《国际清算银行-DeFi杠杆（英）-2024.3.docx》由会员分享，可在线阅读，更多相关《国际清算银行-DeFi杠杆（英）-2024.3.docx（44页珍藏版）》请在三一办公上搜索。
1、DeFi1.everageWearegratefulforhelpfulcommentsfromRaphaelAuer,MikeAlonso,AndreaBarbon,AgostinoCapponi,JonanthanChiu,PedroCuadrosSolas,MarcGlowka,Alfred1.ehar,CyrilMonnet,TheoNijman,AndreasSchrimpf,VladSushko,NikolaTarashev,GaneshViSWanaIh-NaI商，ShihaoYu,ArielZetlin-Jones,ShengxingZhang,MariusZoican,and
2、participantsattheBankforInternationalSettlementsresearchmeeting,PSEworkshoponcentralclearingandmarketinfrastructures:newchallenges,GerzenseesummerworkshopinMoney,Payments,BankingandFinance,theSwissNationalBankTechnologyandFinanceseminar,INQUIREUKconference,theCEM1.AconferenceandtheEBApolicyresearchw
3、orkshop.1.iobaHeimbachacknowledgesthefinancialsupportfromtheINQUIREEuroperesearchgrant.TheviewsareoursandnotnecessarilythoseoftheBankforInternationalSettlements.1.iobaHeimbachETHZurich,hliobaethz.ch.WenqianHuangBankforInternationalSettlements,wenqian.huangbis.org.DeFi1.everageAbstractIndecentralized
4、finance(DeFi),lendingprotocolsaregovernedbypredefinedalgorithmsthatfacilitateautomaticloans-allowinguserstotakeonleverage.ThispaperexaminesDeFileverage-i.e.,theasset-to-equityratio-usingwallet-by-walletdataonmajorlendingplatforms.Theoverallleveragetypicallyrangesbetween1.4and1.9,whilethelargestandmo
5、stactiveusersconsistentlyexhibithigherleverageincomparisontotherest.1.everageismainlydrivenbyloan-to-valuerequirementsandborrowcost,aswellascryptomarketpricemovementsandsentiments.Higherborrowerleveragegenerallyundermineslendingresilience,particularlyincreasingtheshareofoutstandingdebtclosetobeingli
6、quidated.Borrowerswithhighleveragearemorelikelytotilttowardsvolatilecollateralwhentheirdebtpositionsareabouttobeliquidated.JE1.classification:G12,G23,036KeyWords:1.everage,collateralisedborrowing,decentralisedfinance,automatedalgorithm.1IntroductionDecentralizedfinance(DeFi)haswitnessedameteoricrise
7、since2020,disruptingtraditionalfinancialservicesbyofferingusersanalternativewayofconductingtransactions.AmongtheplethoraofDeFiprotocols,lendingplatformshaveemergedasacornerstone,facilitatingcollateralisedborrowingactivitiesonaneconomicallysignificantscale(Aramonte,Huang,andSchrimpf,2021;Chiu,Ozdenor
8、en,Yuan,andZhang,2022).Attheirzenith,theseplatformsheldover$35billionindepositsand$25billioninoutstandingdebt,underscoringtheirsignificancewithintheDeFiecosystem(IOSCO,2022;FSB,2023b).Despiteitsimportance,theintricaciesofuserbehaviorandpooldynamicswithinDeFilendingremainlargelyunexplored.Borrowingwi
9、thcollateralandtheassociatedleverage,however,arenotnewtopicsintraditionalfinance.Theroleofcollateralandleveragehasbeenthoroughlyinvestigatedingeneralequilibriummodels(Geanakoplos,2001;Geanakoplos,2010),infinancialintermediarytheory(AdrianandShin,2010;AdrianandShin,2014)andinasymmetricinformationprob
10、lems(AcharyaandViswanathan,2011),aswellasineventfulmarketssuchastherepurchaseagreements(i.e.,repo)(Infante,2019).However,obtainingdetaileddataonuser-levelleveragehasproventobechallenging,resultinginsporadicempiricalanalyses(withfewexceptionssuchasAng,Gorovyy,andVanInwegen,2011;KahramanandTookes,2017
11、).Thispaperaimstobridgetheseknowledgegapsbyprovidingacomprehensiveanalysisofleveragetakingbehaviorincollateralisedborrowing.Thecontributionofthepaperisthree-fold.Firstly,tothebestofourknowledge,wearethefirsttodocumentindividualDeFiwallets,leverage-whichisdefinedastheasset-to-equityratio(i.e.,theleve
12、rageconceptinAdrianandShin(2010)andAdrianandShin(2014).UsinggranulardatafromtheEthereumblockchain,ourpaperprovidesanextensiveexaminationofDeFileverage,elucidatingitsoveralltrends,groupdisparities,anddrivingfactors.Secondly,ouranalysispresentsnewempiricalevidenceonthesystemicriskimpactofhighleverageo
13、nDeFilendingplatforms.Inparticular,wefocusonlendingresilienceandstrategicsubstitutionbehaviour,basedonindividualwalletsinvestmentportfoliodata.1.astbutnotleast,althoughDeFiremainsapredominantlyself-referentialsystem,thelessonsgleanedfromDeFilending-arealworldlaboratory-couldpotentiallyberelevanttoun
14、derstandfinancialstabilityconcerns,inparticularinrepoandsecuritieslendingmarketsgiventhesimilaritiestheyshare.Wesystemicallyreviewthesimilarities,aswellasthedistinctions,betweenDeFilendingandtraditionalcollateralisedborrowing,suchasrepo.Themainfindingsconsistofthreeparts.Firstly,weexaminetheoverallt
15、rendsofDeFileverage.Throughoutoursampleperiod(January2021-March2023),theoverallleverageofDeFiusersInthecontextofDcFi,whichispseudo-anonymous,wcuse“wallets”andt4uscrs,interchangeably.Becausewalletsarestandaloneunits,largelyisolatedfromthebroaderbalancesheetoftheentity.rangesfrom1.4to1.9.Thisleveraget
16、racksthemarket-widepricemovementswithanapproximately3-monthlag,probablyreflectingspeculativemotivesincryptotrading(Biais,Bisiere,Bouvard,Casamatta,andMenkveld,2020;Auer,Comelli,Doerr,Frost,andGambacorta,2023).Thisasset-to-equityratiotypeofleverageissignificantlylowerthantheleveragepermittedbytheloan
17、-to-value(1.TV)ratio(i.e.,theleverageconceptinGeanakoplos(2001)andFostelandGeanakoplos(2014),whichrangesfrom3.4to4.8.Userleverageexhibitsheterogeneityacrossgroups.Weidentifythreedistinctusergroups:thosewiththelargestoutstandingdebt,themostfrequentusersinteractingwithlendingprotocols,andtheearliestad
18、optersofDeFilendingplatforms.Ourobservationsrevealthatboththelargestandmostactivewalletsexhibithigherleveragecomparedtotheremainingusers,withtheiraverageleverageoftenexceeding2.Wefurtherclassifywalletsintothosewithlongvolatileassetpositionsandthosewithshortvolatileassetpositions.Inotherwords,thelong
19、walletspledgevolatilecoinsascollateralandborrowstablecoinsandtheshortwalletspledgestablecoinsascollateralandborrowvolatilecoins.Theleverageofshortandlongwalletsisnegativelycorrelated.Specifically,whenvolatileassetsappreciate,theleverageoflong(short)walletsdecreases(increases)astheircollateral(debt)v
20、alueincreases.Next,weexplorevarioushypothesestoidentifythefactorsthatareassociatedwithhighborrowerleverage.Ourfindingsindicatethatauser,sleverageishigherwhentheyfacehigher1.TV-permitledleverage,lowernetborrowingcost,andhighermarketsentiment.Inaddition,itisstrikingthattheactualleverageissignificantly
21、lowerthanthe(maximum)leverageallowedbythe1.TVratiorequiredbythelendingplatforms.Thisgapcanbeexplainedbytwochannels.Firstly,sinceborrowersfacesubstantiallossesuponautomaticliquidation,theygenerallyavoidleveragingtothemaximumextent,optinginsteadforamoreconservativeapproachwithasizeablebuffer.Secondly,
22、whenDeFiusersexperienceassetappreciation(i.e.,higherpastreturns),theydepositmorecryptoassetsinthelendingplatformswithouttakingdebts,leadingtohigheramountsofassetsandlowerleverageceterisparibus.OurthirdfindingpertainstothesystemicimpactofhighborrowerleverageonDeFilendingplatforms.1.everagecanpropagat
23、eshocksviarisingliquiditydemandsthatstemfromthefluctuationsofthecollateralvalue(FSB,2023a).WefirstinvestigatehowborrowerleverageaffectsDeFilendingresilience.Weassesslendingresilienceusingtwometrics:value-at-risk(VaR)andliquidationshare.Theformergaugestheshareofloansthatareclosetobeingliquidatedintot
24、alloans,whilethelatterrepresentstheshareofloansthatareliquidated.WefindthathigherborrowerleveragecontributestoincreasedVaR,signifyingheightenedriskwithinlendingpools.IntermsOfliquidationshare,however,borrowerleveragedoesnotappeartohaveasignificantinfluence,asliquidationsarepredominantlyevent-driven(
25、1.eharandParlour,2022).1.astbutnotleast,althoughborrowersinDeFicanadjusttheircollateralportfoliosaslongasmeetingthe1.TVratiorequirements,wefindthatmostborrowersonthebrinkofliquidationdonotshifttowardsmorevolatilecollateral.Thisisprobablyduetothefactthatthe1.TVratiorequirementismorestringentformorevo
26、latileassets.However,conditionalonthosethatdotilttowardsvolatilecollateral,higherborrowerleverageisassociatedwithmoreaggressivestrategiccollateraladjustment.Thisisconsistentwiththeasymmetricinformationproblemstemmingfromthepoolingofcollateralacrossborrowers(Chiu,Ozdenoren,Yuan,andZhang,2022).Wecontr
27、ibutetothreerelevantstrandsofliterature.ThefirstoneistherapidlyevolvingliteratureonDeFilending.AnotherstrandoftheliteraturerelatestotradinginDeFiandincryptoingeneral.InterestedreaderscanrefertoAoyagiandIto(2021);1.eharandParlour(2021);CapponiandJia(2021);BarbonandRanaIdo(2021);Hasbrouck,Rivera,andSa
28、lch(2022);C叩POni,Jia,andYu(2022);1.ocsch,Hindman,Richardson,andWelch(2021);Qin,Zhou,andGervais(2022);Heimbach,Wang,andWattenhofer(2021);Heimbach,Schertenleib,andWattenhofer(2022);MalinovaandPark(2023);Milionis,Moallemi,Roughgarden,andZhang(2022);Milionis,Moallemi,andRoughgarden(2023);Milionis,Moalle
29、mi,andRoughgarden(2023);Fritsch(2021);Berg,Fritsch,Hcimbach,andWattenhofer(2022)andTorres,Camino,etal.(2021)fortheformer,andCongandHe(2019);Biais,Bisiere,Bouvard,Casamatta,andMenkveld(2020);MakarovandSchoar(2021);Schmeling,Schrimpf,andTodorov(2022)andAuer,Comelli,Doerr,Frost,andGambacorta(2023)forth
30、elatter.Aramonte,Huang,andSchrimpf(2021)provideaprimerontheessentialbuildingblocksinDeFi,highlightingtheillusionofdecentralisation.Carapella,Dumas,Gerszten,Swem,andWall(2022)discussthepotentialsandrisksofDeFiplatforms.Chiu,Ozdenoren,Yuan,andZhang(2022)constructatheoreticalmodelofDeFilendingthatcaptu
31、resthedistinctfeatureofcollateralpoolingacrossborrowersandtheassociatedasymmetricinformationproblem.Chaudhary,Kozhan,andViswanath-Natraj(2023)studytheinterestrateparityinDeFilendingandidentifyarelationshipbetweentheinterestratedifferentialandthefuturespremium.Gudgeon,Werner,Perez,andKnottenbelt(2020
32、)empiricallyanalyzethedifferinginterestraterulesofDeFilendingprotocols,whereasRivera,Saleh,andVandeweyer(2023)showthatthepre-determinedinterestratecurvesofDeFilendingarelessefficientcomparedtotraditionallendingplatforms.1.eharandParlour(2022)studythepriceimpactofliquidationsinDeFilending,identifying
33、apotentialsourceoffragilityandspilloverfactorinthisnascentfinancialsegment.Similarly,Perez,Werner,Xu,and1.ivshits(2021)andQin,Zhou,Gamito,Jovanovic,andGervais(2021)empiricallystudyliquidationsinDeFiandtherisksthatstemfromthem.Heimbach,Schertenleib,andWattenhofer(2023)examinearecentepisodeinwhichthea
34、vailableliquidityofalendingpoolwasentirelydepletedandanalysetheunderlyingissuesandCounterfactua1.Yaish,Tochner,andZohar(2022)discusshowcryptocurrencyminerscouldmanipulatetheirinterestrateonDeFiloansbyadjustingtheblockrate,whereasHeimbach,Schertenleib,andWattenhofer(2023)dissectarecentpricemanipulati
35、onattackonalendingprotocol.Finally,Tovanich,Kassoul,Weidcnholzer,andPrat(2023)studyfinancialcontagioninCompound,i.e.,thesecondbiggestDeFilendingprotocolontheEthereumblockchain.OurresearchcomplementsthisstrandOfliteraturebypresentingnewevidenceonDeFileverageanditsconsequenteffectonlendingresilience.T
36、hesecondstrandofliteraturerelatestoleverageintraditionalmarkets.McGuireandTsatsa-ronis(2008)putfortha“regression-based“methodologyforestimatinghedgefundleverageusingpubliclyavailabledata.Ang,Gorovyy,andVanInwegen(2011)provideanin-depthanalysisofhedgefundleveragebasedonsupervisorydata.AdrianandShin(2
37、014)examinetheimpactofleverageonfinancialstability,aswellasitsprocyclicality.KahramanandTookes(2017)usetheuniquefeaturesofIndiasmargintradingsystemtoestablishacausalrelationshipbetweentraders,leverageandastock,smarketliquidity.Utilisingthegranulartransactiondata,wecontributetothisliteraturebyidentif
38、yingthedistinctdrivingfactorsbehindleverageandtheimpactofhighborrowerleverage.Wealsocontributetotheliteratureonrepomarkets.Krishnamurthy,Nagel,andOrlov(2014)findthatrepovolumebackedbyprivateasset-backedsecuritiesfallstonearzeroduringtheglobalfinancialcrisis.Incontrast,Copeland,Marlin,andWalker(2014)
39、presentevidencesuggestingthattherewasnosystem-widerunonrepo,usingconfidentialdataontri-partyrepo.Infante(2019)developsamodelofrepointermediationthatreconcilesthediscrepancyintheaforementionedanalysisonrepomarketruns.Weenrichthisstrandofliteraturebyunderstandinganalternativeapproachtocollateralisedbo
40、rrowing.Therestofthepaperisorganisedinthefollowing.WefirstexplainthemechanicsofDeFilendingandcontrastthattorepotradingandsecuritieslendinginSection2.InSection3,weshowthestylisedfactsonDeFileverage,describingthetrendaswellasthegroupfeatures.Tounderstandthedrivingfactors,werunwallet-daylevelpanelregre
41、ssionstotestseveralhypotheses.Section4looksintotheimpactOfleverageonlendingpoolresilience.Section5examineshowhighleverageaffectsborrowers,collateralselectionwhentheirpositionsareclosetobeingliquidated.Section6concludes.2InstitutionalbackgroundanddataHowDeFilendingworks.OnDeFilendingplatforms,auserca
42、ndepositcryptoassetsintoalendingpoolandreceivesaclaimontheirshareofthepool.Smartcontractsbehindtheseplatformsenabletheusertousethisclaimascollateraltoborrowfromthesameassetorotherassets,subjecttoloan-to-value(1.TV)ratios(Figure1).Table1reportsthe1.TVratioasoftheendofMarch2023formajorcryptoassets.For
43、instance,onAavev2,the1.TVrat100fUSDC-amainstablecoin-is80%,whichmeansthatforacollateralvalueof$100USDC,onecanhaveadebtvalueofupto$80.Thus,theassociatedhaircutrequirementofUSDCis20%andtheimpliedmaximumleverageis5(=l1).The1.TVratiovariesacrosscollateralandtimedependingontherisk-1.TVratiomanagementofth
44、eDeFilendingplatforms.AppendixAprovidesadetaileddescriptionofhowthe1.TVratioandborrowing/tiepositratesaredeterminedinDeFilending.Wenoteherethatthe1.TVratioissetandupdatedbytheprotocolgovernance-adecentralisedautonomousorganisation(DAO)-basedonriskassessmentonsmartcontractsecurity,counterpartyrisk,an
45、dmarketrisk.Therearetwoleverageconcepts.Oneistheleveragerequirementimposedbythelendingplatforms-i.e.,themaximumleverageimpliedbythe1.TVratio-whichisakeyvariableintheleveragecycleproposedbyGeanakoplos(2010);FostelandGeanakoplos(2014).Theotherconceptistheactualleverageinauser*sportfolio,whichmeasurest
46、owhatextenttheuser,sassetsTable1.1.oan-to-va)ucratioandimpliedleverage.1.TVisloan-to-valueratio.Haircutisthecorrespondingdiscountofthecollateralvalue.1.everage7istheimpliedleveragerequirementfromthe1.TVratio.Thedataisfromlendingplatformsasof31March2023.Aavev2Compound1.TVHaircut1.everage71.TVHaircut1
47、.everage7USDC0.8000.2005.0000.8550.1456.897USDT0.0001.0001.0000.0001.0001.000DAI0.7500.2504.0000.8350.1656.061ETH0.8250.1755.7140.8250.1755.714BTC0.7200.2803.5710.7000.3003.333aresupportbytheirownequity-i.e.,theasset-to-equityratio(AdrianandBoyarchenko,2012;AdrianandShin,2014).Figure1showsastylisedb
48、alancesheetofauserthatusesvolatilecoins(VC)ascollateraltoborrowstablecoins(SC).Fromtheuser,sperspective,theirassetsincludethedepositsontheseplatformsandothercoinstheyheldexternaltoDeFilendingplatforms.Theirliabilitiesconsistof(collateralised)debtontheseplatforms.Thedifferencebetweenassetsanddebtcons
49、titutestheuser,sequity.Bothconceptscapturedifferentaspectsofleverage.Giventhefocusofthispaperisontheusers,borrowingbehaviours,theterm“leverage“referstotheasset-to-equityratio,whilet(1.TV-)impliedleverage“referstotheleverage(ormargin)requirementimposedbytheplatform.Figure1.MechanicsofDeFilending.Thisfigureshowsthebalancesheetofauserthatborrowsstablecoins(SC)usingvolatilecoins(VC)ascollateral.Assetliab
