国际清算银行-金融科技与银行信贷：他们对货币政策有何反应？（英）-2023.12_市场营销策划_重点.docx

BISBISWorkingPapersNo1157Fintechvsbankcredit:Howdotheyreacttomonetarypolicy?byGiulioCornelli,FiorellaDeFiore,LeonardoGambacortaandCristinaManeaMonetaryandEconomicDepartmentDecember2023JELclassification:D22,G31zR30Keywords:fintechcredit,monetarypolicy,PVAR,collateralchannelBISWorkingPapersarewrittenbymembersoftheMonetaryandEconomicDepartmentoftheBankforInternationalSettlements,andfromtimetotimebyothereconomists,andarepublishedbytheBank.ThepapersareonSUtyeCtSoftopicalinterestandaretechnicalincharacter.TheviewsexpressedinthemarethoseoftheirauthorsandnotnecessarilytheviewsoftheBIS.ThispublicationisavailableontheBISwebsite(www.bis.org).©BankforInternationalSettlements2023.Allrightsreserved.Briefexcerptsmaybereproducedortranslatedprovidedthesourceisstated.ISSN1020-0959(print)ISSN1682-7678(online)Fintechvsbankcredit:howdotheyreacttomonetarypolicy?GiulioCorneIIi,FiorellaDeFiore,LeonardoGambacortaandCristinaManea*AbstractFintechcredit,whichincludespeer-to-peerandmarketplacelendingaswellaslendingfacilitatedbymajortechnologyfirms,iswitnessingrapidgrowthworldwide.However,itsresponsivenesstomonetarypolicyshiftsremainslargelyunexplored.Thisstudyemploysanovelcreditdatasetspanning19countriesfrom2005to2020andconductsaPVARanalysistoshedsomelightonthedifferentreactionoffintechandbankcredittochangesinpolicyrates.Themainresultisthatfintechcreditshowsalower(evennon-significant)sensitivitytomonetarypolicyshocksincomparisontotraditionalbankcredit.Giventhestillmarginal-althoughfastgrowing-macroeconomicsignificanceoffintechcredit,itscontributioninexplainingthevariabilityofrealGDPislessthan2%,againstaroundonequarterforbankcredit.JELCodes:D22,G31,R30.Keywords:fintechcredit,monetarypolicy,PVAR,collateralchannel.GiulioCornelli(email:giulio.cornellibis.org)iswiththeBankforInternationalSettlements(BIS)andtheUniversityofZurich(UZH).Correspondingauthor.FiorellaDeFiore(email:fiorella.defiorebis.org)andLeonardoGambacorta(email:Ieonardo.gambacortabis.org)arewiththeBISandresearchfellowsofCEPR.CristinaManea(CriStina.maneabis.org)iswiththeBIS.TheauthorsthankMaxCroce,MarcoJacopoLombardiandoneanonymousrefereeforhelpfulcomments.TheviewsexpressedarethoseoftheauthorsanddonotnecessarilyrepresentthoseoftheBankforInternationalSettlements,UZHandCEPR.1. IntroductionCreditmarketsareundergoingaprofoundtransformation.Whiletraditionallenderssuchasbanksandcreditunionscontinuetobetheprimarysourceoffinanceinmosteconomies,withcapitalmarketsalsoplayinganimportantroleinsomecases,newintermediarieshavebeguntomaketheirmark.Inparticular,digitallendingmodelssuchaspeer-to-peerandmarketplacelendinghaveseensignificantgrowthinnumerouseconomiesoverthepastdecade(Claessensetal.,2018).Furthermore,inmorerecentyears,severalprominenttechnology-drivencompanies(oftenreferredas,bigtechs/z)haveventuredintocreditmarkets,providingloanstotheirclientseitherdirectlyorinpartnershipwithfinancialinstitutions(Frostetal.2019).Thesenewtypesofcredit,enabledbyonlineplatformsandbigdataforassessingcreditworthinessarecommonlytermed,fintechcredit". Fintech credit encompasses various innovative credit forms. This includes digital lending models such as peer- to-peer (P2P)marketplace lending and invoice trading, all facilitated by online platforms rather than traditional banks or lending institutions. Another notable form is wbig tech credit1', which is credit extended either directly or in partnership with financial institutions by large firms primarily engaged in the technology sector. For simplicity in this paper we group these two alternative finance forms together, referring to both collectively as wfintech credit".Fintechcreditiswitnessingrapidglobalexpansion,achievingmacroeconomicsignificanceinmanycountriesincludingChina,Korea,Malaysia,andKenyawhereitreachesupto5%oftotalcredit(Cornellietal.r2023).Inlightofthistrend,itbecomesessentialtoinvestigatehowfintechcreditrespondstomonetarypolicyandtoidentifythekeydifferencesinitsmonetarytransmissionmechanismrelativetotraditionalbankcredit. See De Fiore et al (2023) for a model-based analysis of the relative impact of big tech and bank credit on the transmission of monetary policy.Threeprimarydifferencesbetweenfintechandbankcreditcouldinfluencetheirresponsestoamonetarypolicyshock.First,ratherthanrelyingonphysicalcollateraltoaddressagencyissuesbetweenlendersandborrowers,thebusinessmodeloffintechcreditisgroundedindata(Gambacortaetal.,2019).Asaresult,fintechcreditresponsivenesstoassetpricefluctuationstriggeredbyshiftsinmonetarypolicyislower(Gambacortaetal.,2022).Second,fintechplatformsmayoperatewithinregulatoryframeworksdistinctfromthosegoverningtraditionalbanks,enablingthemtoextendcreditundervariedterms.Moreover,thecompetitivedynamicsbetweenfintechplatformsandconventionalbankscanshapecreditofferingsandtheirreactionstomonetarypolicyindifferentways.Astraditionalbankcreditbecomesmoreconstrainedduetomonetarypolicytightening,businessescouldreaddresstheirneedstowardsfintechplatforms(Hasanetal.,2023).Third,thesuperiormonitoringandscreeningcapabilitiesofbigtechlendersrendertheircreditscoringhighlysensitivetochangesinfirms1transactionvolumesandnetworkscores,especiallyforonlinefirms(Gambacortaetal.2022).Therefore,anyalterationinmonetarypolicyaffectinggeneralbusinessconditionscouldswiftlyinfluencecreditsupply.Inparticular,whenmonetarypolicyisrelaxed,bigtechlendersaremorelikelytoestablishnewlendingrelationshipswithfirmsthantheirtraditionalcounterparts(Huangetal”2023).Thissuggeststhatbigtechcreditmightfacilitatethetransmissionofmonetarypolicyviatheextensivemarginrelativetotraditionalbankloans.Insummary,whilethefirsttwodifferencessuggestadiminishedeffectivenessofmonetarypolicythroughfintechcredit,thelatterwouldimplytheopposite.Toshedsomelightonwhichoftheseeffectsdominates,thispaperutilisesnewdatafor19countriesovertheperiod2005-2020(Cornellietal,2023).WeconductaPanelVAR(PVAR)analysistoassesstheresponsesoffintechandbankcredittoamonetarypolicyshock.Ourprimaryfindingisthatfintechcreditexhibitsareduced(evennon-significant)responsivenesstomonetarypolicyshockscomparedtobankcredit.2. DatadescriptionThePVARanalysisisbasedonannualdatafor19countriesovertheperiod2005to2020. CountriesZgeographical areas included in the analysis are: Australia, Brazil, Canada, Chile, China, Euro area, Indonesia, Israel, India, Japan, Korea, Mexico, Russia4 South Africa, Switzerland, Thailand, Turkey, United Kingdom and United States. The behaviour of fintech and bank credit may vary between advanced economies (AEs) and emerging market economies (EMEs). However, due to the limited number of observations available (96 for AEs and 150 for EMEs), we are unable to perform a sample split analysis for the two groups of countries.Theinteractionbetweenmonetarypolicy,thecreditmarketandeconomicactivityisanalysedbymeansofthefollowingvariables:i)thelogarithmofthepropertypriceindex(Pk);ii)thelogarithmofrealGDP(V);iii)thelogarithmoftheconsumerpriceindex(p);iv)thelogarithmofbanklending(£);v)thelogarithmoffintechcredit(F);vi)themonetarypolicyshortterminterestrate(i).ThepropertypriceindexandthebankcreditdataarecompiledbytheBIS.TherealGDPandtheCPIcomefromtheIMF,WorldEconomicOutlook.Theshorttermratehasbeenobtainedfromnationalcentralbanks,Based on data availability, we replace the short-term rate with the shadow rate from UKmfa, UK Limited. For more details see Krippner (2013).whilefintechcreditcomesfromthenewdatasetdevelopedinCornellietal(2023).Toavoidtheproblemofspuriouscorrelations,wehaveconsideredaPVARinfirstdifferences.ThesummarystatisticsofallthevariablesusedintheanalysisarereportedinTable1.Summarystatistics1Table1ObservationsMeanStddevMinMaxLn(propertypriceindex)2740.050.05-0.020.18Ln(realGDP)3040.010.09-0.160.16Ln(CPI)3040.030.030.000.10Ln(bankcredit)3040.070.13-0320.46Ln(fintechcredit)3040.380.73-0.222.43shorttermrate3040.231.56-9.507.771DataWinsorisedatthe5thand95thpercentiles.Sources:Cornellietal(2023);BIS;IMF;nationaldata;authors'calculations.Table2belowreportsunitrootPhillips-Perrontestsforallvariablesinfirstdifferences.Thenullhypothesisthatthevariablescontainunitrootsisalwayslargelyrejected.Unitroottests1Table2Ln(propertypriceindex)Ln(realGDP)Ln(fintechcredit)shorttermratee£U-,-5b方P-valueStatP-valueP-valueStatP-value)P-valueco方P-valueInversechi-squared(38)81.70.0013430.00104.60.00204.70.0C100.10.0C203.50.0CInversenormal-4.00.00-730.00-5.80.00-10.30.0C-5.80.0C-10.80.0CInverselogitt(99)-4.10.00-820.00-6.10.00-12.80.0C-6.00.0C-12.90.0CModifiedinvchi-squared5.00.0011.00.007.60.0019.10.0C7.10.0C19.00.0C1BasedonPhillips-Perrontests.Thenullhypothesisisthatallpanelscontainunitroots.Thesampleincludes19countriesovertheperiod2005-2020.DataWinsorisedatthe5thand95thpercentiles.Sources:Authors'calculations.3. ThePVARModelWemodelasix-variableVARsystem;allthevariables,thatarefoundtobe1(0),aretreatedasendogenous.Therefore,thestartingpointofthemultivariateanalysisis:2ct=c+Zctk+££ctC=I,Nt=lf,TklVVWN(ON)(1)wherezct=pk,Y,p,L,F,iandctisavectorofresiduals.We treat cross-sectional shocks as independent, and we do not model the transmission across borders explicitly. This assumption is aligned with the modelling approach where each country's shocks are not directly influenced by shocks in other countries contemporaneously. This simplification ensures the model's tractability and interpretability, especially given the focus on the effects on fintech and bank credit. The constraint of limited data, especially the time dimensions, further restricts our ability to adopt more sophisticated modelling techniques that could potentially capture cross-country interdependencies. For instance, methods like Global VAR (GVAR) or other multi-country econometric models which are adept at capturing such dynamics require a more extensive dataset as well as additional identifying assumptions to yield reliable estimates. For a discussion of challenges and potential biases introduced by the absence of cross-country interdependencies in PVAR models see Canova and Ciccarelli (2013).Thedeterministicpartofthemodelincludescountryfixedeffects(c),whilethenumberoflags(/)issetto1.TheoptimallagselectioncriteriafollowsAndrewsandLu(2001).Table3belowpresentstheresultsfromthefirst-,second-,third-,andfourth-orderPVARmodelsusingthefirstfourlagsoftheendogenousvariablesasinstruments.Forthefourth-orderpanelVARmodel,onlythecoefficientofdetermination(CD)iscalculatedbecausethemodelisjust-identified.Thefirst-orderPVARisthepreferredmodelbecauseithasthesmallestMBIC,MAIC,andMQIC.Foralagequalto1alsotheCDisminimized.While we also want to minimize Hansen,s J statistic, it does not correct for the degrees of freedom in the model like the MMSC by Andrews and Lu (2001).Thechoiceofthedeterministiccomponent(constantvstrend)hasbeenverifiedbytestingthejointhypothesisofboththerankorderandthedeterministiccomponent(so-calledPantulaprinciple).BeforeperformingtestsonthePVARmodel,wehaveanalysedGrangercausalityamongthezctvariables,focusingonfintechcreditinparticular.Grangertestsverifyifthe×variableisusefulinpredictingthevaluesofanothervariabley,conditionalonpastvaluesofy,thatis,whether%,Granger-causes,y(Granger1969).ThiscanbeimplementedasseparateWaldtestswiththenullhypothesisthatthecoefficientsonallthelagsofanendogenousvariablearejointlyequaltozero;thus,thecoefficientsmaybeexcludedinanequationofthePVARmodel.LagselectionTable3LagsCDJJpvalueMBICMAICMQIC10.86133.590.05-442.86-92.41-228.1620.9756.130.92-328.17-87.87-185.0330.9822.230.96-169.92-49.77-983540.961Thesampleincludes19countriesovertheperiod2005-2020.DataWinsorisedatthe5thand95thpercentiles.Sources:Cornellietal(2023);BIS;nationaldata;Authors'calculations.Table4belowshowsthetestonwhetherthecoefficientsonthelagsofeachvariablearezero.Forexample,theteststhatthechangesinbankcreditormonetarypolicyinterestratesdonotGranger-causethechangeinthelogarithmofthepropertypriceindexarerejectedatthe95%confidencelevel.Interestingly,whilefintechcreditdoesnotGrangercausethepropertypriceindex,itGrangercausesCPIprices,bankcreditandtheshorttermrate.FintechcreditmarginallyGrangercausesrealGDP(p-value0.13)alsoinconsiderationofitsstilllimitedmacroeconomicimpact.PVARGrangertestTable4Equation/excludedLn(propertypriceindex)Ln(realGDP)Ln(CPI)Ln(bankcredit)Ln(fintechcredit)shorttermrate(N毛p-value(N写p-value(N毛p-value(NWp-valueOl毛王p-value(N毛*gQ.Ln(propertypriceindex)0.010.9113.210.001.010.320.110.7713.410.00Ln(realGDP)0.110.710.010.970.410.540.610.452.510.12Ln(CPI)2.510.126.910.016.910.011.010321.010.32Ln(bankcredit)7.110.01,109.910.001.310.262.510.121.610.20Ln(fintechcredit)0.010.892.310.133.110.086.110.013.410.07shorttermrate4.310.048.210.001.010.326.410.010.210.64All26350.00I145.650.0027.150.0028.450.004.350.5018.450.00ThenullhypothesisofthetestisthattheexcludedvariabledoesnotGranger-causetheequationvariable.1Thesampleincludes19countriesovertheperiod2005-2020.DataWinsorisedatthe5thand95thpercentiles.Sources:Authors*calculations.AftercheckingforthestabilityofthePVAR(seeFigureAlintheAppendix),wecalculateorthogonalizedImpulseResponseFunctions(IRFs)andForecastErrorVarianceDecompositions(FEVDs).OrthogonalizedIRFsandFEVDsmaychangedependingonhowtheendogenousvariablesareorderedintheCholeskydecomposition.Specifically,theorderingconstrainsthetimingoftheresponses:shocksonvariablesthatcomeearlierintheorderingwillaffectsubsequentvariablescontemporaneously,whileshocksonvariablesthatcomelaterintheorderingwillaffectonlythepreviousvariableswithalagofoneperiod.BecausetheorderingofvariablesislikelytoaffectorthogonalizedIRFsandtheinterpretationoftheresults,inaccordancewiththetheory,weorderthevariablesasfollows:pk,Y,p,L,F,i.Theinterestrateisorderedlast,soitreactstoallvariableswithinoneyear.ThischoiceisguidedbytheliteraturethatanalysestheeffectivenessofmonetarypolicyshocksusingVARmodels.Graph1reportstheIRFs.ConfidenceintervalsarecalculatedusingMonteCarlosimulationwithp-valuebandsof90%.TheIRFssuggestthatwhileamonetarytighteninghasanegativeeffectonassetpricesandbankcredit,fintechcreditremainsunaffected.A1.1percentagepointincreaseinthemonetarypolicyrate(topleftpanel)isassociatedwitha0.5percentdeclineinassetpricesafterthefirstyearand0.4inthesecondyear(bottomrightpanel).Theeffectbecomesstatisticallynotdifferentfromzerofromthethirdyearonwards,whenalsotheinterestratereturnstowardsthebaseline.Bankcreditdropssignificantlyasaneffectofthemonetarypolic