P aper No. 907 Banks, shadow banks, and business c

5ExternalValidation

Thisarticleclaimsthatvariationsinmarketsentimentareresponsibleforthelockstep motionofcreditspreadsandthebulkofmacroeconomicfluctuations.Takingthis claimseriouslyrequirestohavefaithintheshockitselfaswellasinthetransmission mechanismoftheunderlyingmodel.Inthissection,weoffersupportforeachofthese elementsbasedonexternaldataandadifferenteconometricapproach.

5.1TheMarketSentimentShockandFinancialStress

Underaliteralinterpretation,amarketsentimentshockinourstylizedmodelisan exogenouschangetotheriskpremiumcommandedbyasset-backedsecuritiesoversafe deposits.Wepreferabroaderinterpretation,underwhichtheshockreflectsareversalof aggregatesentimentinthefinancialsector.Tobackthisview,wecompareinFigure4 themarketsentimentshockprocesscomingoutofourestimatedmodeltoameasureof systemicfinancialstressinEurope.ThatmeasureistheCompositeIndicatorofSystemic Stress(CISS)constructedbytheEuropeanCentralBank.Itaggregatesmanyindicators frommoney,bond,equity,andforeignexchangemarkets,aswellbank-specificdataon returnvolatilityandcreditspreads.Weemphasizethatthisserieswasnotusedinthe inferenceaboutourmodel’sparameters.

Thechiefobservationisthatourshockcorrelateswellwiththemeasureoffinancial stress.Thecontemporaneouscorrelationcoefficientis0.67overthesampleperiod. Themarketsentimentshockspikesinthreeoccasions,duringthe2001dot-combubble (whichdidnotleadtoarecessioninEurope),the2008financialcrisis,andthe2012 sovereigndebtcrisis.ThesameistruefortheCISS,withalittlelag.Interestingly,the

15

Figure4:FinancialStress,ModelVersusData

Notes:FinancialstresscorrespondstotheEuropeanCentralBank’sCompositeIndicatorofSystemicStress(CISS).Themarket sentimentshockisscaledtofitthefigure.

correlationisatitshighest,0.72,whentheshockleadstheindicatorbyonequarter.A testofpredictivecausalityindicatesthatthemarketsentimentshockGrangercausesthe financialstressindexatthe0.1-percentconfidencelevel,atuptosixlags.Weconclude thatourtheoreticalobjectisafairlygoodgaugeoffinancialmarketsentiment.

5.2EuropeanBusiness-CycleAnatomy

Inarecentpaper,Angeletos,Collard,andDellas(2020)developanewstrategyto analyzebusinesscycles.TheyestimateaVARonanumberofUSaggregateseries wherejustoneshockisidentifiedsoastomaximizethevolatilityofoneparticular seriesoveraparticularfrequencyband.Theyrepeattheexerciseforeachvariable, hencetakingmultiplecutsofthedataandprovidingan"anatomy."Theirmainresultis interchangeability:whetheronetargetsGDP,consumption,investment,hoursworked, orunemployment,thedifferentshocksproducenearlythesamedynamiccomovements inallvariablesofinterest.Theauthorsconsidertheseshockstobethevariousfacetsof asingleforce,whichtheylabelthe"mainbusiness-cycleshock".

OurgoalinthissubsectionistoconstructaEuropeanmainbusiness-cycleshockand compareitsdynamicpropertiestothoseofourmarketsentimentshock.Wefollowthe proceduredescribedbyAngeletos,Collard,andDellas(2020).Thedataisthesameas intheestimationofourstructuralmodel,exceptthatweaddtheCISSpresentedaboveas anadditionalobservablevariable.Also,weenterallvariablesinlevels,thatiswedon’t detrendordemeananyseries.WeestimateaVARwithtwolagsusingBayesianmethods andaMinnesotaprior.Wethenconstructastructuralshockasalinearcombination oftheVARresiduals.Theidentificationcriterionrequiresthattheshockmaximizethe

16

Figure5:MarketSentimentShockVersusMainBusiness-CycleShock

Notes:ImpulseresponsestothemarketsentimentshockarescaledtomatchthoseoftheVAR.Theshadedareacorrespondstothe 68%highestposteriordensityintervalaroundtheresponsestotheoutput-basedmainbusiness-cycleshock.

contributiontothevolatilityofaparticularvariableoveraparticularfrequencyband. Thatvarianceiscomputedinthefrequencydomain,andthechosenbandisbetween6 and32quarters.Wetargettwovariablessuccessively:outputandtheCISS.Figure5 displaystheresponseoftheVARtothesetwoshocks,alongwiththeresponseofour structuralmodeltotheestimatedmarketsentimentshock.

TwofindingsemergefromtheinspectionofFigure5.First,thetwoVAR-based shocksarevirtuallyindistinguishable.Theybothcauseprocyclicalmovementsin consumption,investment,hours,inflation,nominalrate,andcreditquantities,aswell ascountercyclicalmovementsinspreads.Thus,itappearsthatintheeuroareaover thelasttwodecades,themainbusiness-cycleshock,definedasonethatmaximizes thevariationinoutput,coincideswiththe"mainfinancialshock",definedasonethat maximizesthevariationinsystemicfinancialstress.

Thesecondfindingisthatthemarketsentimentshockgivesrisetoalmostthesame impulseresponsesasthe(twoversionsofthe)mainbusiness-cycleshock.Allendoge- nousvariablesexhibitanicehump-shapedbehavior,peakingafterfourtosixquarters, exceptforthetwocreditspreads,whicharemonotonous.Themodel’sresponsesare aspersistentasthoseoftheVAR.Thisisthankstotheshockprocess’srelativelylow autocorrelationcoefficientof0.939,avaluesmallerthanmostestimatesfoundinthe literatureforfinancialdisturbances.9Thus,theanatomytestvindicatesourmodel’s transmissionmechanism.Inthismodel,shockstoinvestorconfidencetriggerdynamics thatresemblethoseobtainedfromamuchmoreflexiblestatisticalframework.Given howdifferentthetwoapproachesare,theproximityintheiroutcomeprovidesfurther

9Forexample,theriskshockinChristiano,Motto,andRostagno(2014)hasanautocorrelation coefficientof0.972.JermannandQuadrini(2012)find0.969fortheirfinancialshock.

17

supporttothemarketsentimentshockasaleadingimpulsetobusinesscycles.