1. Monetarypolicymatters:Evidencefromnewshocksdata
S. MahdiBarakchian,ChristopherCrowen
GraduateSchoolofManagementandEconomics,SharifUniversityofTechnology,Tehran,Iran
a r t i c l e info
Article history:
Received19February2010
Receivedinrevisedform
24 September2013
Accepted28September2013
Availableonline9October2013
Keywords:
Monetary policy
VAR estimation
Fed Fundsfutures
FOMC
a b s t r a c t
The evidencesuggeststhatmonetarypolicypost1988becamemoreforward-looking,
invalidatingtheidentifyingassumptionsinconventionalmethodsofmeasuringmonetary
policy'seffects,leadingtospuriousandunlikelyresultsforthisperiod.Weproposea
new identificationschemethatusesfactorsextractedfromFedFundsfuturestomeasure
exogenouschangesinpolicy.UsingthisshockseriesinaVAR,werecoverthecontrac-
tionaryeffectofmonetarytighteningonoutput.Moreover,wefindthatasmuchashalf
of thevariabilityinoutputwasdrivenbymonetarypolicyshocks,andthatthereisamild
pricepuzzle.
& 2013ElsevierB.V.Allrightsreserved.
1. Introduction
Identifying theimpactofmonetarypolicyontheeconomyisacentralquestioninempiricalmacroeconomics.Thekey
identification problemissimultaneity.Hence,thefocushasbeenontheexogenousor ‘shock’ component ofpolicychanges.
For theU.S.,aconsensushasemergedonthequalitativeeffectsofamonetarypolicyshock. Christiano etal.(1999)
summarize thisconsensusasfollows:
Afteracontractionarymonetarypolicyshock,shortterminterestratesrise,aggregateoutput,employment,profits
and variousmonetaryaggregatesfall,theaggregatepricelevelrespondsveryslowly,andvariousmeasuresofwages
fall, albeitbyverymodestamounts.Inaddition,thereisagreementthatmonetarypolicyshocksaccountforonly
a verymodestpercentageofthevolatilityofaggregateoutput;theyaccountforevenlessofthemovementsinthe
aggregatepricelevel.
However,thisconsensusissensitivetotheperiodusedforanalysis.Inparticular,itisdependentontheinclusionofthe
1970sandearly1980s,whenshockswerelargeandthepolicymakingenvironmentwasdifferentfromtheonefacedtoday.
When oneattemptstoidentifytheeffectsofmonetarypolicyshocksfortheperiodsincethe1980susingthesame
methodologies oneobtainsquitedifferentresults.Notably,contractionarymonetarypolicyshocksappeartohaveasmall
positiveeffectonoutput.
ThispaperpresentssomeevidenceonchangestothenatureofU.S.monetarypolicyshocksthatwouldcauseconventional
identificationmethodstogivemisleading results.Inparticular,weshowthatU.S. monetarypolicyhasbecomemoreforward
looking. Hence,VARidentificationmethods thatignoretheroleofforecastsinthepolicymaker'sreactionfunctionaremis-
specified.Identificationmethods(suchas RomerandRomer,2004) thatallowforforward-lookingvariablesinthereaction
functionbutdonotallowfortheapparentincreaseintheirrelativeweightwilltendtosufferfromthesameproblem.
Contents listsavailableat ScienceDirect
journal homepage: www.elsevier.com/locate/jme
JournalofMonetaryEconomics
0304-3932/$-seefrontmatter & 2013ElsevierB.V.Allrightsreserved.
http://dx.doi.org/10.1016/j.jmoneco.2013.09.006
n Correspondingauthor.Tel.: +442070710924;fax: +442070710950.
E-mail addresses: c.w.crowe@gmail.com, ccrowe@capulaglobal.com(C.Crowe).
Journal ofMonetaryEconomics60(2013)950–966
2. Weturntofinancialmarketdatainanefforttouncoverameasureofmonetarypolicyshocksthatislesssubjecttothese
criticisms. Following Kuttner(2001), Gürkaynaketal.(2005) and Piazzesi andSwanson(2008) monetary policyshocksare
identified asthe ‘surprise’ component ofmonetarypolicyactions,estimatedusingmovementsinFedFundsfuturescontract
prices onthedayofmonetarypolicyannouncementsfollowingFOMCmeetings.
Factoranalysisisemployedtoefficientlycapturetheinformationcontainedacrossthematurityspectrum,uncovering
the commoninformationfromsixmonthlycontracts:thecurrentmonthandupto5monthsahead.Asin Gürkaynaketal.
(2005) two factorsaresufficienttosummarizetheinformationacrossthesixcontracts.Moreover,inkeepingwiththe
literatureonfactormodelsoftheyieldcurve(e.g. Piazzesi, 2010), thefactorshaveanaturalinterpretationaslevelandslope,
respectively.Theformerisemployedasthemeasureofthepolicyshock.
WeenterthisnewshockmeasureinasimplemonthlyVAR,similarlyto RomerandRomer(2004), estimatedfor
1988:12-2008:06.1 With thisnewmeasure,acontractionarymonetarypolicyshockhasastatisticallysignificantnegative
effect onoutput.Whiletheeffectissmallinabsoluteterms,theforecasterrorvariancedecompositionsuggeststhat,inan
era oflowoveralloutputvolatility,ournewpolicyshockmeasurecanaccountforuptohalfofoutputvolatilityatahorizon
of 3yearsormore—aroundtwicetheproportionusingexistingshockmeasures.Thereissomeevidencefora ‘price puzzle’:
contractionarymonetarypolicyalsoleadstoasmall,andborderlinesignificant,increaseinthegeneralpricelevelata
horizon of1–3 years,althoughthisissubsequentlyreversed.Effortstoeliminatethepricepuzzlebyincludingameasureof
commodity pricesorinflationexpectationsintheVAR,followingsuggestionsintheliterature,arenotsuccessful.
1.1.Therelatedliterature
Our methodologybuildsontheinsightsofanincreasinglyinfluentialliteratureonidentifyingmonetarypolicyshocks
using financialmarketdata. Rudebusch(1998) is anearlypaperadvocatingtheuseofFedFundsfuturesdata,while
Kuttner's(2001) focus onone-daychangesinfuturesprices,ratherthanthedifferencebetweentheimpliedfuturesrateand
the actualpolicyrate,allowsforsharperidentification. Faustetal.(2004) proposeanoveltwo-stageidentificationscheme
in whichtheinformationavailablefromtheFedFundsfuturesisusedtopartiallyidentifyastructuralVAR. Gürkaynaketal.
(2005) use atwofactormodeltocombineinformationfromfuturescontractsatdifferenthorizonsandseparatelyidentify
levelandslopefactors. Hamilton(2008) deriveslevel,slopeandcurvaturefactorsusingthreeFedFundsfuturescontracts,
and estimatestheimpactofthedifferentfactorsonhousingmarketvariables. Thapar (2008) uses 3monthTreasury
Bill futurespricesasaproxyformarketexpectations,inanovelidentificationmethodthatcombinesthesemarket-based
forecastswithGreenbookforecastsofoutputandpricevariables. D'Amico andFarka(2011) uses intradayfuturesdatato
estimatethecontemporaneousrelationbetweenmonetarypolicyandstockpriceswithinaVARframework. Taylor(2010)
carries outaslightlydifferentexercise,usingintradayFedFundsfuturesdatatoidentifytheeffectofmacroeconomicdata
announcements onmarketexpectationsoffuturemonetarypolicychanges.
While thispaperisthereforenotthefirsttoturntoFedFundsfuturesdata,thispaperisthefirsttouseshocksextracted
from futurescontractstoidentifytheresponsesofoutputandinflationtomonetarypolicyshocks.Earliercontributionshave
focused ontheimpactofmonetarypolicyonfinancialratherthanmacrovariables.Because,inourcase,thepolicyshock
is identifiedoutsidetheVAR,onecanavoidsomeoftheweaknessesofstructuralVARestimation.Bycontrast, Faustetal.
(2004) use thestructuralVARmodeltoidentifythemonetarypolicyshockandtoestimatetheimpulseresponsesofthe
macro variablestothepolicyshock,andasaresulttheirmethodissubjecttosomeoftheseweaknesses.Like Kuttner(2001)
and Hamilton (2008), butunlike Rudebusch(1998) and Thapar (2008), thispaperfocusesondailyinnovationsinFed
Fundsfuturesprices.Usingdailydatafrompolicyannouncementdayshelpstoremovetheimpactofothernews(suchas
economic datareleases)andmorecleanlyidentifiestheimpactofexogenouspolicyshocks.Moreover,as Kuttner(2001) has
argued, focusingoninnovationstothefuturespricehelpstostripouttheimpactoffluctuationsintermandriskpremia.
This paperalsocontributestoasmallerliteratureontheinstabilityovertimeofidentifiedimpulseresponsesfromVARs.
Boivin andGiannoni(2006) testforinstabilityinasmallstructuralVAR,andfindevidenceforastructuralbreak. Owyang
and Wall(2009) estimateaggregateandregionalVARsandfindthattheestimatedimpactofmonetarypolicyonoutput
is significantlylowerintheVolcker–Greenspanperiodthanearlier.Bothpapersarguethattheapparentchangeinthe
impact ofmonetarypolicyshocksisarealone,reflectingfundamentalchangesinthetransmissionmechanism.Boivinand
Giannoni arguethatthekeychangeisastrongerFedresponsetoinflationexpectations.OwyangandWallattributethe
changeinresponsivenesstochangesinthepropagationmechanismformonetarypolicy.Bycontrast,ouranalysissuggests
that althoughthereductionintheestimatedimpactreflectsarealchangeinbehavior(forecastsplayingagreaterroleinthe
Fed'sdecision-making),thekeychangeistotheestimatedeffectratherthantheactualeffect,becauseidentification
problemsbecomemorepronouncedwhentheFed'spolicybecomesmoreforwardlooking.
Thenextsectionbrieflyreviewstheliteratureonidentifyingmonetary policyshocksandtheireffects.Itfocusesinparticularon
four identificationschemesthathavereceivedsignificantattention: Christianoetal.'s(1996) recursiveVARidentification; Sims and
Zha's (2006) non-recursiveVAR; Bernanke andMihov's(1998) over-identifiedVAR;and RomerandRomer's(2004) narrative
1 Because theFedFundsfuturesmarketonlystartedtradinginOctober1988,weareunabletoderiveourshockmeasurefortheearlyportionofthe
“great moderation”. However,theresultsfortheotheridentificationstrategieswefollowin Section 2 are broadlythesamewhethertheestimationstartsin
1982,1984or1988.
S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966 951
3. identification.Wecontrastthebaselineresultsintheoriginalpapers withresultsfortherecentperiod(focusingonthepost-1988
periodtoallowacomparisonwithournewmeasure),andthenanalyzehowthenatureofmonetarypolicyshockshaschanged
since theearly1980saspolicyhasbecomemoreforward-looking. Section3 discussestheFedFundsfuturesmarketandoutlines
the newshockmeasure. Section 4 usesthenewmeasuretoestimatetheeffectsofmonetarypolicyshocksinthepost-1988period,
discussestheresultsandoutlinessomerobustnesschecks. Section5 concludes.
2. Thefailureofconventionalidentification schemes
Following Christiano etal.(1999), weidentifyamonetarypolicyshockastheorthogonaldisturbanceterm st in an
equationoftheform
St ¼ f ðΩtÞþst ð1Þ
where St denotesthemonetarystance(ormorenarrowly,theinstrumentofthemonetaryauthority,e.g.theFedFundsrate)
and f is alinearfunctionrelating St to thepolicymaker'sinformationset Ωt.2 The remainderofthissectionillustrateshow
some existingmethodsofidentifyingequation (1) appear tohavebrokendown,andarguethatthismayreflectafailureto
sufficientlycontrolfortheintroductionofmoreforward-lookinginformationinto Ωt overthelasttwotothreedecades.
The fourschemesweconsiderare Christianoetal.'s(1996) recursiveVARapproach, Bernanke andMihov's(1998) over-
identifiedVAR, Sims andZah's(2006) non-recursiveVARand RomerandRomer's(2004) narrativeapproach.Thefulldetailsof
these approachesandoureffortstoreplicatethemaredetailedinthe appendix. Thissectionprovidesabriefoverviewoftheresults.
Estimatedovertheiroriginalsampleperiods—fromthe1960stothemid-1990s—all fourapproachessuggestthatmonetary
policyshockshaveaneffectinlinewiththeconventionalwisdom:amonetarycontractionlowersoutputandotherrealindi-
catorsovertheshorttomediumterm,andhasamoremutedimpact—generallynegative—on inflation.However,estimatingthe
models overthemorerecentperiodyieldsverydifferentresults.Mostworryingly,monetarycontractionsareestimatedtohavea
stimulativeeffectonoutput.
2.1.Fouridentificationschemes
Christiano etal.(1996) estimateaquarterlyVARwithsixvariablesandfourlagsovertheperiod1960Q1–1992Q4.
Their resultsshowthatacontractionaryshockisassociatedwithapersistentdeclineinoutput.Thepriceindexresponds
slowlybuteventuallydeclines.WereplicatetheirresultsandreporttheimpulseresponsesofGDPandtheGDPdeflator
to acontractionarymonetarypolicyshock(Fig. 1 panel a).3 However,whenthesamemodelisestimatedfortherecent
period (1988Q4–2007Q3),neitheroutputnorpricesshowtheexpectedresponse(Fig. 1 panel b).4
Bernanke andMihov(1998) developamonthlymodelinwhichcontemporaneousidentificationrestrictionsareimposed
on monetaryvariablesinordertomodeltheFed'soperatingprocedure.Asinallthemonthlyestimatesinthispaper,output
is measuredbyindustrialproduction. Fig. 2 (panel a)showsthatintheoriginalperiodtheresponsesofindustrialproduction
and pricesareasexpected,andverysimilartoChristiano,EichenbaumandEvans's.5 However,whenthismodelisestimated
for thelaterperiod(panelb),againneitheroutputnorpricesshowtheexpectedresponse.Bothoutputandpricesincrease
significantly—immediatelyinthecaseofoutput,andoverthemediumterminthecaseofprices.6
SimsandZha(2006) includeawidersetofvariables.Wereplicatetheirfindingsfortheiroriginalsample(Fig.3, panela).7
Resultsaresimilartothoseobtainedfrom Christiano etal.'s(1996) recursiveidentificationscheme(theresultsarenot
significant duetothewidestandarderrorbandsobtainedunderthebootstrapalgorithm).However,whenthemodelis
2 Hence Eq. (1) can bethoughtofasthemonetaryauthorities'feedbackruleorpolicyreactionfunction,althoughas Christiano etal.(1999) highlight,
there arepitfallsinidentifyingthecoefficientsin f ðÞ.
3 In thispaperthesizeofthemonetarypolicyshockisalwaysequaltoonestandarddeviationandimpulseresponsesarealwaysreportedwithtwo
standard errorbands.Standarderrorsshownin Figs. 1–4 are obtainedviamultivariatenormalparametricbootstrapping,basedon500replications.
4 Weendoursamplein2007Q3becausenonborrowedreserves(NBR)becomenegativeduringthefourthquarterof2007.Oursampleisalso
truncated (at2007:11)fortheBernankeandMihovestimationforthesamereason.Allestimationfortherecentperiodstartsatend-1988.Thisisbecause
the FedFundsfuturesdatathatwerelyonforidentifyingtheimpactofshocksin Section 4 are onlyavailablefromthisperiodonwards,andwewantto
ensure thattheresultsarecomparableacrossmethods.However,thefindingthatthepreviousmethodologiesappeartobreakdownforthelaterperiod is
robust tostartingthesamplein1982or1984,asalreadynoted.
5 Bernanke andMihovestimatedifferentversionsofthemodel,includingfourthatareover-identifiedandonethatisjust-identified.Wereplicatethe
over-identifiedmodel(FederalFundsratetargetingmodel)sinceBernankeandMihovfindthatthisperformsbestforthepost-1988period.
6 Although thispaperre-estimatesthesameVAR,i.e.amonthlyVARwith13lagsandsixvariables(output,domesticprices,commodityprices,the
FederalFundsrate,totalreservesandNBR),therearesomeminordifferencesbetweenourVARandBernankeandMihov's.TheyinterpolateGDPandthe
GDP Deflatortoconvertaquarterlyseriestoamonthlyseries,whilethispaperusesmonthlyIndustrialProductionandCPIdatainstead.Thispaperalso
employsadifferentcommoditypriceindex.Thesedifferencesareminor,andcomparingtheimpulseresponsesfromtheoriginalperiodsuggeststhatthey
havenosignificanteffectontheresults.
7 Due todataconstraints,thispaperexcludestheirbankruptcymeasurefromtheVAR.Theimpulseresponsesofourmodelestimatedfortheoriginal
period arealmostidenticaltothosein Sims andZha(2006). InfactSimsandZhamentionthatthemeasureofbankruptcymakes “only amodest
contribution” to theresults,while Christiano etal.(1999) also re-estimatetheSimsandZhamodelexcludingthebankruptcymeasure.Havingsaidthis,our
confidence intervalsaresomewhatwiderthanthosereportedbySimsandZha:thisispartlycosmetic(theyreport68%,orapproximatelyonestandard
error,CIs,whereaswereporttwostandarderrorCIs);itmayalsoreflecttheexclusionofthebankruptcymeasureinourestimates,orpossiblydifferences
in thebootstrapalgorithms.
952 S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966
8. targeting theFederalFundsrate.Thethreesubsamplesaretherefore1969:1–1979:10(pre-Volcker);1979:11–1982:10
(nonborrowedreservestargeting);and1982:11–2008:06(FedFundsratetargeting).9
Our firststepistoanalyzethestabilityoftheregressioncoefficientsviaaseriesofChowtestscomparingeachsetof
adjoining subsamples.Thereisclearevidenceofastructuralbreakatbothpotentialbreakpoints.ThissuggeststhatRomer
and Romer'sreactionfunction,thatassumesconstantcoefficientsacrossthewholesample,couldbemisspecified.10 These
resultsareinlinewiththoseof BoivinandGiannoni(2006), whoundertakeasimilarexerciseforasmallstructuralVAR
similar tothesystemsdiscussedin Section 2, andfindstrongevidenceforastructuralbreak.Hence,theVARidentification
methods discussedabove—which likeRomerandRomer'smethodassumetime-invariantcoefficientsinthepolicyreaction
function inordertoidentifymonetarypolicyshocks—are likelytosufferfromverysimilarproblems.
The secondstepistotestwhetherspecificelementsof Ωt havechanged.Thefocusisontwosetsofvariables:theeight
forward-lookingvariables(1-and2-quarteraheadforecasts)andninebackwards-lookingvariables(currentandlastquarter
estimates)includedinRomerandRomer'sspecification,comparingthepost-1988periodwiththerestofthesample.
TableA2intheonlineappendixpresents F testsofthejointsignificanceofthevariablesforthetwosubsamples.
Policymaking appearstobeunambiguouslyforward-lookinginthepost-1988period,butonecannotrejectthenull
hypothesisofnoforward-lookingvariablesin Ωt during thepre-1988period.ThisfindingcorroboratesotheranalysesofFed
policymaking overtheperiod(Orphanides, 2003; Boivin andGiannoni,2006).
These resultsshedsomelightonthefindingspresentedin Section 2. Failuretoallowforstructuralbreaks—under allfour
methods ofidentification—will tendtogivebiasedestimatesoftheshocksthemselves,andhencebiasedestimatesofthe
impact oftheshockonothermacroeconomicvariables.Forinstance,byincreasingthemeasurementerrorassociatedwith
the RomerandRomershockseries,itwillleadtoattenuation(biastowardzero)intheshocks'estimatedmacroeconomic
impact.
The factthatpolicymakingappearstohavebecomemoreforwardlookinginrecentyearshasparticularlyserious
implications fortheVARidentificationmethods,sincethesedonotincludeanyforward-lookingelementsin Ωt. IfFed
policymakers reacttoanexpectedincreaseinoutputgrowthabovetheeconomy'spotentialbytighteningmonetarypolicy
to partiallyoffsetit,thenamonetarycontractionwillappeartocausehighergrowthiftheseanticipatorymovementsare
not explicitlyallowedfor.Sinceanticipatorymovementsappeartohavebecomemoreimportantfortherecentperiodthan
earlier,thismightexplainwhyVARidentificationmethodsidentifytheexpectedcontractionaryimpactofmonetary
tighteningfortheearlierperiod,butforthelaterperiodgeneratethecounterintuitiveexpansionaryeffectsshownin
Section 2. AlthoughRomerandRomer'smethodologyattemptstocontrolforanticipatorymovements,byimposingequal
coefficients throughoutthesampleitmaynotadequatelycapturethestrongereffectsintherecentperiod.
3. AnewFedFundsfutures-basedshockmeasure
Conventionalmethodsofidentifyingmonetarypolicyshocks—which requiretheestimationof(1)withsuitableproxies
for Ωt—will performbadlyifeither Ωt or f ðÞ aremisspecified.Analternativeapproachistousefinancialmarketdatato
obtain theprivatesector'scontemporaneousbeliefsabout f ðΩt Þ at thetimeofeachmeeting,andusetheseasaproxyforthe
true reactionfunctionanditselements.Thiscircumventstheneedtoestimate f ðΩt Þ directly,andthereforedoesnotrequire
that weimposerestrictionsonthevariablesin Ωt or thefunctionalform f ðÞ.
3.1.Overview
Toillustratethisapproachingeneralterms,assumethattherearetwomeasuresoftheprivatesector'sexpectationfor
the policystance St for aparticularpolicymeeting:oneintheimmediaterun-uptothemeeting, t1
bS
t , andoneimmediately
aftertheannouncementofthepolicystancedecidedatthemeeting, t
bS
t . Eachisanoisymeasureoftheprivatesector'strue
expectation:
t1
bS
t
¼ EPt
1
½Stþξt1 ¼ EPt
1
½f ðΩt Þþξt1 ð2Þ
t
bS
t
¼ EPt
½Stþξt ¼ Stþξt ð3Þ
where theprivatesector'sactualexpectationsattime τ of thestanceattime t aredenotedby EPτ
½St . Thenoise ξ can arise
from severalsources,includingtime-varyingriskpremiaaswellasmeasurementorroundingerrors.Wemakethefollowing
twoidentifyingassumptions:
EPt
1
½f ðΩt Þf ðΩtÞ ¼ 0 ð4Þ
9 Bagliano andFavero(1998) identify fiveregimes.Weextendthelastperiodfrom1996:3andstartthefirstperiodin1969:1ratherthan1966:1,
reflecting thecoverageoftheoriginalRomerandRomerseries.Wealsocombinetheirfirsttwoandlasttwoperiods,aswedonotfindthedistinction
meaningful ineithercase.
10 See TableA1intheonlineappendix,whichalsopresentsatestofastructuralbreakattheendof1988,matchingthesubsamplewithavailableFed
Funds futuresdata,thatsuggeststhatoursampleisbroadlyrepresentativeoftheFedFundsratetargetingperiodasawhole.
S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966 957
9. ξtξt1 ¼ 0 ð5Þ
The firstassumption(4)statesthattheprivatesector'sbeliefspriortotheannouncementabouttheFed'sinformationsetare
correct.11 The secondassumption(5)statesthatthenoisetermisunchangedaroundthetimeofthepolicyannouncement.
Then, subtracting(2)from(3)yields
t
bS
t
t1
bS
t
¼ st ð6Þ
This impliesthatasuitableproxyfortheshock, st, isgivenbythechangeinthemeasureoftheprivatesector'sbeliefsabout
the policystancearoundthetimeofapolicyannouncement, t
bS
t
t1
bS
t .
3.2. Fedfundsfuturesdata
Our measuresoftheprivatesector'sbeliefsaboutthepolicystancebS
t arederivedfromFedFundsfuturescontracts.The
FederalFundsfuturesmarketwasestablishedattheChicagoBoardofTrade(CBOT)inOctober1988(see Söderström, 2001,
Kuttner,2001 and Faustetal.,2004 for furtherinformation).Thepriceofacontractformonth mþh (i.e. atahorizon h from
the currentmonth m) isabetonthemonthlyaverageeffectiveFedFundsrateinmonth mþh (denoted re
mþh). Notethatthe
averagetargetFedFundsrate(rmþh) mightdifferfromtheeffectiverateduetotargetingerrorsonthepartoftheFed:
re
mþh
¼ rmþhþɛmþh ð7Þ
These errorsaretypicallysmallandmeanzero.Foragivencontractprice pd
h on day d in month m, thefuturesrate fd
h is
simplygivenby1phd
. Thenstandardno-arbitrageconditionsimplythatthefuturesrateisequaltotheaverageeffectiveFed
Fundsrateinmonth mþh, Edre
mþh, plusarisk(orhedgingorterm)premium ρd
h:
fhd
¼ Edre
mþh
þρhd
ð8Þ
Assuming thattheriskpremium ρd
h remainsconstantandthatthereisalsonochangeintheexpectedaveragetargeting
error Ed½ɛmþh, thenthechangeintheexpectedtargetrateduringsubsequentcalendarmonths(hZ1) followingapolicy
announcement onday d of month m is givenby
ΔEdrmþh ¼ fhd
fhd
1
ð9Þ
while thechangefortheremainderofthecurrentmonth(whoselengthis M days)isgivenby
ΔEdrm ¼ M
Md
f0d
f0d
1
ð10Þ
The innovationtotheexpectedtargetrateinagivenmonththenservesasagoodproxyfortheunderlyingmonetary
policy shock st under fourassumptions.First,theaveragetargetrate rmþh should becorrelatedwiththepolicystance St.
If thisholdsthen fhd
fhd
1 providesanestimateof t
bS
t
t1
bS
t , whilethenoiseterm ξt is givenbythesumoftherisk
premium ρd
h, theexpectedFedtargetingerror Ed½ɛmþh as wellasdataerrors.Second,thereshouldbenopredictablechanges
in thenoisetermsthatmakeup ξt, e.g.duetopredictableeffectsofpolicyannouncementsonriskpremia:thisisanecessary
condition for(5)tohold.Third,thereshouldbenoother ‘news’ that mightaffecttheexpectedfuturesrate(suchasmacro-
economic dataannouncementsthatmighthaveimplicationsforratechangesinthefuture)duringthe24-hourperiod
associatedwiththepolicydecision.Last,thepolicyannouncementitselfshouldnotrevealinformationabouttheFed's
privateinformationset Ωt or itsreactionfunction f ðÞ. Theselasttwoassumptionsarenecessaryfor(4)tohold.12 Assuming
that theseassumptionsarevalid,thenthepolicy ‘surprise’ is agoodmeasureoftheshock.Theevidence,discussedin
Section 3.4, providesstrongsupportforthefirstthreeassumptions,whileevidenceonthefourthismoremixed.
Following Kuttner(2001), theimpactofpolicyannouncements(ornon-announcements)followingFOMCmeetingsis
estimatedbycomparingtheendofdaypriceonthedayfollowingthe(last)dayofthemeetingwiththatonthemeetingday
for meetingsoccurringbeforeFebruary1994,andcomparingthepriceonthedayofthemeetingwiththatonthedaybefore
the meetingforsubsequentmeetings.OuranalysisfocusesonlyonFOMCmeetingdates,ratherthanonalldatesthatthe
Fed announcedchangestothetargetFedFundsrate,includinginter-meetingchanges.13
11 These assumptionsarestatedintheirstrongestformtoclarifytheexposition.Aweakerassumptionwouldbethat,conditionalontherealizationof
Ωt and st, (4) and (5) hold inexpectations.Amoreseriousproblem—simultaneity bias—will ariseif (4) and (5) do notholdevenintheirweaker,conditional
expectations,form,e.g.becausetheprivatesectormakessystematicerrorsinforecastingtheFed'spolicyreactionfunction.Thisissueisaddressed inmore
detail laterinthepaper.
12 For instance,anegativemacroeconomicnewsrelease(onethattendstorevisedownoutputandinflationexpectations)thatoccurredconcurrently
with apolicyannouncementwouldimplylowerratesinthefuture,implyingthat(4)iscontradicted.Similarly,ifapolicyannouncementprovidesnew
information abouttheFed'sinformationset,e.g.sothataratecutsignalsthattheFedexpectsarecession,thentheprivatesector'sbeliefspriorto the
announcement wereincorrectandagain(4)doesnothold.
13 Like Faustetal.(2004), whoalsofocusonregularpolicyannouncements,webelievethatintermeetingchangesaremorelikelytobeassociatedwith
the simultaneousreleaseofmacroeconomicinformationratherthanreflectingexogenousshockstopolicy.
958 S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966
10. 3.3. Constructingtheshockseries
The simplestsignalofthepolicystance St is thefuturesrateforthecurrentmonth,fd
0. However,wearguethatthereare
severalreasonstofocusonarangeofmaturities.First,combiningtheinformationfromseveralsources—essentially takinga
sample meanoftheshockmeasuresobtainedfromcontractsatdifferenthorizons—should helptominimizetheeffectof
noise inaspecificcontract.Thisaveragingmaybeparticularlyimportantsincetheriskpremiumislikelytobemorevolatile
at shorterhorizons(asshowninthedataappendix,themarketforthecurrentmonthcontractisnotthemostliquid,and
intra-monthtradingvolumesareinfactparticularlyvolatileforthiscontract,whichcouldleadtoamorevolatileliquidity
premiumandhenceintroducemorenoiseintotheshockmeasure).Moreover,sincetheFed'spolicydecisionsarerelatively
persistentovertime,apolicychangeinthecurrentperiodwillbereflectedinhigherexpectedratesseveralmonthsahead,
so thatfuturescontractssettlingseveralmonthsinthefuturewillalsocontaininformationaboutthecurrentshock.Indeed,
shockswhichareexpectedtobepermanentmightbeexpectedtohaveagreaterimpactontheeconomy.Butsomeshocks
to currentratesmighthavelittleimpactonlongertermexpectations(forinstance,iftheshockwastotheimmediatetiming
of theratechangeratherthantothelong-termdirectionofrates,as Gürkaynak,2005 argues). Hence,ameasureofshocks
that combinestheinnovationstoratesinthecurrent(spot)monthwiththoseanticipatedinthefutureislikelyabetter
measure oftheoverallpolicystance.Whilecontractsarenowavailableformorethanayearintothefuture,longer-dated
contractshavenotbeenavailableforthewholeperiodandevennowaretypicallyrelativelyilliquid.Hence,wefocuson
contractsforthecurrentmonthandupto5monthsahead.
In ordertocombinetheinformationavailableintheestimatedforecastinnovationsatallsixhorizons,onecanestimatea
simple factormodelviamaximumlikelihood.Denotingthevectorofinnovationsatthesixhorizons(normalisedtohave
mean zeroandvarianceofone)as s, thevectoroffactorsas ϕ, thefactorloadingmatrixas Λ and thevectorofuniquefactors
as e, thefactormodelisgivenby
s ¼ ϕΛ′þe ð11Þ
This methodofextractingthecommonshocksinthecontractpricesatdifferenthorizonshasseveraladvantages.While
one isprincipallyinterestedinextractingthecommonlevelsshock(whichcapturesunexpectedpolicytighteningor
loosening), becausemorethanonefactorisextractedonewecanalsopotentiallyanalyzeshockstothetermstructure.
The methodiswellsuitedtothedata,whichincludesfuturescontractswithdifferinglevelsofliquidityandhencevolatility.
The pricesofsomecontractswilltendtocomovemorethanothers,reflectingaloweruniquevariancefortheseseries.
The factormodelallowsonetocapturethisexplicitly,puttingmoreweightonthoseseriesthatexhibitagreaterdegree
of comovementinextractingthefactors.Atthesametime,thismethodisrelativelysimpleanddoesnotrequireusto
formulateandestimateafullyspecifiedmodelofthetermstructure.Twofactorsadequatelycapturetheinformationinthe
futures shocks.14 The twofactorssummarizethenewinformationonthemediumtermevolutionofpolicyratesthatis
revealedbythepolicyrateannouncement.Indeed,thefactorsturnouttohaveanintuitiveinterpretation.Thefirstfactor,
which ishighlypositivelycorrelatedwithalltheindividualinnovations,canbethoughtofasalevelseffect:thatportionof
the newinformationrelatedtothepolicyannouncementthatcausesverticalshiftsintheexpectedmedium-termtrajectory
for policyrates.Sincethetransmissionofmonetarypolicyisgenerallythoughttooccurviatheimpactofshortratechanges
on longerterm(real)rates,itisthisportionofthenewinformationonratesthatcorrespondsmostcloselytotherelevant
policy shock.Wethereforeusethisfactorasourmeasureoftheunderlyingpolicyshock.
The secondfactor,whosecorrelationwiththeindividualinnovationseriesatdifferentmaturitiesdecreasesmono-
tonicallyfrompositivetonegativeasthematurityincreases,canbethoughtofasaslopeoryieldcurveeffect:thatportion
of thenewinformationrelatingtothepolicyannouncementthatleadstodifferentialeffectsonexpectedpolicyratesinthe
near termandfurtherout.Whilethisfactorcapturesanimportantportionofthenewsrelatingtopolicyannouncements,it
does notcapturethenotionofapolicyshockthatisthefocusofthecurrentpaper.
3.4. Assessingtheshockseries
Our newshockseriesispresentedin Fig. 5. Ourfactor-basedshockmeasurehasameanof0andastandarddeviationof1
by construction.Toaidinterpretation,in Fig. 5 it isscaledtobeaweightedaverageofthedeviationsfromthemeanofthe
six underlyingmonthly “shock” series. Twostandarddeviationbarsareshown,andthe27June2001meetingisindicated
by averticalbartoaidthediscussionin Section 3.5.
The validityofourshockmeasuredependsonthevalidityoftheunderlyingassumptions.Thefirstassumption,that
the FedFundstargetrateattherelevanthorizons(0–5 months)iscorrelatedwiththe ‘true’ monetary stance,seems
uncontroversial. Bernanke andMihov(1998) havedemonstratedthataFedFundstargetingmodelbestdescribesmonetary
policy inthepost-1988period,whileitisintuitivethat,inaneconomywithforward-lookingagentsmakingirreversible
14 Estimating aprincipalfactormodelwithuptosixfactors,thefirstfactoraccountsfor92%ofthetotalvariance,thesecondfactorforafurther9%,and
the thirdfactorfor0.4%.Theeigenvaluesofthefirstthreefactorsare5.2,0.52and0.02,respectively(thelastthreefactorshavenegativeeigenvalues and
make acumulativecontributiontothevarianceof 1%). Hence,amodelwithtwofactorsappearstoadequatelyandparsimoniouslycapturethemain
patterns ofcorrelationinthedata,anditisthisparsimoniousspecificationthatisthenestimatedviaMaximumLikelihood.TablesA3andA4intheonline
appendix presentfurtherdetailsofthefactormodelandtheestimatedshockseries.
S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966 959
11. economic decisions,theoverallstanceofpolicydependsnotonlyonthecurrenttargetratebutalsoontheratesexpectedin
the immediatefuture.Withrespecttothesecondassumption—that thereshouldbenopredictableinnovationstothenoise
component oftheprivatesector'sexpectationsaboutthepolicystanceintheshortrun—Piazzesi andSwanson(2008) show
that anticipatedchangestoriskpremiaintheFedFundsfuturesmarketoccurmainlyatbusinesscyclefrequency.With
respecttothethirdassumption—that otherinformationthatcouldbeconflatedwiththepolicyannouncementandbias
our resultsisnotreleasedonthesameday—Gürkaynaketal.(2005) show thatsomeFOMCmeetingandintermeeting
dates associatedwithpolicyannouncementscoincidewithmacroeconomicdatareleases.However,theyshowthatonly
EmploymentReport releaseshaveanyindependenteffectonFedFundsfutures. Bernanke andKuttner(2005) identify ten
observations,allbefore1994,forwhich EmploymentReport releasescoincidewithpolicyannouncementsorFOMCmeetings.
But ourdecisiontofocusonlyonFOMCmeetingshelpstoalleviatethisproblem,sinceonlythreeofthesedatescoincide
with FOMCmeetings(theotherscoincidewithintermeetingchanges).15 Weprovidesomeempiricalevidencethatthe
inclusion ofthesedatesisnotdrivingourresultsintherobustnesschecksin Section 4.2.
Totestthefourthassumption,onecanregressour(scaled)shockmeasureonthedifferencebetweentheFed's
Greenbookforecastsandhigh-qualityprivatesector(Blue Chip) forecastsforthe17variablesusedin RomerandRomer's
(2004) estimatedreactionfunction,wherethisdifferenceisusedasaproxyfortheFed'sinternalinformation.Sincethe
Greenbookforecastsareonlymadepublicwitha5-yearlag,theshockmeasureshouldonlybecorrelatedwiththeFed's
internalinformationtotheextentthatthelatterisrevealedindirectlybythepolicyrate,theannouncementandanyrelated
communication. Asweshowin Table1, thejointhypothesisofzerocoefficientsonall17variablescannotberejectedatthe
10%level.ThissuggeststhatourshockmeasureshouldberelativelyuncorrelatedwiththeFed'sexclusiveinformation,and
simultaneity biasshouldthereforenotbeasignificantproblem.
However,aninspectionofthecoefficientestimatesin Table1 points toevidencethatourshockmeasuremaybe
contaminatedbytheimpactoftheFedtighteningpolicyinresponsetoneartermoutputandpricepressures,sinceour
shockmeasurerespondspositivelytocurrentquarteroutputandinflationforecasts.Weinvestigatefurthertheimplications
of thisforourresultsin Section 4.3.
Toillustratehowourshockmeasurecomparestoothersintheliterature, Table2 presentscorrelationcoefficientsforour
shockmeasure(New),thechangeinthetargetFederalFundsrate(ΔFF) andRomerandRomer'sshockmeasure(RR; all
on aper-meetingbasis,for157meetings);thefinalrowpresentscorrelationcoefficientsbetweentheper-quarteraverage
of thesethreemeasuresandthemonetarypolicyshockobtainedfromaCholeskydecompositionofChristiano,Eichenbaum
and Evans'squarterlyVARspecification(CEE), for76quarterlyobservations(1988Q4–2007Q3).Ournewshockmeasureis
positivelyandsignificantlycorrelatedwithallthreemeasures(atleastatthe10%level).
3.5. Ournewshockseries:anillustrativeobservation
Our shockmeasure,althoughcorrelatedwithexistingmeasures,candiffersignificantlyfromtheseforsomeobserva-
tions. Thesedifferencescanhelpillustratesomeoftherelativestrengthsofourapproach.Forinstance,theFOMCdecided
at its26-27June2001meetingona25basispointsreductionintheFedFundsrate.Thecutfollowedfivesuccessive50basis
point cuts(threeatthethreeprecedingmeetingsandtwocutsbetweenmeetings),aspartofarate-cuttingcyclethat
sawtheFedFundsratefallfrom6.5%to1.75%overthecourseoftheyear.WhiletheVARandnarrativeidentification
Fig. 5. Newshockseries,inbasispoints.Tomakeitcomparableinsizetothe6underlyingshocks,thefirstfactor(SD¼1 byconstruction)isdividedbythe
sum ofthe6coefficientsfromthefactormodel.Twostandarderrorbandsshownbyhorizontallines;verticallineidentifiestheJune2001FOMCmeeting
discussed in Section 3.5.
15 The threedatesinquestionare7July1989and2July1992(thedayafterthemeeting),and4February1994(thedayofthemeeting).
960 S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966
12. methods identifyanegativepolicyshock,itisclearfromreadingtheFed'sstatementsaswellasfrommarketreactionthat
the Fed'sinterestratecutswerelargelyanendogenousresponsetotheeconomicslowdowninthewakeoftheburstingtech
bubble andconcernsthattheeconomywassettoslowfurther.
By comparison,ourshockmeasureislargeandpositive(almost2standarddeviationbandsabovezero,or10basispoints
when suitablyscaled).Theintuitionforthisisthatmarketparticipantswereanticipatinganother50basispointcutinrates.
Reuters(June28)reportsthat “the markethadpricedintheprospectfor50basispoints.” The smallercuttherefore
representedapositiveshocktoFedFundsrateexpectations.Marketreactiontothemovesupportsourinterpretationofthe
June 27ratecutasapolicytightening.Reuters(June27)reportsthat “the dollarclimbedtoa10-weekhighontheyenon
Thursday,helpedbyaraftoffactors,includingthe...ratecut.” The dollaralsogainedgroundagainsttheeuro.Meanwhile,
bond yieldsrosesignificantly(particularlyfortwo-yeargovernmentpaper).Thesereactionsaremoreconsistentwitha
contractionarythananexpansionarymonetarypolicyshock.
Table1
Regressionresultsand F-test statisticsforpolicyshockmeasureandGreenbookvariables.
Variable Coefficient
Unemployment0 4.26
Output Growth1 1.31
Output Growth0 2.37nnn
Output Growth1 0.783
Output Growth2 1.19
GDP Deflator1 0.92
GDP Deflator0 2.34nn
GDP Deflator1 1.49
GDP Deflator2 0.323
ΔOutputGrowth1 0.541
ΔOutputGrowth0
1.14
ΔOutputGrowth1 0.803
ΔOutputGrowth2
1.44
ΔGDP Deflator1 0.300
ΔGDP Deflator0 1.31
ΔGDP Deflator1 0.117
ΔGDP Deflator2 1.22
Constant 0.610
R2 0.185
F(17) 1.50
p-value 0.132
The dependentvariableisthescaledshockmeasureinbasispoints;theindependentvariablesarethedifferencebetween
the GreenbookandBlueChipforecastsforthe17variablesidentifiedbyRomerandRomer(variablesareestimatesforthe
previousorcurrentquarterorforecastsoneortwoquartersahead,exceptforvariablesdenoted “Δ” which arethechange
in theforecastfromthepreviousmeeting;allvariablesarethendifferencedbetweentheGreenbookandBlueChip
consensus forecasts).Theregressionisrunover113FOMCmeetingsbetween1988and2002.The F-test statisticshownis
for thejointnullhypothesisthatthecoefficientonall17variablesiszero.Standarderrorsarerobusttoheteroskedasticity
(but areomittedfromthetableforbrevity). n10%levelofsignificance.
nn 5% levelofsignificance.
nnn 1% levelofsignificance.
Table2
Correlation betweenshockmeasures.
New ΔFF RR CEE
New1
ΔFF 0.39nnn 1
RR 0.23nnn 0.73nnn 1
CEE 0.22n 0.26nn 0.09 1
Correlation coefficientsforournewshockmeasure(New) andexistingmeasures:thechangeinFedFundsRate
(ΔFF), RomerandRomer'snarrativemeasure(RR), andChristiano,EichenbaumandEvans'smeasure(CEE;
based onCholeskydecompositionofVARresiduals).Coefficientsinrows1–3 basedon157per-meetingvalues;
coefficients inlastrowbasedon76quarterlyvalues.
n 10%levelofsignificance.
nn 5% levelofsignificance.
nnn 1% levelofsignificance.
S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966 961
13. 4. Identifyingtheeffectofmonetarypolicyshocks
FollowingRomerandRomer,weidentifytheeffectofmonetarypolicyshocksusingasmall3variablemonthlyVAR(as
they do,welettheshocksseriestakeavalueofzeroformonthswithoutFOMCmeetings).16 The variablesareorderedso
that monetarypolicyisallowedtorespondto,butnotaffect,outputandinflationcontemporaneously.Weusethelogof
industrial productionasourmeasureofoutputandthelogconsumerpriceindexasourmeasureofprices.17 As withRomer
and Romer'sshockmeasure,ourmeasurecapturesunanticipated changes in policyrates.Hence,likeRomerandRomer,we
enter ourshockmeasurecumulatedintheVAR,sincehereitisthelevel,notthechange,inpolicythatistheappropriate
variable.18 The baselineVARincludes36monthlylags.However,theresultsarefullyrobusttoshorterlagspecificationsthat
match thekindoflagstructureintheotherVARresultscitedin Section 2 and makefewerdemandsonthedatagiven
the relativelyshortsampleavailable.Resultsfor6,12and24months,whicharealmostidenticaltothebaselineimpulse
responses,arediscussedin Section 4.2.
4.1.Baselineresults
Impulse responsefunctionsareshownin Fig. 6. Weshowa95%confidenceintervalestimatedusingasystembootstrap
of theVARandfactormodel(todealwiththegeneratedregressorproblem).Afteralmostoneyear,acontractionary
monetary policyshockshowsasustainednegativeeffectonoutputthathasitsmaximumimpactatahorizonofaroundtwo
years.Theoutputresponseisverysimilartothebaselineresultsfortheearlierperiodreviewedin Section 2 (although with
greaterpersistence),butverydifferentfromtheresultsobtainedforthe1988–2008periodusingthesamemethodologies.
The responseofpricestoamonetarycontractionismoreproblematic.Theeffectbecomessignificantlynegativeonly
afterfouryears;thepositiveresponseoverthemediumterm,althoughsmall,contrastswiththenegativeeffectthat
has generallybeenfoundintheliterature. CastelnuovoandSurico(2006), like Hanson (2004), findevidencethattheprice
puzzle islimitedtothepre-1979period,arguingthatthisisduetoequilibriumindeterminacywhenmonetarypolicy
responds weaklytoinflationexpectations,andthattheinclusionofavariablecapturingthepersistenceofexpectedinflation
under indeterminacycaneliminatethepricepuzzle.However,ourbaselineresultssuggestevidenceforapricepuzzleeven
in thepost-Volckerperiod,whenthereactionofinterestratestoexpectedinflationshouldbesufficientlystrongto
guaranteeequilibriumdeterminacy.Otherstudies(e.g. Christiano etal.,1996) haveincludedameasureofcommodityprices
as ameansofeliminatingthepricepuzzle(althoughtheirargumentforincludingthisvariable,thatcommodityprices
help toforecastinflation,hasbeencriticizedby Hanson, 2004).19 In thefollowingsectionweaddaproxyforinflation
expectationsandacommoditypriceindextoourbaselineVARastwoofaseriesofrobustnesschecks;neitherhelpsto
resolvethepricepuzzle.However,thisapparentlyrobustfindingofasignificantpricepuzzleisconsistentwithotherrecent
workthatusesFedFundsfuturestoidentifypolicyshocks(Thapar,2008).
Receivedwisdomaboutthe “great moderation” period isthatlesspronouncedmonetarypolicyshockshelpedto
contributetothegeneralmoderatinginmacroeconomicvolatility.Inordertoshedsomelightonthisissue,weanalyzethe
percentageoftheforecasterrorvariancesofoutputandpriceswhichcanbeattributedtoourshockmeasureandtwoother
measuresovertherecentperiod,aFederalFundsrateshockandtheRomerandRomershock(Fig. 7).20 Resultsforourshock
measure areshownwithasolidline;resultsforFedFundsrateshock(dashedline)andRomerandRomershock(dotted
line) areshownforcomparison;twostandarderrorbandsforourshockmeasurearealsoshown.
The estimatedimpactofmonetarypolicyshocksonthevarianceofthepricelevelisbroadlysimilaracrossthethree
measures,althoughtheRomerandRomermethodidentifiesthelargesteffect,particularlyatlongerhorizons,whichis
intuitivegiventheimpulseresponseshownin Fig. 4. However,athorizonsofmorethantwoyearstheestimatedimpact
on outputvolatilityisconsiderablyhigherforourshockmeasure—around2timesashighaswitheitherofthealternative
measures.Infact,theresultsusingournewmeasuresuggestthatalmosthalfofforecasterrorvarianceathorizonsofaround
3 yearscanbeaccountedforbymonetarypolicyshocks.Hence,whilemonetarypolicyshocksmayhavemoderatedin
absoluteterms,theirrelativecontributiontooutputvolatilityinrecentyearsmayneedtobereassessed.
16 RomerandRomer'sbaselinespecificationemploysasingleequationapproach.Weapplythismethodologyasoneofaseriesofrobustnesschecksin
Section 4.2
17 This followsmuchoftheliterature,butdiffersfrom RomerandRomer(2004) who usethelogoftheproducerpriceindexforfinishedgoodsastheir
price measure.OurVARsalsoincludeanexogenoustimetrend.
18 An additionalrationaleforusingthecumulatedshockseries,whichisI(1)byconstruction,isthattheoutputandpriceseriesaregenerally
considered I(1);hence,iftheI(0)shockserieswereincludedtheVARwouldbestatisticallyunbalanced,leadingtononstationary,highlypersistent,
residuals. IncludingtheI(1)cumulatedseriesallowsforimplicitcointegrationbetweenthevariablesintheVAR.
19 Giordani (2004) argues thatthepricepuzzlearisesbecausetheVARsystem,byincludingoutputratherthantheoutputgap(whichentersin
theoretical models),ismisspecified.However,sinceourVARmodelincludesalineartimetrend,weareineffectdealingwithanoutputgapmeasure,
assuming that(log)potentialoutputfollowsalineartrend.Thisexplanationisthereforeunlikelytoaccountfortheestimatedpricepuzzleinourmodel.
20 In ordertomaketheresultscomparable,weestimateineachcaseasmallrecursiveVARincludingindustrialproduction,CPIandoneofthree
variables:theFederalFundsrate,theRomerandRomer(cumulated)shockmeasureandour(cumulated)shockmeasure.Thesampleperiodis1988:12–
2008:06.Thisapproachissimilartothatof RomerandRomer(2004), whoestimatethefirsttwoVARstocompareimpulseresponsesusingtheirshock
measure withthoseusingastandardrecursiveVARshockmeasure(withtheFedFundsrateasthemonetaryinstrument).However,weuseCPIasourprice
measure, whereasRomerandRomerusethePPIforfinishedgoods.
962 S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966
14. 4.2. Robustness
Wereporthereresultsforeightrobustnesschecksandonefurthercomparison.Thefirstchangestheorderinginour
baseline VAR,allowingourmonetarypolicyshocktohaveaninstantaneousimpactonoutputandprices.Impulseresponses
remainqualitativelyidentical,althoughthepricepuzzleismorepronounced.ThesecondusesRomerandRomer'sprice
measure (PPIforfinishedgoods).Again,theonly(modest)differenceisinthestrengthandpersistenceofthepricepuzzle.
The thirdmodifiesthelagstructuretoinclude6,12or24lagsratherthan36.Theestimatedimpulseresponsesare
essentiallyunchanged.ThefourthassessessubsamplestabilitybyestimatingthebaselineVAR(withlaglengthreducedto
12inlightoftheshortersample)fortwotruncatedtimeperiods,droppingpre-1990orpost-2001data.Resultsarequalita-
tivelyidenticaltothoseforthesampleasawhole.
As afifthrobustnesscheck,weincludeacommoditypriceindex,orderedfirstintherecursiveVAR.Asalreadydiscussed,
this hashelpedtoeliminatethepricepuzzleinsomestudies.However,thepricepuzzleremains,whiletheoutputresponse
to thepolicyshockisunchanged.Thesixthexerciseincludesameasureofinflationaryexpectationstotesttherobustnessof
the pricepuzzle.Following CastelnuovoandSurico(2006), weuseonequarteraheadexpectedinflationfromtheFed's
Greenbook(replacedbythecorresponding Blue Chip forecastfor2003onwards),andorderthisvariablefirstintherecursive
VAR. Thisexercisedoesnothelptoeliminatethepricepuzzleeither,andtheoutputresponseisalsounaffected.Theseventh
robustnesscheckassesseswhethertheinclusionofFOMCmeetingdatesthatcoincidewith EmploymentReport releasesis
critical totheresults,byincludingdummiesforthesemeetingdates.Theoutputresponsetothepolicyshockremainsthe
same asunderthebaseline.BecauseourshockmeasureisidentifiedoutsidetheVARitseemslikelythatourresultsare
robusttoothermodificationstotheVARframework.
Finally,weestimatesingle-equationsystemsforoutputandpricessimilartothoseestimatedby RomerandRomer
(2004). InkeepingwiththeVARresults,wefindanegativeandpersistenteffectonoutput(withapointestimateof
between1%and2%)andasmallpositiveeffectonthepricelevel(although,duetowideestimatedstandarderrorbands,
botheffectsareonlyattheborderofstatisticalsignificance).21
This sectioncloseswithafinalcomparisonexercise.Toshedsomelightonhowourfactor-derivedshockmeasure
compareswiththesimple Kuttner(2001) spot-monthshock,onecanestimatethebaselinemodelwiththe(cumulated)
spot-monthinnovationinplaceofourshockmeasure.Inthiscase,theimpulseresponseforoutputisclosertothatfor
the otheridentificationschemes,withasmall,albeitinsignificant,positiveoutputresponsetoa ‘contractionary’ policy
shock.Theseresultssupporttheviewthatshockstothespotmonthfuturescontractoftenreflectnewinformationabout
Fig. 6. StructuralVAR(Monthlydata,3endogenousvariablesplusconstantandlineartimetrend,36lags).Variablesorderedasindustrialproduction,
consumer prices,bothseasonallyadjustedandinlogs,andourshockmeasure,cumulated.GraphsshowresponseofindustrialproductionandCPItoaone
standard deviationpositiveshocktothepolicymeasure.StructuralshocksobtainedviaCholeskydecomposition.95%confidenceintervalsproduced by
bootstrappingthecombinedVARandfactormodelsystem(500replications).
21 Resultsofallrobustnesschecksareavailablefromtheauthorsonrequest.
S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966 963
15. the timing,ratherthanthegeneraldirection,ofpolicy.Hence,itisnotsurprisingthattheIRFsassociatedwiththisnoisy
shockmeasureareimpreciselymeasured.Aswiththeotheridentificationschemesdiscussedin Section 2, theapparently
perversesignoftheestimatedeffectofpolicyonoutputissuggestiveofsimultaneitybias,perhapsbecausetimingshocks
are particularlyassociatedwiththeFed'scommunicationofinternalinformation.
4.3. Decomposingourshockmeasure
Section 3.4 presentedevidencethatourshockmeasuremaybecontaminatedbytheFed'sreactiontoitsowninforma-
tion onnearterminflationarypressure. RomerandRomer's(2000) analysisofFedandprivatesectorforecastssuggeststhat
the Fed'sforecastsarelikelytoincludesomeaccurateexclusiveinformation.Toshedsomeadditionallightonthisissue,we
regresstheFed'sexclusiveinformation(thedifferencebetweentheFed'sforecastandtheprivatesectorforecast)on
the privatesector'soverallforecasterror(thedifferencebetweentheactualoutcomeandtheprivatesectorforecast),for
both realGDPandtheGDPdeflatorandatforecasthorizonsof0–2 quarters.TheR2s fromtheseregressionshavethe
interpretationoftheshareoftheFed'sinternalinformationthatturnsouttobecorrectexpost.Thissharevariesfrom1%to
6% forrealGDPandfrom3%to20%fortheGDPdeflator(resultsreportedintheappendix,TableA6).Thispointstopotential
positivebiasinourestimateoftheeffectofpolicyonoutputandinflation.NotethoughthattheFed'saccurateinformation
accounts forarelativelysmallshareofthedifferencebetweenitsforecastandtheprivatesector's,suggestingthatthebias
is small.
Toprovidesomeadditionalevidenceonthelikelyimpactofthisbiasonourresults,wedecomposeourshockmeasure
using theresultsoftheregressionoftheshockontheFed'sinternalinformationpresentedin Table1. Theresidualsfrom
this equationgiveanestimateofthe ‘pure’ shockcomponent,whilethefittedvaluesgiveanestimateofanyremaining
portion ofthesystematiccomponent f ðΩt Þ. However,simultaneitybiasisnottheonlylikelysourceofbiasintheresults.
Attenuationbias(biastowardszero)duetomeasurementerrorisalsolikelytobepresent.Whiletheresidualshouldbe
cleansed ofsimultaneityproblems,if f ðΩt Þ is correctlyspecifiedthenthefittedvaluewillbecleansedofmeasurementerror
(it willallbecapturedbytheresidualterm).Whenthetwodecomposedshockmeasuresareenteredinthebaseline
VAR system(Fig. 8), boththe ‘predicted’ portion oftheshock(bottompanels)andtheresidualportion(toppanels)have
a significantnegativeeffect—of strikinglysimilarmagnitude—on output.Thissuggeststhat,foroutput,thelikelybias
resultingfromthenewsabouttheFed'sowninformationsetbeingincludedinourshockmeasureisofaroundthesame
order ofmagnitudeasthebiasduetomeasurementerror,wherethislatterbiasislikelytobesmall.Moreover,sinceboth
sourcesofbiasshouldtendtodrivetheestimatedeffecttowardszerothetrueeffectislikelysomewhatlarger.Notethatthe
fitted portionofthe “shock” measure accountsfor17%ofoutputvariationata3yearhorizon,whiletheresidualportion
Fig. 7. StructuralVAR(Monthlydata,3endogenousvariablesplusconstantandlineartimetrend,36lags).Variablesorderedasindustrialproduction,
consumer prices,bothseasonallyadjustedandinlogs,andoneofthreepolicymeasures:ourshockmeasure;RomerandRomer'smeasure(both
cumulated); andtheFederalFundsrate.GraphsshowCholeskyFEVDs:thepercentageoftheforecasterrorforoutputandCPIaccountedforbyeachpolicy
measure. TheFEVDforourshockmeasureisshowninbold,withtwostandarderrorbandsproducedbybootstrappingthecombinedVARandfactormodel
system.FEVDsfortheFedFundsrate(dashedline)andRomerandRomershock(dottedline)areshownforcomparison(SEbandsnotshown).
964 S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966
16. accounts for30%,reflectingthefactthatmostofthevariationinourshockmeasurecannotbeaccountedforbytheFed's
internalinformation.
5. Conclusion
ConventionalVARandnon-VARidentificationschemesforestimatingtheeffectofU.S.monetarypolicyshocksonthe
wider economyaresensitivetothesampleperiodunderconsideration.Inparticular,theseschemesgenerateunrealistic
impulse responsefunctionsforoutput,andtoalesserextentprices,forthequartercenturystartinginthemid-1980sknown
as the “great moderation”. TheseapparentlyperverseresultsmaybegeneratedbyafailuretoproperlyidentifytheFed's
reactionfunctiontoallowforchangesinitsparametersovertime,particularlyagreaterweightplacedonforward-looking
variables.
This paperoutlinesanewmeasureofmonetarypolicyshocksderivedfromFedFundsfuturescontractsthatislessprone
to theseproblems.Asaresult,ournewmeasuregeneratesamorerealisticimpulseresponsefunctionforoutput,with
a smallbutstatisticallysignificantnegativeeffectwhosemaximumimpactisfeltatahorizonoftwoyearsfollowing
a monetarycontraction.Thereisalsoevidenceofa “price puzzle” overthemediumterm.Almosthalfofoutputvariability
(at a3yearhorizon)canbeexplainedbymonetarypolicyshocksusingournewidentificationstrategy,twicetheshare
under otheridentificationschemesforthesameperiod.
While ourshockmeasuremaybecontaminatedbytheFed'ssystematicpolicyreactiontoitsinternalforecasts,this
is likelytobiasourestimatedimpulseresponsestowardszero,sothattheestimatedoutputresponsemayrepresentan
underestimate.Moreover,whilethissimultaneitybiasappearstobesmallunderouridentificationscheme,itislikelytobe
more importantforVAR-basedidentificationmethods.
One canrationalizethehighshareofoutputvolatilityaccountedforbyourshockmeasurebyacombinationof
substantiveandeconometricfactors.Substantively,theFedexercisedmoreeffectivecontrolovertheeconomyduringthe
‘great moderation’ period coveredinouranalysis,partlyviaanimprovedfocusonforward-lookingindicators,helpingto
minimize theimpactofexogenousdemandshockssothatagreatershareoftheremainingshocksisaccountedforbypolicy
itself. Althoughtheabsoluteeffectoftheshocksissmall,theirrelativeimpactislargeinaperiodofrelativelylowoverall
volatility.Inaddition,ourshockmeasurecapturesonlypolicychangesthatweretrulyunanticipatedbytheprivatesector,
and itistheseunexpectedmonetarypolicychangesthataregenerallybelievedtohavethelargestimpactonoutput.
Additionaleconometricfactorsincludethefactthatourshockvariableisnotapuremeasureofshocksbutalsoincludesthe
Fed'ssystematicresponsetoitsinternalforecasts.WhiletheinclusionoftheFed'sresponsetoexclusiveinformationwill
tend toreducethemagnitudeoftheestimatedcoefficients,itmayincreasetheoveralleffectbyincreasingthesizeofthe
Fig. 8. StructuralVAR(Monthlydata,4endogenousvariablesplusconstantandlineartimetrend,12lags).Datasample1988:12–2002:12.Variables
orderedasindustrialproduction,consumerprices,bothseasonallyadjustedandinlogs,andthepredictedandresidualcomponentsoftheregression ofour
shock measureontheFed'sprivateinformationdescribedinthetext,bothcumulated.GraphsshowresponseofindustrialproductionandCPItoaone
standard deviationpositiveshocktoeachpolicymeasure.StructuralshocksobtainedviaCholeskydecomposition.TwoStandardErrorbandsproduced by
bootstrappingthecombinedVARandfactorsystem(500replications).
S.M. Barakchian,C.Crowe/JournalofMonetaryEconomics60(2013)950–966 965