Micro Effects of Macro Announcements RealTime Price Di - PDF document

MicroEffectsofMacroAnnouncements:Real-TimePriceDiscoveryinForeignExchangeByTORBENG.ANDERSEN,TIMBOLLERSLEV,FRANCISX.DIEBOLD,ANDCLARAVEGA*Usinganewdatasetconsistingofsixyearsofreal-timeexchange-ratequotations,macroeconomicexpectations,andmacroeconomicrealizations,wecharacterizetheconditionalmeansofU.S.dollarspotexchangerates.Inparticular,weŽndthatannouncementsurprisesproduceconditionalmeanjumps;hencehigh-frequencyexchange-ratedynamicsarelinkedtofundamentals.Thedetailsofthelinkageareintriguingandincludeannouncementtimingandsigneffects.Thesigneffectreferstothefactthatthemarketreactstonewsinanasymmetricfashion:badnewshasgreaterimpactthangoodnews,whichwerelatetorecenttheoreticalworkoninformationprocessingandpricediscovery.(JELF3,F4,G1,C5)Howisnewsaboutfundamentalsincorpo-ratedintoassetprices?Thetopicconfrontedbythisquestion—characterizationofthepricedis-coveryprocess—isofbasicimportancetoallofŽnancialeconomics.Unfortunately,itisalsooneoftheleastwell-understoodissues.Indeed,someinuentialempiricalstudieshavegonesofarastosuggestthatforsomeassets—notablyforeignexchange—pricesandfundamentalsarelargelydisconnected.1Inthispaperweprovideanempiricalexam-inationofpricediscoveryinthechallengingcontextofforeignexchange.Usinganewlyconstructeddatasetconsistingofsixyearsofreal-timeexchange-ratequotations,macroeco-nomicexpectations,andmacroeconomicreal-izations(announcements),wecharacterizetheconditionalmeansofU.S.dollarspotexchangeratesforGermanMark,BritishPound,JapaneseYen,SwissFranc,andtheEuro.Inparticular,weshowthatannouncementsurprises(thatis,thedifferencebetweenexpectationsandrealiza-tions,or“news”)produceconditionalmeanjumps,andweprovideadetailedanalysisofthespeedandpatternofadjustment.Weshowthatconditionalmeanadjustmentsofexchangeratestonewsoccurquickly,effec-tivelyamountingto“jumps,”incontrasttocon-ditionalvarianceadjustments,whicharemuchmoregradual,andthatanannouncement’sim-pactdependsonitstimingrelativetootherrelatedannouncements,andonwhetherthean-nouncementtimeisknowninadvance.WeŽnd,*Andersen:DepartmentofFinance,KelloggSchool,NorthwesternUniversity,andNBER(e-mail:t-andersen@kellogg.nwu.edu);Bollerslev:DepartmentofEconomicsandFinance,DukeUniversity,andNBER(e-mail:boller@econ.duke.edu);Diebold:DepartmentofEconom-ics,Finance,andStatistics,UniversityofPennsylvania,andNBER(e-mail:fdiebold@sas.upenn.edu);Vega:GraduateGroupinEconomics,UniversityofPennsylvania(e-mail:cvega@ssc.upenn.edu).ThisworkwassupportedbytheNationalScienceFoundationandtheWhartonFinancialInstitutionsCenter.WearegratefultoOlsenandAssociatesformakingavailabletheirreal-timeexchange-ratequota-tionsdata,andtoMoneyMarketServicesInternationalformakingavailabletheirnewsannouncementexpectationsdata.Forusefulcommentswethankthreereferees,aswellasRicardoCabellero,DeanCroushore,KathrynDominguez,BernardDumas,MartinEvans,MichaelFleming,JeffFrankel,LindaGoldberg,KenKavajecz,RichLyons,NelsonMark,FrankSchorfheide,NickSouleles,AllanTimmermann,MarkWatson,IngridWerner,andJonathanWright,andparticipantsattheCAFConferenceinDenmarkonMarketMicrostructureandHigh-FrequencyDatainFi-nance,theUniversityofWisconsinConferenceonEmpiri-calModelsofExchangeRates,theFederalReserveBankofPhiladelphiaConferenceonReal-TimeDataAnalysis,theNBERInternationalFinanceandMacroeconomicsProgramMeeting,andseminarsattheUniversityofPennsylvaniaandtheUniversityofHouston.1TheclassicstatementisofcourseRichardA.MeeseandKennethRogoff(1983).Foragoodsurveyofthesubsequentempiricalexchange-rateliteraturethroughtheearly1990’s,seeJeffreyA.FrankelandAndrewK.Rose(1995).Inlaterwork,NelsonC.Mark(1995)andMarkandDonggyuSul(2001)Žndthatfundamentalsmatterinthelongrunbutnotintheshortrun.MartinD.D.EvansandRichardK.Lyons(2002)Žndthatorderowmattersintheshortrunbutfailtolinkorderowtofundamentals.38 moreover,thattheadjustmentresponsepatternischaracterizedbyasigneffect:badnewshasgreaterimpactthangoodnews.Finally,were-lateourresultstorecenttheoreticalandempir-icalworkonassetreturnvolatilityanditsassociationwithinformationprocessingandpricediscovery.Thepaperrelatestoearlierworkinintriguingways,butatleastthreefeaturesdifferentiateourŽndingsfrompreviousresultsalongimportantdimensions.Theseincludeourfocusonforeignexchangemarkets,ourfocusonconditionalmeanasopposedtoconditionalvariance,dy-namicsandthelengthandbreadthofoursampleofexchange-rateandannouncementdata.Letusdiscussthembrieyinturn.First,wefocusonforeignexchangemarketsasopposedtostockorbondmarkets,andweaddressthecentralopenissueinexchange-rateeconomics—thelinkbetweenexchangeratesandfundamentals.Itiscomforting,however,thatanumberofrecentpapersfocusinglargelyonbondmarketsreachconclusionssimilartoours.PierluigiBalduzzietal.(2001),forexam-ple,examinetheeffectsofeconomicnewsonpricesintheU.S.interdealergovernmentbondmarket,Žndingstrongnewseffectsandquickincorporationofnewsintobondprices,whileMichaelJ.FlemingandEliM.Remolona(1997,1999)showthatthelargestbondpricemovementsstemfromthearrivalofnewsannouncements.2Second,wefocusprimarilyonexchange-rateconditionalmeansasopposedtoconditionalvariances.Thatis,wefocusprimarilyonthedeterminationofexchangeratesthemselves,asopposedtotheirvolatility.Wemaintainthisfocusbothbecausetheconditionalmeanisofintrinsicinterest,andbecausehigh-frequencydiscrete-timevolatilitycannotbeextractedac-curatelyunlesstheconditionalmeanismodeledadequately.HenceourworkdiffersinimportantrespectsfromthatofLouisH.EderingtonandJaeHaLee(1993),RichardPayne(1996),AndersenandBollerslev(1998),andBollerslevetal.(2000),forexample,whoexaminecalen-darandnewseffectsinhigh-frequencyasset-returnvolatilitybutdonotconsidertheeffectsofnewsonreturnsthemselves.Third,weuseanewdatasetwhichspansacomparativelylongtimeperiod,andincludesabroadsetofexchangeratesandmacroeconomicindicators.Notwithstandingtheimprovementsobtainedthroughtheaboveconsideration,ourresultsarequiteconsistentwithpriorrelatedwork.Indeed,severalstudieshavelinkedmacroeconomicnewsannouncementstojumpsinexchangerates,andourŽndingsmaybeviewedaspro-vidingconŽrmationandelaboration.CharlesGoodhartetal.(1993),forexample,examineoneyearofhigh-frequencyDollar/Poundex-changeratesandtwospeciŽcnewsevents—aU.S.tradeŽgureannouncementandaU.K.in-terest-ratechange—andconcludeineachcasethatthenewscausedanexchange-ratejump.Similarly,AlvaroAlmeidaetal.(1998)intheirstudyofthreeyearsofhigh-frequencyDM/Dollarexchangeratesandalargersetofnewsannouncements,documentsystematicshort-livednewseffects.Finally,KathrynM.Dominguez(1999)arguesthatmostlargeexchange-ratechangesoccurwithintensecondsofamacroeco-nomicnewsannouncement,andthatclosetimingofcentralbankinterventionstonewsannounce-mentsincreasestheireffectiveness.Weproceedasfollows.InSectionIwede-scribeourhigh-frequencyexchange-rateandmacroeconomicexpectationsandannounce-mentsdata.InSectionIIwecharacterizethespeedandpatternofexchange-rateadjustmenttomacroeconomicnews,andwedocument,amongotherthings,thesigneffects(i.e.,alargerexchange-rateresponsetobadthangoodnews).InSectionIIIwerelatethesigneffectstorecenttheoriesofinformationprocessingandpricediscovery.WeconcludeinSectionIV.I.Real-TimeExchangeRates,ExpectedFundamentals,andAnnouncedFundamentalsThroughoutthepaperweusedataonexchange-ratereturnsinconjunctionwithdataonexpecta-tionsandannouncementsofmacroeconomicfundamentals.Thedataarenovelinseveralre-spects,suchasthesimultaneoushighfrequencyandlongcalendarspanoftheexchange-ratere-turns,aswellasthereal-timenatureoftheexpectationsandannouncementsoffundamen-tals.Herewedescribetheminsomedetail.2Also,inconcurrentrelatedworkforT-bondfutures,NikolausHautschandDieterHess(2001)reporthighlysigniŽcant,butshort-lived,priceandvolatilityimpactsinresponsetonewandrevisedemploymentŽgures.39VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS A.Exchange-RateDataTheraw5-minuteCHF/$,DM/$,Euro/$,Pound/$,andYen/$returnserieswereobtainedfromOlsenandAssociates.Thefullsampleconsistsofcontinuouslyrecorded5-minutere-turnsfromJanuary3,1992throughDecember30,1998,or2,189days,foratotalof2,189z2885630,432high-frequencyforeignex-change(FX)returnobservations.AsinUlrichA.Mu¨lleretal.(1990)andMichelM.Dacorognaetal.(1993),weusealloftheinterbankquotesthatappearedontheReutersscreenduringthesampleperiodtoconstructour5-minutereturns.Eachquoteconsistsofabidandanaskpricetogetherwitha“timestamp”tothenearestsecond.AfterŽlteringthedataforoutliersandotheranomalies,weobtaintheaveragelogpriceateach5-minutemarkbylinearlyinterpolatingtheaverageofthelogbidandthelogaskatthetwoclosestticks.Wethenconstructcontinuouslycom-poundedreturnsasthechangeinthese5-minuteaveragelogbidandaskprices.Goodhartetal.(1996)andJonDanielssonandPayne(1999)Žndthatthebasiccharacteristicsof5-minuteFXreturnsconstructedfromquotescloselymatchthosecalculatedfromtransac-tionprices(whicharenotgenerallyavailablefortheforeignexchangemarket).Itiswellknownthattheactivityinthefor-eignexchangemarketslowsdecidedlyduringweekendsandcertainholidaynontradingperi-ods;seeMu¨lleretal.(1990).Hence,asisstan-dardintheliterature,weexplicitlyexcludedanumberofdaysfromtheraw5-minutereturnseries.Wheneverwedidso,wealwayscutfrom21:05GMTthenightbeforeto21:00GMTthatevening.ThisparticulardeŽnitionofa“day”wasmotivatedbytheebbandowinthedailyFXactivitypatternsdocumentedinBollerslevandIanDomowitz(1993)andkeepsthedailyperiodicityintact.Inadditiontothethinweek-endtradingperiodfromFriday21:05GMTuntilSunday21:00GMT,weremovedseveralŽxedholidays,includingChristmas(December24–26),NewYear’s(December31–January2),andJulyFourth.Wealsocutthemovingholi-daysofGoodFriday,EasterMonday,MemorialDay,JulyFourth(whenitfallsofŽciallyonJuly3),andLaborDay,aswellasThanksgivingandthedayafter.Althoughourcutsdonotaccountforalloftheholidaymarketslow-downs,theycapturethemostimportantdailycalendareffects.Finally,wedeletedsomeofthereturnscon-taminatedbybrieflapsesintheReutersdatafeed.Thisproblem,whichoccursalmostexclu-sivelyduringtheearliestpartofthesample,manifestsitselfassequencesofzeroorconstant5-minutereturnsinplaceswheremissingquoteshadbeeninterpolated.Toremedythis,wesim-plyremovedfromeachexchange-rateseriesthedayscontainingthe15longestzeroandconstantruns.BecauseoftheoverlapamongsetsofdaysdeŽnedbythiscriterion,weactuallyremovedonly25days.Intheendweareleftwith1,724daysofdata,containingT51,724z2885496,512high-frequency5-minutereturnobservations.Standarddescriptivestatisticsrevealthatthe5-minutereturnshavemeansthatarenegligibleanddwarfedbythestandarddeviations,andthattheyareapproximatelysymmetricbutdistinctlynon-Gaussian,duetoexcesskurtosis.Ljung-Boxstatisticsindicateserialcorrelationinbothreturnsandabsolutereturns.Toassesstheeconomicrelevanceofthetem-poraldependenciesinthereturnseries,weturntotheautocorrelationsincolumnoneofFigure1.Therawreturnsdisplaytiny,butneverthelesssta-tisticallysigniŽcant,serialcorrelationattheveryshortestlags,presumablyduetomicrostructureeffects.However,theshort-lagreturnserialcorre-lationisnegligiblerelativetothestrongserialcorrelationinabsolutereturns,shownincolumntwoofFigure1.Thesampleautocorrelationsofabsolutereturnsdisplayveryslowdecayandpro-nounceddiurnalvariation,inlinewiththeresultsofDacorognaetal.(1993)andAndersenandBollerslev(1998).Interestingly,notonlytheshapesbutalsotheamplitudesofthediurnalpat-ternsinabsolutereturnautocorrelationsdifferno-ticeablyacrosscurrencies.B.ExpectedFundamentals,AnnouncedFundamentals,andNewsWeusetheInternationalMoneyMarketSer-vices(MMS)real-timedataonexpectedandrealized(“announced”)macroeconomicfunda-mentals,deŽning“news”asthedifferencebe-tweenexpectationsandrealizations.Everyweeksince1977,MMShasconductedaFridaytelephonesurveyofabout40moneymanagers,collectedforecastsofallindicatorstobere-40THEAMERICANECONOMICREVIEWMARCH2003 FIGURE1.SAMPLEAUTOCORRELATIONFUNCTIONSRETURNSANDABSOLUTERETURNSNotes:WeplotthesampleautocorrelationsofreturnsandabsolutereturnsforŽvecurrencies,togetherwithBartlett’sapproximate95-percentconŽdencebandsunderthenullhypothesisofwhitenoise.Ineachgraph,theverticalaxisisthesampleautocorrelationandthehorizontalaxisisdisplacementin5-minuteintervals.Toavoidcontaminationfromshiftsinandoutofdaylightsavingtime,wecalculatethesampleautocorrelationsusingonlydayscorrespondingtoU.S.daylightsavingtime.41VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS leasedduringthenextweek,andreportedthemedianforecastsfromthesurvey.Numerousinuentialstudies,fromearlyworksuchasThomasJ.UrichandPaulWachtel(1984)throughrecentworksuchasBalduzzietal.(2001),haveveriŽedthattheMMSexpecta-tionscontainvaluableinformationabouttheforecastedvariable,andinmostcasesareunbi-asedandlessvariablethanthoseproducedfromextrapolativebenchmarkssuchasARMAmodels.Table1providesabriefdescriptionofsalientaspectsofU.S.andGermaneconomicnewsannouncements.Weshowthetotalnumberofobservationsinournewssample,theagencyreportingeachannouncement,andthetimeoftheannouncementrelease.NotethatU.S.an-nouncementstimesareknowninadvance,whereasonlythedayfortheGermanannounce-mentsareknowninadvancebuttheirtimingwithinthedayisvariableandunknownapriori.ThetargetFedfundsratedeservesspecialmention.TheFederalOpenMarketCommit-tee’s(FOMC)announcementofthefederalfundsratetarget,althoughlikelyproducingim-portantnews,isnonstandardandhenceisnottypicallyexamined.ItisnonstandardbecausepriortoFebruary1994itwasnotannounced;instead,theFOMCsignaledthetargetrate,butdidnotstateitexplicitly,throughopenmarketoperationsperformedfrom11:30to11:35A.M.EasterntimeonthedayoftheFOMCmeeting.InFebruary1994,theFOMCbegantoan-nouncechangesinthetargetrateonmeetingdays,albeitatirregulartimes,andfrom1995onwarditannouncedthetargetrateonmeetingdaysregularlyat2:15P.M.Easterntime,asdescribedinKennethN.Kuttner(2001).ToassesstheeffectsofFOMCnews,weneedtoknowannouncementdaysandtimes,aswellasthemarket’sexpectedFedfundsratetargetandtheannounced(orsignaled)value.Determinationofannouncementdaysandtimesisrelativelystraightforward.Wecollectedtheirregular1994announcementtimesfromReuters.3Before1994weusean11:30A.M.Easternannouncementtime,andafterMarch,1995weusea2:15P.M.announcementtime.4Determinationofmarketex-pectationsissimilarlystraightforward:weuseMMSsurveydataonexpectedfederalfundratetargetsfromJanuary1992toDecember1998.5Theannouncementsthemselvesaretrickiertoconstruct,duetothepre-1994FOMCsecrecy;weusetheannouncementdataconstructedbyMichaelW.Brandtetal.(2001),kindlypro-videdbyKennethKavajecz.InFigure2weshowthepatternofU.S.releasedatesthroughoutthemonth.6Thisisofpotentialimportance,becausethereissomere-dundancyacrossindicators.Forexample,con-sumerandproducerpriceindexes,althoughofcoursenotthesame,areneverthelessrelated,andFigure2revealsthattheproducerpriceindexisreleasedearlierinthemonth.Henceonemightconjecturethatproducerpricenewswouldexplainmoreexchange-ratereturnvari-ationthanconsumerpricenews,asthetypicalamountofconsumerpricenewsrevealedmayberelativelysmallgiventheproducerpricenewsrevealedearlierinthemonth.Becauseunitsofmeasurementdifferacrosseconomicvariables,wefollowBalduzzietal.(2001)inusingstandardizednews.Thatis,wedividethesurprisebyitssamplestandarddevi-ationtofacilitateinterpretation.Thestandard-izednewsassociatedwithindicatorkattimetisSkt5Akt2Ekt sˆk,whereAktistheannouncedvalueofindicatork,EktisthemarketexpectedvalueofindicatorkasdistilledintheMMSmedianforecast,andsˆkisthesamplestandarddeviationofAkt2Ekt.Useofstandardizednewsfacilitatesmeaningful3Theannouncementtimeswere11:05A.M.Easterntimeon02/04/94,2:20P.M.on03/22/94and07/06/94,2:30P.M.on11/15/94,2:26P.M.on05/17/94,2:23P.M.on12/20/94,1:17P.M.on08/16/94,2:22P.M.on09/27/94,2:24P.M.on02/01/95.4TheFOMCcanalsosurprisethemarketbychangingtheFedfundstargetbetweenFOMCmeetings.Becausethisdoesnothappenofteninoursample(5outof62times)andwedonothavetheexacttimingofsuchpolicychanges,wedonotaccountforthem.Similarly,datalimitationspreventusfrominvestigatingtheeffectofFedopenmarketopera-tions;seeCampbellR.HarveyandRogerD.Huang(2002)forarecentanalysisinvolvingtheearlier1982–1988timeperiod.5OnecouldalsoattempttoinferexpectationsfromFedfundsfuturesprices,asinGlennD.Rudebusch(1998)andKuttner(2001).6ThedesignoftheŽgurefollowsChart2ofFlemingandRemolona(1997).42THEAMERICANECONOMICREVIEWMARCH2003 TABLE1—U.S.ANDGERMANNEWSANNOUNCEMENTS AnnouncementNumberofobservationsaSourcebDatescAnnouncementtimedU.S.AnnouncementsQuarterlyAnnouncements1.GDPadvance47BEA05/22/87–10/30/988:30A.M.2.GDPpreliminary46BEA06/17/87–12/23/988:30A.M.3.GDPŽnal47BEA01/22/87–11/24/988:30A.M.MonthlyAnnouncementsRealActivity4.Nonfarmpayrollemployment144BLS12/05/86–12/04/98e8:30A.M.5.Retailsales145BC12/11/86–12/11/988:30A.M.6.Industrialproduction145FRB12/15/86–12/16/989:15A.M.7.Capacityutilization145FRB12/15/86–12/16/989:15A.M.8.Personalincome142BEA12/18/86–12/24/98f10:00/8:30A.M.g9.Consumercredit129FRB04/04/88–12/07/983:00P.M.hConsumption10.Personalconsumptionexpenditures143BEA12/18/86–12/24/98i10:00/8:30A.M.j11.Newhomesales117BC03/02/89–12/02/9810:00A.M.Investment12.Durablegoodsorders143BC12/23/86–12/23/98k8:30/9:00/10:00A.M.l13.Constructionspending128BC04/01/88–12/01/98m10:00A.M.14.Factoryorders127BC03/30/88–12/04/98n10:00A.M.15.Businessinventories129BC04/14/88–12/15/9810:00/8:30A.M.oGovernmentPurchases16.GovernmentbudgetdeŽcit124FMS04/21/88–12/21/98p2:00P.M.NetExports17.Tradebalance128BEA04/14/88–12/17/988:30A.M.Prices18.Producerpriceindex145BLS12/12/86–12/11/988:30A.M.19.Consumerpriceindex145BLS12/19/86–12/15/988:30A.M.Forward-looking20.ConsumerconŽdenceindex90CB07/30/91–12/29/9810:00A.M.21.NAPMindex107NAPM02/01/90–10/01/9810:00A.M.22.Housingstarts145BC12/30/86–12/30/988:30A.M.23.Indexofleadingindicators145CB12/30/86–12/30/988:30A.M.Six-WeekAnnouncementsFOMC24.Targetfederalfundsrate62FRB2/5/92–12/22/982:15P.M.qWeeklyAnnouncements25.Initialunemploymentclaims384ETA07/18/91–12/31/988:30A.M.26.Moneysupply,M1628FRB12/04/86–12/31/984:30P.M.27.Moneysupply,M2563FRB03/03/88–12/31/984:30P.M.28.Moneysupply,M3563FRB03/03/88–12/31/984:30P.M.GermanAnnouncementsrQuarterlyAnnouncements29.GDP24GFSO03/09/93–12/03/98VariesMonthlyAnnouncementsRealActivity30.Employment59FLO04/06/93–12/08/98Varies31.Retailsales59GFSO04/14/93–12/10/98Varies32.Industrialproduction63GFSO05/04/93–12/07/98Varies43VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS comparisonsofresponsesofdifferentexchangeratestodifferentpiecesofnews.Operationally,weestimatetheresponsesbyregressingassetreturnsonnews;becausesˆkisconstantforanyindicatork,thestandardizationaffectsneitherthestatisticalsigniŽcanceofresponseestimatesnortheŽtoftheregressions.Beforeproceeding,wepausetodiscussingreaterdetailthepossibilitythattheMMSfore-castsmaynotcaptureallinformationavailableimmediatelybeforetheannouncement.SurelyinformationdoesnotstopowingbetweenthetimethattheMMSforecastisproducedandthetimethatthemacroeconomicindicatorisreal-ized;hencetheMMSforecastsmaybe“stale.”Justhowstaletheyare,however,isanempiricalmatter.ThisissuehasbeeninvestigatedalreadyinthecontextofnewseffectsoninterestratesbyBalduzzietal.(1998),whoregresstheactualannouncement,Ai,onthemedianforecastoftheMMSsurvey,Fi,andthechangeinthe(veryannouncement-sensitive)ten-yearnoteyieldfromthetimeofthesurveytothetimeoftheannouncement,Dy:TABLE1—Continued. AnnouncementNumberofobservationsaSourcebDatescAnnouncementtimedInvestment33.Manufacturingorders62GFSO04/06/93–12/07/98Varies34.Manufacturingoutput64GFSO03/02/93–12/07/98VariesNetExports35.Tradebalance61GFSO07/13/93–12/11/98Varies36.Currentaccount61BD07/13/93–12/11/98VariesPrices37.Consumerpriceindex68GFSO03/01/93–12/23/98Varies38.Producerprices65GFSO03/18/93–12/22/98Varies39.Wholesalepriceindex68GFSO03/16/93–12/16/98Varies40.Importprices68GFSO03/26/93–12/21/98VariesMonetary41.MoneystockM366BD03/18/93–12/18/98VariesNotes:WegrouptheU.S.monthlynewsannouncementsintosevengroups:Realactivity,thefourcomponentsofGDP(consumption,investment,governmentpurchases,andnetexports),prices,andforward-looking.Withineachgroup,welistU.S.newsannouncementsinchronologicalorder.aTotalnumberofobservationsintheannouncementssample.bBureauofLaborStatistics(BLS),BureauoftheCensus(BC),BureauofEconomicAnalysis(BEA),FederalReserveBoard(FRB),NationalAssociationofPurchasingManagers(NAPM),ConferenceBoard(CB),FinancialManagementOfŽce(FMO),EmploymentandTrainingAdministration(ETA),GermanFederalStatisticalOfŽce(GFSO,StatistischesBundesamtDeutschland),FederalLaborOfŽce(FLO,Bundesanstaltfu¨rArbeit),Bundesbank(BD).cStartingandendingdatesoftheannouncementssample.dEasternStandardTime.DaylightsavingtimestartsontheŽrstSundayofAprilandendsonthelastSundayofOctober.e10/98isamissingobservation.f11/95,2/96,and03/97aremissingobservations.gIn01/94,thepersonalincomeannouncementtimemovedfrom10:00A.M.to8:30A.M.hBeginningin01/96,consumercreditwasreleasedregularlyat3:00P.M.Priortothisdatethereleasetimesvaried.i11/95and2/96aremissingobservations.jIn12/93,thepersonalconsumptionexpendituresannouncementtimemovedfrom10:00A.M.to8:30A.M.k03/96isamissingobservation.lWheneverGDPisreleasedonthesamedayasdurablegoodsorders,thedurablegoodsordersannouncementismovedto10:00A.M.On07/96thedurablegoodsordersannouncementwasreleasedat9:00A.M.m01/96isamissingobservation.n10/98isamissingobservation.oIn01/97,thebusinessinventoryannouncementwasmovedfrom10:00A.M.to8:30A.M.p05/88,06/88,11/98,12/89,and01/96aremissingobservations.qBeginningin3/28/94,theFedfundsratewasreleasedregularlyat2:15P.M.Priortothisdatethereleasetimesvaried.rPriorto1994thedatareferonlytoWestGermany.Beginningin1994,thedatarefertotheuniŽedGermany.ThetimingoftheGermanannouncementsisnotregular,buttheyusuallyoccurbetween2:00A.M.and8:00A.M.EasternStandardTime.44THEAMERICANECONOMICREVIEWMARCH2003 Ait5a0i1a1iFit1a2iDyt1eit.Thisparticularregressionfacilitatesthetest-ingofseveralhypotheses.First,ifthereisinformationcontentintheMMSsurveydata,thecoefŽcientestimatesa1ishouldbeposi-tiveandsigniŽcant.Second,ifthesurveyinformationisunbiased,thea0icoefŽcientestimatesshouldbeinsigniŽcant,andtheslopetermsa1ishouldbeinsigniŽcantlydif-ferentfromunity.Finally,ifexpectationsarerevisedbetweenthesurveyandtheannounce-ment,thereshouldbeareactioninthebondpriceatthetimeoftheforecastrevision,andweshouldseearelationshipbetweenthechangeinyieldandtheannouncement.Asalreadymentioned,Balduzzietal.(2001)Žnd,ashavemanyothers,thatmostoftheMMSforecastscontaininformationandareunbiased.7Moreimportantlyfortheissueathand,however,theyalsoŽndthatformostindicatorsthehypothesisthata2i50cannotberejected,indicatingthattheMMSforecastsdonotappearsigniŽcantlystale.II.ExchangeRatesandFundamentalsWewillspecifyandestimateamodelofhigh-frequencyexchange-ratedynamicsthatallowsforthepossibilityofnewsaffectingboththeconditionalmeanandtheconditionalvariance.Ourgoalistodeterminewhether7Inadditiontobeingunbiased,DouglasK.PearceandV.VanceRoley(1985)andGrantMcQueenandRoley(1993)alsoŽndthattheMMSsurveysaremoreaccurate,inthesenseofhavinglowermeansquarederrors,thantheforecastsfromstandardautoregressivetime-seriesmodels.FIGURE2.U.S.MACROECONOMICANNOUNCEMENTRELEASEDATESDATAFORMONTHXNotes:WeshowthesequenceofannouncementdatescorrespondingtodataformonthX,formostoftheeconomicindicatorsusedinthepaper.Forexample,March(monthX)consumercreditdataareannouncedbetweenMay(monthX12)5andMay10.GDPdataarespecial,becausetheyarereleasedonlyquarterly.Hence,theGDPdatareleasedinagivenmonthareeitheradvance,preliminary,orŽnaldependingonwhetherthemonthistheŽrst,second,orthirdofthequarter.Forexample,ŽrstquarterQ1GDPadvancedataareannouncedbetweenApril(monthX11)27andMay4,ŽrstquarterGDPpreliminarydataareannouncedbetweenMay(monthX12)27andJune4,andŽrstquarterGDPŽnaldataareannouncedbetweenJune(monthX13)27andJuly4.Thetableisbasedon2001ScheduleofReleaseDatesforPrincipalFederalEconomicIndicators,producedbytheU.S.OfŽceofManagementandBudgetandavailableathttp://clinton4.nara.gov/textonly/OMB/ pubpress/pei2001.html. 45VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS high-frequencyexchange-ratemovementsarelinkedtofundamentals,andifsohow.Ourmotivationsaretwofold.TheŽrstmotivationisobviouslythepossibilityofreŽningourunder-standingofthefundamentaldeterminantsofexchangerates,thecentralandstilllargelyun-resolvedquestionofexchange-rateeconomics.Thesecondmotivationisthepossibilityofim-provedhigh-frequencyvolatilityestimationviaallowanceforjumpsduetonews,asmisspeci-Žcationoftheconditionalmean(forexamplebyfailingtoallowforjumps,ifjumpsareinfactpresent)willproducedistortedvolatilityesti-matesindiscretetime.8A.ModelingtheResponseofExchangeRatestoNewsWemodelthe5-minutespotexchangerate,Rt,asalinearfunctionofIlaggedvaluesofitself,andJlagsofnewsoneachofKfunda-mentals:(1)Rt5b01Oi51IbiRt2i1Ok51KOj50JbkjSk,t2j1«t,t51,...,T.Asdiscussedearlier,K541andT5496,512.WechoseI55andJ52basedontheSchwarzandAkaikeinformationcriteria.9Weallowthedisturbanceterminthe5-minutereturnmodel(1)tobeheteroskedastic.FollowingAndersenandBollerslev(1998),weestimatethemodelusingatwo-stepweightedleast-squares(WLS)procedure.WeŽrstesti-matetheconditionalmeanmodel(1)byordi-naryleast-squaresregression,andthenweestimatethetime-varyingvolatilityof«tfromtheregressionresiduals,whichweusetoper-formaweightedleast-squaresestimationof(1).Weapproximatethedisturbancevolatilityusingthemodel:(2)z«ˆtz5c1csˆd~t! 2881Ok51KOj950J9bkj9zSk,t2j9z1XOq51QXdqcosXq2pt 288D1fqsinXq2pt 288DD1Or51ROj050J0grj0Dr,t2j0D1ut.Theleft-hand-sidevariable,z«ˆtz,istheabsolutevalueoftheresidualofequation(1),whichproxiesforthevolatilityinthe5-minuteintervalt.Asrevealedbytheright-handsideofequation(2),wemodel5-minutevolatilityasdrivenpartlybythevolatilityoverthedaycontainingthe5-minuteintervalinquestion,sˆd(t),partlybynewsSk,t,andpartlybyacalendar-effectpatternconsistinglargelyofintradayeffectsthatcapturethehigh-frequencyrhythmofdevia-tionsofintradayvolatilityfromthedailyaver-age.SpeciŽcally,wesplitthecalendareffectsintotwoparts.TheŽrstisaFourierexibleformwithtrigonometrictermsthatobeyastrict8Inthispaper,weareprimarilyinterestedinexchange-ratevolatilityonlyinsofarasitisrelevantforinferenceregardingexchange-rateconditionalmeandynamics.Wehavereservedforfutureworkadetailedanalysisofvola-tilityinrelationtoconditionalmeanandvariancejumps.Foradiscussionoftheeffectsofconditionalmeanandvariancejumpsonrealizedvolatility,seeAndersenetal.(2003).9WealsotriedallowingfornegativeJ,toaccountforannouncementleakagebeforetheofŽcialtime,andmoregenerallytoaccountforthefactthattheMMSforecastsmightnotcaptureallinformationavailableimmediatelybeforetheannouncement,butdoingsoprovedunnecessary.Thisaccordswiththeearlier-discussedŽndingofBalduzzietal.(2001)that,toagoodapproximation,theMMSforecastsdocaptureallinformationavailableimmediatelybeforetheannouncement.Moreover,ifleakageispresent(andintrospection,ifnottheempirics,suggeststhatthereis likelysomeleakage,howeversmall)thenourestimatednewsresponsecoefŽcients,whichcorrespondonlytotheimpactatthetimeoftheofŽcialannouncement,arelowerboundsforthetotalnewsimpact.46THEAMERICANECONOMICREVIEWMARCH2003 periodicityofoneday.10ThesecondisasetofdummyvariablesDr,tcapturingtheJapaneselunch,theJapaneseopen,andtheU.S.lateafternoonduringU.S.daylightsavingtime.Letusexplainingreaterdetail.ConsiderŽrstthedailyvolatility,sˆd(t),whichistheone-day-aheadvolatilityforecastfordayd(t)(thedaythatcontainstimet)fromasimpledailycon-ditionallyGaussianGARCH(1,1)modelusingspotexchange-ratereturnsfromJanuary2,1986throughDecember31,1998.Becausesˆd(t)isintendedtocapturethe“average”levelofvol-atilityondayd(t),itmakessensetoconstructitusingaGARCH(1,1)model,whichisroutinelyfoundtoprovideaccurateapproximationstodailyassetreturnvolatilitydynamics.11NowconsidertheFourierpartforthecalen-dareffects.Thisisaveryexiblefunctionalformthatmaybegivenasemi-nonparametricinterpretation(A.RonaldGallant,1981).TheSchwarzandAkaikeinformationcriteriachosearatherlowQ54forallcurrencies,whichachievesparametriceconomyandpromotessmoothnessintheintradayseasonalpattern.Finally,considerthenewseffectsSandnon-FouriercalendareffectsD.Topromotetracta-bilitywhilesimultaneouslymaintainingexibility,weimposepolynomialstructureontheresponsepatternsassociatedwiththebkj9andgrj0param-eters.12Forexample,ifaparticularnewssur-priseaffectsvolatilityfromtimet0totimet01J9,wecanrepresenttheimpactovertheeventwindowt50,1,...,J9byapolynomialspeciŽcation,p(t)5c01c1t1...1cPtP.ForP5J9thiswouldimplytheestimationofJ911polynomialcoefŽcientsandwouldnotconstraintheresponsepatterninanyway.Useofalower-orderedpolynomial,however,con-strainstheresponseinhelpfulways:itpromotesparsimonyandhencetractability,retainsexi-bilityofapproximation,andfacilitatestheim-positionofsensibleconstraintsontheresponsepattern.Forexample,wecanenforcethere-quirementthattheimpacteffectslowlyfadestozerobyimposingp(J9)50.PolynomialspeciŽcationsensurethatthere-sponsepatternsarecompletelydeterminedbytheresponsehorizonJ9,thepolynomialorderP,andtheendpointconstraintimposedonp(J9).FornewseffectsS,wetakeJ9512,P53,andp(J9)50.13Thelastconditionleadstoapolynomialwithonelessparameter;substitutingt512intop(t)wehave,p(t)5c0[12(t/12)3]1c1t[12(t/12)2]1c2t2[12(t/12)].Weestimateeachpolyno-mialseparatelyforallannouncementsandforeachexchangerate.Forexample,payrollem-ploymentpolynomialparameterestimatesare(cˆ0,cˆ1,cˆ2)5(0.177175,20.0645,0.008367)fortheDM/$,(0.163146,20.05544,0.00704)fortheCHF/$,(0.114488,20.03795,0.00477)forthePound/$,(0.108867,20.03186,0.004003)fortheEuro/$,and(0.11717,20.04619,0.006289)fortheYen/$.Finally,bkj95gkpk(j9),wheregkisthecoefŽcientestimateinequation(2).Asforthenon-Fouriercalendar-effectresponsepat-ternsD,fortheJapanesemarketopeningweuseJ056,P51,p(J0)50,fortheJapaneselunchhourweuseJ050(i.e.,astandarddummyvariablewithnopolynomialresponse),andfortheU.S.lateafternoonduringU.S.daylightsavingtimeweuseJ0560,P52,andp(0)5p(J0)50.14Inclosingthissubsection,wenotethatwecouldhavehandledthevolatilitydynamicsdif-ferently.Inparticular,insteadofestimatingex-plicitparametricmodelsofvolatilitydynamics,wecouldhavesimplyestimatedequation(1)usingheteroskedasticity-andserial-correlation10WealsotranslatetheFouriertermsleftwardasappro-priateduringU.S.daylightsavingtime.(OnlyNorthAmer-icaandEuropehavedaylightsavingtime.)11ForsurveysofGARCHmodelinginŽnancialenvi-ronments,seeBollerslevetal.(1992)andDieboldandJoseLopez(1995).Otherpossibilities,alsoexploredwithlittlechangeinqualitativeresults,includeuseofdailyrealizedvolatilitiesasinAndersenetal.(2001)andAndersenetal.(2001a,2003).12Thisisparticularlyimportantinthecaseofcondi-tionalvarianceasopposedtoconditionalmeandynamics,becauseconditionalvariancesturnouttoadjusttoshocksmoreslowlythandoconditionalmeans,therebyinvolvinglongerdistributedlags,aswewillsubsequentlyemphasize.Hence,althoughtractabilitydidnotrequiretheimpositionofpolynomialshapeontheconditionalmeandistributedlags,itgreatlyenhancestheaccuracyoftheconditionalvarianceestimates.13The“constraint”thatvolatilitynewseffectslingerforatmostanhour(J9512)isnonbinding.Initialexperi-mentationallowingforJ9536revealedthatonehourwasenoughforfulladjustment,forallindicatorsandcurrencies.14TheJapaneseopeningisat8P.M.Easterndaylightsavingtime,theJapaneselunchhouris11P.M.through12:30A.M.Easterndaylightsavingtime,theU.S.lateafternoonduringdaylightsavingtimeisdeŽnedtostartat3P.M.Easterndaylightsavingtime.47VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS consistent(HAC)standarderrors.WeŽndthatapproachlessattractivethantheoneweadopted,foratleastthreereasons.First,weareinterestednotonlyinperformingheteroskedasticity-robustinferenceaboutthecoefŽcients(donebothbyourWLSandbyHACestimation)inequation(1),butalsoinobtainingthemostefŽcientestimatesofthosecoefŽcients.Second,al-thoughHACestimationisasymptoticallyro-busttoresidualheteroskedasticityofunknownform,itsgeneralrobustnessmaycomeatthepriceofinferiorŽnite-sampleperformancerel-ativetotheestimationofawell-speciŽedpara-metricvolatilitymodel.15Third,despitethefactthattheyarenotcentraltotheanalysisinthepresentpaper,boththeintra-andinterdayvol-atilitypatternsareofintrinsicŽnancialeco-nomicinterestandhenceonemaywantestimatesoftheseinothersituations.NotwithstandingalloftheseaprioriargumentsagainsttheuseofHACestimationinthepresentcontext,asacheckontherobustnessofourresults,wealsoperformedalloftheempiricalworkrelatedtothemeaneffectsusingHACestimation,withnochangeinanyofthequalitativeresults(al-thoughanumberofthecoefŽcientswerenolongerstatisticallysigniŽcant).B.NewsEffectsI:NewsAnnouncementsMatter,andQuicklyThemodel(1)–(2)providesanaccurateap-proximationtobothconditionalmeanandcon-ditionalvariancedynamics.Sincethemodelcontainssomanyvariablesandtheirlags,itwouldprovecounterproductivetosimplyreportalloftheparameterestimates.Instead,Figure3showstheactualandŽttedaverageintradayvolatilitypatterns,whichobviouslyagreefairlyclosely.Further,inFigure4wepresentgraph-icallytheresultsforthemostimportantindica-tors,andwediscussthoseresults(andsomeothers,notshownintheŽgure)inwhatfollows.LetusŽrstconsidertheeffectsofU.S.mac-roeconomicnews.Throughout,newsexertsagenerallystatisticallysigniŽcantinuenceonexchangerates,whereasexpectedannounce-mentsgenerallydonot.Thatis,onlyunantici-patedshockstofundamentalsaffectexchangerates,inaccordancewiththepredictionsofra-tionalexpectationstheory.ManyU.S.indica-torshavestatisticallysigniŽcantnewseffectsacrossallcurrencies,includingpayrollemploy-ment,durablegoodsorders,tradebalance,ini-15SeeC.RadhakrishnaRao(1970)andAndrewChesherandIanJewitt(1987).FIGURE3.ACTUALANDFITTEDINTRADAYVOLATILITYPATTERNSNotes:Thesolidlineistheaverageintradaypatternoftheabsoluteresidualreturnz«ˆtzoverthe2885-minuteintervalswithintheday,where«ˆtistheresidualfromtheexchange-rateconditionalmeanmodel(1)inthetext.ThedashedlineistheŽttedintradaypatternofz«ˆtzfromtheexchange-ratevolatilitymodel(2)inthetext.Toavoidcontaminationfromshiftsinandoutofdaylightsavingtime,weconstructtheŽguresusingonlydayscorrespondingtoU.S.daylightsavingtime.48THEAMERICANECONOMICREVIEWMARCH2003 tialunemploymentclaims,NAPMindex,retailsales,consumerconŽdence,andadvanceGDP.Thegeneralpatternisoneofveryquickex-change-rateconditionalmeanadjustment,char-acterizedbyajumpimmediatelyfollowingtheannouncement,andlittlemovementthereafter.FavorableU.S.“growthnews”tendstoproducedollarappreciation,andconversely.Thisiscon-sistentwithavarietyofmodelsofexchange-ratedetermination,fromsimplemonetarymodels(e.g.,Mark,1995)tomoresophisticatedframe-worksinvolvingaU.S.centralbankreactionfunctiondisplayingapreferenceforlowina-tion(e.g.,JohnB.Taylor,1993).16OnecanseefromthecenterpaneloftheŽrstrowofFigure4,forexample,thataonestandarddeviationU.S.payrollemploymentsurprisetendstoappreciate(ifpositive)ordepreciate(ifnegative)thedollaragainsttheDMby0.1616Formostofourmacroeconomicindicators,includingthoseonwhichweprimarilyfocus,thesignofa“goodshock”isclear:movementsassociatedwithincreasedrealU.S.economicactivityaregoodforthedollar.Sometimes,however,itisnotobviouswhichdirectionshouldbeviewedasgood,asperhapswithconsumercredit.FIGURE4.EXCHANGE-RATERESPONSESTOU.S.NEWSNotes:WegraphthethreenewsresponsecoefŽcientsassociatedwiththeexchange-rateconditionalmeanregression(1),correspondingtoresponsesattheannouncement,Žveminutesaftertheannouncement,andtenminutesaftertheannounce-ment.Wealsoshowtwostandarderrorbands,underthenullhypothesisofazeroresponse,obtainedusingtheweightedleast-squaresestimationmethoddescribedinthetext.49VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS percent.17Thisisasizeablemove,frombothstatisticalandeconomicperspectives.Onthestatisticalside,wenotethatonly0.7percentofour5-minutereturnsshowanappreciationordepreciationbiggerthan0.10percent.Ontheeconomicside,wenotethat0.16percentisalsolargerelativetotheaverageDM/$spread,whichtendstobearound0.06percentduringtheperiodwestudy(see,HendrikBessem-binder,1994,andJoelHasbrouck,1999,Table1).Itisimportanttonotethat,althoughcloselytimednewseventsarehighlycorrelated,thecorrelationdoesnotcreateaseriousmulticolin-earityproblemexceptinafewspeciŽcin-stances.Forexample,industrialproductionandcapacityutilizationarereleasedatthesametime,andtheyarehighlycorrelated(0.64).Ingeneral,however,theeventthattwoannounce-mentswithinthesamecategory(e.g.,realac-tivity)arereleasedsimultaneouslyisrare.NowletusfocusontheDM/$rateinsomedetail.ItisofparticularinterestbothbecauseofitscentralroleintheinternationalŽnancialsystemduringtheperiodunderstudy,andbecausewehavenewsdataonbothU.S.andGermanmacro-economicindicators.18FirstconsidertheeffectsofU.S.macroeconomicnewsontheDM/$rate.NewsannouncementsonavarietyofU.S.indica-torssigniŽcantlyaffecttheDM/$rate,includingpayrollemployment,durablegoodsorders,tradebalance,initialclaims,NAPMindex,retailsales,consumerconŽdence,CPI,PPI,industrialproduc-tion,leadingindicators,housingstarts,construc-tionspending,federalfundsrate,newhomesales,andGDP(advance,preliminary,andŽnal).NowconsidertheeffectofGermanmacroeco-nomicnewsontheDM/$rate.19InsharpcontrasttothelargenumberofU.S.macroeconomicindi-catorswhosenewsaffecttheDM/$rate,onlyveryfewoftheGermanmacroeconomicindicatorshaveasigniŽcanteffect(M3andindustrialpro-duction).Weconjecturethatthedisparitymaybeduetothefact,detailedinTable1,thatthereleasetimesofU.S.macroeconomicindicatorsareknownexactly(dayandtime)butonlyinexactlyforGermany(daybutnottime).Uncertainreleasetimesmayresultinlessmarketliquidity(andtrading)aroundtheannouncementtimes,henceresultinginsmallernewseffectsaroundthean-nouncements,ultimatelyproducingamoregrad-ualadjustment,perhapsforafewhoursaftertheannouncements.Alternatively,greaterprean-nouncementleakageinGermanymayresultinadjustmentstakingplacegraduallyinthedayspriortotheactualannouncement.Mostoftheexplanatorypoweroftheex-change-rateconditionalmeanmodel(1)comesfromthelaggedvaluesofthedependentvari-ableandthecontemporaneousnewsannounce-ment.Hence,although58percentofthedaysinoursamplecontainanewsannouncement,toagoodapproximationthenewspredictsonlythedirectionandmagnitudeoftheexchange-ratemovementduringthe5-minutepost-releasein-tervals,whichcorrespondtoonlytwo-tenthsofonepercentofthesampleobservations.Tofo-cusontheimportanceofnewsduringannounce-mentperiods,wenowestimatethemodel(3)Rt5bkSkt1«t,whereRtisthe5-minutereturnfromtimettotimet11andSktisthestandardizednewscorrespondingtoannouncementk(k51,...,41)attimet,andtheestimatesarebasedononlythoseobservations(Rt,Skt)suchthatanannouncementwasmadeattimet.WeshowtheestimationresultsinTable2,whichcontainsanumberofnoteworthyfea-tures.First,newsonmanyofthefundamentalsexertsasigniŽcantinuenceonexchangerates.Thisisofcourseexpected,givenourearlierestimationresultsforequation(1)assumma-rizedinFigure4.NewsfromFOMCdelibera-tions,forexample,clearlyinuencesexchangerates:thelargeandstatisticallysigniŽcantco-efŽcients,andthehighR2’s,arestriking.Theirpositivesignsindicatethat,asexpectedforex-ampleinastandardmonetarymodel,Fedtight-eningisassociatedwithdollarappreciation.2017Weinterpretaone-standard-deviationsurpriseas“typical.”18Germannewsistheonlynon-U.S.newsthatisreadilyavailablefromMMS.19ToaŽrstapproximation,GermannewsisrelevantonlyforDM/$determination,incontrasttoU.S.news,whichisrelevantforthedeterminationofallU.S.dollarexchangerates.20Itwouldbeinteresting(withalongersampleofdata)toexaminethestabilityoftheresponsecoefŽcientoverdifferentstagesofthebusinesscycle;see,e.g.,McQueenandRoley(1993).AccordingtothestandardU.S.business-cyclechro-nologyproducedbytheNationalBureauofEconomicRe-search,theUnitedStateswasinanexpansionfromMarchof1991untilMarchof2001;henceourentiresample.50THEAMERICANECONOMICREVIEWMARCH2003 TABLE2—U.S.ANDGERMANCONTEMPORANEOUSNEWSRESPONSECOEFFICIENTSANDR2VALUES AnnouncementPound/$Yen/$DM/$CHF/$Euro/$bkR2bkR2bkR2bkR2bkR2U.S.AnnouncementsQuarterlyAnnouncements1.GDPadvance0.0290.0980.0360.1020.08*0.3010.079*0.3070.061*0.4202.GDPpreliminary0.0380.1340.0220.0810.055*0.1850.057*0.2070.0170.0483.GDPŽnal20.0040.0040.0190.0480.0170.0290.0100.0070.0060.010MonthlyAnnouncementsRealActivity4.Nonfarmpayrollemployment0.098*0.1890.084*0.2140.161*0.2370.144*0.2690.08*0.2325.Retailsales0.048*0.2250.0190.0660.067*0.2410.059*0.1700.041*0.1936.Industrialproduction0.020*0.1050.019*0.0780.029*0.1310.034*0.1470.018*0.0867.Capacityutilization0.0170.0610.0160.0550.0210.0460.0230.0580.0180.0418.Personalincome0.0070.0150.0010.0000.0060.0070.0030.00120.0050.0059.Consumercredit0.0020.0020.0090.0190.0040.0120.0020.00220.0020.004Consumption10.Personalconsumptionexpenditures20.0030.0030.0050.00620.0070.01020.0110.0120.0070.00811.Newhomesales0.0020.0020.0110.0300.010.01520.0020.0010.0050.003Investment12.Durablegoodsorders0.055*0.2660.027*0.0810.088*0.3630.085*0.3550.043*0.23713.Constructionspending0.019*0.0870.01*0.0260.031*0.0910.017*0.0340.0150.03014.Factoryorders0.0110.0240.0060.0060.0180.0380.0190.0410.031*0.10215.Businessinventories20.0040.0080.010.0290.0090.0120.0020.0010.0070.015GovernmentPurchases16.GovernmentbudgetdeŽcit0.007*0.0570.0080.0380.0020.0030.0100.0500.0030.006NetExports17.Tradebalance0.092*0.5290.112*0.3700.138*0.5850.124*0.4800.084*0.414Prices18.Producerpriceindex0.0050.0030.0000.0000.0190.0200.0170.0170.018*0.04619.Consumerpriceindex0.0160.0480.0120.0330.031*0.1010.035*0.1040.0150.027Forward-looking20.ConsumerconŽdenceindex0.037*0.1740.022*0.1030.058*0.2220.054*0.2140.035*0.18921.NAPMindex0.028*0.1990.012*0.0360.039*0.1410.036*0.1460.025*0.07422.Housingstarts0.0060.0080.0050.0070.0170.0280.02*0.0330.0080.00923.Indexofleadingindicators0.0120.0310.0090.0060.0120.0090.0110.00520.0050.005Six-WeekAnnouncements24.Targetfederalfundsrate0.048*0.2290.050*0.1620.072*0.2590.072*0.2300.0320.142WeeklyAnnouncements25.Initialunemploymentclaims20.014*0.02520.012*0.01920.022*0.03620.026*0.04620.019*0.05826.Moneysupply,M10.0000.0000.0000.0000.004*0.0200.004*0.0190.002*0.00927.Moneysupply,M20.0000.00020.0010.0010.004*0.0190.005*0.0300.002*0.01328.Moneysupply,M30.0000.0000.0010.0020.0020.0040.004*0.0230.002*0.011GermanAnnouncementsQuarterlyAnnouncements29.GDP20.0040.04220.0020.00120.0070.02220.0110.06820.0040.015MonthlyAnnouncementsRealActivity30.Employment0.0000.0000.0020.0010.0000.0000.01*0.0450.0030.00331.Retailsales0.0010.0010.0040.00820.0030.00420.0020.00320.01*0.09132.Industrialproduction20.011*0.05920.0090.03620.017*0.17220.015*0.10520.0050.01551VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS Second,unliketheR2valuesforequation(1),whicharetypicallyverysmall,theR2valuesforequation(3)areoftenquitehigh.Newsannouncementsoccurcomparativelyrarelyandhaveanonnegligiblebutshort-livedimpactonexchangerates;hencetheR2inanequationsuchas(3)mustbelowwhencomputedacrossall5-minuteobservations.Incontrast,onenaturallyexpectshigherR2val-ueswhencomputedusingonlyannouncementobservations,althoughtheprecisesizeisofcourseanempiricalmatter.Table2revealsR2valuesthatareoftenaround0.3andsome-timesapproaching0.6.Finally,itisinterestingtonotethattheresultsofYin-WongCheungandClementYuk-PangWong(2000)andCheungandMenzieDavidChinn(2001),obtainedbysur-veyingtraders,coherereassuringlywiththemodel-basedresultsdocumentedhere.Inpar-ticular,CheungandChinn(2001)reportthattradersbelievethatexchangeratesadjustal-mostinstantaneouslyfollowingnewsan-nouncements,andthatnewsregardingrealvariablesismoreinuentialthannewsre-gardingnominalvariables,whichisentirelyconsistentwiththeempiricalresultsreportedinTable2.C.NewsEffectsII:AnnouncementTimingMattersOnemightwonderwhether,withinthesamegeneralcategoryofmacroeconomicindicators,newsonthosereleasedearliertendtohavegreaterimpactthanthosereleasedlater.Toevaluatethisconjecture,wegroupedtheU.S.indicatorsintoseventypes:realactivity,con-sumption,investment,governmentpurchases,netexports,prices,andforward-looking.Withineachgroup,wearrangedtheannouncementsinthechronologicalorderdescribedinFigure2.TheconjectureisgenerallyveriŽed.Inthees-timatesofequation(3)withineachindicatorgroup,theannouncementsreleasedearliesttendtohavethemoststatisticallysigniŽcantcoefŽ-cientsandthehighestR2values.21InFigure5weplottheR2ofequation(3)withineachindicatorgroup,asafunctionoftheannounce-menttiming.Theclearlyprevalentdownwardslopesrevealthattheearlyannouncementsdoindeedhavethegreatestimpact.Thefactthat“announcementtimingmatters”21Oneexceptionisthenominalgroup;theconsumerpriceindexseemsmoreimportantthantheproducerpriceindex,despiteitsearlierreleasedate.TABLE2—Continued. AnnouncementPound/$Yen/$DM/$CHF/$Euro/$bkR2bkR2bkR2bkR2bkR2Investment33.Manufacturingorders20.0070.02520.0080.02920.0110.06120.010.04220.0020.00234.Manufacturingoutput20.0010.00120.017*0.09120.0070.04120.0090.04820.0070.034NetExports35.Tradebalance20.0040.0180.0010.0000.0000.0000.0010.00120.0050.01936.Currentaccount20.0030.0090.0060.01920.0060.03520.0060.03320.0060.031Prices37.Consumerpriceindex20.020*0.15920.0040.0160.0000.0000.0070.01620.0010.00138.Producerprices20.0020.0030.0030.01220.0030.00320.0040.01120.0080.01539.Wholesalepriceindex0.0000.0000.0030.00320.0110.03920.0030.0050.0040.01240.Importprices0.0070.07920.0090.0490.0030.0050.0060.01920.0030.003Monetary41.MoneystockM320.02*0.2150.0000.00020.033*0.18120.02*0.11320.023*0.161Notes:Weestimatethecontemporaneousexchange-ratenewsresponsemodel,Rt5bkSkt1«t,whereRtisthe5-minutereturnfromtimettotimet11andSktisthestandardizednewscorrespondingtoannouncementk(k51,...,41)madeattimet.Weestimatetheregressionusingonlythoseobservations(Rt,Skt)suchthatanannouncementwasmadeattimet.WereportthebˆkandR2values,andwemarkwithanasteriskthosecoefŽcientsthatarestatisticallysigniŽcantatthe5-percentlevel,usingheteroskedasticity-andautocorrelation-consistentstandarderrors.52THEAMERICANECONOMICREVIEWMARCH2003 helpswiththeinterpretationofourearlier-reportedempiricalresultsinTable2,whichindicatethatonlysevenofthe40announce-mentssigniŽcantlyimpactedallthecurrencyspeciŽcations.Thereasonisthatmanyoftheannouncementsaretosomeextentredundant,andthemarketthenonlyreactstothosereleasedearlier.Hence,forexample,U.S.durablegoodsFIGURE5.U.S.NEWSEFFECTSASAFUNCTIONOFRELEASETIMENotes:Weestimatethecontemporaneousexchange-ratenewsresponsemodel,Rt5bkSkt1«t,whereRtisthe5-minutereturnfromtimettotimet11andSktisthestandardizednewscorrespondingtoannouncementk(k51,...,17)madeattimet.Weestimatetheregressionusingonlythoseobservations(Rt,Skt)suchthatanannouncementwasmadeattimet.OntheverticalaxiswedisplaytheR2values,andonthehorizontalaxiswedisplaymacroeconomicnewsannouncementsinthechronologicalorderdocumentedinTable2.The“newsnumbers”areasfollows: GDPRealActivityInvestmentForward-Looking1.GDPadvance4.Payrollemployment10.Durablegoodsorders14.ConsumerconŽdence2.GDPpreliminary5.Retailsales11.Constructionspending15.NAPMindex3.GDPŽnal6.Industrialproduction12.Factoryorders16.Housingstarts7.Capacityutilization13.Businessinventories17.Indexofleadingindicators8.Personalincome9.Consumercredit53VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS ordersmatterforallcurrencypairsbutU.S.factoryorders,whicharereleasedlater,donot.D.NewsEffectsIII:VolatilityAdjuststoNewsGraduallyAsdiscussedpreviouslyanddocumentedinFigure4,exchangeratesadjusttonewsimme-diately.Itisinterestingtonote,however,thatexchange-ratevolatilitiesadjustonlygradually,withcompleteadjustmentoccurringonlyafterJ95125-minuteperiods,oronehour.WeprovidedetailsinTable3.Asalreadynoted,andasshownagaininthetoppanelofthetable,thecontemporaneousreturnre-sponsecoefŽcientsaresizeableandstatisti-callysigniŽcant,andthefullresponseoccursimmediately.Incontrast,thecontemporane-ousvolatilityresponsecoefŽcients,althoughstatisticallysigniŽcant,aresmaller,asshowninthemiddlepanelofthetable.Importantly,however,thecompleteresponseofvolatilitytonewsoccursonlyafteranhourorso,anditisnoticeablylargerthaneitherthecontempo-raneousvolatilityresponseorthecontem-poraneousreturnresponse,asshowninthebottompanelofthetable.E.NewsEffectsIV:PureAnnouncementEffectsarePresentinVolatilityItispossiblethatthemerepresenceofanannouncementmightboostvolatility,quiteapartfromthesizeoftheassociatedsurprise.Toexplorethispossibilityweaddtothere-turnsequation(1)JlagsofannouncementperioddummiesoneachofKfundamentals,andwealsoaddtothevolatilityequation(2)J9lagsofannouncementperioddummiesoneachofKfundamentals.AsshowninTable4,theannouncementdummiesaregenerallyinsigniŽcantinthereturnsequation(1)butgenerallysigniŽcantinthevolatilityequation(2),inlinewithearlierresultsforbondmar-ketssuchasFlemingandRemolona(1997,1999).Newseffectsarestillimportant,how-ever,inbothconditionalmeanandvariancedynamics.TABLE3—RETURNANDVOLATILITYNEWSRESPONSECOEFFICIENTS AnnouncementPound/$Yen/$DM/$CHF/$Euro/$ContemporaneousReturnResponse,bk0Nonfarmpayrollemployment0.092*0.072*0.159*0.115*0.081*Durablegoodsorders0.055*0.029*0.084*0.083*0.041*Tradebalance0.083*0.115*0.142*0.131*0.084*Initialunemploymentclaims20.010*20.009*20.018*20.024*20.017*ContemporaneousVolatilityResponse,bk0Nonfarmpayrollemployment0.058*0.053*0.084*0.077*0.058*Durablegoodsorders0.017*0.010*0.027*0.018*0.018*Tradebalance0.023*0.040*0.034*0.031*0.026*Initialunemploymentclaims0.003*0.004*0.010*0.010*0.005*CumulativeVolatilityResponse,¥j950J9bkj9Nonfarmpayrollemployment0.356*0.328*0.519*0.476*0.356*Durablegoodsorders0.106*0.060*0.163*0.114*0.108*Tradebalance0.139*0.244*0.210*0.191*0.161*Initialunemploymentclaims0.021*0.023*0.059*0.060*0.033*Notes:Weestimatetheexchange-rateconditionalmeanmodel(1),Rt5b01¥i51IbiRt2i1¥k51K¥j50JbkjSk,t2j1«t,andwereportestimatesofthecontemporaneousresponseofexchange-ratereturnstonews,bk0.Wealsoestimatethedisturbancevolatilitymodel(2),z«ˆtz5c1csˆd~t! 2881Ok51KOj950J9bkj9zSk,t2j9z1XOq51QXdqcosXq2pt 288D1fqsinXq2pt 288DD1Or51ROj050J0grj0Dr,t2j0D1ut,andwereportestimatesofthecontemporaneousresponseofexchange-ratevolatilitytonews,bk05gkpk(0),asdescribedinthetext.Finally,wealsoreportestimatesofthecumulativevolatilityresponse,¥j95012gkpk(j9),asdescribedinthetext.AsterisksdenotestatisticalsigniŽcanceatthe5-percentlevel.54THEAMERICANECONOMICREVIEWMARCH2003 F.NewsEffectsV:AnnouncementEffectsareAsymmetric—ResponsesVarywiththeSignoftheNewsWehaveseenthatnewsaboutmacroeconomicfundamentalssigniŽcantlyaffecthigh-frequencyexchangerates.Thusfarwehaveallowedonlyforconstantnewseffects,butitisnaturaltogofartherandaskwhetherthenewseffectsvarywiththesignofthesurprise.Toaddressthisissuewegeneralizeequation(3)byallowingtheimpactresponsecoefŽcientbktobealinearfunctionofthenewssurpriseSkt,allowingforadifferentconstantandslopeoneachsideoftheorigin,(4)bk55b0k1b1kSktifSt#0b2k1b3kSktifSt.0.Inserting(4)into(3)yieldstheimpactresponsespeciŽcation,(5)Rt55b0kSkt1b1kSkt21«tifSt#0b2kSkt1b3kSkt21«tifSt.0.FollowingRobertF.EngleandVictorK.Ng(1993),wecalltheunionofb0kSkt1b1kSkt2totheleftoftheoriginandb2kSkt1b3kSkt2totheTABLE4—RETURNANDVOLATILITYNEWSRESPONSECOEFFICIENTSANDANNOUNCEMENTDUMMYCOEFFICIENTS AnnouncementPound/$Yen/$DM/$CHF/$Euro/$ContemporaneousReturnResponseNonfarmpayrollemploymentbk00.091*0.071*0.159*0.115*0.079*uk00.0180.0080.029*20.0020.020Durablegoodsordersbk00.052*0.028*0.082*0.083*0.039*uk020.023*20.00420.023*20.01720.010Tradebalancebk00.086*0.121*0.144*0.131*0.085*uk00.0130.0290.0130.0040.012Initialclaimsbk020.009*20.009*20.017*20.023*20.017*uk00.00120.010*20.00520.00220.006ContemporaneousVolatilityResponseNonfarmpayrollemploymentbk00.0173*0.0216*0.0215*0.0169*0.015*uk00.0566*0.0436*0.0873*0.0837*0.0597*Durablegoodsordersbk00.014*0.0098*0.023*0.0148*0.0144*uk00.00420.00020.00480.00460.0043Tradebalancebk00.0226*0.0255*0.0214*0.0149*0.0153*uk00.00010.0174*0.0162*0.0198*0.0141*Initialclaimsbk00.000520.00050.0039*0.0062*0.002uk00.0035*0.0048*0.0062*0.0038*0.0032*Notes:Weaddtoequation(1)JlagsofannouncementperioddummiesoneachofKfundamentals,Rt5b01¥i51IbiRt2i1¥k51K¥j50JbkjSk,t2j1¥k51K¥j50JukjDk,t2j1«t,andwereportestimatesofthecontemporaneousreturnresponsetonewsandtoannouncementperiods,bk0anduk0,respectively.Wealsoaddtoequation(2)J9lagsofannouncementperioddummiesoneachofKfundamentals,z«ˆtz5c1csˆd~t! 2881Ok51KOj950J9bkj9zSk,t2j9z1Ok51KOj950J9ukj9Dk,t2j91XOq51QXdqcosXq2pt 288D1fqsinXq2pt 288DD1Or51ROj050J0grj0Dr,t2j0D1ut,andreportestimatesofthecontemporaneousreturnresponsetonewsandtoannouncementperiods,bk0anduk0,respectively.AsterisksdenotestatisticalsigniŽcanceatthe5-percentlevel.55VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS rightoftheoriginthe“newsimpactcurve.”22InthetoprowofFigure6weshowthenewsimpactcurvesaveragedacrossallmacroeconomicfunda-mentals,k51,...,41.Itisclearthat,onaverage,theeffectofmacroeconomicnewsoftenvarieswithitssign.Inparticular,negativesurprisesoftenhavegreaterimpactthanpositivesurprises.23Itisinterestingtoseewhetherthesigneffectprevailswhenwelookseparatelyatthemost22DespitethesuperŽcialresemblanceintermsofdocu-mentingasymmetricresponsestonews,ourworkisverydifferentfromthatofEngleandNg(1993)andmanysubsequentrelatedstudies.Inparticular,theEngle–Ngnewsimpactcurvetracksthevarianceofequityreturnscondi-tionaluponthesignandsizeofpastreturns(withnoallowanceforatime-varyingconditionalmeanreturn),whereasournewsimpactcurvetracksthemeanofforeignexchangereturnsconditionaluponthesignandsizeofmacroeconomicnews.23Tothebestofourknowledge,suchsigneffectshavenotpreviouslybeendocumentedfortheforeignexchangemarket.Evidenceofasymmetricconditional-meannewseffectsexistsinothercontexts,however.Forexample,JenniferConradetal.(2001)Žndasymmetriceffectsofearningsnewsonstockreturns,whilerecentconcurrentworkbyHautschandHess(2001)detailsanasymmetricresponsetoemploymentnewsintheT-bondfuturesmarket.FIGURE6.U.S.NEWSIMPACTCURVESNotes:Inthetoprowweshowthenewsimpactcurvesaveragedacrossallmacroeconomicfundamentals,k51,...,41.Intheremainingrowsweshowthenewsimpactcurvesforpayrollemployment,tradebalance,durablegoodorders,andinitialclaims.Seetextfordetails.56THEAMERICANECONOMICREVIEWMARCH2003 importantnewsannouncements.Intheremain-ingrowsofFigure6weshowthenewsimpactcurvesforpayrollemployment,tradebalance,durablegoodsorders,andinitialclaims.Thesigneffectisgenerallymaintained,althoughthereissomevariationacrossindicatorsandcurrencies.AsymmetryintheYen/$,DM/$,andCHF/$responsetopayrollemploymentandtradebalancenews,forexample,isverypro-nounced,whereasitislargelyabsentinthePound/$andEuro/$response.Inthenextsectionweexploremoredeeplytheeconomicsbehindtheasymmetricresponse.Recenttheoreticalmodelssuggestthattheasymmetrymaybedriven,inpart,bythedy-namicsofuncertaintyregardingtheunderlyingstateoftheeconomy.ItturnsoutthatourMMSdatasetcontainsnotonlyexpectations,butalsoameasureofthecross-sectionaldispersionofexpectations,thestandarddeviation.Usingthecross-sectionalstandarddeviationasaproxyforstateuncertainty,wecanthereforedirectlyas-sessakeymechanismthoughttogenerateasymmetricresponse,towhichwenowturn.III.AsymmetricResponse,InformationProcessing,andPriceDiscoveryTwostrandsofliteratureimplyasymmetryintheresponseofexchangeratestonews.Inpar-ticular,theyimplythatbadnewsin“goodtimes”shouldhaveanunusuallylargeimpact,aviewthatisalsocommoninthepractitionercommunity,asemphasizedbyConradetal.(2001).Notethatourentiresampletakesplaceingoodtimes—1992through1998.Hencethetheoreticalpredictionthat“badnewsingoodtimesshouldhaveunusuallylargeeffects,”de-generatesinoursampleperiodto“badnewsshouldhaveunusuallylargeeffects,”which,toareasonablygoodapproximation,iswhatwefoundearlier.TheŽrststrandoftheliteratureis“behavior-al”andfocusesprimarilyonequities,attheŽrmlevel.NicholasBarberisetal.(1998),forexam-ple,modelinvestorsasbelievingthatŽrmearn-ingsfollowatwo-stateregime-switchingprocess—erroneously,asearningsactuallyfol-lowarandomwalk—withmean-revertingearn-ingsinstate0andupward-trendingearningsinstate1.Henceaseriesofpositiveearningsleadsinvestorstoinferthatstate1holds,withtheconcomitantexpectationofadditionalpositiveearnings.Insuchasituation,badnewsgener-atesalargenegativeresponsebecauseitisasurprise,whereasgoodnewsgenerateslittlere-sponsebecauseitisanticipated.Thesecondrelevantstrandoftheliteratureusesarational-expectationsequilibriumap-proachandfocusesmoreonthemarketlevelasopposedtotheŽrmlevel,asinPietroVeronesi(1999),TimothyC.Johnson(2001a,b),andAlexanderDavidandVeronesi(2001).Veronesi(1999),inparticular,modelsinvestorsas(cor-rectly)believingthattheeconomyfollowsatwo-stateregime-switchingprocess,with“low”and“high”statescorrespondingtorecessionsandexpansions.Agentssolveasignalextrac-tionproblemtodeterminetheprobabilityp(t)ofbeinginthehighstate,andequilibriumassetpricescanbeshowntobeincreasingandconvexfunctionsofp(t).Theintuitionforthiskeyresultissimple.Supposethatp(t21)'1,i.e.,investorsbelievethatthehighstatealmostsurelyprevails.Thenifbadnewsarrivesattimet,twothingshappen:Žrst,expectedfutureassetvaluesdecrease,andsecond,p(t)decreases(i.e.,stateriskincreases).Risk-averseinvestorsrequireadditionalreturnsforbearingthisaddi-tionalrisk;hencetheyrequireanadditionaldiscountontheassetprice,whichdropsbymorethanitwouldinapresent-valuemodel.Conversely,supposeinvestorsareconŽdentthatthelowstateprevails,i.e.,p(t21)'0.Thenifgoodnewsarrivesattimet,expectedfutureassetvaluesincrease,butp(t)alsoincreases(i.e.,stateriskagainincreases).Asbefore,investorsrequireadditionalreturnsforbearingthisadditionalrisk;hencetheyre-quireadiscountontheassetprice,whichincreasesbylessthanitwouldinapresent-valuemodel.Foranumberofreasons,itisnotourinten-tionheretoexplicitlytestthepractitionerclaimthatpricesrespondmoststronglytobadnewsingoodtimes,ortodirectlyimplementVeronesi’smodelortocombineitwiththeBarberis-Shleifer-Vishymodel.First,ourdatasetisnotwell-suitedtothatpurpose;asmentionedabove,itcontainsonlytheexpansionary1990’s.Sec-ond,Conradetal.(2001)havealreadymadeadmirableprogressinthatregard,Žndinggen-eralsupportfortheassertionthat(stock)pricesrespondmoststronglytobadnewsingoodtimes.Third,theBarberis-Schleifer-Vishnymodelisnotparticularlywell-suitedtotheforex57VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS contextrelevanthere,asitfocusesontheearn-ingsstreamforanindividualŽrm.Instead,wetakeastruethepractitionerclaimthatpricesrespondmoststronglytobadnewsingoodtimes,andwefocusontheexplanationembodiedinVeronesi’smodel.Weuseanin-terestingfeatureofourMMSexpectationsdatatoassessthekeyallegedmechanismthroughwhichbadnewsingoodtimestranslatesintolargepricemoves:increaseduncertaintyaboutthestateoftheeconomy.Inparticular,wehavedatanotonlyonthemedianexpectationsofmacroeconomicfundamentals,butalsoontheassociatedstandarddeviationsacrosstheindi-vidualforecasters.Hencewecancheckdirectlywhetheruncertaintyaboutthestateoftheecon-omy,asproxiedbythestandarddeviationofexpectationsacrosstheindividualforecasters,increasesfollowingthearrivalofbadnewsingoodtimes.24Beforeproceedingtoexaminetheeffectofbadnewsarrivalsonsubsequentfore-castdispersion,however,twoissuesarise.First,itisnotclearwhattiminginthedatamatchesthegenerictiminginthemodel.Clearly,badnewsattimet21meansthatexpectationsfortimetareformedinabadnewsenvironment,butwhatifthenewsatt22wasbadandthenewsatt21wasnot?Perhapsagentshaveamemorythatlastslongerthanoneannouncementperiod,sothateventhelattercasecouldbeviewedasabadnewsenviron-ment.Ingeneral,wemightsaythatweareinabadnewsenvironmentifthenewswasbadatanyoftimest21,t22,...,t2d,forsomed.Second,toenhanceourchancesofdetectingthe“Veronesieffect,”ifitexists,wemaynotwanttotrackthearrivalofallbadnews,butratheronlybadnewsthatexceedssomemini-malthreshold,saythepthpercentileofthedistributionofbadnews,wherep,liked,mustbechosen.Asabenchmark,wesimplysetd51andp550percent(i.e.,themedian).Figure7plotsthecorrespondingstandarddeviationoftheMMSpayrollemployment,du-rablegoodsorders,andtradebalanceforecasts.Theshadedareasindicateabadnewsenviron-mentusingthecriteriad51andp550percent.Analystforecastdispersionisindeedhigherfollowingbadnewsthanatothertimes;speciŽcally,theuncertaintyofpayrollemploy-mentis30percenthigher,theuncertaintyofdurablegoodsordersis6percenthigher,andtheuncertaintyofthetradebalanceis12percenthigher.Theseeffectsarerobusttoreasonablevariationinpandd.IV.ConcludingRemarksandDirectionsforFutureResearchThegoaloftheresearchonwhichthispaperreportsistodeepenourunderstandingofthelinks24Ofcourse,thenotionofforecastuncertaintyandtheforecastdispersionacrossforecastersarenotexactlyequivalentconcepts.VictorZarnowitzandLouisA.Lambros(1987)show,however,thattheyaregenerallypositivelycorrelated. FIGURE7.FORECASTUNCERTAINTYNotes:Weplotthetimeseriesofcross-sectionalstandarddeviationsoftheMoneyMarketServicesforecasts.Theshadedareasdenote“badnews”times.Seetextfordetails.58THEAMERICANECONOMICREVIEWMARCH2003 betweenexchange-ratemovementsandnewsaboutfundamentals.Tothatend,inthispaperwehavedocumentedimportantnewseffects,withasymmetricresponsepatterns.Letusconcludebyrelatingourresultstoworkonorderowanddrawingimplicationsforfutureresearch.Inrecentinnovativework,EvansandLyons(2002)showthatsignedorderowisagoodpredictorofsubsequentexchange-ratemove-ments.Thisworkisimportantinthatiten-hancesourunderstandingofthedeterminantsofhigh-frequencyexchange-ratemovements,butlesssatisfyinginthatitremainsignorantaboutthedeterminantsofhigh-frequencyorderow.We,incontrast,haveshownthatnewsaffectsexchangerates.Combiningourperspectivesfo-cusesattentiononthecausallinksamongnews,orderow,andforexmovements,whichinourviewisaprimecandidateforfutureresearch.Itwillbeofinterest,forexample,todeterminewhethernewsaffectsexchangeratesviaorderoworinstantaneously.25Inworkdonesubse-quentlytotheŽrstdraftofthispaper,EvansandLyons(2001)andKennethA.FrootandTarunRamadorai(2002)tacklepreciselythatissue.AsecondkeydirectionforfutureresearchispushingfartherwiththeimplicationsofVeronesi(1999)fortheanalysisofhigh-frequencynewseffects.PresentlywehaveveriŽedthatthekeymechanismthatampliŽestheeffectsofbadnewsingoodtimesinVeronesi’smodel—increasedstateuncertainty—isoperativeinthedata.However,onecouldpotentiallygofartherandexploitthebroaderimplicationsofVeronesi’sworkforourapproach,namelythatnewseffectsareingeneralafunctionofstateuncertainty,byincludinginteractionsofnewswithstateuncer-taintyinbothourconditionalmeanandcondi-tionalvariancespeciŽcations.Thiswouldbeparticularlyinterestingifdatawereavailableonexchangeratesandfundamentalsspanningbadaswellasgoodtimes,butasofthiswriting,suchdataremainelusive.Third,itwouldbeofinteresttoexplorenotonlytheeffectsofregularlyscheduledquanti-tativenewsonmacroeconomicfundamentals,butalsotheeffectsofirregularlyscheduled,qualitative“headlinenews,”asprices,andper-hapsorderow,mayreasonablybeexpectedtorespondtoboth.26Itisnotobvious,however,howtodosoinacompellingway;boththeconceptualandthepracticalcomplicationsseemdaunting.Fourth,itwillbeofinteresttoattemptananalysisofstructuralstability,asthemarketmaychangeitsviewaboutwhichnewsisim-portantforexchangerates,orabouthowtointerpretthesignofasurprise.Insomeinter-pretations,forexample,apositiveU.S.inationsurprisewouldtendtoproducedollardepreci-ation(e.g.,whentheU.S.centralbankreactionfunctionassignsrelativelylowweighttothelevelofination),whereasinotherinterpreta-tionsitwouldproducedollarappreciation(e.g.,whentheU.S.centralbankreactionfunctionshowsstrongpreferenceforlowination,asinTaylor,1993).Finally,welookforwardtocharacterizingthejointresponsesoftheforeignexchange,stock,andbondmarketstoreal-timenewssurprises.Responseshavenowbeenstudiedforeachmar-ketinisolation:FlemingandRemolona(1999)andBalduzzietal.(2001)studythebondmar-ket,MarkJ.FlanneryandArisProtopapadakis(2002)studythestockmarket,andthispaper,ofcourse,studiestheforeignexchangemarket.Amultivariateframework,however,willfacilitateanalysisofcross-marketmovementsandinter-actions,orlackthereof,whichmayforexampleshedlightonagents’viewsregardingcentralbankreactionfunctions.REFERENCESAlmeida,Alvaro;Goodhart,CharlesA.E.andPayne,Richard.“TheEffectsofMacroeco-nomicNewsonHighFrequencyExchangeRateBehavior.” JournalofFinancialand QuantitativeAnalysis ,September1998, 33 (3),pp.383–408. Andersen,TorbenG.andBollerslev,Tim.“Deut-scheMark-DollarVolatility:IntradayActiv-ityPatterns,MacroeconomicAnnouncements,andLongerRunDependencies.” Journalof Finance ,February1998, 53 (1),pp.219–65. Andersen,TorbenG.;Bollerslev,Tim;Diebold,25Forinstance,KennethR.FrenchandRichardRoll(1986)andFlemingandRemolona(1999)botharguethatpubliclyavailablenewsmaybeincorporatedinpricesin-stantaneously,evenwithouttrading.26IndirectevidenceisprovidedbyDirkEddelbu¨ttelandThomasH.McCurdy(1998),whoreportsigniŽcantlyheight-enedforeignexchange-ratevolatilityinresponsetothemerefrequencyofheadlinenewsitemsontheReutersnewsscreen.59VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS FrancisX.andEbens,Heiko.“TheDistribu-tionofRealizedStockReturnVolatility.” JournalofFinancialEconomics ,July2001, 61 (1),pp.43–76. Andersen,TorbenG.;Bollerslev,Tim;Diebold,FrancisX.andLabys,Paul.“TheDistributionofRealizedExchangeRateVolatility.” Jour- naloftheAmericanStatisticalAssociation , March2001a, 96 (453),pp.42–55. .“ModelingandForecastingRealizedEconometrica,2003,71(forth-coming).Balduzzi,Pierluigi;Elton,EdwinJ.andGreen,T.Clifton.“EconomicNewsandtheYieldCurve:EvidencefromtheU.S.TreasuryMarket.”SalomonCenterWorkingPaperNo.S-98-5,NewYorkUniversity,1998. .“EconomicNewsandBondPrices:EvidencefromtheU.S.TreasuryMarket.” JournalofFinancialandQuantitativeAnal- ysis ,December2001, 36 (4),pp.523–43. Barberis,Nicholas;Shleifer,AndreiandVishny,Robert.“AModelofInvestorSentiment.” JournalofFinancialEconomics ,September 1998, 49 (3),pp.307–43. Bessembinder,Hendrik.“Bid–AskSpreadsintheInterbankForeignExchangeMarkets.” JournalofFinancialEconomics ,June1994, 35 (3),pp.317–48. Bollerslev,Tim;Cai,JunandSong,FrankM.“IntradayPeriodicity,LongMemoryVolatil-ity,andMacroeconomicAnnouncementEf-fectsintheU.S.TreasuryBondMarket.” JournalofEmpiricalFinance ,May2000, 7 (1),pp.37–55. Bollerslev,Tim;Chou,RayY.andKroner,KennethF.“ARCHModelinginFinance:AReviewoftheTheoryandEmpiricalEvi-dence.”JournalofEconometrics,April–May1992,52(1–2),pp.5–59.Bollerslev,TimandDomowitz,Ian.“TradingPat-ternsandPricesintheInterbankForeignEx-changeMarket.”JournalofFinance,September1993,48(4),pp.1421–43.Brandt,MichaelW.;Edelen,R.andKavajecz,K.A.“LiquidityintheU.S.TreasuryMarket:AsymmetricInformationandInventoryEf-fects.”Unpublishedmanuscript,UniversityofPennsylvania,2001.Chesher,AndrewandJewitt,Ian.“TheBiasofaHeteroskedasticityConsistentCovarianceMatrixEstimator.”Econometrica,September1987,55(5),pp.1217–22.Cheung,Yin-WongandChinn,MenzieDavid.“CurrencyTradersandExchangeRateDy-namics:ASurveyoftheU.S.Market.” Jour- nalofInternationalMoneyandFinance , August2001, 20 (4),pp.439–71. Cheung,Yin-WongandWong,ClementYuk-Pang.“ASurveyofMarketPractitioners’ViewsonExchangeRateDynamics.” Jour- nalofInternationalEconomics ,August 2000, 51 (2),pp.401–19. Conrad,Jennifer;Cornell,BradfordandLandsman,Wayne.“WhenIsBadNewsReallyBadNews?”Unpublishedmanuscript,UniversityofNorthCarolina-ChapelHill,2001.Dacorogna,MichelM.;Mu¨ller,UlrichA.;Nagler,RobertJ.;Olsen,RichardB.andPictet,OlivierV.“AGeographicalModelfortheDailyandWeeklySeasonalVolatilityintheForeignExchangeMarket.” JournalofInternational MoneyandFinance ,August1993, 12 (4),pp. 413–38. Danielsson,JonandPayne,Richard.“RealTrad-ingPatternsandPricesinSpotForeignEx-changeMarkets.”Unpublishedmanuscript,LondonSchoolofEconomics,1999.David,AlexanderandVeronesi,Pietro.“InationandEarningsUncertaintyandtheVolatilityofAssetPrices:AnEmpiricalInvestigation.”Unpublishedmanuscript,UniversityofChi-cago,2001.Diebold,FrancisX.andLopez,Jose.“ModelingVolatilityDynamics,”inKevinHoover,ed.,Macroeconometrics:Developments,tensionsandprospects.Boston,MA:KluwerAca-demicPress,1995,pp.427–72.Dominguez,KathrynM.“TheMarketMicro-structureofCentralBankIntervention.”Na-tionalBureauofEconomicResearch(Cambridge,MA)WorkingPaperNo.7337,1999.Eddelbu¨ttel,DirkandMcCurdy,ThomasH.“TheImpactofNewsonForeignExchangeRates:EvidencefromHighFrequencyData.”Un-publishedmanuscript,UniversityofToronto,1998.Ederington,LouisH.andLee,JaeHa.“HowMarketsProcessInformation:NewsReleasesandVolatility.”JournalofFinance,Septem-ber1993,48(4),pp.1161–91.Engle,RobertF.andNg,VictorK.“MeasuringandTestingtheImpactofNewsonVolatili-ty.”JournalofFinance,December1993,48(5),pp.1749–78.60THEAMERICANECONOMICREVIEWMARCH2003 Evans,MartinD.D.andLyons,RichardK.“WhyOrderFlowExplainsExchangeRates.”Un-publishedmanuscript,UniversityofCalifor-nia-Berkeley,2001. .“OrderFlowandExchangeRateDy-namics.” JournalofPoliticalEconomy ,Feb- ruary2002, 110 (1),pp.170–80. Flannery,MarkJ.andProtopapadakis,Aris.“MacroeconomicFactorsDoInuenceAg-gregateStockReturns.” ReviewofFinancial Studies ,Summer2002, 15 (3),pp.751–82. Fleming,MichaelJ.andRemolona,EliM.“WhatMovestheBondMarket?”EconomicPolicyReview,FederalReserveBankofNewYork,December1997,3(4),pp.31–50. .“PriceFormationandLiquidityintheU.S.TreasuryMarket:TheResponsetoPub-licInformation.” JournalofFinance ,Octo- ber1999, 54 (5),pp.1901–15. Frankel,JeffreyA.andRose,AndrewK.“Em-piricalResearchonNominalExchangeRates,”inGeneGrossmanandKennethRogoff,eds.,Handbookofinternationaleco-nomics.Amsterdam:North-Holland,1995,pp.1689–1729.French,KennethR.andRoll,Richard.“StockReturnVariances:TheArrivalofInformationandtheReactionofTraders.”JournalofFi- nancialEconomics ,September1986, 17 (1), pp.5–26. Froot,KennethA.andRamadorai,Tarun.“Cur-rencyReturns,InstitutionalInvestorFlows,andExchange-RateFundamentals.”NationalBureauofEconomicResearch(Cambridge,MA)WorkingPaperNo.9101,2002.Gallant,A.Ronald.“OntheBiasinFlexibleFunctionalFormsandanEssentiallyUnbi-asedForm:TheFourierFlexibleForm.” JournalofEconometrics ,February1981, 15 (2),pp.211–45. Goodhart,CharlesA.E.;Hall,StephenG.;Henry,S.G.BrianandPesaran,Bahram.“NewEffectsinaHigh-FrequencyModeloftheSterling-DollarExchangeRate.”JournalofAppliedEconometrics,January–March1993,8(1),pp.1–13.Goodhart,Charles;Ito,TakatoshiandPayne,Richard.“OneDayinJune1993:AStudyoftheWorkingoftheReuters2000–2Elec-tronicForeignExchangeTradingSystem,”inJeffreyA.Frankel,GiampaoloGalli,andAl-bertoGiovannini,eds.,Themicrostructureofforeignexchangemarkets.Chicago:Univer-sityofChicagoPressforNBER,1996,pp.107–79.Harvey,CampbellR.andHuang,RogerD.“TheImpactoftheFederalReserveBank’sOpenMarketOperations.” JournalofFinancial Markets ,April2002, 5 (2),pp.223–57. Hasbrouck,Joel.“SecurityBid/AskDynamicswithDiscretenessandClustering:SimpleStrategiesforModelingandEstimation.” JournalofFinancialMarkets ,February 1999, 2 (1),pp.1–28. Hautsch,NikolausandHess,Dieter.“What’sSur-prisingaboutSurprises?TheSimultaneousPriceandVolatilityImpactofInformationArrival.”Unpublishedmanuscript,Univer-sityofKonstanz,Germany,2001.Johnson,TimothyC.“ReturnDynamicsWhenPersistenceIsUnobservable.” Mathematical Finance ,October2001a, 11 (4),pp.415–45. .“Volatility,Momentum,andTimeVaryingSkewnessinForeignExchangeRe-turns.”Unpublishedmanuscript,LondonBusinessSchool,2001b.Kuttner,KennethN.“MonetaryPolicySurprisesandInterestRates:EvidencefromtheFedFundsFuturesMarket.” JournalofMonetary Economics ,June2001, 47 (3),pp.523–44. Mark,NelsonC.“ExchangeRatesandFunda-mentals:EvidenceonLong-HorizonPredict-ability.”AmericanEconomicReview,March1995,85(1),pp.201–18.Mark,NelsonC.andSul,Donggyu.“NominalExchangeRatesandMonetaryFundamen-tals:EvidencefromaSmallPost-BrettonWoodsPanel.” JournalofInternationalEco- nomics ,February2001, 53 (1),pp.29–52. McQueen,GrantandRoley,V.Vance.“StockPrices,News,andBusinessConditions.” Re- viewofFinancialStudies ,Fall1993, 6 (3),pp. 683–707. Meese,RichardA.andRogoff,Kenneth.“Empir-icalExchangeRateModelsoftheSeventies:DoTheyFitOutofSample?”JournalofInternationalEconomics,February1983,14(1–2),pp.3–24.Mu¨ller,UlrichA.;Dacorogna,MichelM.;Olsen,RichardB.;Pictet,OlivietV.;Schwarz,Mat-thiasandMorgenegg,Claude.“StatisticalStudyofForeignExchangeRates,EmpiricalEvidenceofaPriceChangeScalingLaw,andIntradayAnalysis.”JournalofBankingandFinance,December1990,14(6),pp.1189–208.61VOL.93NO.1ANDERSENETAL.:MICROEFFECTSOFMACROANNOUNCEMENTS Payne,Richard.“AnnouncementEffectsandSeasonalityintheIntra-dayForeignEx-changeMarket.”Unpublishedmanuscript,LondonSchoolofEconomics,1996.Pearce,DouglasK.andRoley,V.Vance.“StockPricesandEconomicNews.” Journalof Business ,January1985, 58 (1),pp.49–67. Rao,C.Radhakrishna.“EstimationofHet-eroscedasticVariancesinLinearModels.”JournaloftheAmericanStatisticalAssocia-tion,March1970,65(1),pp.161–72.Rudebusch,GlennD.“DoMeasuresofMone-taryPolicyinaVARMakeSense?”Interna-tionalEconomicReview,November1998,39(4),pp.907–31.Taylor,JohnB.“DiscretionversusPolicyRulesinPractice.”Carnegie-RochesterConferenceSeriesonPublicPolicy,December1993,39(0),pp.195–214.Urich,ThomasJ.andWachtel,Paul.“TheEf-fectsofInationandMoneySupplyAn-nouncementsonInterestRates.”JournalofFinance,September1984,39(4),pp.1177–88.Veronesi,Pietro.“StockMarketOverreactiontoBadNewsinGoodTimes:ARationalEx-pectationsEquilibriumModel.”ReviewofFi-nancialStudies,Winter1999,12(5),pp.975–1007.Zarnowitz,VictorandLambros,LouisA.“Con-sensusandUncertaintyinEconomicPredic-tion.” JournalofPoliticalEconomy ,June 1987, 95 (3),pp.591–621. 62THEAMERICANECONOMICREVIEWMARCH2003