What Explains the Stock Market's Reaction to Federal Reserve Policy?

What Explains the Stock Market’s Reaction to Federal Reserve Policy? ∗ Ben S. Bernanke Kenneth N. Kuttner March 2004 Abstract This paper analyzes the impact of changes in monetary policy on equity prices, with the objectives both of measuring the average reaction of the stock market and alsoofunderstanding theeconomic sources ofthatreaction. Wefindthat,onaverage, ahypothetical unanticipated 25-basis-point cut inthefederal funds ratetarget isassociatedwithaboutaonepercentincreaseinbroadstockindexes. AdaptingamethodologyduetoCampbell(1991)andCampbellandAmmer(1993),wefindthattheeffects of unanticipated monetary policy actions on expected excess returns account for the largest partoftheresponse ofstock prices. JELcodes: E44,G12. ∗ Board of Governors of the Federal Reserve System and Princeton University (Bernanke) and Oberlin CollegeandNBER(Kuttner). CorrespondencetoKenKuttner,EconomicsDepartment,RiceHall,10North ProfessorStreet,Oberlin,OH44074,e-mailkenneth.kuttner@oberlin.edu.ThankstoJohnCampbellforhis advice; to Jon Faust, Refet Gu¨rkaynak, Martin Lettau, Sydney Ludvigson, Athanasios Orphanides, Glenn Rudebusch, Brian Sack, Chris Sims, Eric Swanson, an anonymous referee and the associate editor of the JournalofFinancefortheircomments;andtoPeterBondarenkoforresearchassistance.Theviewsexpressed herearesolelythoseoftheauthors,andnotnecessarilythoseoftheFederalReserveSystem.

1 Introduction Theultimateobjectivesofmonetarypolicyareexpressedintermsofmacroeconomicvariablessuchasoutput,employment,andinflation. However,theinfluenceofmonetarypolicy instrumentsonthesevariablesisatbestindirect. Themostdirectandimmediateeffectsof monetarypolicyactions,suchaschangesinthefederalfundsrate,areonthefinancialmarkets; by affecting asset prices and returns, policymakers try to modify economic behavior inwaysthatwillhelptoachievetheirultimateobjectives. Understandingthelinksbetween monetary policy and asset prices is thus crucially important for understanding the policy transmissionmechanism. This paper is an empirical study of the relationship between monetary policy and one of the most important financial markets, the market for equities. According to the conventional wisdom, changes in monetary policy are transmitted through the stock market via changes in the values of private portfolios (the “wealth effect”), changes in the cost of capital, and by other mechanisms as well. Some observers also view the stock market as an independent source of macroeconomic volatility, to which policymakers may wish to respond. For these reasons, it will be useful to obtain quantitative estimates of the links between monetary policy changes and stock prices. In this paper we have two principal objectives. First, we measure and analyze in some detail the stock market’s response to monetary policyactions, both in the aggregate and at the levelof industryportfolios. Second,wetrytogainsomeinsightsintothereasonsforthestockmarket’sresponse. Estimating the response of equity prices to monetary policy actions is complicated by thefactthatthemarketisunlikelytorespondtopolicyactionsthatwerealreadyanticipated. Distinguishing between expected and unexpected policy actions is therefore essential for discerning their effects. A natural way to do this is to use the technique proposed by Kuttner(2001), which uses federal funds futures data to constructa measure of “surprise” ratechanges.1 Toexplaintheeconomicreasonsfortheobservedmarketresponsetopolicy 1Cochrane and Piazzesi (2002) proposed using the change in term eurodollar rates to measure policy surprises,whileRigobonandSack(2002)utilizedtheeurodollarfuturesrate. Whilethesemeasuresprovide informative gauges of interest rate expectations over a slightly longer horizon, Gu¨rkaynak et al. (2002) 1

surprisesrequiresanassessmentofhowthosepolicysurprisesaffectexpectationsoffuture interest rates, dividends,and excess returns. To do this,we adapt the procedure developed byCampbell(1991)andCampbellandAmmer(1993),whichusesavectorautoregression (VAR)tocalculaterevisionsinexpectationsofthesekeyvariables. Theresultspresentedinsection2ofthepapershowthatthemarketreactsfairlystrongly to surprise funds rate changes. Specifically, for a sample consisting of the union of days with a change in the target funds rate target and days of meetings of the Federal Open Market Committee (FOMC), we estimate that the CRSP value-weighted index registers a one-day gain of roughly one percent in response to a hypothetical surprise 25-basis-point easing. The market reacts little, if at all, to the component of funds rate changes that are anticipatedbyfuturesmarketparticipants. Acomparablereactionisobservedatamonthly unitofobservation. These results are broadly consistent with those of other studies which have looked at the link between monetary policy and the stock market. Thorbecke (1997), for example, documented a response of stock prices to shocks from an identified vector autoregression (VAR);inasimilarvein,Jensenetal. (1996)andJensenandMercer(1998),examinedthe market’sresponsetodiscountratechanges. Thispaperimprovesontheseearliereffortsby usingameasureofmonetarypolicybasedonfuturesdata,whichmorecleanlyisolatesthe unanticipatedelementofpolicyactions. Inthatsense,thispaperresemblesthemorerecent workofRigobonandSack(2002),whoreportedasignificantresponseofthestockmarket tointerestratesurprisesderivedfromeurodollarfutures. Thatpaper’smaininnovationwas the use of a novel, heteroskedasticity-based estimator to correct for possible simultaneity bias,anapproachsubsequentlyextendedbyCraineandMartin(2003). Theanalysisinthis papertakesamoreconventionalevent-studyapproach,whilecontrollingdirectlyforcertain kinds of information jointly affecting monetary policy and stock prices. Section 2 also includes an assessment of the results’ sensitivity to potential outliers, and an exploration ofcertainkindsofasymmetriesinthemarket’sresponse. Additionalanalysisdistinguishes showed that federal funds futures are the best predictors of target funds rate changes one to five months ahead. 2

between policy actions that affect the expected level of future interest rates, versus those thataffect onlythetimingofratechanges. Section 3 takes up the question of what explains equity prices’ response, an issue not addressedbyanyofthepaperscitedabove. Theapproachtakenhereisanadaptationofthe VAR method proposed by Campbell (1991) and Campbell and Ammer (1993). The main finding is that policy’s impact on equity prices comes predominantly through its effect on expectedfutureexcessequityreturns. Specifically,wefindthatwhileanunanticipatedrate cut (for example) generates an immediate rise in equity prices, it tends to be associated with an extended period of lower-than-normal excess returns. Some effect of policy on equity returns can be traced to revisions in cash flow forecasts, but very little is directly attributable to changes in expected real interest rates. One interpretation of this result is that monetary policy surprises are associated with changes in the equity premium, a point wediscussfurtherbelow. Butintheabsenceofafully-developedassetpricingmodel,itis impossibletodistinguishthisinterpretationfroma simplemarketoverreaction. Relativelyfewpaperstodatehaveattemptedtoprovideanexplanationforthemarket’s reaction to monetary policy. One effort along these lines is that of Patelis (1997), who also used the Campbell-Ammer framework to perform a decomposition similar to ours. Goto and Valkanov (2000) used a somewhat different VAR-based method to focus on the covariance between inflation and stock returns. Both relied on policyshocksderivedfrom identifiedVARs,however,ratherthanthefutures-basedsurpriseusedinouranalysis. Boyd et al. (2001) also considered the linkage between policy and stock prices. Their analysis focused on the market’s response to employment news, rather than to monetary policy directly,however. 3

2 The reaction of equity prices to changes in the target federal funds rate Thissectionfocusesontheimmediateimpactofmonetarypolicyonequityprices,bothfor broad stock market indices and for industryportfolios. As noted in the introduction, however, one difficulty inherent in measuringpolicy’seffects is that asset markets are forward looking and hence tend to incorporate any information about anticipated policy changes. Someeffortisthereforerequiredtoisolatetheunexpectedpolicychangewhichmightplausibly generate a market response. This does not say that asset prices respond to monetary policyonlywhentheFed surprisesthe markets,ofcourse. Naturally,assetpriceswillalso respond to revisions in expectations about future policy, which in turn may be driven by newsaboutchangingeconomicconditions. Ourfocusonunexpectedpolicyactionsallows ustocircumventdifficultissuesof endogeneityandsimultaneity,anddiscernmoreclearly thestockmarketreactiontomonetarypolicy. One convenient,market-basedway toidentifyunexpectedfundsrate changes relieson the price of federal funds futures contracts, which embody expectations of the effective federalfundsrate,averagedoverthesettlementmonth.2 KruegerandKuttner(1996)found thatthefederalfundsfuturesratesyieldedefficientforecastsoffundsratechanges. Kuttner (2001) subsequently used these futures data to estimate the response of the term structure to monetary policy. The analysis in this section employs a similar method to gauge the response of equity prices to unanticipated changes in the federal funds rate from 1989 through2002. 2.1 Measuring the surprise element of policy actions Ameasureofthesurpriseelementofanyspecificchangeinthefederalfundstargetcanbe derivedfromthechangeinthefuturescontract’spricerelativetothedaypriortothepolicy 2Thecontracts,officiallyreferredtoas“30DayFederalFundsFutures,”aretradedontheChicagoBoard ofTrade.Theimpliedfuturesrateis100minusthecontractprice. 4

action. Foraneventtakingplaceondayd ofmonthm,theunexpected,or“surprise”target funds rate change can be calculated from the change in the rate implied by the currentmonthfuturescontract. Butbecausethecontract’ssettlementpriceisbasedonthemonthly average federal funds rate, the change in the implied futures rate must be scaled up by a factorrelatedtothenumberofdaysinthemonthaffectedbythechange, (cid:1) (cid:2) D ∆iu = D−d f m 0 ,d − f m 0 ,d−1 , (1) where∆iu istheunexpectedtargetratechange, f0 isthecurrent-monthfuturesrateandD m,d isthenumberofdaysinthemonth.3 Theexpectedcomponentoftheratechangeisdefined astheactualchangeminusthesurprise,or ∆ie = ∆i− ∆iu . (2) Getting the timing right is, of course, crucial for event-study analysis. Before 1994, when the Fed instituted its current policy of announcing changes in the funds rate target, market participants generally became aware of policy actions on the day after the FOMC’s decision, when it was implemented by the Open Market Desk. Following Rudebusch (1995) and Hilton (1994), we assign most pre-1994 rate changes to the date of the Desk’s implementation. As documented in Kuttner (2003), however, the sample contains several minor deviations from this pattern. Six of these correspond to days on which the Deskallowedthefundsratetodriftdownwardinadvance(andpresumablyinanticipation) of the FOMC’s decision, with the full awareness that its inaction would be interpreted as aneasingofpolicy. Aseventhexceptionoccurred onDecember18,1990,whentheBoard of Governors made an unusual late-afternoon announcement of a cut in the discount rate, 3Because the monthlyaverageofthe effectivefederalfundsrate onwhichthe contractis basedis very close to the averagetargetrate, this methodgenerallyprovidesa goodgaugeof the surprisechangein the targetfederalfundsrate. Inordertominimizetheeffectofanymonth-endnoiseintheeffectivefundsrate, however,theunscaledchangeintheone-monthfuturesrateisusedtocalculatethefundsratesurprisewhen thechangefallsononeofthelastthreedaysofthemonth.Also,whentheratechangeoccursonthefirstday ofthemonth, f1 isbeusedinsteadof f0 . SeeKuttner(2001)fordetails. m−1,D m,d−1 5

fromwhichmarketobservers(correctly)inferreda25-basis-pointrate cut. Thepolicyofannouncingtargetratechanges,whichbeganinFebruary1994,eliminates virtually all of the timing ambiguity associated with rate changes in the earlier part of the sample. Moreover, because the change in the target rate is usually announced prior to the closeof thefuturesmarket,theclosingfuturesprice generallyincorporatestheday’snews about monetary policy. The only exception is October 15, 1998, when a 25-basis-point rate cut was announced after the close of the futures markets. In this case, the difference between the opening rate on the 16th and the closing rate on the 15th is used to calculate thesurprise. 2.2 Baseline event study results One approachtomeasuringtheimpactofFederal Reserve policyonthestockmarketisto calculatethemarket’sreactiontofundsratechangesonthedayofthechange. Themarket may of course also react to the lack of a change in the funds rate target, if a change had beenanticipated. Becausethisapproachinvolveslookingattheresponsetospecificevents, it might be described as an “event-study” style of analysis. For the purpose of this paper, the relevantsample of events is defined as the unionof all days when the funds rate target was changed, and days corresponding to FOMC meetings. The first “event” in the sample is the June 1989 25-basis-point rate cut, and the last corresponds to the FOMC meeting in December 2002. The 17 September 2001 observation is excluded from the analysis, as thatday’srate cutoccurred on thefirst day of tradingfollowingtheSeptember 11 terrorist attacks. Altogether,thesamplecontains131observations. Table 1 presents a selection of descriptive statistics on the policy surprises and stock returns in our sample. The statistics are reported both for the pre-1994 period, when changes in the funds rate target were generally unannounced and frequently occurred between scheduled FOMC meetings, and the post-1994 period when all rate changes were announced, and mostcoincided withFOMC meetings. As measured bythe standard deviation,thetypicalfundsratesurpriseinbothperiodsisroughly10basispoints;bycontrast, 6

equity prices are half again as volatile post-1994 as pre-1994. In both subsamples, equity returns are roughly ten percent more volatile on the monetary policy“event” days than on “non-event”days,consistentwithpolicyactionsinducingamarketreactionofsomekind. BaselineestimatesofthereactionofequitypricestomonetarypolicyappearinTable2. Theresultsincolumn(a)ofthetablearebasedonaregressionoftheCRSPvalue-weighted returnontherawchangeinthefederalfundsratetarget, H =a+b∆i + ε , (3) t t t making no distinction between surprise and expected changes; H represents the stock ret turn,andi isthefundsratetarget. Theregressionusedfortheresultsincolumn(b) t H =a+be ∆ie+bu ∆iu+ ε , (4) t t t t distinguishesbetween expected and unexpected funds rate changes, ∆ie and ∆iu, usingthe t t decompositiondescribedaboveinsection2.1. In both specifications, the error term ε represents factors other than monetary policy t that affect stock prices on event days. These factors are assumed to be orthogonal to the changesinthefederalfundsrateappearingontheright-handsideoftheregression. Section 2.3belowdiscussesthevalidityofthisassumptioninsomedetail,andsection2.4presents resultsthatcontroldirectlyforoneobservablesourceofendogeneity. Although it has the expected negative sign, the response to the raw target rate change reported incolumn(a) of Table 2.2issmalland insignificant. Whenthe target rate change is broken down into its expected and surprise components, however, the estimated stock market response to the latter is negativeand highly significant: the results reported in column(b)implya−4.68%one-dayreturninresponsetoaonepercentagepointsurpriserate cut.4 The R2 indicates that 17% of the variance in equity prices on these “event” days is associated with news about monetary policy. While Fed policy accounts for a nontrivial 4VerysimilarresultsareobtainedusingtheS&P500inplaceoftheCRSPvalue-weightedreturn. 7

portion of the variance of stock returns on event days, clearly it is far from the only piece ofnewinformationaffectingstockreturns. The negativerelationshipbetween fundsrate surprises and stock returns is readily visible in Figure 1. Also apparent, however, are a number of observations characterized by very large changes in equity prices — some exceeding three standard deviations in magnitude. This naturally raises the question of whether the results reported in the first two columnsofTable2aresensitivetotheinclusionoftheseobservations. To determine which observations might have an unduly large effect on the regression results, we computed influence statistics for each observation in the sample. These statistics are calculated from the quadratic form ∆bˆ(cid:2) Σ ˆ−1∆bˆ , where ∆bˆ is change in the vector t t t ofregressioncoefficientsresultingfromdroppingobservationt,and Σ ˆ istheestimatedcovariance matrix of the coefficients. The distributionof these statistics, plotted in Figure 2, confirms that six observations, all with statistics in excess of 0.3, exert an unusually large influence on the estimates; the comparable statisticsfor the remainingobservationsare all well below0.2 and mostare less than 0.05. The sixobservationsassociated withthe large influencestatisticsarelabeledinFigure1: 8August1991,2July1992,15October1998,3 January2001,20March2001,and18April2001. Thefirsttwooftheseareassociatedwith events other than monetary policy actions, while the most recent four arguably represent unusualreactionstomonetarypolicyactions. Eachisinitsownwayisrevealing. Allthreeofthecandidateoutliersoccurringduringtheeasingcyclethatbeganin2001 areclassifiedassuchbecauseoftheirabnormallylargereactionstothefundsratesurprises. Theunexpected50-basis-pointintermeetingratereductionson3Januaryand18Aprilwere both greeted euphorically, with one-day returns of 5.3% and 4.0% respectively. The 50basis-pointrate cut on20 March was receivedless enthusiastically,however. Eventhough the cut was more or less what the futures market had been anticipating, financial press reportedthatmanyequitymarketparticipantswere“disappointed”theratecuthadn’tbeen anevenlarger75basispointaction. Consequently,themarketlostmorethan2%. Another unusually vehement reaction to a Fed action is associated with the 25-basis- 8

point intermeeting rate cut on 15 October 1998, which was taken in response to unsettled conditions in the financial markets — specifically, the deteriorating situations in Asia and Russia. Forwhateverreason,theunexpectedintermeetingcutliftedequitiesover4%. Thestockmarketfelllessthan0.3%on2July1992. Whatmakesthisreactionunusual, however,isthefactthatitcameonadaywhentheFedunexpectedlycutthefundsratetarget by 50 basis points. The decision to cut was no doubt influenced by that day’s unusually bleakemploymentreport,inwhichreportedpayrollemploymentfellby117thousand. This raisestheissuethatsomeofthe“surprise”ratechangesinthesamplemayinfactrepresent endogenous responses to economic news, such as the employmentreport. This possibility isinvestigatedingreater detailbelowinsection2.4. The final candidate outlier is 21 August 1991, when the CRSP value-weighted index rose2.7%onadayassociatedwithanFOMCmeeting. Thefuturesmarkethadapparently priced insomepossibilityof a rate cuton thatday,butthe FOMC’sdecisiontoleave rates unchanged generated a small,positivesurprise. The financial press reported that the stock marketjumpwasaresponsetotheresolutionoftheattemptedcoupinRussia—clearlyan eventwithnodirectrelationtothatday’sFOMC decision. Two additionalobservationsare highlightedin Figure 1: 17 May and 16 August1994. While their relatively low influence statistics (0.05 and 0.04) do not qualify them as outliers, they stand out as unusual instances in which equities rose in spite of significant, positive funds rate surprises. As noted in Kuttner (2001), a similarly anomalous response isobservedintheresponseofbondyieldsonthosedates. Thereasonseemstobethatboth oftheselarger-than-expected50-basis-pointratehikeswereaccompaniedbystatementsby the FOMC suggestingthat further rate increases were notimminent. Thisinterpretation is consistent with the results reported below in section 2.6 indicating that the three-monthaheadfuturesratesfellonthedatesinquestion. Columns (c) and (d) of Table 2 show the effect of dropping the six candidate outliers identified above. (The two observations from 1994 are retained.) The estimated response tofundsrate surprisesisstillnegativeandsignificant,butsmallerinmagnitude: −2.55,as 9

opposedto−4.68. Theresponsetotheexpectedcomponentissmaller(andnownolonger significantat the0.05level),asis theresponseto theraw fundsrate change incolumn(c). Excludingthesixoutliersalsodecreases theR-squaredfrom0.17to0.05. 2.3 Orthogonality revisited As noted above, the event-study results reported in section 2.2 rely on the assumption that the error term is orthogonal to funds rate changes. One reason for a violation of this condition would be a contemporaneous response of monetary policy to the stock market. Thereare,however,noclearexamplesofinstancesinwhichadropinequitypricesledthe FOMC to cut rates, or the inverse. Even in monthly data, evidence for such a systematic reaction is elusive.5 Moreover, to the extent the FOMC did respond in this way, it would tendtoreducethesizeoftheestimatedresponsetothefundsratesurprise. The orthogonality condition would also fail to hold if monetary policy and the stock market both responded jointly (and contemporaneously) to new information. For example,therelease ofdataindicatingweaker-than-expectedeconomicgrowthwouldplausibly cause the stock market to decline, and make a cut in the funds rate target more likely.6 As in the case of a direct policy response to the stock market, the resulting tendency for rate cutstobeassociatedwithstockmarketdeclineswouldleadtoadownwardbiasinthesize of policy’s estimated market impact. A similarly attenuated reaction would be observed if surprise policy actions were thought to reveal private information about the state of the economy.7 Instancesofdirect,same-daypolicyresponsestoeconomicnewsarerareinoursample — at least in recent years, when the FOMC meeting schedule dictated the timing of most policy actions. During the pre-1994 subsample, however, it was not uncommon for the 5See, for example, Bernanke and Gertler (1999)and Fuhrer and Tootell (2003). Some evidenceto the contrarywasobtainedbyRigobonandSack(2003),however. 6The17September2001isanextremeexampleofjustsuchajointresponse:theFed’s50basispointrate cutandthestockmarket’ssharpdropwerebothclearlyspurredbythepreviousweek’sterroristattacks. 7Romer and Romer (2000) suggested that such an information advantage could account for the bond market’sresponsetomonetarypolicy.HoweverFaustetal.(2003)foundlittleevidencetosupportthisview. 10

FOMC to cut rates on the heels of weaker-than-expected employment data. In fact, ten of the 23 rate cuts from June 1989 through July 1992 coincided with the release of the employment report. The analysis in section 2.4 below addresses this issue directly by allowingforadifferentmarketresponseonemploymentrelease days. Recent studies have proposed two generic solutions to the endogeneity and jointresponse issues. One is to use intraday data in a relatively narrow “event window” surrounding the FOMC’s announcement, thus distinguishingthe impact of the policy change from the effects of news arriving earlier or later in the day. Applying this approach, Gu¨rkaynak et al. (2004) reported, in work subsequent to ours, an equity price response thatwasvirtuallyidenticaltothatobtainedfromdailydata. (Usingintradaydatadid,however, result in a considerable improvementin the R2.) Also in subsequentwork, D’Amico and Farka (2003) uncovered a similar reaction using an approach incorporating intraday datainaVARspecification. The other generic solution is more statistical in nature. Rigobon and Sack (2002), for example, used an estimator which, by exploiting the heteroskedasticity introduced by exogenous monetary policy actions, yields consistent estimates of the market’s response. Inarelatedapproach,CraineandMartin(2003)developedamultivariatefactormodelthat allows all asset prices to respond to common, unobserved information shocks. In the end, however, both studiesreport results that are very close to those obtained from event-study methods. A correlation between the error term and the regressors in (4) could also arise if the regressors were measured with error. This possibilitywas explored by Poole et al. (2002) inthecontextofTreasuryyields’responsetomonetarypolicy. Theyassumedthemeasurement error in the funds rate surprise was uncorrelated with other factors affecting yields, turningitintoaclassicalerrors-in-variablesproblem. Togaugethesizeofthemeasurement error, Poole et al. calculated thevariance ofthe futuresrate on dayswhen theactual funds ratechange wasinlinewiththeconsensusmarketexpectationsreported bytheWallStreet Journal;usingthisestimate, theyfoundthe attenuationin thebond market’sresponsewas 11

typicallyontheorderoffivetotenpercent. Overall,thealternativeeconometricmethodsthathavebeenusedtocorrectformismeasurement of the funds rate surprises uniformly yield results similar to those relying on the event-studymethodologyused in section 2.2. Moreover, to the extent that the event-study resultsarebiased,thatbiastendstounderstatethetrueresponsetomonetarypolicy. Thus, it seems safe to proceed using the event-studyapproach, bearing in mind that it may yield slightlyconservativeestimatesofthestockmarket’sreactiontomonetarypolicy. 2.4 Employment releases and subsample stability This section investigatesthe robustness of the results reported in section 2.2 along two dimensions. One issue has to do with the joint response of monetary policy and the stock market to economic news. As noted above, ten funds rate cuts in the pre-1994 part of the sampleoccurredonthesamedayastheemploymentreport. After1994,withratechanges moreorlessdictatedbytheexogenousschedulingofFOMCmeetings,thisbecomeslessof anissue. AsisevidentinFigure1,theseobservationsarecharacterizedbylittle,ifany,correlationbetween thefunds rate surprise and the stockreturn. In these instances, the “good news” for the stock market represented by the Fed’s actions seems to have been almost exactly offset by the “bad news” about economic activity contained in the employment report. Another issue concerns the stability of the estimated relationship. The avoidance of a same-day response to employment reports (and other economic news) is one possible reason the relationship might have changed in the early 1990s. The FOMC’s practice of explicitly announcing rate changes, which began in February 1994, may also have altered thestockmarket’sresponsetomonetarypolicy. Toexplorethepossibilityofdifferentresponseseitherpost-1994,orondaysassociated withemploymentreleases,weinteractthesurpriseratechangewithdummyvariables: one equal to 1 starting with the 4 February 1994 observation, and another equal to 1 on the days of pre-1994 employment releases. Table 3 reports the response of the CRSP value- 12

weightedindextosurpriseratechangesinthepresenceoftheseinteractivedummies. Like Table 2, columns (a) and (b) give the results for the full sample, and columns (c) and (d) givetheresultsforthesampleexcludingthesixcandidateoutliersidentifiedabove. At first glance, the results in column (a) appear to show that the entire equity price response can be traced to the post-1994 period. The coefficient on the surprise itself is only−1.25andinsignificant;thatonthesurpriseinteractedwiththepost-1994dummyisa highlysignificant−6.87. Thisconclusionwouldbepremature,however,asthisregression neglects the possibilityof endogeneity in the policy response prior to 1994. Including the surprise interacted with the employment release dummy as in column (b), increases the magnitude of the surprise response, and the positive interaction term implies a near-zero response to policy when it coincides with an employment release. These coefficients are statistically significant at only the 0.10 level however, and the post-1994 interaction term remainslarge andhighlysignificant. The significance of the post-1994 term is heavily influenced by the six outliers identified above, however. With those observations excluded, as in column (c), post-1994 rate surpriseshaveonlyaslightlylargereffect,andthedifferenceisnotstatisticallysignificant. The coefficient on the surprise itself −2.29, and significant at the 0.05 level. But when theemploymentinteractiontermisincluded,thesurprisecoefficientgrowsto−3.57. This effect is almost exactly offset for employment release days by the 3.33 coefficient on the interaction term. Both are now highly significant, and the post-1994 dummy remains insignificant. Thus,ifthesixcandidateoutliersarediscardedasunrepresentative,thereisno evidenceofabreakin1994. Furthermore,theresultsconfirmthattheendogeneityproblem discussedabovereducestheOLS estimatesofthemarket’sresponsetopolicysurprises. 2.5 Asymmetries Another set of questions concerns asymmetries, broadly defined: the possibility that the equity price response to monetary policy depends on the direction of the action, or on the contextinwhichitoccurred. Asaboveinsection2.4,interactivedummyvariablesareused 13

intheregressiontoinvestigatethesequestions. One possibility is that the magnitude of the market’s response depends the sign of the surprise. To allow for this, a dummy variable was set to 1 for those 37 observations with positivesurprises. An interaction term involving this dummy and the surprise rate change wasthenincludedintheregression. Theinteractiveterminvolvingtheemploymentrelease is also included, in order to pick up the smaller impact of funds rate surprises on employment release days. As above, the regressions are run with and without the six candidate outliers identified earlier. The results reported in Table 4 provide weak support at best for this form of asymmetry. For the full sample, in column (a) of the table, the coefficient on theinteractiontermindicatesasmallereffectofpositivesurprises,butthedifferenceisnot statisticallysignificant. Thereisvirtuallynodifferencefortheno-outliersample,shownin column(d). Arelatedkindofasymmetrycanbemodeledbyincludinginteractivedummiesforrate changes associated with increases in the funds rate, and with surprises associated with no changeinthefundsrate. Thefullsamplecontains14observationsoftheformer,and76of thelatter. The resultsofthisexerciseappear incolumns(b) and(e) ofthetablefor thefull andno-outliersamples. Again,thestatisticallyinsignificantcoefficientinthe rate increase interaction variable suggests the direction of movement is not an important determinant of the market’s reaction. The positive and statistically significant estimated coefficient on the“no change” interactionvariabledoes,however,indicatethatthe marketrespondsvery little,ifatall,topolicy“inactions.” Thispresumablymeansthatthefailuretomoveatany specificFOMC meetingmaybeviewedlargelyaspostponingtheinevitable. A third sort of asymmetry has to do with the context of the rate decision: whether it was taken at an FOMC meeting (109 observations), or represented a change in the direction of short-term interest rates (five observations). Interaction terms involving suitablyconstructeddummyvariablesareagainusedtocapturepossibledifferencesinthemarket’s response. The sign of the FOMC interaction term is unclear a priori. Decisions taken at FOMC meetings may be less subject to the sort of endogeneity issues discussed above, 14

whichwouldtendtoincrease theimpactofratechangesonthesedays. Ontheotherhand, intermeeting changes (at least those not associated with employment reports) may convey anurgencyonthepartoftheFOMCwhichwouldtendtoincreasethesizeoftheresponse. Totheextentthatinterestratereversalshavealargerimpactonexpectedfutureinterestrates thanotherratechanges,thesechangesinthetargetfederalfundsratewouldbeexpectedto elicitalarger stockmarketresponse. Columns (c) and (f) of Table 4 show the results from a regression that includes the FOMC and reversal interaction terms, along with employment report and post-1994 regressors. The coefficient on the surprise term remains an economically and statistically significant−3.97forthefullsample,and−3.67fortheno-outliersample. Inthefullsample,themeasuredresponseissmalleronFOMCdays;thisdifferencedisappears,however, whenthecandidateoutliersareexcluded. Reversalsseemtohavealargeadditionalimpact on the stock market: −6.33 for the full sample, and −17.62 for the no-outlier sample. In the lattercase, theimplausiblylarge estimateisdrivenalmostentirelyby thesingleobservation in the southeast corner of Figure 1, corresponding to the first rate increase in 1994. Clearly,reversalsinthedirectionofratechangeshaveoccasionallybeenmetwithextreme marketreactions,whichaccountsfortheexaggeratedresponse. Withonlyfiveobservations inthesample,however,inferenceontheadditionalstockmarketimpactofreversalsishazardous at best. “Dummying out” these observations at least provides further confirmation thatthebaselineresultsare notdependentontheinclusionoftheseevents. Taken together, the results presented above confirm the existence of a strong one-day reactionofthestockmarkettounanticipatedchangesinfederalfundsrate. Justhowstrong thisresponseisdependsonwhetherthehandfulofpotentiallyanomalousobservationsare viewed as representative, or discarded as outliers. The estimated response is stable over time,oncethetendencyfortheFOMCtoreacttoemploymentnewsintheearlypartofthe sample is controlled for. The estimated reaction does, however, appear to be smaller (or nonexistent)forpolicysurprisesassociatedwithnochangeinthefundsratetarget. 15

2.6 Timing versus level surprises Whiletheresultspresentedaboveareconsistentwithastrongresponseofequityreturnsto funds rate surprises, that response is anything but uniform. In some cases, the reaction is muted, while in others the reaction seems out of proportion with the size of the measured surprise. One explanation for the lack of uniformity is that funds rate surprises differ in their impact on expected future short-term interest rates. Many of the surprises in the sample may have been interpreted as an advancement or postponement of a more-or-less inevitable change in policy, while others were viewed as altering the expected path of the fundsrateformonthstocome. Surpriseswithamoredurableonpolicyexpectationswould naturally tend to have a larger effect on equity prices than those which merely altered the timingofpolicyactions. One way to gauge policy surprises’ impact on expected future short-term rates is to examine the relationship between the surprises and the change in the fed funds futures rates in subsequent months. This relationship is depicted in Figure 3, which plots the change in the three-month-ahead futures rate against the funds rate surprise for the 131 observations in our June 1989 through December 2002 sample. The 45 degree line in the figure corresponds to a one-for-one response of the three-monthfutures rate to the current monthfunds rate surprise. Observationslying alonga shallowerline (i.e., those belowthe 45degreelineinthenortheastquadrantandaboveinthesouthwestquadrant)aretherefore those associated with a less than one-for-one effect on three-month-ahead expectations; those lying along a steeper line had a greater-than one-for-one effect. As noted above, the announcementsaccompanyingthetworatehikesinMayandAugust1994actuallylowered three-month-ahead interest rate expectations, and as a result those two observationsfall in thesoutheastquadrant. Regressing the change in the three-month-ahead futures rate on the policy surprise yields an estimated slope coefficient of 0.65, as shown in column (a) of Table 5. This suggeststheimpactofpolicysurprisesonexpectationsistypicallymuchlessthanone-forone; the difference is significant at the 0.01 level. A plausible interpretation of this result 16

is that many of the unexpected funds rate changes in the sample are to a large extent surprises only with regard to the timing of policy actions. As shown in columns (b) through (d),FOMCmeetingsand“nochange”surprisestendtobeassociatedwithanevensmaller responseofexpectations. Inorder todeterminetheextenttowhichdifferencesinpolicysurprises’impactonexpectationscanhelpexplainthestockmarket’sresponse,ourapproachistodefineavariable reflecting the difference between the surprises’ effects on current and three-month-ahead interest rate expectations, and include this term in the equity return regressions. Specifically, our “timingsurprise” variableis defined as the difference between the change in the three-month-aheadfuturesrateandthecurrentfundsratesurprise,i.e.,theverticaldistance from each observation to the 45 degree line in Figure 3. The timing surprise for an action with equal effects on current and expected future interest rates would thus be zero; those witha smaller effect on expected future interestrates wouldbe negative. Resultsfrom the stockreturnregressionsthatincludethetimingsurprisetermappearinTable6. Forcomparisonpurposes,columns(a)and(c)ofthetablesimplyreproducethebaseline results reported earlier in Table 2, with and without the six candidate outliers. Columns (b) and (d) report the regression results when the timing surprise term is added to the regression. The inclusion of this term increases the magnitude of the coefficient on the current-month surprise, which goes from −4.68 to −6.20 for the full sample. Because this coefficient can now be interpreted as the impact of a funds rate surprise that changes expectationsby the sameamount(i.e., withthe timingsurprise equalto zero), thisimplies a larger stock price response to those policy surprises that affect the level of interest rates expectedtoprevailthree monthshence. Similarly, the statistically significant, negative coefficient on the timing surprise term saysthatsurpriseswithaless-thanone-for-oneimpactonexpectations(i.e.,thoseforwhich thechangeinthethree-monthfuturesrate issmallerthanthecurrent-monthsurprise)have a correspondingly smaller effect on stock prices. In the extreme case of a “pure” timing surprisewithnoeffectontheexpectedlevelofrates,theresponseisgivenbythedifference 17

betweenthetwocoefficients: −1.91forthefulland0.09fortheno-outliersample. (Neither is statistically significant at even the 0.10 level.) The results therefore show that policy actions affect stock returns only to the extent that they alter the expected level of rates in themonthsahead. 2.7 Results based on monthly data An alternative way to define the policy surprise is to focus on the expected change in policy at a regular, monthly horizon. Unlike the event study approach, the regular timing isamenabletothetimeseriesanalysisemployedbelowinsection3toassessthecausesof themarket’sresponse. Itisworthnotingthatinthisapproach,anymonthcouldpotentially contain a surprise policy action, and that a failure to change the funds rate target in any month could represent a policy surprise. Consequently, the monthly time-series approach is less susceptible to any sample selection issues that might arise in the context of the event-studymethodology. The use of monthly data calls for a slightly different gauge of unanticipated policy actions. Sincethepriceofthefederalfundsfuturescontractisbasedonthemonthlyaverage federalfundsrate,anaturaldefinitionofthemonth-t surprisewouldbe 1 D ∆ ¯i t u ≡ D ∑ i t,d − f t 1 −1,D , (5) d=1 where i t,d isthe fundsrate target onday d of montht,and f t 1 −1,D isthe rate corresponding totheone-monthfuturescontractonthelast(Dth)dayofmontht−1.8 Theexpectedfunds ratechangeisdefinedanalogouslyas ∆ ¯i t e ≡ f t 1 −1,D −i t−1,D . (6) The sum of the two is the average funds rate target in montht minus the target on the last 8Thesettlementpriceofthefederalfundsfuturescontractisdeterminedbytheaverageoverthecalendar month,carryingthepriorbusinessday’srateovertoweekendsandholidays. 18

day of month t−1. (The notation ∆ ¯ is used to distinguish this from the conventionallydefinedfirstdifferenceoperator.) This definition of the funds rate surprise raises a time aggregation issue. Measuring the surprise in terms of the average funds rate will tend to attenuate the size of the policy surprises,asdiscussedindetailinEvansandKuttner(1998). Unfortunately,withoutmaking specific assumptions about the days of possible rate changes, there is no clean way to correct for this problem.9 Consequently, some caution is required when interpreting the magnitudeofthesurprisesmeasuredinthisway. Itisalsoimportanttonotethattheendogeneity issue discussed above in section 2.4 is almost certainly going to be more relevant to monthly funds rate surprises than it was for the day-ahead surprises. Rate changes that wereunanticipatedasoftheendofthepriormonthmaywellincludeasystematicresponse toeconomicnews,suchasemployment,outputandinflation. The results shownin Table 7 supportthe viewthat the month-aheadsurprisesincorporate an endogenous reaction to economic developments. The table reports the parameter estimatesandR2 fromaregressionofthemonthlypolicysurprisesonthesurpriseelement ofkeyeconomicreports,calculatedasthedifferencebetweenthenumberreleasedandthe consensus expectation for that number, compiled by Money Market Services.10 Over the full May 1989 through December 2002 sample, there appears to be a significant withinmonthimpactofseveraldatareleasesonthefundsratetarget: nonfarmpayrolls,industrial production, retail sales, and core PPI, although these latter two have the “wrong” (i.e., negative)sign. This relationship seems to be much stronger in the early part of the sample, however. The second column of the table shows the results for the same regression estimated from May 1989 through September 1992 (the date of the last rate cut associated with an employment report). The Fed’s reaction to bad payroll employment news is now particularly pronounced. Moreover,theregressionaccountsfornearlyhalfofthevarianceofthefunds 9One solution would have been to assume that post-1994 rate changes were always expected to occur at scheduled FOMC meetings. The three intermeeting rate cuts in 2001 have made this assumption less plausible,however. 10WeareindebtedtoEricSwansonforhisassistancewiththesedata. 19

rate surprises. By contrast, in the more recent February 1994 through December 2002 subsample, there is very little evidence of a within-month reaction to economic news, as shown in the third column of the table. Only retail sales is significant, and the regression nowaccountsfor amuchsmallershareofthevarianceoffundsratesurprises.11 Table8reportstheresultsfromaregressionofthemonthlyCRSPvalue-weightedreturn theexpectedandunexpectedcomponentsofmonthlyfundsratechanges, H =a+be ∆ ¯ie+bu ∆ ¯iu+ ε . (7) t t t t Column (a) reports the estimates for the full sample, consisting of all 164 months from May 1989 through December 2002. As in the earlier results, there is a strong, statistically significantnegativeresponsetounanticipatedrateincreases,andlittleornoresponsetothe anticipated actions. The R¯2 indicates that nearly 7% of the monthly stock return variance canbetracedtounanticipatedpolicyactions. It is interesting to note that the magnitudeof the response, −11.43, is about twice that found in the event-study analysis. This difference in magnitudes is readily explained by the time aggregationissue alluded to earlier. In fact, if funds rate changes on average take place in the middle of the month (for example, if rate changes were distributed uniformly overthedaysofthemonth),thenthemagnitudeoftheestimatedmonthlysurpriseswillbe attenuatedby one-half,which wouldexplainthedoublingof the estimatedresponse ofthe stockprice. The negative relationship between policy surprises and stock returns is also evident in thescatterplotof thedatainFigure4. Asinthedailydata, anumberofobservationsstand out as potential outliers, again raising the question of whether the results are sensitive to their inclusion. As above, influence statistics were calculated for each observation in the sample; those with statistics in excess of 1.5 are flagged as outliers in the plot. (The most 11Again,retailsalesis significantwiththe“wrong”sign. Butthis resultis dueentirelytoananomalous 7%jumpinretailsalesinNovember2001,whichhappenedtooccurinamonthinwhichtheFedalsocutthe fundsratetarget. 20

conspicuous of these is the data point deep in the southwest quadrant, which corresponds toSeptember2001.) Droppingthesetenobservationsmakeslittledifferencetotheresults, however. Infact,asshownincolumn(b)ofTable8,theestimatedcoefficientof−14.26is somewhatlarger thanitisforthefullsample,andthe R¯2 risesto0.096. Themonthlydatacontainverylittleevidenceforthesortsofasymmetriesuncoveredin the daily data. As shown in columns(c) and (d), there is no indicationthat the stock price response dependson the signof the surprise,or on the directionof the rate change. Nor is there any evidence of a different response to policy reversals, or to the MMS employment surprises.12 Wehavesofarfocusedontheresponsesofbroadequityindexes,butofcourseitisalso possibletoexaminetheresponsesofmoredisaggregatedindexes. Table9reportsestimates of(7)forthetenindustryportfoliosconstructedfromCRSPreturnsasinFamaandFrench (1988).13 The most responsive industries are high tech and telecommunications, with coefficientshalf againas large thatfor the overallvalue-weightedindex. Onthe otherend of thespectrum,energyandutilitiesare onlyhalfasresponsiveastheoverallmarket,andthe relevant coefficients are statistically insignificant.14 The low R2s indicate that very little of those industries’ variance is associated with unexpected policy actions. The estimates’ precision is, however, not sufficient to reject the hypothesisof an equal reaction for all 10 industries. Anaturalquestionisthedegreetowhichthepatternofresponsesofindustryportfolios isconsistentwiththeimplicationsoftheCAPM—thatis,whethertheobservedresponses are proportional to the industries’ market “betas”. A straightforward way to address this questionistoobtainindustrybetasfromaregressionoftheexcessreturninindustryi,y i,t , 12Interestingly,theemploymentsurpriseisnegativeandsignificantinaunivariateregression(notreported), butbecomesinsignificantoncethefederalfundssurpriseisincluded. Thisisconsistentwiththefindingsof Boydetal. (2001),andcorroboratestheirconjecturethatthepolicyresponseaccountsforequities’perverse response. 13TheFama-Frenchportfoliodataareavailablefrommba.tuck.dartmouth.edu/pages/faculty/ken.french. 14Usingmethodssimilartoours,Guo(2002)foundthattheimpactofmonetarypolicyonstockpricesdoes notseemtodependonfirmcapitalization. 21

onthemarketexcessreturn,y M,t , y i,t = α + β i y M,t + ν t (8) estimated on the same May 1989 to December 2002 sample used to estimate the response to monetary policy.15 The industry response implied by the CAPM can then be expressed as: bˆu = β ˆ ×bˆu (9) i i wherebˆu istheestimatedresponseoftheCRSP value-weightedexcessreturntofundsrate surprises. Figure5plotsthesefittedresponsestomonetarypolicyagainsttheestimatedresponses, bˆu,reportedinTable9,alongwiththe80%confidenceintervalsassociatedwiththoseestii mates. Alsoplottedisthe45-degreelinethatthepointswouldlieoniftheCAPMperfectly accounted for variationacross industries. Althoughthe fit isnot perfect, the pointslineup reasonablywellalongthe45-degreeline,suggestingthattheone-factorCAPMdoesagood jobofexplainingtheobservedindustryvariation. High-tech’smeasuredsensitivitytomonetarypolicyisinfactsomewhatlessthanitsbetawouldimply,whiletelecommunications’ is somewhat greater. On the other end of the spectrum, the utilitiesand energy industries’ low market betas for the most part account for their muted response to monetary policy. The CAPM-implied response represented by the 45-degree lies within the confidence intervals associated with the estimated responses, although given the imprecision of those estimates,thisisclearlynotapowerfultest. 15Usingbetasbasedonthesumofcontemporaneousandlaggedmarketcovariances,asinCampbelland Vuolteenaho (2003), makes virtually no difference to the results. Campbell and Vulteenhaho’s proposed two-factordecompositionalsoyieldsverysimilarresultstothosereportedinthetext. 22

3 Policy, fundamentals and stock prices Having documented the reaction of equity returns to Federal Reserve policy in section 2 above, we now turn to the more difficult question of what explains the observed reaction. Therearethreebroadreasonswhyanunexpectedfundsrateincreasemayleadtoadecline instockprices: itmaybeassociatedwitha decrease inexpectedfuturedividends,a risein thefutureexpectedrealinterestratesusedtodiscountthosedividends,oranincreaseinthe expectedexcessreturns(i.e.,theequitypremiums)associatedwithholdingstocks. Simple regressions of equity returns on surprise changes in the federal funds rate are silent on the question,soamorestructuredapproachisrequiredtodisentanglethevariouseffects. The approach of this paper is an adaptation of the method used by Campbell (1991), andCampbellandAmmer(1993). Inbrief,theirmethodusesalog-linearapproximationto decompose excess equity returns into components attributable to news about real interest rates, dividends, and future excess returns, then employs a VAR methodology to obtain proxies for the relevant expectations.16 We take the Campbell-Ammer framework one step further, however, by relating the proxies for expectations to the news about the path of monetary policy embodied in the surprises derived from federal funds futures. This allows us to estimate the impact of federal funds surprises on expected future dividends, real interest rates, and expected future excess returns. It turns out that the largest effects come from revisionsto expectationsof future excess returns, and toexpectationsof future dividends. Realinterestrateshaveaverysmalldirectimpact. The object of this analysis is the (log) excess return on equities, denoted y t+1 . This is defined as the total return on equities (price change plus dividends), minus the risk-free rate (the one-month Treasury bill yield). The return dated t+1 is measured over period t, i.e., from the beginning of period t to the beginning of period t+1. Let e y represent t+1 the unexpected (relative to expectationsformed at the beginningof periodt) excess return duringperiodt,i.e.,y t+1 −E t y t+1 . 16BecauseVARsrequireperiodictimeseriesdata,thesubsequentanalysiswillusethemonthlymeasure ofthefundsratesurprises. 23

Using the linearization developed by Campbell and Shiller (1988), we can express the period t unexpected excess return on equity in terms of the revision the expectation of discounted future dividends, the real interest rate, and future excess returns. (A sketch of thederivationcanbefoundintheappendix.) Thedecompositioncanbewrittenas: e y =e˜d −e˜r −e˜ y (10) t+1 t+1 t+1 t+1 wheretheesrepresenttherevisioninexpectationsbetweenperiodst andt+1,andthetilde denotesadiscountedsum,sothat ∞ e˜ t d +1 = (E t+1 −E t )∑ρ j ∆d t+1+j j=0 ∞ e˜ t r +1 = (E t+1 −E t )∑ρ jr t+1+j (11) j=0 ∞ e˜ t y +1 = (E t+1 −E t )∑ρ jy t+1+j . j=1 Thediscountfactorρ,whichcomesoutofthelinearization,representsthesteady-stateratio oftheequitypricetothepriceplusdividend;followingCampbellandAmmer(1993),this is set to 0.9962. As emphasized by Campbell (1991), (10) is really nothing more than a dynamicaccountingidentityrelatingthecurrentexcessreturntorevisionsinexpectations. As such, it containsno real economic content, muchless any specific asset pricingmodel; such a model would be required to provide a link between the conditional expectations of futurereturnsandeconomicvariables(e.g.,consumption). Implementing this decomposition requires empirical proxies for the expectations appearing in (10). The approach of Campbell (1991) and Campbell and Ammer (1993) is to model expectations using a Vector Autoregression (VAR) involving the variables of interest (excess returns and the real interest rate) along with any other indicators that might be helpful in forecasting those variables. Calculating the discounted sum of the revisions in expectations is straightforward; to do so involves writing the n variable, p lag VAR as a 24

first-ordersystem, z t+1 =Az t +w t+1 , (12) where z t+1 is an appropriately stacked np×1 vector containing the excess equity return, the real interest rate, and any additional indicators. With the VAR expressed in this form, theingredientsof(10)are givenby e t y +1 = s y w t+1 , e˜ t y +1 = s y ρA(1− ρA)−1w t+1 , e˜ t r +1 = s r (1− ρA)−1w t+1 and (13) e˜d = e y +e˜ y −e˜r , t+1 t+1 t+1 t+1 wheres ands areappropriate1×npselectionmatrices. y r Two features of the Campbell-Ammer method deserve further comment. One is its parametric approach to constructing long-horizon expectations of stock returns: one has to assume that the dynamics of equity returns many years in the future are adequately captured by a parsimonious VAR model. To a large extent, this parametric approach is forceduponus,astherelativelyshortexperiencewithfederalfundsfuturesisnotsufficient to directly estimate the long-horizon impact on stock asset returns, particularly in light of thequestionablesmall-samplepropertiesoflong-horizonregressions(seeNelsonandKim (1993)). Butasdiscussedbelow,theuseoftheVARdoesallowustoestimatethedynamics ofstockreturnsoveralongersamplethantheperiodforwhichfuturesdataareavailable. Asecondimportantfeatureoftheapproachisthatdividendsarenotincludedexplicitly as a variable to be forecast; given e y , e˜ y and er , ed is backed out from (10). In t+1 t+1 t+1 t+1 principle, it would be possible to forecast dividendsdirectly in the VAR, and instead back y outan implied e˜ . In practice, however,thisiscomplicatedby a strongseasonalpattern, t+1 andarootnearunityinthedividendprocess. Itisimportanttonotethattotheextentthatthe VARunderstatesthepredictabilityofexcessreturns,treatingdividendsasaresidualmeans 25

thatthemethodwillendupattributingtoomuchofthereturnvolatilitytodividends.17 3.1 The forecasting VAR The first step is to set up a VAR to capture the dynamic correlations between the excess equityreturnandtherealinterestrate(calculatedastheone-monthbillyieldminusthelog differenceinthenon-seasonally-adjustedCPI). TheVARmustthereforeincludethesetwo variables at a minimum, plus whatever other variables that might be useful in forecasting them. (One important constraint, of course, is that these variables are available in real time.) We follow Campbell and Ammer (1993) in using a six-variable one-lag system that included, besides the real rate and equity return: the relative bill rate (defined as the three-month bill rate minus its 12-month lagged moving average), the change in the bill rate, the (smoothed) dividend price ratio, and the spread between the 10-year and onemonth Treasury yields. For comparability with the Campbell-Ammer (1993) results, we useJanuary1973asthestartingdateforestimation. 3.2 A variance decomposition of equity returns Equation (10) expresses the current month’s excess equity returns into three components, which may be correlated with one another. The variance of the current excess return can therefore be broken down into the sum of the three variances, plus (or minus) the relevant threecovariances, Var(e y ) = Var(e˜d )+Var(e˜r )+Var(e˜ y )− t+1 t+1 t+1 t+1 2Cov(e˜d ,e˜r )−2Cov(e˜d ,e˜ y )+2Cov(e˜ y ,e˜r ) , (14) t+1 t+1 t+1 t+1 t+1 t+1 givingasenseoftherelativecontributionsofnewsaboutrealinterestrates,dividends,and expectedfutureexcessreturnstofluctuationsinthecurrentexcessreturn. Theresultsofthis 17AusefulcheckontheCampbell-Ammerprocedurewouldbetocompareitsimplieddividendforecasts with the observed behavior of dividends. Such a comparison is beyond the scope of the present paper, however. 26

decomposition appear in Table 10. For comparison, the table displays results for the full 1973–2002sample and for the subsamplebeginningin May 1989, when the federal funds futures data became available. The columns labeled “total” show the total contribution, and those labeled “share” expresses that contribution as a percentage of the excess return variance,i.e.,normalizingbyVar(e y ). t+1 The results for the full 1973–2002 sample are similar to those reported by Campbell andAmmer(1993)fortheir1973–87sample. Inparticular,thevarianceinexpectedfuture excess returns accounts for the majority of the variance of the current equity return: 76%, compared with Campbell and Ammer’s 101%. Dividends make a correspondingly larger contributionof24.5%,asopposedtoCampbellandAmmer’s14%. Inbothcases,thecontribution of the real interest rate is negligible (0.3% and 3% respectively) and statistically insignificant. The1989–2002subsampleyieldssomewhatdifferentresults,asshownintheright-hand portionofthetable. Considerablylessvarianceisattributedtorevisionsinexpectationsof future excess returns, and the dividendcomponent now plays a somewhatlarger role. The main reason for this seems to be a decline in the forecastability of equity returns in recent years, consistent with the observed fall in the adjusted R-squared from 0.04 to basically zero. Withreturnslessforecastable,theCampbell-Ammermethodologybydefaultassigns moreoftheexcessreturnvariancetodividendnews. 3.3 The effects of federal funds surprises The moststraightforward way to analyze the impact of monetary policywithin the frameworkintroducedaboveistoincludethefederalfundssurprisesintheVARasanexogenous variable z t+1 =Az t + φ∆ ¯i t u +1 +w t ⊥ +1 (15) where φ is an n×1 vector capturing the contemporaneous response of the elements of z t+1 to the unanticipated rate change period t +1. The new disturbance term w t ⊥ +1 is 27

by construction orthogonal to the funds rate surprise. This effectively breaks the VAR’s one-month-ahead forecast error into a component having to do with news about monetary policy,φ∆ ¯iu andacomponentincorporatinginformationaboutthingsotherthanpolicy. t+1 Because ∆iu represents a prediction error from a rational forecast made at time t, it t+1 shouldbeorthogonaltoz .18 ConsistentestimatesofbothAandφcanthereforebeobtained t by first estimating the VAR’s parameters, and then regressing the VAR’s one-step-ahead forecast errors on the funds rate surprises. Normally, there would be no advantage to the two-stepprocedureoversimplyestimating(15)directly. Butinourcase,usingthetwo-step procedure allows us to estimate the VAR dynamics (i.e., the coefficients in the A matrix) over a sample longer than the period for which federal funds futures are available.19 The longersamplewillofcoursetendtoimprovetheestimates’precision. 3.3.1 The dynamicresponse tofunds ratesurprises Incorporating the federal funds surprises into the VAR in this way allows us to do two things. First, because it extracts an orthogonal element from the w forecast error, we can t use it to calculate the dynamic responses of the variables in the VAR to the orthogonal component. Thek-monthresponsetoaone-percentage-pointsurpriseincreaseinthefunds ratecanbecalculatedquitesimplyasAkφ. An obvious question to arise at this point concerns the relationship between these futures-based funds rate surprises and the more familiar monetary policy shocks derived from an identified VAR. The methods used to construct the one-month-ahead funds rate forecasts differ, of course, with one using the futures market’s implicit forecast, and the other using a reduced-form econometric model. Forecast methodologies aside, however, theorthogonalizationproceduredescribedaboveisconceptuallyequivalenttoorderingthe federal funds rate first in a VAR system. Since this precludes any contemporaneous re- 18KruegerandKuttner(1996)showedthatinpractice,thefederalfundsfuturespredictionerrorsaregenerallyuncorrelatedwithlaggedinformation. 19Faustetal.(2002)usedasimilarprocedure.Specifically,theyestimatetheVARparametersoverthefull sample,butchooseanorthogonalizationbasedontheresponseofinterestratesoverthepost-1989subsample. 28

action of the funds rate to economic news, the surprises calculated in this way may well incorporateanendogenouspolicyresponsetoinformationarrivingwithinthemonth. Consequently, the impulse responses may represent the effects of things other than monetary policyperse. One way tominimizethisproblemwouldbe to purge the futures-based funds rate surprises of any contemporaneousresponse to the economyby projectingthem ontothe relevantinformationvariables,suchasthedatanewsobtainedfromtheMMSsurvey. Alternatively,sincetheresultsaboveinsection2.7indicatetherehasbeenlittle,ifany,correlation between the funds rate surprises and data news since 1994, the φ estimated only on the post-1994 subsample should be relatively free from this endogeneity problem. This is the approachtakenintheresultspresentedbelow. The upper-left-hand panel of Figure 6 displaysthe dynamic responseof excessreturns calculated in this way. The initial decline of 11.6% (not shown, because of the difference in scale) is followed by another month of negative returns, and then by several months of near-zeroexcessreturns.20 Aftersixmonths,equitiesbegintoexhibitsmallpositiveexcess returns, peaking at 0.16% per month (1.9% at an annual rate), and continuingfor a period measuredinyears. Thecontractionaryfundsratesurprisealsoleadstoasizableincreaseintherelativebill rate, which persists several months (essentially by construction). The real T-bill rate rises sharply at first, but the increase is relatively short-lived, and all but disappears after four months. Inthenearterm,thedynamicsofequityexcessreturnsaredominatedbytheeffects ofrisinginterestrates. Butastheseeffectsdieout,thelong-runeffectofthedividend-price ratio, which rises as a result of the fall in equity prices, reasserts itself. This leads to the highlypersistent,positiveexcessreturnsvisibleintheimpulseresponsefunction. 20This11.6%responsediffersslightlyfromtheresultsinsection2.7becausethedependentvariableisthe forecasterrorinthelogexcessreturn,ratherthantherawnominalreturn. 29

3.3.2 Explainingthe stockmarket’s reaction toFedpolicy The secondthingthisapproach allowsusto dois calculate theimpact ofthe federal funds surprises on the discounted sums of expected future excess returns, interest rates, and dividends. And since it is these sums that are related to the current excess return through (10),thisprovidesa naturalwaytodeterminethesource(orsources)ofthestockmarket’s reactiontomonetarypolicy. Onewaytoassesspolicy’seffectonthesediscountedsumsissimplytousetheVARto calculate e˜d , e˜r , and e˜ y , which represent the revisionsin expectationsof the relevant t+1 t+1 t+1 present values, and regress these variables in turn on ∆iu . Although this would provide t+1 theanswerweareafter,thestandarderrorswouldbemisleading,astheywouldfailtotake intoaccountthedependenceofthee˜sontheestimatedparametersoftheVAR. Analternativewaytodothesamecalculationistowriteoutthee˜sintermsoftheVAR y coefficients. Takinge˜ asanexample: t+1 e˜ t y +1 = s y ρA(1− ρA)−1w t+1 or = s y ρA(1− ρA)−1( φ∆ ¯i t u +1 +w t ⊥ +1 ) . (16) Theresponseofthepresentvalueofexpectedfutureexcessreturnstothefederalfundsrate surpriseisjust s ρA(1− ρA)−1 φ . (17) y Thus, the response of expected future excess returns depends not onlyon the φ vector, but also on the VAR dynamics represented by A. Similarly, the response of the present value ofcurrentandexpectedfuturerealreturnsis s (1− ρA)−1 φ , (18) r 30

andtheimpliedresponseofthepresentvalueofcurrentandexpectedfuturedividendsis s φ +s ρA(1− ρA)−1 φ +s (1− ρA)−1 φ (19) y y r oralternatively (s +s )(1− ρA)−1 φ . (20) y r Thestandarderrors fortheseresponsesare calculatedusingthedeltamethod,asinCampbellandAmmer(1993). The results of these calculations appear in Table 11. With the VAR estimated over the entire 1973–2002 sample, funds rate surprises have a large, marginally significant impact onthediscountedsumoffutureexcessreturns,accountingforjustoverhalfofthecontemporaneousresponseexcessreturns,equalto−11.55. Thereasonforthislargecontribution isreadily understoodintermsof the impulseresponsesplottedin Figure6. Thoughsmall, fundsrateshocksareestimatedtohaveahighlypersistentpositiveeffectonexcessreturns. Discounting these future positive excess returns back using a discount factor near unity yields a large negative impact on the current excess return. The −4.82 impact of funds rate surprises on dividends is nearly as large as that of future excess returns, and it too is significant at the 0.10 level. The impact on the discounted sum of real rates is very small, however,accountingforlessthanonepercentagepointoftheexcessreturnresponse. The resultsare qualitativelysimilarwhen the VAR isestimated overthe shorter 1989– 2000 sample. The only noteworthy difference is the smaller impact on expected future excessreturns,whichnowaccountforastatisticallyinsignificant3.29percentagepointsof the −11.01% response. The reason for this can be traced to the smaller amount of longrunforecastabilityinexcessreturns inthe post-1989sample. In fact,theimpulseresponse functions from this truncated sample (not shown) are nearly identical to those for the full 1973–2002 sample, shown above. The main difference is that the response of the excess return is negligible after six months or so, and it is this difference that accounts for the smallercontributionoffutureexcessreturns. 31

4 Conclusions Thisstudyhas documenteda relativelystrongand consistentresponse of the stockmarket to unexpected monetary policy actions, using federal funds futures data to gauge policy expectations. For broad stock market gauges like the CRSP value-weighted index, an unexpected 25-basis-point rate cut would typically lead to an increase in stock prices on the order of one percent. The result is robust to the exclusion of outliers and to the choice of windows for measuring the stock market’s response. There is some evidence of a larger market responseto policychanges thatare perceivedto be relativelymore permanent,and a smaller response to unexpected inaction on the part of the FOMC. We also find that reactions to monetary policy surprises tend to differ across industry-based portfolios, with the high-tech and telecommunications sectors exhibiting a response half again as large as that of the broad market indices. Other sectors, such as energy and utilities, seem not to be significantly affected by monetary policy. The industry responses to monetary policy changesseembroadlyconsistentwiththepredictionsofthestandardCAPM. Althoughwehavefoundaneffectofmonetarypolicyonthestockmarketofreasonable size, we should emphasize that monetary policy surprises are responsible for only a small portion of the overall variability of stock prices. Our method also does not allow us to determine the role played by anticipated monetary policy in stock price determination. Stocks are claims to real assets, so if monetary neutrality holds stock values should be independent of monetary policy in the very long run. In the medium term, however, real andnominalvolatilityinducedbytheformofthemonetarypolicyrulemaywellinfluence stockvalues. A more difficult question is why stock prices respond as they do to monetary policy. We have tried to make progress on this question by asking whether monetary policy affects stock values through its effects on real interest rates, expected future dividends, or expectedfuture stockreturns. The resultspresentedin thispaper showed,perhaps surprisingly,thatthereactionofequitypricestomonetarypolicyis,forthemostpart,notdirectly attributable to policy’s effects on the real interest rate. This finding is the result of the 32

relativelytransitorymovementsinrealinterestratesinducedbysurprisepolicyactions. Instead,theimpactofmonetarypolicysurprisesonstockpricesseemstocomeeitherthrough its effects on expected future excess returns or on expected future dividends. (The exact breakdownbetweenthesetwochannelsdependssomewhatonthechoiceofsample,which appearstoaffectthelong-horizonforecastabilityofexcessreturns.) Economically, how should we interpret the result that monetary policy affects stock prices in significant part by affecting expected excess returns? Taken literally, this result suggeststhattightmoney(for example)lowersstockprices by raisingtheexpected equity premium. This could come about in at least two ways. First, tight money could increase the riskinessof stocks directly, for example, by raising the interest costs or weakening the balance sheetsof publiclyownedfirms. Second, tightmoneycouldreduce thewillingness of stock investors to bear risk, for example by reducing expected levels of consumption, as in Campbell and Cochrane (1999), or because of its association with higher inflation, as in Brandt and Wang (2003). These linkages open up the possibility of new ways in which monetary policy may affect real activity — for example, by affecting the level of precautionarysaving. Analternativeinterpretationofourresultsisthatthelargemovementsinexcessreturns associatedwithmonetarypolicychanges reflect excesssensitivityoroverreactionof stock pricestopolicyactions. Amoretightlystructuredanalysisthatencompassesa widerclass of assetsmay helpto differentiatethese interpretations. In anycase, further explorationof the link between monetary policy and the excess return on equities is an intriguing topic forfutureresearch. 33

Appendix: deriving equation 10 This appendix provides a brief sketch of the derivation of the log-linearized relationship betweenthecurrentexcessreturn,expectedfutureexcessreturns,dividendgrowth,andreal interestratesgivenin(10). ThederivationroughlyfollowsCampbellandShiller(1988)and Campbell(1991). Thestartingpointissimplythedefinitionofthestockreturn,H t+1 : 1+H t+1 ≡ P t+1 +D t (1) P t wherePisthestockpriceandDisthedividend. Takinglogsandlettingh t+1 =ln(1+H t+1 ) yields: h t+1 =ln(P t+1 +D t )−ln(P t ) . (2) The next step is to derive a log-linear approximation to ln(P t+1 +D t ). One way to do this is to first-difference, and express the change in the log of the sum as the weighted sum of thelogdifferences ∆ln(P t+1 +D t )≈ ρ∆p t+1 +(1− ρ ) ∆d t (3) whereρ isthesteady-stateP/(D+P). “Integrating”thisexpressiongives ln(P t+1 +D t )≈k+ ρp t+1 +(1− ρ )d t , (4) substituting this into the expression for h t+1 , substituting δ t for d t−1 −p t , and combining termsgives h t+1 ≈ k− ρδ t+1 + δ t + ∆d t (5) ≈ k+(1− ρL −1) δ + ∆d . (6) t t Thenextstepistosolveforward,giving δ t = (1− ρL −1)−1(h t+1 − ∆d t −k) (7) ∞ = ∑ρ i(h t+1+i − ∆d t+i )−k/(1− ρ ) . (8) i=0 34

Substitutingthis,andasimilarexpressionfor δ t+1 ,into(5)andcollectingtermsyields: ∞ ∞ h t+1 −E t h t+1 =−∑ρ i(E t+1 −E t )h t+1+i +∑ρ i(E t+1 −E t ) ∆d t+1+i (9) i=1 i=0 whichcorrespondstoequation1inCampbell(1991). Abreakdownofexcessreturnscanthenbederivedbyexpressingtheequityreturnh t+1 asthesumofa risk-freerateandanexcessreturn h t+1 =r t+1 +y t+1 . (10) Becauseitisassumedthatr t+1 isknownattimet,the“excessreturnsurprise”y t+1 −E t y t+1 isthesameastheoverallreturnsurpriseh t+1 −E t h t+1 . Sotherisk-freeratecanbeincluded inthetwo-waybreakdownasfollows: ∞ ∞ y t+1 −E t y t+1 =−∑ρ i(E t+1 −E t )(y t+1+i +r t+1+i )+∑ρ i(E t+1 −E t ) ∆d t+1+i (11) i=1 i=0 oras ∞ y t+1 −E t y t+1 = −∑ρ i(E t+1 −E t )y t+1+i − i=1 ∞ ∞ ∑ρ i(E t+1 −E t )r t+1+i +∑ρ i(E t+1 −E t ) ∆d t+1+i . (12) i=1 i=0 Again, because E t r t+1 = r t+1 , it doesn’t matter whether the summation involving the rs y begins at 0 or 1. Finally, letting e represent the “excess return surprise” and replacing t+1 thesummationswiththecorrespondinge˜syields(10). 35

References Bernanke, Ben S., & Gertler, Mark. 1999. Monetary Policy and Asset Price Volatility. FederalReserveBankofKansasCityEconomicReview,17–51. Boyd, John H., Jagannathan, Ravi, & Hu, Jian. 2001 (Dec.). The Stock Market’s Reaction to UnemploymentNews: Why BadNews Is UsuallyGoodForStocks. WorkingPaper 8092.NBER. Brandt, MichaelW., &Wang, KevinQ. 2003. Time-varyingriskaversion and unexpected inflation. JournalofMonetaryEconomics,1457–1498. Campbell, John Y. 1991. A Variance Decomposition for Stock Returns. The Economic Journal,101(Mar.),157–79. Campbell, John Y., & Ammer, John. 1993. What Moves the Stock and Bond Markets? A VarianceDecompositionforLong-TermAssetReturns. JournalofFinance,48(Mar.), 3–37. Campbell, John Y., & Cochrane, John. 1999. By Force of Habit: A Consumption-Based ExplanationofAggregateStockMarketBehavior. JournalofPoliticalEconomy,107, 205–51. Campbell, John Y., & Shiller, Robert J. 1988. The Dividend-Price Ratio and Expectations ofFutureDividendsandDiscountFactors. ReviewofFinancialStudies,1,195–228. Campbell, John Y., & Vuolteenaho, Tuomo. 2003. Bad Beta, Good Beta. Manuscript, HarvardUniversity. Cochrane,JohnH.,&Piazzesi,Monika.2002(Mar.). TheFedandInterestRates: AHigh- FrequencyIdentification. WorkingPaper8839.NBER. Craine, Roger, & Martin, Vance. 2003. Monetary Policy Shocks and Security Market Responses. Manuscript,UniversityofCaliforniaatBerkeley. D’Amico, Stefania, & Farka, Mira. 2003. The Fed and the Stock Market: A Proxy and InstrumentalVariableIdentification. Manuscript,ColumbiaUniversity. Evans, Charles L., & Kuttner, Kenneth N. 1998 (Dec.). Can VARs Describe Monetary Policy? WorkingPaper98-19.Federal ReserveBankofChicago. Fama, Eugene F., & French, Kenneth R. 1988. Permanent and Temporary Componentsof StockPrices. JournalofPoliticalEconomy,96,246–273. 36

Faust, Jon, Swanson, Eric, & Wright, Jonathan H. 2002 (Feb.). Identifying VARs Based on High Frequency Futures Data. International Finance Discussion Paper 2002-720. Board ofGovernorsoftheFederalReserve System. Faust,Jon,Swanson,Eric,&Wright,Jonathan.2003. DoFederalReservePolicySurprises RevealPrivateInformationAbouttheEconomy? Manuscript,Board of Governorsof theFederal ReserveSystem. Fuhrer, Jeff, & Tootell, Geoffrey. 2003. Eyes on the Prize: How Did the Fed Respond to theStockMarket? Manuscript,FederalReserve BankofBoston. Goto,Shingo,&Valkanov,Rossen.2000. TheFed’sEffectonExcessReturnsandInflation isMuchBiggerThanYouThink. Manuscript,UCLAAndersonSchool. Guo, Hui. 2002. Stock Prices, Firm Size, and Changes in the Federal Funds Rate Target. Manuscript,Federal ReserveBankofSaintLouis. Gu¨rkaynak, Refet S., Sack, Brian P., & Swanson, Eric T. 2002. Market-Based Measures of Monetary Policy Expectations. Manuscript, Board of Governors of the Federal Reserve System. Gu¨rkaynak, Refet S., Sack, Brian P., & Swanson, Eric T. 2004. The Effects of Monetary Policy on Asset Prices: An Intraday Event-Study Analysis. Manuscript, Board of GovernorsoftheFederal ReserveSystem. Hilton, Spence. 1994. Expected Federal Funds Rate Data. Unpublished memorandum, Federal ReserveBankofNewYork,June17. Jensen, Gerald R., & Mercer, Jeffrey M. 1998. Monetary Policy and the Cross-Section of ExpectedStockReturns. Manuscript,NortheasternIllinoisUniversity. Jensen,GeraldR.,Johnson,R.R.,&Mercer,JeffreyM.1996. BusinessConditions,Monetary Policy, and Expected Security Returns. Journal of Financial Economics, 40, 213–37. Krueger, JoelT., &Kuttner, KennethN. 1996. The Fed Funds Futures Rate as a Predictor ofFederal ReservePolicy. JournalofFuturesMarkets,16(8),865–879. Kuttner,KennethN.2001. Monetarypolicysurprisesandinterestrates: Evidencefromthe Fed fundsfuturesmarket. JournalofMonetaryEconomics,47(3),523–44. Kuttner, Kenneth N. 2003. Dating Changes in the Federal Funds Rate, 1989–92. Manuscript,Federal ReserveBankofNewYork. 37

Nelson,CharlesR.,&Kim,MyungJ.1993. PredictableStockReturns: TheRoleofSmall SampleBias. JournalofFinance,48(June),641–61. Patelis,AlexD.1997.StockReturnPredictabilityandtheRoleofMonetaryPolicy.Journal ofFinance,52(Dec.), 1951–72. Poole,William,Rasche,Robert,&Thornton,Daniel.2002. MarketAnticipationsofMonetaryPolicyActions. FederalReserve BankofSt.LouisReview,84,65–93. Rigobon, Roberto, & Sack, Brian. 2002 (Jan.). The Impact of Monetary Policy on Asset Prices. FinanceandEconomicsDiscussionSeries 2002-4.Board ofGovernorsofthe Federal ReserveSystem. Rigobon, Roberto, & Sack, Brian. 2003. Measuring the Reaction of Monetary Policy to theStockMarket. QuarterlyJournalofEconomics,118,639–70. Romer, Christina D., & Romer, David H. 2000. Federal Reserve Information and the BehaviorofInterestRates. AmericanEconomicReview,90,429–457. Rudebusch, Glenn. 1995. Federal Reserve Interest Rate Targeting, Rational Expectations, andtheTermStructure. JournalofMonetaryEconomics,35(2),245–274. Thorbecke, William. 1997. On Stock Market Returns and Monetary Policy. Journal of Finance,52(June),635–54. 38

Table1 Descriptive statistics ThetablereportsselecteddescriptivestatisticsforfederalfundsratesurprisesandtheCRSP value-weightedequityreturn overthesamplesgivenin the columnheadings. Allstatistics excludethe17September2001observation. May1989– February 1994– January1994 December2002 Numberofevents: ratechanges 55 76 andFOMCmeetings Standarddeviationof 10.4 9.5 federal fundssurprise,basispoints Standarddeviationofequityreturn 0.80 1.26 oneventdays,% Standarddeviationofequityreturn 0.71 1.11 onnon-eventdays,% Proportionofratechanges 0.67 0.95 takingplaceatFOMCmeetings 39

Table2 The response ofequity pricestofederal funds ratechanges The table reports the results from regressions of the one-day CRSP value-weighted equity return on changes in the federal funds rate (columns a and c), and on the surprise and expectedcomponentsofthefundsratechange(columnsbandd). Allvariablesareexpressed in percentage terms. The full sample consists of the 55 target rate changes and the 77 FOMC meeting dates over the period from June 1989 through December 2002, excluding the 17 September 2001 observation,for a total of 131 observations. The outliers excluded from the regressions in columns c and d correspond to the six observationswith influence statisticsinexcessof0.3,leaving125usableobservations. Parenthesescontaint-statistics, calculatedusingheteroskedasticity-consistentestimatesofthestandarderrors. Fullsample Excluding outliers Regressor (a) (b) (c) (d) Intercept 0.23 0.12 0.17 0.11 (2.58) (1.35) (2.14) (1.37) Rawfunds ratechange −0.61 ... −0.11 ... (1.06) (0.31) Expected change ... 1.04 ... 0.67 (2.17) (1.62) Surprise change ... −4.68 ... −2.55 (3.03) (2.79) R¯2 0.007 0.171 −0.007 0.049 40

Table3 Testsforsubsample stabilityandendogeneity The table reports the results from regressions of the one-day CRSP value-weighted equity return onthe surpriseand expected componentsof the change inthe federal fundsrate, all expressedinpercentageterms. Thepost-1994dummyissetto1forobservationsbeginning with4February1994. Theemploymentdummyissetto1forpre-1994observationswhen a change in the target funds rate coincided with an employment release. The full and nooutliersamplesarethesameasthoseusedfortheresultsappearinginTable2. Parentheses containt-statistics,calculatedusingheteroskedasticity-consistentestimatesofthestandard errors. Fullsample Excluding outliers Regressor (a) (b) (c) (d) Intercept 0.16 0.16 0.12 0.12 (1.80) (1.76) (1.43) (1.43) Expected change 1.09 1.09 0.69 0.69 (2.26) (2.24) (1.68) (1.67) Surprise change −1.25 −2.55 −2.29 −3.57 (1.14) (1.70) (2.28) (3.77) Surprise change× post-1994 −6.87 −5.58 −0.78 0.50 (3.59) (2.61) (0.45) (0.29) employment report ... 2.67 ... 3.33 (1.82) (2.55) R¯2 0.280 0.283 0.042 0.054 41

Table4 Testsforasymmetries The table reports the results from regressions of the one-day CRSP value-weighted equity return on the surprise and expected components of the change in the federal funds rate, all expressed in percentage terms. The positive surprise dummy is set to 1 when the surprise change in the funds rate is greater than zero. The no rate change and positive rate changedummiesequal1whenthefundsratetargetisunchangedorincreased. TheFOMC meeting dummy is set to 1 for those observations coinciding with FOMC meetings. The reversaldummyequals1forratechangesthatreversethedirectionofthepreviouschange. The post-1994 dummy is set to 1 for observations beginning with 4 February 1994. The employmentdummyissetto1forpre-1994observationswhenachangeinthetargetfunds ratecoincidedwithanemploymentrelease. Thefullandno-outliersamplesarethesameas those used for the results appearing in Table 2. Parentheses containt-statistics, calculated usingheteroskedasticity-consistentestimatesofthestandarderrors. Fullsample Excluding outliers Regressor (a) (b) (c) (d) (e) (f) Intercept −0.02 0.12 0.14 0.12 0.12 0.13 (0.17) (1.34) (1.72) (1.32) (1.42) (1.63) Expected change 0.84 1.56 1.03 0.72 0.97 0.72 (1.58) (3.24) (2.24) (1.67) (2.00) (1.76) Surprise change −7.57 −8.34 −3.97 −3.26 −4.49 −3.67 (4.67) (5.73) (2.98) (3.30) (4.91) (3.14) Surprise change× employment 7.05 8.11 2.54 3.05 4.14 3.46 (4.43) (5.26) (1.47) (2.36) (3.19) (2.35) positive surprise 7.39 ... ... −0.34 ... ... (1.59) (0.10) noratechange ... 10.42 ... ... 4.00 ... (3.81) (2.25) positive ratechange ... 3.05 ... ... 0.58 ... (0.76) (0.15) FOMCmeeting ... ... 4.25 ... ... 0.67 (2.75) (0.39) reversal ... ... −6.33 ... ... −17.62 (3.09) (4.08) post-1994 ... ... −4.61 ... ... 0.80 (2.48) (0.44) R¯2 0.260 0.323 0.369 0.053 0.065 0.098 42

Table5 The response ofinterest rate expectationstofederal funds ratesurprises The table reports the results from regressions of the one-day change in the three-monthahead federal fundsfutures rate on the surpriseand expected componentsof thechange in the federal funds rate, all expressed in percentage terms. The no rate change and positive rate change dummies equal 1 when the funds rate target is unchanged or increased. The FOMCmeetingdummyissetto1forthoseobservationscoincidingwithFOMCmeetings. The reversal dummy equals 1 for rate changes that reverse the direction of the previous change. Thefullandno-outliersamplesarethesameasthoseusedfortheresultsappearing in Table 2. Parentheses contain t-statistics, calculated using heteroskedasticity-consistent estimatesofthestandarderrors. Regressor (a) (b) (c) (d) Intercept −0.01 −0.01 −0.01 −0.01 (1.46) (1.55) (1.34) (1.40) Expected change 0.07 0.05 0.07 0.07 (2.10) (1.32) (2.29) (2.08) Surprise change 0.65 0.70 0.73 0.66 (13.37) (14.71) (14.54) (12.83) Surprise change× noratechange ... −0.36 ... ... (3.24) FOMCmeeting ... ... −0.21 ... (2.07) reversal ... ... ... −0.12 (2.24) R¯2 0.726 0.745 0.744 0.727 43

Table6 The stockmarket response tolevelversustimingsurprises The table reports the results from regressions of the one-day CRSP value-weighted equity return on the surprise and expected components of the change in the federal funds rate, and the timing surprise, all expressed in percentage terms. The timing surprise is defined as thedifference betweenthe changein thethree-month-aheadfuturesrate and thecurrent month’ssurprise. Thefullandno-outliersamplesarethesameasthoseusedfortheresults appearing in Table 2. Parentheses containt-statistics, calculated using heteroskedasticityconsistentestimatesofthestandarderrors. Fullsample Excluding outliers Regressor (a) (b) (c) (d) Intercept 0.12 0.09 0.11 0.09 (1.35) (1.09) (1.37) (1.11) Expected change 1.05 1.34 0.67 0.94 (2.17) (2.92) (1.62) (2.46) Surprise change −4.68 −6.20 −2.55 −4.17 (3.03) (3.80) (2.79) (4.20) Timingsurprise ... −4.29 ... −4.27 (2.20) (3.25) Effectof“pure” ... −1.91 ... 0.09 timingsurprise (0.91) (0.08) R¯2 0.171 0.192 0.049 0.085 44

Table7 The impact ofeconomic newsonfederal funds ratesurprises The table reports the results from regressions of the monthly federal funds rate surprise on the unexpected components of the data releases listed in the row headings, over the sampleindicatedinthecolumnheadings. SurveydatagatheredbyMoneyMarketServices are used to calculate the data surprises. Asterisks denote statistical significance based on heteroskedasticity-consistentestimatesofthestandarderrors: ***forthe0.01level,**for the0.05level,and*forthe0.01level. Subsample Datasurprise Fullsample 5/89–9/92 2/94–12/02 HeadlineCPI 0.016 −0.124 −0.010 CoreCPI −0.058 −0.012 0.152 HeadlinePPI 0.001 −0.027 −0.024 CorePPI −0.085 ∗∗ −0.304 ∗∗∗ −0.022 Nonfarmpayrolls 0.203 ∗∗ 0.624 ∗∗∗ −0.009 Industrialproduction 0.069 ∗ 0.136 0.028 Retailsales −0.031 ∗∗ −0.061 −0.035 ∗∗∗ Retailsales,xautos 0.023 0.093 0.021 R2 0.128 0.454 0.087 R¯2 0.082 0.304 0.012 45

Table8 The monthlyresponse ofequityprices tofederalfunds rate surprises The table reports the results from regressions of the one-month CRSP value-weighted equity return on the surprise and expected components of the one-month change in the federal funds rate, all expressed in percentage terms. The full sample includes 164 monthly observations spanning May 1989 through December 2002. The no-outlier sample contains154observations. Parenthesescontaint-statistics,calculatedusingheteroskedasticityconsistentestimatesofthestandarderrors. Full No Testsforasymmetries sample outliers Regressor (a) (b) (c) (d) (e) Intercept 0.13 −0.03 −0.01 −0.07 0.10 (0.32) (0.09) (0.02) (0.16) (0.24) Expected change −1.11 0.96 −1.07 −2.72 −1.09 (0.37) (0.35) (0.36) (0.72) (0.36) Surprise change −11.43 −14.26 −12.46 −11.01 −10.49 (3.95) (5.43) (3.69) (3.46) (2.53) Surprise change× positive surprise ... ... 6.82 ... ... (0.63) noratechange ... ... −4.88 ... (0.75) positive ratechange ... ... ... 6.59 ... (0.52) reversal ... ... ... ... 3.52 (0.50) post-1994 ... ... ... ... −3.77 (0.50) Employment surprise ... ... ... ... −0.69 (0.10) R¯2 0.065 0.096 0.061 0.056 0.049 Standard error 4.28 3.85 4.30 4.30 4.31 Durbin-Watson statistic 2.02 2.09 2.02 2.02 2.03 46

Table9 The response ofFama-Frenchindustry portfoliostofederalfunds rate surprises Thetablereportstheresultsfromregressionsoftheone-monthreturnsontheFama-French industryportfoliosindicatedintherowheadingsonthesurpriseandexpectedcomponents of the one-month change in the federal funds rate, all expressed in percentage terms. The regressions also include an intercept, whose coefficient is not reported. The full sample includes 164 monthly observations spanning May 1989 through December 2002. Parentheses contain t-statistics, calculated using heteroskedasticity-consistent estimates of the standarderrors. Responsetofederalfundsratechanges: Market Index anticipated unanticipated R¯2 SE DW beta CRSPvalueweighted −1.11 −11.43 0.065 4.28 2.02 1 (0.37) (3.95) Nondurables −0.85 −9.65 0.046 4.17 2.00 0.60 (0.25) (2.88) Durables −1.47 −12.45 0.048 5.56 1.97 1.02 (0.38) (3.04) Manufacturing −2.02 −8.82 0.035 4.26 2.03 0.85 (0.61) (2.81) Energy 0.20 −4.03 −0.003 4.71 2.12 0.55 (1.02) (1.24) Hightech 0.06 −14.73 0.025 8.22 2.00 1.61 (0.01) (2.72) Telecommunications 0.35 −16.10 0.065 6.16 1.85 1.16 (0.60) (3.31) Wholesale/retail −4.75 −11.97 0.056 4.85 1.95 0.90 (1.47) (3.64) Healthcare −1.04 −8.04 0.017 4.96 2.15 0.72 (0.29) (1.80) Utilities −1.24 −5.42 0.006 4.21 1.97 0.32 (0.48) (1.55) Other −1.21 −11.08 0.051 4.62 2.09 0.92 (0.35) (3.61) 47

Table10 Avariancedecompositionofexcessequity returns The table reports the decomposition of the variance of the current excess equity returns intothevariancesofrevisionsinexpectationsofdividends,realinterestrates,futureexcess returns, and the covariances between these three components. The excess equity return is the difference between the CRSP value-weighted return and the one-month Treasury bill rate. A six-variable VAR(1) is used to construct forecasts of future real interest rates and excess returns. The VAR includes the excess equity return, the real interest rate, the relative bill rate (defined as the three-month bill rate minus its 12-month lagged moving average), the change in the three-month bill rate, the smoothed dividend price ratio, and the spread between the 10-year and one-month Treasury yields. Parentheses contain tstatistics,calculatedusingthedeltamethod. 1973–2002 1989–2002 Total Share(%) Total Share(%) Var(excess return) 21.5 19.0 Var(dividends) 5.3 24.5 6.1 31.9 (6.2) (1.8) Var(realrate) 0.3 1.4 0.1 0.6 (2.4) (1.5) Var(future returns) 16.4 76.0 7.2 38.0 (1.8) (1.2) −2Cov(dividends, realrate) −0.4 −2.1 −0.6 −3.2 (0.8) (0.7) −2Cov(dividends, future excessreturn) 0.2 1.0 7.2 −37.7 (0.0) (2.3) 2Cov(future excess return,realrate) −0.2 0.8 1.0 5.1 (0.1) (1.1) R¯2 fromexcessreturn equation 0.040 -0.003 48

Table11 The impactofmonetarypolicyondividends, interest rates,andfuture returns The table reports the impact of monetary policy surprises on the current excess equity return, and the discounted sums of future excess equity returns, current and future real interestrates, andcurrent andfuture dividends. Thesix-variableVAR(1)used toconstruct real interest rate and excess equity return forecasts is estimated over the sample indicated in the column headings, and the contemporaneous response to the funds rate surprises is estimated on the February 1994 to December 2002 subsample. Parentheses contain tstatistics,calculatedusingthedeltamethod. SampleusedforVAR 1/73–12/02 5/89–12/02 Currentexcessreturn −11.55 −11.01 (3.87) (3.72) Futureexcessreturns 6.10 3.29 (1.74) (1.10) Realinterestrate 0.64 0.77 (1.03) (1.87) Dividends −4.82 −6.96 (1.73) (2.35) 49

6 5 4 3 2 1 0 -1 -2 -3 -4 -0.50 -0.25 0.00 0.25 Federal funds rate surprise, % % ,nruter dethgiew-eulav PSRC 1/3/2001 10/15/1998 4/18/2001 8/21/1991 5/17/994 8/16/1994 7/2/1992 FOMC meeting Intermeeting 3/20/2001 Employment report Reversal Figure 1. Federal funds rate surprises and equity returns, daily data. The figure is a scatterplotofone-day CRSP value-weightedequityreturnsagainstthe surpriseelementof changes in the federal funds rate, for the 131 event days in the sample. Observations are distinguishedby their association with FOMC meetings, intermeeting target rate changes, the release of employmentreports, and changes in the direction of rate movements(reversals). Thesixobservationswithboldfacedatelabelsarethoseflaggedascandidateoutliers on the basis of regression influence statistics. The two observations with italicized date labelsare thoseassociatedwithunusualannouncementsbytheFOMC. 50

120 100 80 60 40 20 0 0.05 0.10 0.15 0.20 0.25 0.30 > 0.30 Upper bound of bin snoitavresbo fo rebmuN Figure 2. Distribution of regression influence statistics. The statistics are based on the changes in the estimated parameters from a regression of one-day CRSP value-weighted equity returns on the surprise and expected components of the federal funds rate change, droppingeachobservationinturnfromthesample. 51

0.25 0.00 -0.25 -0.50 -0.50 -0.25 0.00 0.25 Federal funds rate surprise,% %,etar serutuf sdnuf def htnom-3 niegnahC 1-for-1effect on 3-monthexpectations Greater than 1-for-1effect Less than 1-for-1effect Perverse effect 8/16/1994 5/17/1994 1-f ore 1 x e p f e f c e t c a t ti o n s n o Figure 3. Federal funds rate surprises and funds rate expectations. The figure is a scatterplot of one-day changes in the three-month-ahead federal funds futures rate against the surprise element of changes in the federal funds rate, for the 131 event days in the sample. Observationsare distinguishedaccording to whether the reaction of three-monthaheadexpectationsaregreaterthan,lessthan,equalto,oroppositeinsignfromthefederal fundsratesurprise. Thetwoobservationswithdatelabelsarethoseassociatedwithunusual announcementsbytheFOMC. 52

15 10 5 0 -5 -10 -15 -20 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 Federal funds rate surprise, % % ,nruter dethgiew-eulav PSRC candidate outliers Figure4. Federalfundsratesurprisesandequityreturns,monthlydata. Thefigureisa scatterplotofone-monthCRSP value-weightedequityreturnsagainstthe surpriseelement of changes in the federal funds rate, for the 164 months in the sample. Ten candidate outliers,identifiedonthebasisofregressioninfluencestatistics,are distinguished. 53

0 -5 -10 -15 -20 -20 -15 -10 -5 0 Industry response implied by the CAPM, % % ,esnopser yrtsudni detamitsE Energy Utilities Nondurables High-tech Telecom Figure 5. Estimated industry responses and CAPM implications. The figure depicts theone-monthresponsesof theFama-French industryportfoliostoa one percentage point federal funds rate surprise. The values on the horizontal axis are the industry stock return responsesimpliedbytheCAPM.Theverticalaxisvaluesare theestimatedindustryreturn responses reported in Table 9. The vertical lines represent the 80% confidence intervals associatedwiththeestimatedindustryresponses. 54

excess equity return 10-year to 1-month spread 0.30 0.00 0.00 -0.04 -0.30 -0.60 -0.08 -0.90 Initial response = -11.6% -1.20 -0.12 0 5 10 15 20 0 5 10 15 20 real interest rate dividend/price ratio 0.40 0.03 0.30 0.02 0.20 0.10 0.01 0.00 -0.10 0.00 0 5 10 15 20 0 5 10 15 20 change in bill rate relative bill rate 0.09 0.09 0.06 0.06 0.03 0.03 0.00 0.00 -0.03 -0.03 0 5 10 15 20 0 5 10 15 20 Months following federal funds rate surprise Figure 6. The dynamic responses of excess equity returns, interest rates, and the dividend-price ratio to federal funds rate surprises. Each panel depictsthe response of theindicatedvariabletoaonepercentagepointfederalfundsratesurprise. ThecontemporaneousresponsetothefundsratesurprisesisestimatedontheFebruary1994toDecember 2002subsample. A six-variableVAR(1),estimatedoverthe 1973–2002sample,isusedto projectthe futurepathof each variable. Because of thelarge difference inscale, theinitial excess return response is not shown. Each variable is experessed in monthly percentage terms. 55