES all excl. all excl. all excl. all excl. all excl. all excl. all excl. states AR&IA states AR&IA states AR&IA states AR&IA states AR&IA states AR&IA states AR&IA A. Rate -67.1 -68 -44.4** -42.1** 9.46 7.2 1.58 1.37 2.51 2.19 -12.1** -12.3** -24** -24.4** (45.677) (47.155) (21.139) (20.630) (10.657) (11.064) (2.237) (2.301) (2.011) (2.042) (5.712) (5.910) (10.330) (10.727) R-squared 0.527 0.521 0.173 0.166 0.193 0.198 0.024 0.024 0.026 0.026 0.048 0.047 0.097 0.096 B. SL 73.9 76 47.8** 49** 48*** 48.3*** 6.97** 7.14** 10.5*** 10.7*** 1.86 1.76 -41.2*** -40.9*** (47.039) (48.310) (23.603) (23.592) (15.115) (15.601) (3.191) (3.281) (3.166) (3.249) (6.019) (6.188) (9.220) (9.340) SH 1.77 2.88 -19.8** -17.2** 58.6*** 57.7*** 13.7*** 13.6*** 18.9*** 18.8*** -12*** -12.2*** -57.5*** -57.9*** (7.917) (7.888) (8.704) (8.177) (12.439) (12.285) (3.282) (3.268) (3.938) (3.920) (3.225) (3.253) (11.647) (11.702) NL 61.1 60.8 79.1*** 73.7*** 6.58 9.58 3.51 3.79 .746 1.2 6.22 6.22 -35*** -33.7*** (42.729) (44.367) (22.596) (21.554) (12.579) (12.906) (2.544) (2.612) (1.962) (2.016) (4.596) (4.735) (7.777) (7.888) NH 3.91 4.62 35.2*** 37.1*** 16.5*** 15.9*** 7.95*** 7.89*** 3.15* 3.12* -4.09* -4.23* -54.9*** -55.1*** (6.284) (6.299) (8.279) (8.174) (4.248) (4.144) (2.357) (2.347) (1.578) (1.567) (2.121) (2.143) (10.611) (10.665) LL 72.4 74.8 72.3*** 72.7*** -4.38 -2.6 4.28* 4.39* -.0283 .253 13*** 13*** -12.8* -12.9* (45.338) (46.713) (22.742) (22.792) (13.292) (13.861) (2.413) (2.507) (1.869) (1.951) (4.609) (4.793) (7.021) (7.275) LH 7.45 7.68* 38.2*** 39*** -1.06 -1.37 1.98** 1.97** -1.47* -1.47* 2.53** 2.47** -32.7*** -32.9*** (4.531) (4.562) (8.316) (8.373) (3.601) (3.654) (0.784) (0.786) (0.819) (0.825) (1.181) (1.185) (7.485) (7.523) HL 74 75.4 27.9 28.4 -10.9 -9.24 3.99* 4.06* .932 1.19 12.7** 13.1** 39.4*** 37.9** (45.417) (46.463) (22.885) (22.325) (13.553) (14.127) (2.298) (2.378) (1.992) (2.082) (5.907) (6.069) (14.313) (14.364) R-squared 0.527 0.521 0.185 0.177 0.199 0.204 0.026 0.026 0.032 0.032 0.049 0.049 0.119 0.117 Controls YES YES YES YES YES YES YES YES YES YES YES YES YES YES TimeFE YES YES YES YES YES YES YES YES YES YES YES YES YES YES Obs. 105,420 104,084 105,420 104,084 105,420 104,084 105,420 104,084 105,420 104,084 105,402 104,084 105,420 104,084 Note:*,**,and***denotesignificanceatthe10%,5%,and1%levels,respectively.Thistablereportsthecoefficientsandstandarderrorsclusteredatthestatelevelobtainedusingequations 1(panelA)and2(panelB).Thedependentvariablesrepresentthemailvolumeper10,000individualswithcreditscores,inaggregateandforeachofthesixlendercategoriesanalyzed: FinTech-Bankpartnerships,financecompanies,FinTechlenders,paydaylenders,specialistbanks,andmainstreambanks. RateisthekeyexplanatoryvariableinpanelAandequals1for high-ratestates,and0otherwise.ThekeyexplanatoryvariablesforpanelBareindicatorvariablesforthecreditscorecategory(subprime,VantageScorecreditscorelowerthanorequalto660; nearprime,VantageScorecreditscorehigherthan660andlowerthanorequalto719;lowprime,VantageScorecreditscorehigherthan719andlowerthanorequalto792;andhighprime, VantageScorecreditscorehigherthan792)andthelevelofthestateconsumerfinancerateceiling(withhighandlow).Controlvariablesincludefinancialanddemographiccharacteristics suchasage,incomelevel,thepresenceofchildreninthehousehold,whetherthehomeisownedorrented,ethnicity,residencetype,householdhead’slevelofeducation,andthestate-level GDP.TheVantageScorecreditscoreisincludedonlyamongcontrolsusedforregressionresultspresentedinpanelA.Source:MintelComperemedia,BureauofEconomicAnalysis,andFederal ReserveBankofNewYorkConsumerCreditPanel/Equifax. 43