BIS-大型科技公司與貨幣政策的信貸渠道

BISWorking

PapersNo

1088BigtechsandthecreditchannelofmonetarypolicybyFiorellaDeFiore,LeonardoGambacortaandCristina

ManeaMonetaryandEconomic

DepartmentApril2023JELclassification:E44,E51,E52,G21,

G23.Keywords:BigTechs,monetarypolicy,credit

frictions.BISWorkingPapersarewrittenbymembersoftheMonetaryandEconomicDepartment

of

the

Bank

for

International

Settlements,

and

from

time

to

time

by

othereconomists,andarepublishedbytheBank.Thepapersareonsubjectsoftopicalinterest

and

are

technical

in

character.

The

views

expressed

in

them

are

those

of

theirauthorsandnotnecessarilytheviewsofthe

BIS.ThispublicationisavailableontheBISwebsite

().? BankforInternationalSettlements2023.Allrightsreserved.Briefexcerptsmaybereproducedortranslatedprovidedthesourceis

stated.ISSN1020-0959

(print)ISSN1682-7678

(online)BigTechsandtheCreditChannelofMonetary

PolicyF.

De

Fiore? L.

Gambacorta? C.

Manea?§March9,

2023AbstractWedocumentsomestylizedfactsonbigtechcreditandrationalizethemthroughthelensofamodelwherebigtechsfacilitatematchingonthee-commerceplatformandextendloans.Thebigtechreinforcescreditrepaymentwiththethreatofexclusionfromtheplatform,whilebankcreditissecuredagainstcollateral.Ourmodelsuggeststhat:(i)ariseinbigtechs’matchingefficiencyincreasesthevalueforfirmsoftradingontheplatformandtheavailabilityofbigtechcredit;(ii)bigtechcreditmitigatestheinitialresponseofoutputtoamonetaryshock,whileincreasingitspersistence;(iii)theefficiencygainsgeneratedbybigtechsarelimitedbythedistortionaryfeescollectedfrom

users.JELclassification:E44,E51,E52,G21,

G23Keywords:BigTechs,monetarypolicy,credit

frictions?BankforInternationalSettlementsandCEPR.Email:

Fiorella.DeFiore@.?BankforInternationalSettlementsandCEPR.Email:

Leonardo.Gambacorta@.?BankforInternationalSettlements.Email:

Cristina.Manea@.§ThisresearchprojectwaspartiallycompletedwhileC.Maneawasworking

for

theresearchcenterof

the

DeutscheBundesbank.TheviewsexpressedinthispaperareourownandshouldnotbeinterpretedasreflectingtheviewsoftheBankforInternationalSettlements,theDeutscheBundesbank,ortheEurosystem.WethankK.Adam,F.

Alvarez,J.Gal′?,R.ReisandH.Uhligforusefuldiscussionswhiledevelopingtheanalyticalframeworkofthisproject,toF.SmetsandM.Bussi`erefordiscussingourpaperattheCEBRAandtheAnnualSNBMonetaryEconomicsConference,aswellasforusefulcommentstoJ.

Benchimol,F.Boissay,B.Bundick,P.Cavallino,K.Dogra,M.Gertler,A.Glover,M.Hoffmann,T.Holden,D.Lee,V.Lewis,M.Lombardi,

G.Lombardo,

Y.Ma,J.Matschke,L.Melosi,E.

Mertens,T.Mertens,E.Moench,M.Rottner,J.Sim,M.Spiegel,I.Vetlov,L.ZhengandseminarandconferenceparticipantsattheBankforInternationalSettlements,DeutscheBundesbank,CentralBankofIsrael’sVIMACROseminarseries,KansasCityFed,BSESummerForum,IFRMP,PadovaMacro,RAD,CEBRA,OsloMacroconference,ASSA2023,andTinbergenInstitute.WearegratefultoG.CornelliandA.Maurinforexcellentresearchassistance,andtoV.Shreetiforsharingdataonbigtechs’fees.121 IntroductionLargetechnologyfirmssuchasAlibaba,Amazon,FacebookorMercadoLibre(bigtechs)haverecentlyventuredinfinancialmarketsbyprovidingloanstovendorsontheire-commerceplatforms.Bigtechcredithasrapidlyexpandedovertherecentyears,reachingvolumesofUSD530billionin2019,upfromonlyaroundUSD11billionin2013.Thepaceofincreaseinbigtechcreditcanbeexpectedtoexceedthatofbankcreditinsomecountries.Forinstance,during2020-21,bigtechcreditinChinarecordedanaverageannualgrowthrateof37%,comparedto13%forbank

credit.Thesechangesinfinancialintermediationcanshapethetransmissionofmonetarypolicyinnotableways.Thebusinessmodelofbigtechsreliesonthecollectionanduseofvasttrovesofdataratherthancollateraltosolveagencyproblemsbetweenlendersandborrowers.Creditscoringgeneratedusingmachinelearningandbigdataareabletoidentifyfirms’creditworthinesswithmoreprecisionthantraditionalcreditbureauratings(Frostetal.(2020)).Moreover,thethreatofbeingbannedfromthee-commerceplatformorevenofhavingone’sreputationtarnishedservesasanextra-legalbuthighlyeffectivemeansofcontractenforcementforbigtechcompanies(Gambacorta,KhalilandParigi(2022)).Thecrucialroleofdatainthecreditscoringprocessandthethreatofexclusionfromthebigtechecosystemreducetheneedforfirmstopledgecollateralinloancontracts.Thisexplainswhybigtechcreditisuncorrelatedwithrealestatevalues,butitishighlycorrelatedwithfirm-specificcharacteristics,suchastransactionvolumesonthebigteche-commerceplatform(Gambacortaetal.(2022)).Astheshareofbigtechcreditrises,monetarypolicywillaffectcreditsupplylessviaassetprices(thetraditional”physicalcollateralchannel”`alaKiyotakiandMoore(1997)),andmoreviarepaymentincentivecompatibilityconstraintswithinBigTechs’ecosystems(thenovel”networkcollateralchannel”thatwe

highlight).Ourpaperaimstoshedsomelightontheeffectsofbigtechs’entryintofinanceonthemacroeconomyandonmonetarypolicytransmission.Wedevelopamodelthatcanreplicatetwokeyempiricalfactsaboutbigtechs.First,usingmacrodataforChinaandtheUS,andextendingpreviousevidencebasedonChinesemicrodata,weshowthatbigtechcreditdoesnotreacttochangesinassetpricesandlocaleconomicconditions,unlikebankcredit.Second,weuselocalprojectionstoshedlightontheimportanceofthephysicalcollateralchannelrelativetothenetworkcollateralchannelforthetransmissionofmonetarypolicy.Keydriversofthestrengthofthesechannels

isthesensitivityofcommercialpropertypricesande-commercesalestomonetarypolicy.

Weshowthatcommercialpropertypricesrespondmorestronglythane-commercesalestoamonetarypolicyshock,althoughless

persistently.Ourmodelisconsistentwiththisevidenceandcanbeusedtoevaluatehowtheadventofbigtechcreditwillimpactthemonetarypolicytransmission.Theanalysisfocusesonbusiness-to-business(B2B)transactions(i.e.transactionsbetweenfirms),whichaccountfor80%ofglobalonline

transactions.1Inourframework,abigtechplatformfacilitatesthesearchandmatchingbetweenintermediategoodsfirmsandfinalgoodsfirms,andextendsworkingcapitalloanstotheformer.2Intermediategoodsfirmsmayfinancetheirworkingcapitalwithbothsecuredbankcreditandbigtechcredit,butcannotcommittorepaytheirloans.Thecrucialdifferencebetweenbigtechcreditandbankcreditrelatestoborrowers’opportunitycostofdefault.Firmsthatdefaultonbankcreditlosetheircollateral(realestate).Incontrast,thosethatdefaultonbigtechcreditloseaccesstobigtechs’e-commerceplatform,andhencetheirfutureprofits.Anincentivecompatiblecontractthuslimitsthetotalamountofcredittothesumofphysicalandnetworkcollateral.Nominalwagesaresticky,andmonetarypolicyaffectstherealeconomy.Whensearchfrictionsinthegoodsmarketsandcreditfrictionsinthefinancialmarketsaresettozero,themodelcollapsestothebasicNewKeynesianmodelwithsticky

wages.3Weobtainthreesetsofresults.First,anexpansioninbigtechs,ascapturedbyanincreaseinmatchingefficiencyonthecommerceplatform,raisesthevalueforfirmsoftradingintheplatformandtheavailabilityofbigtechcredit.Thisinturnrelaxesfinancingconstraintsandincreasesfirms’production,drivingaggregateoutputclosertotheefficientlevel.Second,thereactionofcreditandoutputtoamonetarypolicyshockcruciallydependsonthesensitivityoffirms’opportunitycostofdefaultonbigtechcredit(thestreamoffutureprofitsfromoperatingonthebigtechplatform)comparedtothatofdefaultingonbankcredit(thevalueofphysicalcollateral).Inourbaseline1According

to

the

United

Nations

Conference

on

Trade

and

Development

(UNCTAD),

the

average

share

of

B2B

inglobal

e-commerce

sales

over

the

period

2017-19

was

equal

to

83.8%.2Bigtechs’businessmodelischaracterizedbyamutuallyreinforcingdata-network-activityfeedbackloopwhichhelpsincreasethespeedandaccuracywithwhichtheplatformisabletoconnectbuyersandsellers.Thehigherthematchingefficiency,themoreseamlessandconvenienttheplatformis,andthemorelikelyusersaretousetheplatformfortheirtransactions(Boissayetal.

(2021)).3Forsimplicity,stickywagesaretheonlysourceofnominalrigiditiesinthemodel.Apartfromrenderingmonetarypolicynon-neutral,stickywagesarenecessaryforcreditconstraintstoamplifytheimpactofmonetarypolicy

shocks.34calibration,theintroductionofbigtechcreditmitigatestheinitialresponsesofaggregatecredit

andoutputtoamonetaryshock,butincreasesthepersistenceoftheeffectofmonetarypolicyonthemacroeconomy.Third,bigtechs’macroeconomicefficiencygainsarelimitedbythedistortionarynatureofthefeescollectedfromtheir

users.Ourpapercontributestotheliteratureonthefinancialacceleratorwherephysicalcollateralplaysacrucialroleintheamplificationofmacroeconomicfluctuationsandthetransmissionofmonetarypolicy(e.g.GertlerandGilchrist(1994)).Ariseincollateralvaluesduringtheexpansionaryphaseofthebusinesscycletypicallyfuelsacreditboom,whiletheirsubsequentfallinacrisisweakensboththedemandandsupplyofcredit,leadingtoadeeperrecession.ThecollateralchannelwasarelevantdriveroftheGreatDepression(Bernanke(1983)),andofthemorerecentfinancialcrisis(MianandSufi(2011),Bahajetal.(2019),OttonelloandWinberry(2020)andIoannidouetal.(2022)).Ourpapercontributesbyanalysinghowbigtechs’useofbigdataforcreditscoringandofnetworkcollateralinsteadofphysicalcollateralaffectthelinkbetweenassetprices,creditandthebusiness

cycle.Inoursetup,bigtechcreditsupplyisultimatelyconstrainedbyfirms’expectedprofits.Ouranalysisthereforealsorelatestotheliteratureonthemacroeconomiceffectsofintangiblecollateral(Amable,ChatelainandRalf(2010),Nikolov(2012))andearnings-basedborrowingconstraints(Drechsel(2022),LianandMa

(2021)).Finally,ourpaperrelatestoarecentliteratureonfinancialinnovationandinclusionbyshowinghowariseinmatchingefficiencybetweenbuyersandsellersoncommercialplatformscanleadtoanoverallexpansionofcreditsupply.Theempiricalevidencesuggeststhatfintechandbigtechcreditaregrowingwherethecurrentfinancialsystemisnotmeetingdemandforfinancialservices(Bazarbash(2019),HaddadandHornuf(2019)).Bechetal.(2022)findthatcreatingadigitalpaymentfootprintenablessmallfirmstoaccesscreditfrombigtechcompanies,andthatthishasspillovereffectsfortheirabilitytoaccessbankcreditaswell.Frostetal.(2020)usedataforMercadoCredito,whichprovidescreditlinestosmallfirmsinArgentinaonthee-commerceplatformMercadoLibre.Theyfindthat,whenitcomestopredictinglossrates,creditscoringtechniquesbased

on

big

data

and

machine

learning

have

so

far

outperformed

credit

bureau

ratings.Thepaperproceedsasfollows.Section2describesthestylisedfactsonbigtechcredit.Section3

describesourtheoreticalframeworkwithaspecialfocusonthedualroleofthebigtech

firm5ascommerceplatformandfinancialintermediary.Section4describestheparametrizationof

themodel.Section5showsthesteady-stateequilibriumasafunctionofthematchingefficiencybetweensellersandbuyersonthecommerceplatform.Section6studiestheeffectsofbigtechcreditonthedynamicresponsestoamonetarypolicyshock.Section7

concludes.Stylisedfactsonbig

techsOverthelastdecade,bigtechplatformshaveexpandedtheiractivitygloballyandstartedventuringintocredit

provision.Expansionofbigtechcreditand

e-commerceBigtechcredithasrapidlyexpandedinthelastyearsbecomingmacroeconomicallyrelevantinChina,KenyaandIndonesia(Cornellietal.(2022)).TheexpansionhasbeenparticularlystrongduringtheCovid-19pandemic,duetotheincreaseine-commerceactivitythathasalsoincreasedthedemandforcredit.E-commercerevenueshaverisenfromanestimated$1.4trillionin2017to$2.4trillionin2020,whichamountsto2.7%ofglobaloutput(Figure1,left-handpanel).Recentestimatesindicatethat3.5billionindividualsglobally(about47%ofthepopulation)usede-commerceplatformsin2022.Chinaisthelargestmarket,followedbytheUnitedStates,Japan,theUnitedKingdomandGermany.Mostoftheactivityisbusiness-to-business(B2B)transactions(i.e.transactionsbetweenfirms),whichaccountfor80%ofglobalonlinetransactions(Figure1,right-hand

panel).Thefeestructureofbigtechsgeneratesaroundonethirdoftheirtotalrevenues(Boissayetal.(2021)).Thesefeescanbechargedfordifferentservices,includingplatformaccessfeesforthird-partymerchantsandconsumers,subscriptionfeesforpremiumservices,andadvertisingfeesforreachingawideraudience.E-commerceplatformfeesaretypicallydividedinafixedcomponentandavariableone.Thefixedfeescoveranumberofservicesprovidedbytheplatformforproductadvertisementandareoftennegligibleorabsentformerchants.Thevariablefeeisapercentageofthesalepricechargedbybigtechstothird-partymerchantsforusingtheirplatformstoreachcustomers.Forexample,Amazonchargesthird-partysellersareferralfee,whichrangesfrom6%to45%ofthesaleprice,dependingontheproductcategory.TableA1intheAppendixreportsthestructureofthee-commerceplatformfeesforaselectednumberofbigtechs.Theaveragevariableplatformfeeis

8.5%.Figure

1:

Upward

trend

in

e-commerce,

mostly

via

Big

Tech

platformsNotes:

Source:V.Alfonso,C.Boar,J.Frost,L.GambacortaandJ.Liu(2021):E-commerceinthepandemicandbeyond,BISBulletins,no36,January(leftpanel);UNCTADwithsharescorrespondingtoaveragescalculatedovertheperiod2017-19(right

panel)Bigtechs’rapidexpansionincreditprovisionmirrorstheevolutionoftheirrevenues.Dueto

theirlargeprofitsbigtechshaveasubstantialamountofliquiditythatcanbeusedtofinancelendingtofirmsandconsumers.Boissayetal.(2020)showthatbigtechfirmsaremoreprofitableandcapitalisedthanglobalsystemicallyimportantfinancialinstitutions(G-SIFIs)andhavealargeramountofassetsinliquidform.PriortotheCovidshock,theaverageearning-to-assetratioforbigtechswas24%,against4%forG-SIFIs.Thelargeramountofprofitswasalsoreflectedinahigherequity-to-totalassetratio(52%against8%)andcash-to-totalassetratio(11%and7%,respectively).2.2Bigtechcreditvsbank

creditBigtechcreditisnotcollateralisedandhasashortermaturitythanbankcredit.ForthecaseofChina,bigtechcredithasanaveragematurityoflessthanoneyearandistypicallyrenewedseveraltimes,asfarasthecreditapprovalremainsinplace(Gambacortaetal.(2022)).Whiletwothirdsofbigtechcredithasamaturityofoneyearorless,thissharedropsto43%forbankcredit.Similarcharacteristics

are

detected

for

Mercado

Libre

in

Mexico

(Frost

et

al.

(2020)).6Duetolackofcollateralbigtechcreditislesscorrelatedwithhousepricesthanbank

credit.Moreover,asfirmsoperateone-commerceplatforms,thedemandforbigtechcreditislesscorrelatedwithlocalbusinessconditions,wherethefirmisheadquartered.Gambacortaetal.(2022a)comparetheelasticityofdifferentcredittypestohousepricesandlocalGDP.Themainresult(reportedinFigure2)isthatbigtechcreditdoesnotcorrelatewithlocalbusinessconditionsandhousepriceswhencontrollingfordemandfactors,whileitreactsstronglytochangesinfirmcharacteristics,suchastransactionvolumesandnetworkscoresusedtocalculatefirmcreditratings.Bycontrast,bothsecuredandunsecuredbankcreditreactsignificantlytolocalhouseprices,whichincorporateusefulinformationontheenvironmentinwhichclientsoperateandontheir

creditworthiness.(a)Elasticitiestotransactions

(1)(b)Elasticitiestohouse

prices0.149***0.150.120.089***0.090.060.030.00Big

tech

credit Secured

bank

credit Big

tech

credit SecuredbankcreditFigure2:Creditelasticitytotransactions(left)andhouseprices

(right)Notes:Significancelevel:???p<0.01.Quarterlypaneldataforover2millionChineseSMEsfrom2017to2019withaccesstobothbankcreditandbigtechcreditfromthefinancialarmofAlibabaGroup(AntGroup).(1)Transactionvolumesincludesalesone-commerceplatformsbyonlinefirmsandrevenuesforsalesbyofflinefirmsoperatingwithinthebigtechecosystem.Theelasticityofbigtechcreditwithrespecttoe-commercesalesaloneis0.407***.Source:Gambacortaetal.

(2022)Weextendtheanalysisbycomputingunconditionalelasticitiesofbigtechcreditandbankcredittohousepricesandtoe-commercesales,respectively,basedonmacroeconomicdataforChinaandtheUnitedStates,overtheperiod2013-2020.OurresultsuncoverpatternssimilartothoseemergingfromChinesemicrodata(Table1).Inbothregions,bankcreditismorecorrelatedtohousepricesthanbigtechcredit,whereastheoppositeistruefore-commercesales.Thisevidencesuggeststhatawideruseofbigtechcreditmightdecreasethesignificanceofthecollateral

channel.0.05970.542***0.00.10.20.30.40.5Bigtechcredittohouse

priceChina0.56United

States0.18Bankcredittohouse

price1.40???1.02???Bigtechcredittoe-commerce

sales5.39???3.75???Bankcredittoe-commerce

sales0.39???0.25???Table1:Creditelasticitytohousepricesandtoe-commercesales(macro

data)Notes:

Unconditionalelasticities.Estimationperiod2013-2020.***Significanceatthe1%level.Sources:dataonbigtechs

are

from

Cornelli

et

al

(2022a),

on

e-commerce

sales

are

from

Statista

and

on

house

prices

are

from

the

BIS.ModelThemodelischaracterizedbythreemainbuildingblocks:creditfrictionsintheproductionsector,searchandmatchingalongtheproductionchain,andnominalrigiditiesintheformofsticky

wages.Theeconomyispopulatedby(1)alargenumberofidenticalhouseholdswhoconsume,investandwork,(2)intermediategoodsfirmswhichproduceusinglaborandcapital,(3)retailerswhichproducefinalgoodsusingintermediategoodsasinputs,(4)abigtechfirmwhichfacilitatestransactionsbetweenfirmsandretailers,andextendscredittotheformer,(5)bankswhichgivesecuredloanstofirms,(6)agovernmentwhichissuesrisk–freenominalbonds,and(7)acentralbankwhichsetsthenominalinterest

rate.Firmssellintermediategoodstoretailersviaabigtechcommerceplatformwherebuyersandsellersneedtosearchforandmatchwithoneanother.Intermediategoodsfirmsfinancetheirworkingcapitalinadvanceofsaleswithbothsecuredbankcreditandbigtech

credit.HouseholdsTheeconomyispopulatedbyalargenumberofidenticalinfinitely-livedhouseholds.Eachhouseholdismadeupofacontinuumofmembers,eachspecializedinadifferentlaborservice,andindexedbyj∈[0,1].Incomeispooledwithineachhousehold.AtypicalhouseholdchooseseachperiodhowmuchtoconsumeCtandhowmuchtoinvestinnominalrisk–freepublicbondsBtandequityEttomaximizeitsintertemporal

utility,E0Σ. .∞ 1?σt=0βt tC ?

11?

σ?

χ10tL

(j)1+φ1+

φ8dj∫ ΣΣsubjecttothesequenceofbudget

constraintst

thttet∫10PC+B+E

Q

≤ W

(t tj)L

(ht?1j)dj

+

B (1+

it?1tet)+ED+

Et?1etQ+

Υt(1)tfort=0,1,2...,takingemploymentchoicesL

(∫10j)and

labor

income W

(t tj)L(j)djasgiven.Individually,eachhouseholdhasnoinfluenceonnominalwageratesWt(j)setbyunions,

ortemploymentlevelsLt(j)determinedbyfirms.Ptisthepriceofafinalconsumptiongood,Qeis

thetunit

price

of

equity,

it

is

the

nominal

interest

rate

paid

at

t

+

1

on

public

bonds

bought

at

t,

De

istg pbt t tthedividendpaidonequity,Υ≡Υ+Υ+Υareaggregate(net)lump-sumtransfers

receivedbythehouseholds,whereΥgarelump-sumnettransfersbythegovernment,Υpare

lump-sumt tnetpay–outsbytheprivatesector(i.e.byintermediategoodsfirmsandretailers)

andΥbaretlump-sumnettransfersbythebigtechfirm.4Thehouseholdreceivesthewagesforalltypes

oflaborservicesasbankdepositsatthebeginningofperiodtandusesthemwithintheperiodtobuyfinalconsumptiongoods.ThemaximizationproblemissubjecttostandardsolvencyconstraintsrulingoutPonzischemesonbondsand

equity0,TBT≥

0, lim

E0T

→∞0,TTE

Qh eTPT PTlim

E0.Λ Σ .Λ ΣT

→∞≥

0,(2)where

Λ ≡

βTC?σT0,T C?σt.Households’optimalityconditionsconcerntheoptimalintertemporal

alloca-tionofconsumptiondescribedby

theEulerequations,1

=

E

Λt t,t+1?1t+1t,Π (1+i)

,(3)(4)e et t,Q

=

D

+

E

Λt t,t+1Π Q?1 et+1

t+1,,togetherwiththesequenceofbudgetconstraintsin(1)

fort=0,1,2,...,and

thetransversalityconditionsin(2),where

Λ

t+1t,t+1 C?σttC?σ≡

β istherealstochasticdiscountfactor,andΠ

≡

Pt

Pt?1is

the(gross)inflationratebetweent?1and

t.Thewagesettingproblemandnominalwagerigiditiesarestandard(seeforinstanceGal′?(2015),Chapter6):eachperiodworkersspecializedinagiventypeoflabor(ortheunionrepresentingthem)setwagessubjecttotheCalvo-typenominalrigidities.Specifically,workersspecialisedinany

given4Equityinvestmentisusedtofinancecapitalintheintermediategoodssector.Fortractability,capitalentersproductionrightaway(seedetailsinSection3.3.2),andhence,dividendsarepaidinthesameperiodwhentheequityinvestmentis

made.9labortypecanresettheirnominalwageonlywithprobablity1?θweachperiod,independently

ofthetimeelapsedsincetheylastadjustedtheirwage.Equivalently,eachperiodthenominalwageforworkersofanygiventyperemainsunchangedwithprobabilityθw.Inthisenvironment,thewagedynamicsaredescribeduptoafirst-orderlog-linearapproximation

bywttwt+1^π =

βE

{π }?

λwμtw(5)wtwwhere

π iswageinflationrate,λ

≡(1?θw

)(1?βθw

)θw

(1+?wφ)w,

with

? theelasticityofsubstitution

amonglabortypesindexedbyj,and

μ^twtw w≡μ

?

μ denotesthedeviationsoftheeconomy’s(lo