系统性风险与宏观经济实证分析

SystemicRiskandtheMacroeconomy:AnEmpiricalEvaluationStefanoGiglioyBryanKellyySethPruittzyUniversityofChicagoBoothSchoolofBusinesszArizonaStateUniversityW.P.CareySchoolofBusinessJanuary2015AbstractThisarticleevaluatesalargecollectionofsystemicriskmeasuresbasedontheirabilitytopredictmacroeconomicdownturns.Weevaluate19measuresofsystemicriskintheUSandEuropespanningseveraldecades.Weproposedimensionreductionestimatorsforconstructingsystemicriskindexesfromthecrosssectionofmeasuresandprovetheirconsistencyinafactormodelsetting.Empirically,systemicriskindexesprovidesignificantpredictiveinformationout-of-sampleforthelowertailoffuturemacroeconomicshocks.Keywords:systemicrisk,quantileregression,dimensionreductionWethankLarsHansenformanyhelpfulconversationsaboutthisproject.WealsothankTobiasAdrian,GadiBarlevy,JohnCochrane,DougDiamond,RochelleEdge,ArvindKrishnamurthy,NellieLiang,SergioRebelo,AmirSufi,AmitSeru,AllanTimmermann,JonathanWrightandseminarparticipantsattheFederalReserveBoard,theFederalReserveBankofSanFrancisco,UniversityofChicago(Booth),NorthwesternUniversity(Kellogg),theMidwestEconomicAssociation,theWesternFinanceAssociation,andtheNBER2013SummerInstituteEFWWgroupforhelpfulcomments.WethankXiaoQiaoforexcellentresearchassistanceandRogerKoenker,AndrewLo,ChristianBrownleesandMarkFloodforsharingMatlabcode.1IntroductionTheabilityoffinancialsystemstresstotriggersharpmacroeconomicdownturnshasmadesystemicriskafocalpointofresearchandpolicy.Manysystemicriskmeasureshavebeenproposedintheaftermathofthe2007-2009financialcrisis.Inthispaperwehavethreecomplementaryobjectivesforestablishinganunderstandingofsystemicriskmeasuresandtheirempiricalassociationwithrealmacroeconomicoutcomes.Ourfirstgoalistoprovideabasicquantitativedescriptionofacompendiumofexistingsystemicriskmeasures.Whileindividualmeasuresareexploredinseparatepapers,therehasbeenlittleempiricalanalysisofthemasagroup.Weexamine19previouslyproposedmeasuresofsystemicriskintheUSand10measuresfortheUKandEurope.1Inbuildingthesemeasures,weusethelongestpossibledatahistory,whichinsomecasesallowsustousetheentirepostwarsampleintheUS.Totheextentthatsystemicallyriskyepisodesarerarelyobservedphenomena,ourlongtimeseriesandinternationalpanelprovideempiricalinsightsoverseveralbusinesscycles,incontrasttootherliterature’semphasisonthelastfiveyearsintheUS.Theabsenceofaclearcriteriontojudgetheperformanceofsystemicriskmeasureshasmadeitdifficulttoestablishempiricalpatternsamongthemanypapersinthisarea.Ofcourse,therearenumerouspotentialcriteriaonecouldconsider,suchastheusefulnessforriskmanagementbyfinancialinstitutionsortheabilitytoforecastassetpricefluctuations.Wefocusouranalysisontheinteractionsbetweensystemicriskandthemacroeconomytohighlightwhichmeasuresarevaluableasaninputtoregulatoryorpolicychoices.Therefore,oursecondobjectiveistoevaluatesystemicriskmeasureswithrespecttoaspecificempiricalcriterion:Howwelldoriskmeasuresforecastachangeinthedistributionoffuturemacroeconomicshocks?Ourhopeis1Bisiasetal.(2012)provideanexcellentsurveyofsystemicriskmeasures.Theiroverviewisqualitativeinnature–theycollectdetaileddefinitionsofmeasuresbutdonotanalyzedata.Ourgoalistoprovideaquantitativedescriptionofriskmeasuresandstudytheirassociationwitheconomicdownturns.1toidentifyasubsetofsystemicriskmeasures,ifany,thatareinformativeregardingfutureproduction,employmentorconsumption.Thiswouldallowustoshedlightonthelinksbetweenfinancialdistressandmacroeconomicrisks.Tooperationalizethiscriterionweusepredictivequantileregression,whichesti-mateshowaspecificpartofthemacroeconomicshockdistributionrespondstosystemicrisk.Wearguethataquantileapproachisappropriateforevaluatingthepotentiallyasymmetricandnonlinearassociationbetweensystemicriskandthemacroeconomythathasbeenemphasizedinthetheoreticalliterature.2Thesetheoriespredictthatdistressinthefinancialsystemcanamplifyadversefundamentalshocksandresultinseveredownturnsorcrises,whiletheabsenceofstressdoesnotnecessarilytriggeramacroeconomicboom.3Quantileregressionisaflexibletoolforinvestigatingtheimpactofsystemicriskonmacroeconomicshocks’tail,andinparticularlowertail,behavior,separatelyfromtheircentraltendency.Ourthirdgoalistodeterminewhetherstatisticaldimensionreductiontechniqueshelpdetectarobustrelationshipbetweenthelargecollectionofsystemicriskmeasuresandthemacroeconomy,aboveandbeyondtheinformationinpotentiallynoise-riddenindividualmeasures.Dimensionreductiontechniqueshavebeenwidelystudiedintheleastsquaresmacro-forecastingliterature,andweextendthesetothequantileregressionsetting.Weposethefollowingstatisticalproblem.Supposeallsystemicriskmeasuresareimperfectlymeasuredversionsofanunobservablesystemicriskfactor.Furthermore,supposethattheconditionalquantilesofmacroeconomicvariablesalsodependontheunobservedfactor.Howmayweidentifythislatentfactorthatdrivesbothmeasuredsystemicriskandthedistributionoffuturemacroeconomicshocks?2See,forexample,BernankeandGertler(1989),KiyotakiandMoore(1997),Bernanke,GertlerandGilchrist(1999),BrunnermeierandSannikov(2010),GertlerandKiyotaki(2010),Mendoza(2010),andHeandKris