Lending Standards, Business Loans, and Output: The U.S. and Canadian Experience* (2024)

LendingStandards,BusinessLoans,andOutput:

TheU.S.andCanadianExperience*

PierreL.Siklos,WilfridLaurierUniversityBradyLavender,WilfridLaurierUniversity[FirstDraft:December2009:ThisDraft:JUNE2011]PRELIMINARYandINCOMPLETE–NOTTOBEQUOTEDORCITEDWITHOUTPERMISSION

*ResearchforthisstudywasundertakenwhileBradyLavenderwasagraduatestudentatWilfridLaurierUniversity.FinancialassistancefromtheSocialScienceandHumanitiesResearchCouncilisgratefullyacknowledged.Resultsnotshownarerelegatedtoaseparateappendixavailableonrequest.BothauthorsaregratefultoStephenMurchisonoftheBankofCanadaformakingavailablerealtimerealGDPandpotentialrealGDPforCanada.

ABSTRACT

Current economicconditions,combined with limited creditavailabilityandan aggressivemonetarypolicy,haveonceagainhighlightedtheimportanceofunderstandingtherelationshipbetweencreditavailabilityandrealeconomicactivity.Thepurposeofthispaperistoestimatevectorautoregressive(VAR)andvectorerrorcorrectionmodels(VECM)forCanadaandtheUnitedStatesandtouseinnovationaccountingtoinvestigatetherelationshipbetweenchangesin non‐price lending standards, business loans and output. We then estimate a factor‐augmentedVARtoaskwhethermacroeconomicandfinancialmarketconditionsintheU.S.affectearlierfindingsforCanada.Correspondingauthor:PierreSiklos,DepartmentofEconomics,WilfridLaurierUniversity,75UniversityAve.,Waterloo,ON,CANAD,N2L3C5e‐mail:psiklos@wlu.caKeywords:macro‐financiallinkages,creditstandards,LoanOfficerSurveyJELClassificationCodes:E32,E5,G21

1. Introduction

Currenteconomicconditions,togetherwithconcernsovercreditavailabilitycombinedwithanaggressivemonetarypolicy,haveremindedpolicymakersandothersoftheimportanceofunderstandingthelinksbetweencreditavailabilityandrealeconomicactivity.Roosa’s(1951)classicarticlestatesthatcreditavailabilityisanessentialpartofmonetarypolicyeffectiveness.Whilesomepostulatethatcreditavailabilityaffectstherealeconomy(BlinderandStiglitz1983),creditmarketdynamicsarequiteperplexingasasymmetricinformationleadstocreditrationing,potentiallymanifesting itselfasa ‘non‐price’characteristicof credit conditions(StiglitzandWeiss1981).

ThepurposeofthispaperistoestimatetimeseriesmodelsforCanadaandtheUnitedStatesandtouseinnovationaccountingtoinvestigatetherelationshipbetweenchangesinlendingstandards,businessloansandoutput.Inlightofthefinancialcrisis,anditsaftermath,therehasbeensurprisinglylittleresearchontheinfluenceofloanofficers’viewsofa*ggregateeconomicconditionsandhowthesestandardsmightinfluenceaggregateeconomicoutcomes.Interestinthesequestionshasalsobeenstoked,particularlyintheUnitedStates,duetofearsthatthebuild‐upofexcessreservesintheU.S.bankingsystemsincetheFederalReserveengagedinquantitativeeasingmayleakintothebroadereconomyandgenerateinflationinfuture.1We are especiallyinterestedintheempirical significanceandmagnitude of twochannelsthroughwhichnon‐priceaspectoflendingarelinkedtotherealsideoftheeconomyandmonetarypolicy.Moreprecisely,weinvestigatewhethercreditstandardsinfluencereal

1_{Agoodsourcefortheongoingdebateoverthepotentialimpactofexcessreservescanbefoundat}

economicactivityandwhetherthelinkisrobusttosampleselectionandmodelspecification.Next,weexaminethelinksbetweenthesestandards,thevolumeofcommercialloansandtheconductofmonetarypolicy.Dopolicyrateshocksinfluenceloansstandards,anddoloanstandardsalsohaveaneffectonmonetarypolicy?Ifstatisticallysignificanteffectsarefoundandarerobustthenmacro‐modelsneedtobeaugmentedwithameasureofchangingcreditstandardsasaproxyforatypeoffinancialfrictionthatinfluencesrealeconomicoutcomes.

TheUSFederalReserve’sSeniorLoanOfficerSurvey(hereafterSLOS)andBankofCanadadataareusedtoproxythenon‐priceaspectsoflending.Theimportanceofnon‐pricecreditconditionshaslongbeendeemedcriticalashighlightedbyRoosa(1951),JaffeeandStiglitz(1990),andtheStiglitzandWeiss(1981)creditrationingmodel.Creditmarketsdonotreachequilibriumbasedonpricealoneasthereisanon‐priceelementthatisveryimportantdue to moral hazard and adverse selection problems caused by imperfect information(BlanchardandFischer1989).InthepresentpapertheseareproxiedbythestandardsforlendingasviewedfromtheperspectiveofseniorLoanOfficers.

WhiletherehavebeenafewstudiespublishedbasedonU.S.FederalReservedata(Lownet.al.2000,LownandMorgan2006),toourknowledgenoonehasanalyzedCanada'sSLOSdatainthistypeofframework.ThecurrentliteraturerelyingontheFederalReserve’ssurvey explores the effects of tightening ‘non‐price’ credit conditions using vectorautoregressions(VAR) butfails to testfor cointegration. Ifcointegration isevident,VARestimationswouldresultinamodelmisspecificationasavectorerrorcorrectionmodel(VECM)is needed to account for thelong‐run equilibrium relationship. Of course, cointegrationpresumesthatthereissomeunderlyingequilibriumeconomicrelationshipthathasbeen

omitted.Second,therehasbeennoattempttoexaminewhethertheinterpretationofthesurvey’simpactisinfluencedbyeconomicrestrictionsofthekindeconomiststypicallyimposeintheirmodels.Finally,giventhedifferencesineconomicoutcomesbetweenCanadaandtheU.S.,especiallysincetheglobalfinancialcrisisof2007‐2009,acomparisonofthetwocountriescanyieldusefulinsightsaboutthecontributiontheSLOStocreditconditionsandoutcomes.Additionally, from the Canadian perspective, while it is common to consider how U.S.macroeconomicshocksaffecttheCanadianeconomyitwouldseemnaturaltoextendthelinkstoincludethoseemanatingfromtheU.S.financialsystem.Theso‐called‘global’financialcrisisputspaidthenotionthatU.S.influencesonCanadianmacroeconomicoutcomesoperatesolelythroughtherealsideoftheeconomy.Accordingly,theseparateestimatesfortheU.S.andCanadaareaugmentedwithafactor‐augmentedVAR(FAVAR)whichattemptstocapturetheessenceofrealandfinancialshocksontheCanadianeconomyemanatingfromCanada.

Theremainderofthispaperproceedsasfollows.Section2presentsthemethodologyandfindingsoftheliteraturethat motivatedthisstudy.Section 3motivatestheuseofVAR/VECM/FAVARanalysesanddescribesthemethodology.Section4describesthedatausedandprovidesbasicanalysisontheFederalReserve’sandBankofCanada’sSeniorLoanOfficerSurveys.Wethenestimatethedynamicrelationshipbetweentighteningcreditstandards,businessloans,andoutput.WealsoreplicatetheVARestimatedinLownet.al.(2000),thenestimateVARsandmoresophisticatedtimeseriesmodelsusingbyextendingtheiroriginalsample.Empiricalresultsbasedonextensionstothebasicmodelarealsoprovided.

Briefly,thispaperfindsanegativerelationshipbetweentighteningnon‐pricelendingstandards,loans,output,andinterestratesoverallsampleperiodsforbothCanadaandthe

UnitedStates.Althoughthisanalysisdoesnotshowadirectcausalrelationship,asloanofficersmaybeawellinformedgroupthatchangeslendingstandardsbasedontheirexpectationoftheeconomy,thesurveydataappearstocontainsomecriticalinformationinthespecificationofmacromodelsandshouldbeconsideredasacandidateforinclusioninmodelsthat,untilrecently,gaveshortshrifttoaroleforchangingcreditconditions.Moreover,unexpectedtighteningofcreditstandardsresultindisequilibriainourcointegratingequationwhichaffectsbusinessloans,output,andothermacroeconomicvariables.Theseconclusionsareconsistentwithcreditrationingtheory,asnon‐pricelendingstandardsarelikelytoaffectcreditavailabilityregardlessoftheinterestrate. Finally,theFAVARestimatesforCanadasuggestthatmacro‐modelswhichincorporateaninfluencefromU.S.realshocksbutomitfinancialshocksarepotentiallymis‐specified.Section5concludesandoffersideasforfurtherresearch.

2. CreditConditionsandMacroeconomicOutcomes:LiteratureReview

Disequilibriaincreditmarketshavebeendiscussed,modeled,andempiricallytested,fordecades. Roosa (1951) proposes the availability doctrine, namely the notion that creditavailabilityimpactstheeffectivenessofmonetarypolicy.Fuerst(1994)furtherrefinesRoosa’savailabilitydoctrinebydefiningtwodistinctcomponents,namelycreditrationing,acommonphenomenoninvirtuallyallcreditmarkets,andaroleformonetarypolicywhichcanimpactthesupplyofcredit.

BlanchardandFischer(1989)definetwotypesofcreditrationing.Typeonecreditrationingoriginateswhenindividualscannotborrowasmuchastheywantatthegoinginterestrate.Typetwocreditrationingariseswhen,amongidenticalborrowers,someindividualsareabletoborrowwhileothersareunabletodoso.

Basiceconomictheorysuggeststhatacreditshortagewouldresultinhigherinterestratesuntilthemarketclears.Ifcreditrationingexists,marketsarenotclearingaspricesarenottheonlyrelevantsignal.Consequently,disequilibriaincreditmarketscanpersist.StiglitzandWeiss(1981)presentacreditrationingmodelformarketswithimperfectinformationwhichattributesshort‐rundisequilibriatoexogenousshocksandpricestickiness.Theysuggestthatlong‐rundisequilibriacanbeexplainedbygovernmentintervention.

StiglitzandWeiss’s(1981)modeldemonstratesthathigherinterestratesleadthemostriskaversefirmstodropoutofthepotentialborrowingpool,creatinganadverseselectionproblem.Additionally,itprovidesanincentiveforborrowerstoengageinriskybehaviour,resultinginamoralhazardproblem.Asinterestratesincrease,theprobabilitythatborrowerswillsuccessfullypaybackloansdecreasesbecauseofimperfectinformationinthecreditmarkets.Therefore,sincebanksareprofitmaximizinginstitutions,aninterestrateceilingwouldbeset,leadingtoacreditmarketdisequilibrium.Aprofitmaximizingpriceceilingduetoapositivecorrelationbetweenthepriceofborrowingandtheprobabilityofloandefaultistheliterature’sexplanationforwhybanksdonotincreasepricestoclearcreditmarkets.Instead,banksresorttocreditrationing.

SchreftandOwens(1991)suggestthatthemonetarypolicyauthority(viz.,theUSFederalReserveinthiscase)takestheviewthat,asthecostofavailablefundsincreases,interestratesappliedtobankloanslagchangesinnon‐pricelendingstandards.Therefore,informationonbanks’non‐pricelendingstandardshelpsexplaintheimpactmonetarypolicyhas onthebanking sector.Bernanke and Blinder (1992) empiricallydemonstrate thata

reductioninavailablefundscausebankstoselloffsecuritiesintheshort‐run.However,inthelong‐run,securitiesarereplenishedresultinginadecreaseinloans.Hence,itisimportanttounderstand the transmission mechanism between changes in available funds and loanavailability.

ItcomesasnosurprisethenthattheU.S.FederalReserveandtheBankofCanada,amongothers,conductsurveystomeasurehowbanklendingstandardshavechangedoverthepreviousthreemonths.Aswithallsurveys,theSeniorLoanOfficerSurveycontainssomeinherentbiases.AlthoughLownet.al.(2000)believethesurveydataareinformativeandhelpimproveforecastaccuracyforsomereal economicvariables in macromodels, they alsohighlightthesurvey’squalitativenature,smallsamplesize,andpossiblereportingbiasasadditionalpotentialpitfallswhenusedineconometricmodels.SchreftandOwens(1991)alsorecognize the possibility of bias in survey data caused by reports not being madeanonymously.

Lownet.al.(2000),LownandMorgan(2006),Swiston(2008),andBeatonet.al.(2009)examinetheU.S.FederalReserve’sSLOStoseeifthedatarepresentareasonableproxyforcreditavailability.Allofthesestudiesconcludethatthesurveyisareasonableproxyforeconomywidenon‐pricecreditconditions.

Lownet.al.(2000)estimateaVARwithrealGDP,GDPdeflator,commodityprices,thefederalfundsrate,amountofcommercialandindustrialloans,andU.S.SLOSdata.Relyingonasampleof1974Q1to1984Q4and1990Q2‐1998Q4,Lownet.al.(2000)findthatanunexpectedonestandarddeviationtighteningintheSLOSdataresultsinU.S.commercialandindustrial

loansdecliningby2.5%.Lownet.al.(2000)thencontendthat,ifthedataareareasonableproxyforcreditconditions,thesurveymightshedsomelightoncreditmarketdisequilibria,potentiallyinformingmonetarypolicyofficialsaswellastheprivatesector,especiallyasthesurvey’sreleasedateprecedessomeeconomicandfinancialdata.

LownandMorgan(2006)extendtheanalysisofLownet.al.(2000)byestimatinga

vectorautoregressivemodelconsistingofrealGDP,GDPdeflator,commodityprices,thefederalfundsrate,amountofcommercialandindustrialloans,andU.S.SLOSdata,appliedovera disjointed sample period of 1969Q1‐1984Q1 and 1990Q2‐2000Q2. Impulse responsefunctionsrevealthatashockinU.S.FederalReserve’sSLOSdataresultsinan8%nettighteningofstandardsanddecreasesoutputbyabout0.5%atit*trough.TheyalsofindthatthefederalfundsrateseemstodecreaseafteraSLOSshock,fallingbyabout50basispointsafterthreequarters.

Swiston(2008)incorporatesU.S.FederalReserveSLOSdatainafinancialconditionsindexasherecognizesthedataishighlycorrelatedwithrealactivityandfinancialmarketvariables.UsingaVAR,thestudyfindsthatatighteningofSLOSstandardsresultinnon‐residentialfixedinvestmentdeclinesforovertwoyears,eventuallyfallingbyover2%.Ashaveothers, Swiston concludes that the survey significantly explains economic growth afteraccountingforforward‐lookingfinancialmarketinformationsuchasequityreturnsandhighyieldbondspreads.TheBankofCanada’sfinancialconditionsindex(FCI)alsoincorporatesaneffectfromtheSLOS.

Beatonet.al.(2009)addsU.S.FederalReserveSLOSdatatoafinancialconditionsindex.Asaresult,theyfindthataonestandarddeviationshocktoSLOSstandardsdata,equivalenttoanettighteningof8.6%,reducesGDPbyroughly0.6%aftertwoyears.TheauthorsalsoestimateaVARandconcludethatatighteningofbusinesslendingstandardscontributessignificantly to a contraction in business investment. Guichard and Turner (2008) alsoincorporateU.S.SLOSdataintoafinancialconditionsindex.Inthisstudy,variableweightsareconstructedusingareduced‐formoutputgapequationandanunrestrictedVAR.TheyreportthattheSLOSdataisstatisticallysignificantwitha1%percentnettighteningintheSLOSleadingtoadeclineinGDPgrowthofapproximately0.25percent.

Increasesinloansandinvestmentcouldbeexplainedbyeasinglendingstandardsorincreasingloandemand.Thus,whenanalyzingthepredictivepropertiesoftheSLOSdata,itisessentialtocontrolforloandemand.Thereisanidentificationproblemaschangesinloanprice,thegoinginterestrateonloanablefunds,reflectsbothdemandandsupplyfactorswhichoperatesimultaneously.Assuch,itisnecessarytouseappropriateinstrumentswhentryingtoisolatethedemandandsupplyfactors.

Analyzing the affect SLOS non‐price standards have on loans, output, and othereconomicvariablesiscriticaltounderstandingthetransitionmechanismsofthecreditmarkets.

Cunningham(2006)testsSLOSdatatoseeifloanofficer'suniquepositioninthebanking

industryprovidesadditionalinformationoverandabovethatwhichthewellinformedpublicisable to collectanddigest.Cunningham(2006) uses abatteryofstandard testssuchasregressinglaggedvaluesofSLOSdataandavectorofcontrolvariablesonninedifferent

dependentvariables,suchasrealGDPand realGDPgrowthlessprivateinvestment.HeconcludesthatFederalReserveSLOSsurveydatacausedstatisticallysignificantchangesinlendingandrealeconomicactivity.

AsfarasweareawarethereisnostudyexaminingtheBankofCanada’sSLOSdatainaneconometricmodel.

3. EstimationApproachandData

a. MethodologyWebeginwithastandardVARwhichiswrittenasfollows:0 1 1t   t  ty A A y ε (1)

whereyisavectorofendogenousvariables,allofwhichareobservableandrepresentthemacroeconomicmodelthatservesasthebasisforinvestigatingtheroleofcreditstandardswhileεisanerrortermwiththeusualproperties.Exogenousvariablescanalsobeaddedtoequation(1)buttherelevanttermisomittedhereforsimplicity.Previousstudiesdidnotconsider the possibility, assuming that the span of the data set is sufficiently long, ofcointegrationbetweensomeofthecorevariablesinequation(1).Thus,forexample,thereispossiblyanequilibriumrelationshipbetweenthestanceofmonetarypolicy,proxiedbyacentralbankpolicyrate,andthevolumeofloans.Similarly,atighteningofmonetarypolicymaywellbelinkedinthe‘long‐run’withthedegreetowhichcreditstandardsaretightenedorloosened.Similarly,iftherealandfinancialsectorsarelinkedtoeachotherthenastatistical

attractionbetweenrealeconomicactivity(i.e.,realGDP)andloansisalsolikelytoexist.Unfortunately,ifcointegrationispresentinthedatathen(1)ismis‐specifiedandthevectorerrorcorrectionform(VECM)representsthecorrectspecification.Equation(1)writteninVECMformis'0 1t  t t y A  y ε (2)whereyremainsthevectorofendogenousvariables,andπ,whoserankrepresentsthenumberofcointegratingrelationships,is(A1–I),thatis,itisobtainedbydifferencingbothsidesofequation(1).Theneedtodifferencetheseriesstems,ofcourse,fromthenonstationarybehaviorof manymacroeconomictime series. Assuch,differencing is notnecessarytoestimateandgenerateusefulinferencesfromVARssuchas(1).Instead,itisthecointegrationpropertythatrequiresthesecondterminequation(2)sincey_t10isthelongrun.

Returningtothe standardform forthe VAR(i.e., equation(1))wecannextaddobservablesthatdefinecreditconditions,followingLownandMorgan(2006),whichresultsinanexpandedVARoftheform0 1 1 2 1t   t  t  ty A A y A z ε (3)wherezisthevectorofvariablesthatproxycreditconditions.Equation(2)thenrepresentsthebenchmarkorcoremodelthatisestimatedandcapturestheessenceofthelinksbetweentherealandfinancialsectorsoftheeconomy.

Aspointedoutearlier,itisconceivablethattheseriesmeasuringcreditconditionsconflate(loan)demandandsupplyfactors.Thus,forexample,atighteningofcreditstandards

canbothleadtoafallinloansand,hence,economicactivity,orinsteadrepresentaresponsetoanongoingeconomicslowdown.AgainfollowingLownandMorgan(2006),itisusefulthereforetoconsiderothervariablesthatmayhelpusidentifydemandfromsupplyfactorsininfluencingthevolumeofloansandthedeterminationofcreditstandards.AnextendedVARisspecificed,whereinweaddforward‐lookingvariableswhich,atleastintheory,arethoughttoprimarilyaffectloandemand.Theyare:forecastsofrealGDPgrowth,thetermspread,andanaggregateindicatoroffinancialconditions.Allthreeoftheseserieshavebeenusedinmanypreviousstudiestocapturefutureeconomicconditionsand,therefore,arelikelytodirectlyaffectloandemand.Inotherwords,ifloanstandardsretaintheirsignificantinfluenceonmacroeconomicactivity,aftercontrollingforloandemand,thentheSLOScapturestheroleofchangingcreditconditions.Consequently,creditstandardshaveanimportantplaceinmacroeconomicmodelsthatexplicitlycapturetherelationshipbetweenrealandfinancialsectorsofaneconomy.WecanwritetheextendedVARstraightforwardlyasfollows

* ' ' * '

0 1 1 2 1

t   t  t  t

y A A y A z ε (4)

whereallthetermswerepreviouslydefinedexceptthaty*nowrepresentsthebenchmarkendogenous variables augmentedwithadditional variables thatdrive loandemand. It isstraightforwardtomodifyequation(4)towriteitinVECMform.

Next,wearealsointerestedintheinfluenceofbothmacroeconomicandfinancialfactorsemanating fromtheU.S. onCanadian macroeconomic andfinancial conditionsapossiblewayofdoingsoiseithertoestimateequations(1)through(4)inapanelsettingor,instead,augmentthespecificationswithaproxyfortheimpactofU.S.factorsontheCanadian

economy.Unfortunately,anyextensiontotheexistingspecificationsislikelytoexhaustthelimited number of available degrees of freedom. A more practical approach under thecirc*mstances then is to estimate afactor vector autoregressive model as outlined, forexample,inBernanke,Boivin,andEliasz(2005).IfthefactorssummarizingU.S.macroeconomicandfinancialconditionsaresummarizedbyFthen,inafirststep,thefollowingequationdescribestheprincipalcomponentsextractedfromtheVARbasedonU.S.datawhichcanbewrittenas

US US

t   t  t

y F e (5)

whereyisthevectorofendogenousUSvariables,Λarethefactorloadingsandeisazeromean,constantvarianceerrorterm.TheresultingjointdynamicsdescribingtheFAVARcanthenbeexpressedas11( )t ttt t            F FLy y (6)whereψ(L)isapolynomialororderd.

b. DataandStylizedFacts

ForseveraldecadestheFederalReservehasconductedasurveyofbankofficialsknownastheSeniorOfficerLoanSurvey(http://www.federalreserve.gov/boarddocs/SnloanSurvey/).Foratimeitwasfeltthatthe‘non‐price’elementcapturedbythesurveywerebecomingunimportantasinterestratesbecamemarketclearingandappearedtoserveastheprimary

mechanisminmakingcommercialloandecisions.Subsequenteventsledtoarethinkoftheroleofnon‐pricefactorsincreatingcreditcrunchtypeconditionsandthesurveywasre‐instated.SLOSdatarepresentabalanceofopinionbasedonthefollowingquestion:

“Overthepastthreemonths,howhaveyourbank’screditstandardsforapproving

loanapplicationsforC&Iloansorcreditlikes–excludingthosetofinancemergers

andacquisitions–changed?1)Tightenedconsiderably2)tightenedsomewhat3)

remainedbasicallyunchanged4)easedsomewhat5)easedconsiderably”(LownandMorgan2006)2

Since1999theBankofCanadahasconducteditsownquarterlysurveyofSeniorLoanOfficer Survey on business lending practices (http://www.bankofcanada.ca/publications‐

research/periodicals/slos/).Thesurveysample,whichhasbeenlargelyunchangedsince1999,consistsof11financialinstitutionswithatotalmarketshareofroughly60%.TheSLOSfocusesoncorporate,commercialandsmallbusinessloans.

Inthesurvey,financialinstitutionsareasked:

"Howhaveyourinstitution’sgeneralstandards(i.e.yourappetiteforrisk)andtermsfor

approvingcreditchangedinthepastthreemonths?”(Faruquiet.al.,2008)

Surveyrespondentsindicatewhethertheirbusinesspracticeshavetightened,eased,orremained unchanged with respect to pricing of credit,general standard, limit ofcapitalallocation,andtermsofcredit.Thenon‐pricecreditstandardsdataincorporatesgeneralstandards,limitofcapitalallocation,andtermsofcredit.Non‐pricedataisabalanceofopinionwhichiscalculatedusingthepercentageoftighteningresponsesminusthepercentageofeasingresponses.Inspiteofthedistinctionbetweenpriceandnon‐priceelementsinthe

measurementofcreditstandardsthedifferencesbetweenthetwoindicatorsdoesnotappearsubstantive.3

OtherremainingmacroeconomicandfinancialassetdatawereobtainedfromCANSIM,the U.S. Bureau of Economic Analysis, Federal Reserve Economic Data(http://research.stlouisfed.org/fred2/;FREDII)andtheBoardofGovernorsoftheU.S.FederalReserveSystem.Dataareseasonallyadjustedatthesource,unlessotherwiseindicated.TheloansvariableisannualizedCanadiancharteredbankbusinessloans.RealGDPisexpenditurebasedusing2002dollars.Consumerpriceindexisbasedona2005basketofgoods.CommoditypricesistheBankofCanada’scommoditypriceindex(1982‐1990=100)evaluatedinUSdollars.TheovernightrateistheindicatorofBankofCanadamonetarypolicy.RealtimeGDPandpotentialrealGDPdataarefromtheBankofCanada.

U.S.realGDPisseasonallyadjustedatannualratesand2005prices.TheGDPdeflatorwascalculatedusingreal(2005)andnominalGDP.Themonetarypolicyindicatoristhefederalfundsrate.Toproxycommoditypricesweexperimentedwithtwoseries,namelyoilprices(WestTexasintermediatecrudepriceperbarrel)aswellastheProducerPriceIndex(PPI;1982=100).InwhatfollowsallresultsrelyontheoilpriceproxyastheconclusionswereessentiallythesamewhenusingthePPI.

Severalotherserieswerealsocollectedsince,aspreviouslynoted,adifficultywiththehypothesesofinterestisthatitisdifficulttoempiricallyidentifyloandemandfromloansupplyfactorsininterpretingtheroleandinfluenceof the Senior OfficerLoan Surveys (SLOS).

3_{Anappendixprovidesatimeseriesplotofthetwosurveyindicators.Theseriesappearhighlyattractedtoeach}

otherinthesensethattheyarecointegrated.Hence,intheempiricalworkthatfollowsweusethebalanceof

FollowingLownandMorgan(2006)wealsocollectedseriesforexpectedrealGDPgrowth,proxiedusingtheoneyearaheadConsensusForecastsfortheU.S.andCanada,thetermspread,andanindicatoroffinancialstressinbothcountries.Thetermspreadisthe3monthcommercialpaperbill–TreasurybillspreadforboththeU.S.andCanada.FinancialconditionsareevaluatedusingtheChicagoFederalReservesNationalFinancialConditionsIndex(NFCI)

“measures risk, liquidity and leverage in money markets and debt and equity markets as well as in

the traditional and “shadow” banking systems”. Additional details can be obtained from

http://www.chicagofed.org/webpages/publications/nfci/index.cfmbutitisnoteworthythatthe

indexdoesseemroughlycoincidentwithrecessions,thatis,thevalueoftheindexrisessharplywiththeonsetofarecession(notshown).Similarly,forCanada,thebankofCanada’sFinancialConditionsIndex(FCI)servesasthemeasureoffinancialstress(http://credit.bank‐banque‐

canada.ca/financialconditions/fci).NotethatthereissomeoverlapbetweentheFCIandsomeofthevariablesusedintheestimatedmodel.Forexample,theSLOSandinterestratesarebothcomponentsoftheindexalthoughtheiroverallweightissmall.Nevertheless,astheFCIalsoincludeshousingandequitypricesaswellastherealexchangerate,theserepresentelementsthatcanassistusinseparatelyidentifyingdemandfromsupplyfactorsincreditconditionsandtheirlinktomacroeconomicconditions.

Theempiricalresultspresentedbelowrelyonasampleofquarterlydatafortheperiod1972:1‐2011:1fortheU.S.and1999:2‐2011:1forCanadiandata.TheSLOSinCanadaisofamorerecentoriginthanthecomparableU.S.survey.TheoriginaldatasetusedbyLownandMorganisslightlylongersincetheirsamplebeginsin1968:1butwewereunabletoobtainthe

dataforthe1968‐1971period.TheU.S.dataalsocontainsagapwhensurveydataarenotavailable.

Figures1aand1bplottheSLOSsurveydataagainstloansfortheU.S.andCanada,respectively.Alsoshownareshadedareasrepresentingrecessiondatesinbothcountries.Therecessiondates for the U.S. are from the NBER’s business cycle chronology(http://www.nber.org/cycles.html) while the dates for Canada were obtained from thechronology established by the Economic Cycle Research Institute(http://www.businesscycle.com/).TheverticaldashedinFigure1arepresentstheendofthesampleconsideredbyLownandMorgan(2006).Atotalof6recessionswererecordedintheU.S.dataandtheriseintheSLOSstandards,indicatingnettighteningofcreditconditions,isreadilyapparent.Noticealsothattheindicatorreachesitshighestpointduringthefinancialcrisisof2008‐2009.Onlytheperiodfollowingthefirstoilpriceshockof1974‐1975comesclosetoyieldingasimilartighteningofcreditconditions.Canadaalsoexperiencesasharptighteningofcreditconditionsbuttherisein2008‐2009,acasualtyofthe‘global’financialcrisis,appearstohaveitsoriginsalreadyin2007whenothereventsthatprecipitatedtheeventualglobalcrisistookplace.4. EmpiricalResults

Thevariousspecificationspreviouslydescribedwereestimatedforthreesamples,namelytheoriginalsamplespecifiedbyLownandMorgan(2006),whichexcludestheperiodwhennoSLOSdataareavailable,afullsampleendingin2011:1,aswellassamplethatconsidersdataonlysince1990.Oneadvantageofconsideringdatasince1990isthatitoverlapswiththe

periodwhentheFed’smonetarypolicywasrelativelymorepredictablethanduringthe1970sand1980s.LaglengthsforallVARs,VECMsandFAVARswerechosenbyrelyingeitherontheAICorHQcriteria.Sincethelossofdegreesoffreedomisanimportantbarriertoestimation,especiallyfortheextendedVARs,wetypicallyerredonthesideofparsimony.Forthemostpart,however,whenthevariousVARsandtheirvariantswereestimatedforlongerlags,theconclusionswereunchanged.Allspecificationsusedthefollowingordering



y p p i, , , , ,c  s



(7)

wherethemacrovariables,namelythelogarithmofrealGDP,thelogarithmoftheGDPdeflator,thelogarithmofcommodityprices,thepolicyrate,arefollowedbythefinancialvariables,namelythelogarithmofcommercialorbusinesscredit,andthecreditstandardssurveyindex.TheextendedVARsareorderedjustasin(7)exceptthatrealGDPgrowthforecasts,atermspread,andthefinancialconditionsindexprecedethecreditvariables.Althoughitisgenerallyagreedthatmacrovariablesshouldbeorderedbeforethecreditvariablesthereisnotnecessarilyanyconsensusontheparticularorderingofvariableswithineachgroup.Hence,inadditiontoestimatingconventionalimpulseresponses,wealsoestimategeneralizedimpulseresponsesastheseareinsensitivetothechosenordering.

All impulse response functions (IRFs) are evaluated over a 12 quarter horizon andconfidencebandsareestimatedviaMonteCarlo(100replications).Figures2and3presenttheIRFsforequations(1)and(3),thatis,thebenchmarkandextendedVARsinthelevelsorloglevelsoftheseries.GiventhedimensionoftheVARsthefiguresfocusonthreeIRFsnamelytheresponseofrealGDPtoashockinloanstandards,aswellastheIRFsfortheimpactofloans

standardsonloanstandardsandtheimpactofloansonloanstandards.Figure2aattemptstoreplicateportionsofFigure2inLownandMorgan(2006).WhileweareabletogenerallyreproducetheirresultsthereareatleasttwonotabledifferencesbetweentheirVARsandours.First,thelatestvintageofdataforrealGDPisutilizedwhichislikelydifferentthantheseriesemployedintheirstudy.Second,thesampleperiodusedtogeneratetheIRFsisslightlyshorter,asnotedpreviously,sincewewereunabletocollecttheSLOSdatafortheperiod1968‐197,inclusive.Nevertheless,theresultsdoshowthatrealGDPreactsnegativelytoanettighteningofcreditstandards(i.e.,ariseintheindex)andthatloansdodeclinewhenstandardsareindeedtightened(middleplot).Incontrast,however,toLownandMorgan(2006,Figure2)wefindnoimpactfromloanstostandards.Hence,theconcernthattheSLOSdataconflatesbothdemandandsupplyfactorsmaynotbewarranted,atleastbasedonthebenchmarkmodel.Whendatacoveringthefullsampleisconsidered(seeFigure2b)theresultsaresimilar.Note,however,thatSLOSshockshavearelativelylargerimpactonrealGDPthanintheshortersamplewhilethenegativeimpactofapositiveSLOSshockonloansismoremutedthanwhendatafromthemostrecentdecadeareomitted.RecallthatthelastdecadehasbeenbesetbyaweakeningofstandardsandtheemergenceofastrongerlinkbetweentherealandfinancialsectorsoftheU.S.economy.Nevertheless,theadditionofalmostadecadeofdatahasnotfundamentallyalteredtheroleoftheSLOSininfluencingrealGDPorthevolumeofcommercialloansintheU.S.

Next,weconsidertheimpactofSLOSshockswhenwerelyonrealtimedata.Figure2cillustratestheresultsbasedonthe2009Q3vintageforrealGDP,thatis,onethatlikelyreflectstheimpactofthefinancialcrisisonmacroeconomicconditions.Justasitisnowwidelyaccepted

thatthestanceofmonetarypolicyisdictatedbytheinformationatthedisposalofpolicymakersitisreasonabletoassumethatcreditstandardswillalsobeinfluencedbywhatthedataportendtoseniorloanofficersatthetimetheysetstandards.TheimpactonrealGDPofashocktostandardscontinuestobenegativeandislargelyunaffectedbytheresorttorealtimedata.Similarly,loansdeclineinresponsetoanSLOSshock.However,wecannowreportapositiveimpulseresponsefromashocktoloansoncreditstandardsimplyingthatariseinthevolumeofloanshastheeffectofleadingtoanettighteningofstandards.ThisisthepreciselywhatLownandMorganalsoreport.Yet,theresultistobeexpectedsincetheU.S.wasinthethroesofafinancialcrisis.Whenthesameexerciseisrepeatedforaselectionofothervintages(resultsnotshown)theimpactfromloanstocreditstandardslargelydisappears.Thevintageswerecarefullychosentotakeadvantageofthestylizedfact,observedinFigure1,ofapositiverelationshipbetweenatighteningofcreditstandardsandtheonsetofarecession.Therefore,the2000Q4and2007Q3vintagesprecedebusinesscyclepeakswhilethe2002Q3,asisthecasewiththe2009Q3vintagedisplayedinFigure2c,comeimmediatelyafterbusinesscycletroughs.Finally,Figure3showstheIRFsfortheextendedVARrepresentedbyequation(3).Inspiteoftheweakevidencethatthereisaconflatingofdemandandsupplyfactorsintheinterpretationoftheimpactofchangingcreditstandardsonthemacro‐economytheadditionofforward‐lookingvariablesdrivingthedemandforloansdoesnotchangetheIRFslinkingshockstocreditstandardstorealGDPorthevolumeofloans.Wenowalsoreportatighteningofcreditstandardsinresponsetoashockfromloans.

TurningtotheevidenceforCanada,showninFigure4,theresultsarereminiscentoftheevidencejustpresentedfortheU.S.Moreover,estimatesbasedonrealtimedatadonotseem

toimpactthefindingsbasedonreviseddataorevenU.S.data.SinceCanadadidnotexperiencethefinancialcrisisonthesamescaleasdidtheU.S.itisdifficulttoblametheearlierresultsasbeingdrivenbythescaleofthefinancialcrisis.Hence,wehaveadditionalevidencemakingtheimportantlinkbetweentherealandfinancialsectorsoftheeconomy.Finally,itisworthpointingoutthatnoneoftheIRFsaresensitivetotheorderingofthevariablesbasedonanexaminationoftheGeneralizedIRFs(notshown).

Wenowturntothevariancedecompositions(VD)whicharedisplayedinTables1and2.These are all based on the full sample estimates (1972‐2011). These reveal some veryinterestingfeaturesaswellashighlightingsomeU.S.‐Canadadifferences.TheVDsforthecoreU.S.modelconfirmthestatisticalimportanceoftheSLOSinexplainingnotonlymovementsinrealGDPbutalsothevolumeofcommercialloans.WealsofindthattheVDsfortheSLOSdonotexplainthefedfundsratebutthatalmostathirdofthevolumeofloansisexplainedisexplainedbytheFed’spolicyrateafter12quarters.Additionally,thefedfundsrateexplainsmoreofthevarianceoftheSLOSthananyothervariableotherthanitsownlags.Whenthecoremodelisre‐estimatedusingrealtimedatatheimpactcanbedramaticasseeninTable1b.Forexample,theproportionofthevarianceofthefedfundsrateexplainedbytheSLOSisapproximately25timeslargerthanwhenfinalreviseddatawereused.Similarly,theVDsfortheimpactofSLOSonloans,after12quarters,isover3timeslargerthaninthecoremodelcase.Needlesstosaythe2009Q3vintageisexceptionalbuttheresultsdomakeabundantlyclearthatinferencecanbehighlysensitivetotheuseofrealtimedata.Finally,whenweturntotheVDsfortheextendedmodeltheresultsaremixed.Thus,whiletheSLOShasanevenlargerimpactthaninthecoremodelonmonetarypolicy(i.e.,thefedfundsrate)theexplanatory

poweronrealGDPisroughlyhalfaslargeasinthecoremodelcase.Incontrast,theextendedmodelrevealsthatmonetarypolicy’simpactonSLOSisnotaslargeasinthemorerestrictivemodelbutthatthevolumeofloansexplainsapproximatelyfivetimesmoreofthevariationinSLOSthaninthecoremodel.

TheCanadianevidenceshowninTable2mirrorsmanyofthefeaturesfoundintheU.S.case.Nevertheless,therearetwonotabledifferences.First,SLOSexplainsamuchgreaterportionofthevarianceofallofthevariablesthaninthecomparablecoreU.S.model.Totheextentthatdifferencesinregulationandsupervisionplayaroleininfluencinghowtightorloosearecreditstandards,theVDsoffersomeevidencethatsuggestsamarkedcontrastbetweentheU.S.andCanadianexperiencesinrecentyears.Second,unlikeitsU.S.cousin,thetermspreadexplainsamuchlargerportionofthevarianceofallthemacrovariables.Inotherwords,whilethereisstrongevidenceforCanadathatthetermspreadhasanimportantinfluenceonrealGDPthesamecannotbesaidforU.S.data.

5. Conclusions[tobeadded]

References

Adrian,T.,andH.Shin(2011),“FinancialIntermediariesandMonetaryEconomics”,inB.FriedmanandM.Woodford(Eds.),HandbookofMonetaryEconomics(Amsterdam:Elsevier),pp.601‐650.Bayoumi,T.,andR.Darius(2011),“ReversingtheFinancialAccelerator:CreditConditionsandMacro‐FinancialLinkages”,IMFworkingpaper11/26,February.Beaton,K.,R.Lalonde,andC.Luu(2009),“AFinancialConditionsIndexfortheUnitedStates”BankofCanadaDiscussionPaper.Bernanke,B.S.(1993),“CreditandtheMacroeonomy”,FederalReserveBankofNewYorkQuarterlyReview(Spring).

Bernanke,BenS.,JeanBoivinandPiotrEliasz(2005)"MeasuringTheEffectsOfMonetaryPolicy:AFactor‐AugmentedVectorAutoregressive(FAVAR)Approach,"QuarterlyJournalofEconomics120(1,Feb),387‐422.Bernanke,B.S.,andS.Gilcrhist(1999),“TheFinancialAcceleratorinaQuantitativeBusinessCycleFramework”,inJ.TaylorandM.Woodford(Eds.),HandbookofMonetaryEconomics,vol.1C(Amsterdam:Elsevier),pp.1341‐93.

Bernanke,B.S.,andM.Gertler(1995),“InsidetheBlackBox:TheCreditChannelofMonetaryPolicyTransmission”,JournalofEconomicPerspectives9(Fall):27‐48.

Bernanke,B.S.,andA.S.Blinder(1992),“TheFederalFundsRateandtheChannelsofMonetaryTransmission.”AmericanEconomicReview83.4(1992):901‐921.

Blanchard,O.J.,andS.Fischer(1989),LecturesonMacroeconomics(Cambridge:TheMITPress).

Blinder,A.S.,andJ.E.Stiglitz(1983)“Money,CreditConstraints,andEconomicActivity”AmericanEconomicReview73(2):297‐302.

Ciccarreli,A.Maddaloni,andJ‐L.Peydró(2010),“TrustingtheBankers:ANewLookattheCreditChannelofMonetaryPolicy”,workingpaper,EuropeanCentralBank.Cunningham,T.J.(2006)“ThePredictivePoweroftheSeniorLoanOfficerSurvey:DoLendingOfficersKnowAnythingSpecial?”,FederalReserveBankofAtlantaWorkingPaper24.Engle,R.F.,andC.W.J.Granger(1987)“Co‐integrationandErrorCorrection:Representation,Estimation,andTesting.”Econometrica55(2):251‐276.

Faruqui,U.,P.Gilbert,andW.Kei(2008)“TheBankofCanada’sSeniorLoanOfficerSurvey.”BankofCanadaReview.

FederalReserveBoard(2009),“SupportingStatementfortheSeniorLoanOfficerOpinionSurveyonBankLendingPractices”

SeniorLoanOfficerOpinionSurveyonBankLendingPractices,http://www.federalreserve.gov/boarddocs/snloansurvey/about.htm

Fuerst,T.S.(1994),“TheAvailabilityDoctrine.”JournalofMonetaryEconomics34(1994):429‐443.

Gertler,M.,andN.Kiyotaki(2011),“FinancialIntermediationandCreditPolicyinBusinessCycleAnalysis”,inB.FriedmanandM.Woodford(Eds.),HandbookofMonetaryEconomics

(Amsterdam:Elsevier),pp..

Guichard,S.,andD.Turner(2008),“QuantifyingtheEffectsofFinancialConditionsonUSActivity.”OECDEconomicDepartmentWorkingPapers,635.

Jaffee,D.,andJ.E.Stiglitz(1990),“CreditRationing.”In:HandbookofMonetaryEconomics:

VolumeTwo,B.M.FriedmanandF.H.Hahn(Eds.),(Amsterdam:NorthHolland),pp.838–888.

Kashyap,A.,J.Stein,andD.Wilcox(1993),“MonetaryPolicyandCreditConditions:EvidencefromtheCompositionofExternalFinance”AmericanEconomicReview83(1):78‐98.

Lown,C.,andD.P.Morgan(2006),“TheCreditCycleandtheBusinessCycle:NewFindingsUsingtheLoanOfficerOpinionSurvey.”JournalofMoney,Credit,andBanking,38(6):1575‐1597.

Lown,C.,D.P.Morgan,andS.Rohatgi(2000)“ListeningtoLoanOfficers:TheImpactofCommercialCreditStandardsonLendingandOutput.”FederalReserveBankofNewYorkEconomicPolicyReview,6(2).

Martin,A.,McAndrews,J.,andD.Skeie(2010),“BankReserves,Lending,andtheInterbankMarket”,workingpaper,FederalReserveBankofNewYork.Owens,RaymondE.,andStaceyL.Schreft(1991)“SurveyEvidenceofTighterCreditConditions:WhatDoesItMean?”EconomicReview77(2):29‐34.

Roosa,R.V.(1951),“InterestRatesandtheCentralBank”,InMoney,TradeandEconomic

Growth:EssaysinHonorofJ.H.Williams(NewYork:Macmillan),pp.207‐295.

Stiglitz,J.E.,andA.Weiss(1981),“CreditRationinginMarketswithImperfectInformation.”AmericanEconomicReview,71(3),393‐410.

Figure1aSeniorOfficerLoanSurveyandCommercialLoans,1972‐2011:U.S.

Figure1bSeniorOfficerLoanSurveyandCommercialLoans,1999‐2011:Canada

‐40‐200204060801001975 1980 1985 1990 1995 2000 2005 2010

Nettighteningof standards Commercial loan growth

Percent(annual)No standards reported1984- 90‐40‐20020406080100‐1.0‐0.50.00.51.01.52.02.51999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

BalanceofOverall credit standards Growth in business credit

SLOSindexAnnualprecentchange

APPENDIX

TheNFCI:U.S.

TheFCI:Canada

‐4‐3‐2‐10121999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010BOC_FCI+=improvingfinancialconditions/‐=deterioratingfinancialconditions

CoreSeries:U.S.Data 8.48.68.89.09.29.49.61975 1980 1985 1990 1995 2000 2005 2010USre a lGDP3.23.64.04.44.81975 1980 1985 1990 1995 2000 2005 2010GDPdefl a tor123451975 1980 1985 1990 1995 2000 2005 2010Oi lPri ce0481216201975 1980 1985 1990 1995 2000 2005 2010

USfederalfundsrate

7.07.58.08.59.09.51975 1980 1985 1990 1995 2000 2005 2010Commercialloans‐40‐200204060801001975 1980 1985 1990 1995 2000 2005 2010SLOSLogarithmofthelevelsLogarithmofthelevels(index)LogarithmofthelevelsPercentLogarithmofthelevels(Billionsof$)+indicatesnettighteningofstandards‐indicatesnetlooseningofstandards