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

AAdditionalEvidence

A.1CreditSpreadsinEuroAreaCountriesandtheUnitedStates

AsmentionedintheIntroduction,Figure6plotstheevolutionofhouseholdandbusiness creditspreadsintwelvememberstatesoftheeuroareaaswellasintheUnitedStates.In everycountryexceptGreece,theseriesarehighlycorrelated.Inallinstances,spreads spikeduringeachrecessionofthesample.Thus,coordinatedandcountercyclical spreadsisastylizedfactofbusinesscyclesinadvancedeconomies.

A.2SwitchingOfftheShadowBanks

Themaincontributionofthispaperistodeviseamechanismthatrationalizescorre- latedcreditspreads.Wearguethatthenonbankfinancialsector,withitsvolatileand procyclicalfundingstructure,isacrucialconduitthroughwhichinvestorconfidence propagatestothemainstreambankingsector,andthen,totherestoftheeconomy.

Onewaytohighlighttheroleplayedbyourmechanismistodeactivateit.We estimateaversionofthemodelwithoutshadowbanks.Traditionalbankscontinue intermediatingfundsbetweenpatienthouseholdsontheonehandandimpatienthouse- holdsandentrepreneursontheother.Butnowwemakethestandardassumptionthat theyareabletoperfectlydiversifyidiosyncraticrisk.Thisrenderssecuritizationmoot: incaseofdefault,bankssimplyseizethepledgedcollateralandsellitbackatmarket price.Thefinancialfrictionresidesinabankruptcycostbanksmustpaytoaudittheir borrowers.Thus,theversionwithoutshadowbanksboilsdowntoastandardcostly stateverificationmodelalaBGG,butwithtwofinancialacceleratormechanisms—on householdsandfirms—insteadofone.Specifically,thebanks’participationconstraints

22

Figure6:HouseholdandFirmCreditSpreadsinEuroAreaCountriesandtheUnitedStates

Notes:Creditspreadsintheeuroareaarethedifferencebetweeninterestrateonbankloansandtheshort-termeurointerbankrate. SpreadsintheUnitedStatesarewithrespecttothefederalfundsrate.Numbersinbracketindicatethecorrelationbetweenthetwo series.Shadedbarsshowrecessions.

(Equations(6)and(7)inthemaintext)recovertheirusualform

[1−Fⁱ(ω¯_tⁱ₊₁)]R_tⁱ₊₁B_tⁱ+(1−µ^h)Gⁱ(ω¯_tⁱ₊₁)R_t^h₊₁Q_t^hH^¯_tⁱ≥R_t+1B_tⁱ,
[1−F^e(ω¯_t^e₊₁)]R_t^e₊₁B_t^e+(1−µ^k)G^e(ω¯_t^e₊₁)R_t^k₊₁Q_t^kK^¯_t≥R_t+1B_t^e,

whereµ^handµ^kdenotethemonitoringcostsofimpatienthouseholdsandentrepreneurs,

respectively.
Weestimatethealternativemodelonthesamedataset,usingtheexactsameprior

distributionsasinthebaselinecase.Notethatthemarketsentimentshockbecomes irrelevantintheabsenceofitspropagationmechanism,sowedropit.Instead,weadd ahouseholdriskshockσ_tⁱ,mirroringthefirmriskshockσ_t^e.Inparticular,σ_tⁱdenotes thestandarddeviationoflogωⁱandismeanttoreflectidiosyncrasiesinlocalhous-

ingmarkets(localemployment,publicinfrastructure,weatherconditions,population dynamics).10

Withthemarketsentimentshockoutoftherace,wefindthatthefirmriskshock

10Wefirstestimatedourbaselinemodelwiththehouseholdriskshock.Butitscontributiontoall variables,includingthehouseholdspread,wasclosetonil,sowedecidedtodropit.

23

Figure7:ShockContributiontoCreditSpreads

Notes:Solidlinescorrespondtoactualcreditspreadsintheeuroareafrom2003Q1to2019Q4.Verticalbarsshowthecontribution ofthedifferentshockstotheevolutionofspreads.Dashedlinescorrespondtocreditspreadsimpliedbytheestimatedmodel between1999Q1and2002Q4,periodforwhichdataonspreadsisnotavailable.

becomesthemaindrivingforceoftheEuropeanbusinesscycle.11Thisechoesthe resultsinChristiano,Motto,andRostagno(2014)withUSdata.However,thefirmrisk shockstrugglestomatchanumberofvariablesrelatedtohouseholds,mostimportantly consumption,creditquantity,andthecreditspread.Asithappens,thehouseholdrisk shock(partially)fillsthegap.Figure7summarizesthesefindingsbydecomposing thecontributionofprominentshockstotheevolutionofthehouseholdandfirmcredit spreads.Thetoprowpresentstheoutcomeforthebaselinemodel,iewiththemarket sentimentshock,whilethebottomrowshowstheequivalentforthemodelwithout shadowbanks.

Thetwoeconomiesareinstarkcontrast.Inourbaselinemodel,themarketsentiment shockmakesupthelion’sshareofbothcreditspreadsmovements.Asexplained inthemaintext,thisisbecausebanksrelyonshadowbanksaspartoftheirrisk managementstrategy.Wheninvestorsentimentdeteriorates,shadowbanksretrench fromthesecuritizedloanmarket,pushingdownthepriceofasset-backedsecurities, andobligingbankstooffsetthisdropinincomebyincreasinglendingratesonhousehold

11Thefirmriskshockaccountsfor31,9,42,and22percentofthevarianceinoutput,consumption, investment,andhours,respectively.Parameterestimatesandacompletevariancedecompositionare availableuponrequest.

24

Table2:ContributionofThreeShocks,6–32Quarters

MarketsentimentFirmriskHouseholdrisk

shockν_t shockσ_t^e shockσ_tⁱ

A.BaselineModel
Variancehouseholdspread 89 5 –

Variancefirmspread 85 0 –
CovarianceHspread–Fspread 99 0 –

B.ModelwithNoShadowBanks
Variancehouseholdspread – 0 81

Variancefirmspread – 92 0
CovarianceHspread–Fspread – –243 –784

Note:Varianceandcovariancecontributionsarecomputedonbandpass-filtereddatageneratedbythetwo modelsevaluatedatthemodeoftheposteriordistribution.

andbusinessborrowersaltogether.Inthemodelwithoutshadowbanks,thetworisk shocksemergeasthecentralforcebehindcreditspreads.Buteachriskshockaccounts forits’own’spread:thehouseholdriskshockdrivesthehouseholdspread,thefirmrisk shockdrivesthefirmspread.Neitherdisturbancehasanyrolewhatsoeveronthe’other’ spread.Theupshotisthat,inordertomatchthejointdynamicsofcreditspreadsinthe data,theestimationprocedureattributesahighdegreeofcorrelationof0.69between theexogenousprocesses.Thisresultisatoddswiththeassumptionthatshocksare structuralandindependentfromoneanother.

Table2reinforcesthesepoints.Wereportthecontributionofsentimentandrisk shockstothevarianceandcovarianceinhouseholdandfirmcreditspreadsatbusiness- cyclefrequency.Asthetoppanelmakesclear,themarketsentimentshockinthebaseline modelaccountsforvirtuallyallthecovarianceinspreads.Inthemodelwithoutshadow banks(bottompanel),eachriskshockcontributesnegativelytothatcovariance.

Toconclude,absentthetransmissionchannelthatshadowbanksconstitute,the theoryisunabletogenerateafactualresponseofcreditspreadswithasingleimpulse, andthereforefallsshortofprovidingaplausibleexplanationforbusinesscycles.This underscoresthecontributionofourpapertothedebate.

25