Micro Effects of Macro Announcements: Real-Time …boller/Published_Papers/aer_03.pdfMicro Effects...

Micro Effects of Macro Announcements

Real-Time Price Discovery in Foreign Exchange

By TORBEN G ANDERSEN TIM BOLLERSLEV FRANCIS X DIEBOLD AND CLARA VEGA

Using a new data set consisting of six years of real-time exchange-rate quotationsmacroeconomic expectations and macroeconomic realizations we characterize theconditional means of US dollar spot exchange rates In particular we nd thatannouncement surprises produce conditional mean jumps hence high-frequencyexchange-rate dynamics are linked to fundamentals The details of the linkage areintriguing and include announcement timing and sign effects The sign effect refersto the fact that the market reacts to news in an asymmetric fashion bad news hasgreater impact than good news which we relate to recent theoretical work oninformation processing and price discovery (JEL F3 F4 G1 C5)

How is news about fundamentals incorpo-rated into asset prices The topic confronted bythis questionmdashcharacterization of the price dis-covery processmdashis of basic importance to all of nancial economics Unfortunately it is alsoone of the least well-understood issues Indeedsome in uential empirical studies have gone sofar as to suggest that for some assetsmdashnotably

foreign exchangemdashprices and fundamentals arelargely disconnected1

In this paper we provide an empirical exam-ination of price discovery in the challengingcontext of foreign exchange Using a newlyconstructed data set consisting of six years ofreal-time exchange-rate quotations macroeco-nomic expectations and macroeconomic real-izations (announcements) we characterize theconditional means of US dollar spot exchangerates for German Mark British Pound JapaneseYen Swiss Franc and the Euro In particularwe show that announcement surprises (that isthe difference between expectationsand realiza-tions or ldquonewsrdquo) produce conditional meanjumps and we provide a detailed analysis of thespeed and pattern of adjustment

We show that conditional mean adjustmentsof exchange rates to news occur quickly effec-tively amounting to ldquojumpsrdquo in contrast to con-ditional variance adjustments which are muchmore gradual and that an announcementrsquos im-pact depends on its timing relative to otherrelated announcements and on whether the an-nouncement time is known in advance We nd

Andersen Department of Finance Kellogg SchoolNorthwestern University and NBER (e-mail t-andersenkelloggnwuedu) Bollerslev Department of Economicsand Finance Duke University and NBER (e-mailbollerecondukeedu) Diebold Department of Econom-ics Finance and Statistics University of Pennsylvania andNBER (e-mail fdieboldsasupennedu) Vega GraduateGroup in Economics University of Pennsylvania (e-mailcvegasscupennedu) This work was supported by theNational Science Foundation and the Wharton FinancialInstitutions Center We are grateful to Olsen and Associatesfor making available their real-time exchange-rate quota-tions data and to Money Market Services International formaking available their news announcement expectationsdata For useful comments we thank three referees as wellas Ricardo Cabellero Dean Croushore Kathryn DominguezBernard Dumas Martin Evans Michael Fleming JeffFrankel Linda Goldberg Ken Kavajecz Rich LyonsNelson Mark Frank Schorfheide Nick Souleles AllanTimmermann Mark Watson Ingrid Werner and JonathanWright and participants at the CAF Conference in Denmarkon Market Microstructure and High-Frequency Data in Fi-nance the University of Wisconsin Conference on Empiri-cal Models of Exchange Rates the Federal Reserve Bank ofPhiladelphia Conference on Real-Time Data Analysis theNBER International Finance and Macroeconomics ProgramMeeting and seminars at the University of Pennsylvaniaand the University of Houston

1 The classic statement is of course Richard A Meeseand Kenneth Rogoff (1983) For a good survey of thesubsequent empirical exchange-rate literature through theearly 1990rsquos see Jeffrey A Frankel and Andrew K Rose(1995) In later work Nelson C Mark (1995) and Mark andDonggyu Sul (2001) nd that fundamentals matter in thelong run but not in the short run Martin D D Evans andRichard K Lyons (2002) nd that order ow matters in theshort run but fail to link order ow to fundamentals

38

moreover that the adjustment response patternis characterized by a sign effect bad news hasgreater impact than good news Finally we re-late our results to recent theoretical and empir-ical work on asset return volatility and itsassociation with information processing andprice discovery

The paper relates to earlier work in intriguingways but at least three features differentiate our ndings from previous results along importantdimensions These include our focus on foreignexchange markets our focus on conditionalmean as opposed to conditional variance dy-namics and the length and breadth of our sampleof exchange-rate and announcement data Let usdiscuss them brie y in turn

First we focus on foreign exchange marketsas opposed to stock or bond markets and weaddress the central open issue in exchange-rateeconomicsmdashthe link between exchange ratesand fundamentals It is comforting howeverthat a number of recent papers focusing largelyon bond markets reach conclusions similar toours Pierluigi Balduzzi et al (2001) for exam-ple examine the effects of economic news onprices in the US interdealer government bondmarket nding strong news effects and quickincorporation of news into bond prices whileMichael J Fleming and Eli M Remolona(1997 1999) show that the largest bond pricemovements stem from the arrival of newsannouncements2

Second we focus primarily on exchange-rateconditional means as opposed to conditionalvariances That is we focus primarily on thedetermination of exchange rates themselves asopposed to their volatility We maintain thisfocus both because the conditional mean is ofintrinsic interest and because high-frequencydiscrete-time volatility cannot be extracted ac-curately unless the conditional mean is modeledadequately Hence our work differs in importantrespects from that of Louis H Ederington andJae Ha Lee (1993) Richard Payne (1996)Andersen and Bollerslev (1998) and Bollerslevet al (2000) for example who examine calen-dar and news effects in high-frequency asset-

return volatility but do not consider the effectsof news on returns themselves

Third we use a new data set which spans acomparatively long time period and includes abroad set of exchange rates and macroeconomicindicators

Notwithstanding the improvements obtainedthrough the above consideration our results arequite consistent with prior related work Indeedseveral studies have linked macroeconomicnews announcements to jumps in exchangerates and our ndings may be viewed as pro-viding con rmation and elaboration CharlesGoodhart et al (1993) for example examineone year of high-frequency DollarPound ex-change rates and two speci c news eventsmdashaUS trade gure announcement and a UK in-terest-rate changemdashand conclude in each casethat the news caused an exchange-rate jumpSimilarly Alvaro Almeida et al (1998) in theirstudy of three years of high-frequency DMDollar exchange rates and a larger set of newsannouncements document systematic short-livednews effects Finally Kathryn M Dominguez(1999) argues that most large exchange-ratechanges occur within ten seconds of a macroeco-nomic news announcement and that close timingof central bank interventions to news announce-ments increases their effectiveness

We proceed as follows In Section I we de-scribe our high-frequency exchange-rate andmacroeconomic expectations and announce-ments data In Section II we characterize thespeed and pattern of exchange-rate adjustmentto macroeconomic news and we documentamong other things the sign effects (ie alarger exchange-rate response to bad than goodnews) In Section III we relate the sign effects torecent theories of information processing andprice discovery We conclude in Section IV

I Real-Time Exchange Rates ExpectedFundamentals and Announced Fundamentals

Throughout the paper we use data on exchange-rate returns in conjunction with data on expecta-tions and announcements of macroeconomicfundamentals The data are novel in several re-spects such as the simultaneous high frequencyand long calendar span of the exchange-rate re-turns as well as the real-time nature of theexpectations and announcements of fundamen-tals Here we describe them in some detail

2 Also in concurrent related work for T-bond futuresNikolaus Hautsch and Dieter Hess (2001) report highlysigni cant but short-lived price and volatility impacts inresponse to new and revised employment gures

39VOL 93 NO 1 ANDERSEN ET AL MICRO EFFECTS OF MACRO ANNOUNCEMENTS

A Exchange-Rate Data

The raw 5-minute CHF$ DM$ Euro$Pound$ and Yen$ return series were obtainedfrom Olsen and Associates The full sampleconsists of continuously recorded 5-minute re-turns from January 3 1992 through December30 1998 or 2189 days for a total of 2189 z288 5 630432 high-frequency foreign ex-change (FX) return observations As in UlrichA Muller et al (1990) and Michel M Dacorognaet al (1993) we use all of the interbankquotes that appeared on the Reuters screenduring the sample period to construct our5-minute returns Each quote consists of a bidand an ask price together with a ldquotime stamprdquoto the nearest second After ltering the datafor outliers and other anomalies we obtainthe average log price at each 5-minute markby linearly interpolating the average of thelog bid and the log ask at the two closestticks We then construct continuously com-pounded returns as the change in these 5-minuteaverage log bid and ask prices Goodhartet al (1996) and Jon Danielsson and Payne(1999) nd that the basic characteristics of5-minute FX returns constructed from quotesclosely match those calculated from transac-tion prices (which are not generally availablefor the foreign exchange market)

It is well known that the activity in the for-eign exchange market slows decidedly duringweekends and certain holiday nontrading peri-ods see Muller et al (1990) Hence as is stan-dard in the literature we explicitly excluded anumber of days from the raw 5-minute returnseries Whenever we did so we always cut from2105 GMT the night before to 2100 GMT thatevening This particular de nition of a ldquodayrdquowas motivated by the ebb and ow in the dailyFX activity patterns documented in Bollerslevand Ian Domowitz (1993) and keeps the dailyperiodicity intact In addition to the thin week-end trading period from Friday 2105 GMTuntil Sunday 2100 GMT we removed several xed holidays including Christmas (December24ndash26) New Yearrsquos (December 31ndashJanuary 2)and July Fourth We also cut the moving holi-days of Good Friday Easter Monday MemorialDay July Fourth (when it falls of cially on July3) and Labor Day as well as Thanksgivingand the day after Although our cuts do notaccount for all of the holiday market slow-

downs they capture the most important dailycalendar effects

Finally we deleted some of the returns con-taminated by brief lapses in the Reuters datafeed This problem which occurs almost exclu-sively during the earliest part of the samplemanifests itself as sequences of zero or constant5-minute returns in places where missing quoteshad been interpolated To remedy this we sim-ply removed from each exchange-rate series thedays containing the 15 longest zero and constantruns Because of the overlap among sets of daysde ned by this criterion we actually removedonly 25 days

In the end we are left with 1724 days of datacontaining T 5 1724 z 288 5 496512high-frequency 5-minute return observationsStandard descriptive statistics reveal that the5-minute returns have means that are negligibleand dwarfed by the standard deviations and thatthey are approximately symmetric but distinctlynon-Gaussian due to excess kurtosis Ljung-Box statistics indicate serial correlation in bothreturns and absolute returns

To assess the economic relevance of the tem-poral dependencies in the return series we turn tothe autocorrelations in column one of Figure 1The raw returns display tiny but nevertheless sta-tistically signi cant serial correlation at the veryshortest lags presumably due to microstructureeffects However the short-lag return serial corre-lation is negligible relative to the strong serialcorrelation in absolute returns shown in columntwo of Figure 1 The sample autocorrelations ofabsolute returns display very slow decay and pro-nounced diurnal variation in line with the resultsof Dacorogna et al (1993) and Andersen andBollerslev (1998) Interestingly not only theshapes but also the amplitudes of the diurnal pat-terns in absolute return autocorrelations differ no-ticeably across currencies

B Expected Fundamentals AnnouncedFundamentals and News

We use the International Money Market Ser-vices (MMS) real-time data on expected andrealized (ldquoannouncedrdquo) macroeconomic funda-mentals de ning ldquonewsrdquo as the difference be-tween expectations and realizations Everyweek since 1977 MMS has conducted a Fridaytelephone survey of about 40 money managerscollected forecasts of all indicators to be re-

40 THE AMERICAN ECONOMIC REVIEW MARCH 2003

FIGURE 1 SAMPLE AUTOCORRELATION FUNCTIONS

RETURNS AND ABSOLUTE RETURNS

Notes We plot the sample autocorrelations of returns and absolute returns for ve currencies together with Bartlettrsquos approximate95-percent con dence bands under the null hypothesis of white noise In each graph the vertical axis is the sample autocorrelationand the horizontal axis is displacement in 5-minute intervals To avoid contamination from shifts in and out of daylight saving timewe calculate the sample autocorrelations using only days corresponding to US daylight saving time


leased during the next week and reported themedian forecasts from the survey Numerousin uential studies from early work such asThomas J Urich and Paul Wachtel (1984)through recent work such as Balduzzi et al(2001) have veri ed that the MMS expecta-tions contain valuable information about theforecasted variable and in most cases are unbi-ased and less variable than those produced fromextrapolative benchmarks such as ARMAmodels

Table 1 provides a brief description of salientaspects of US and German economic newsannouncements We show the total number ofobservations in our news sample the agencyreporting each announcement and the time ofthe announcement release Note that US an-nouncements times are known in advancewhereas only the day for the German announce-ments are known in advance but their timingwithin the day is variable and unknown a priori

The target Fed funds rate deserves specialmention The Federal Open Market Commit-teersquos (FOMC) announcement of the federalfunds rate target although likely producing im-portant news is nonstandard and hence is nottypically examined It is nonstandard becauseprior to February 1994 it was not announcedinstead the FOMC signaled the target rate butdid not state it explicitly through open marketoperations performed from 1130 to 1135 AMEastern time on the day of the FOMC meetingIn February 1994 the FOMC began to an-nounce changes in the target rate on meetingdays albeit at irregular times and from 1995onward it announced the target rate on meetingdays regularly at 215 PM Eastern time asdescribed in Kenneth N Kuttner (2001)

To assess the effects of FOMC news we needto know announcement days and times as well asthe marketrsquos expected Fed funds rate target andthe announced (or signaled) value Determinationof announcement days and times is relativelystraightforward We collected the irregular 1994announcement times from Reuters3 Before 1994we use an 1130 AM Eastern announcementtime and after March 1995 we use a 215 PM

announcement time4 Determination of market ex-pectations is similarly straightforward we useMMS survey data on expected federal fund ratetargets from January 1992 to December 19985

The announcements themselves are trickier toconstruct due to the pre-1994 FOMC secrecywe use the announcement data constructed byMichael W Brandt et al (2001) kindly pro-vided by Kenneth Kavajecz

In Figure 2 we show the pattern of USrelease dates throughout the month6 This is ofpotential importance because there is some re-dundancy across indicators For example con-sumer and producer price indexes although ofcourse not the same are nevertheless relatedand Figure 2 reveals that the producer priceindex is released earlier in the month Henceone might conjecture that producer price newswould explain more exchange-rate return vari-ation than consumer price news as the typicalamount of consumer price news revealed maybe relatively small given the producer pricenews revealed earlier in the month

Because units of measurement differ acrosseconomic variables we follow Balduzzi et al(2001) in using standardized news That is wedivide the surprise by its sample standard devi-ation to facilitate interpretation The standard-ized news associated with indicator k at time t is

S kt 5A kt 2 Ekt

sk

where Akt is the announced value of indicator kEkt is the market expected value of indicator kas distilled in the MMS median forecast and sk

is the sample standard deviation of Akt 2 EktUse of standardized news facilitates meaningful

3 The announcement times were 1105 AM Easterntime on 020494 220 PM on 032294 and 070694 230PM on 111594 226 PM on 051794 223 PM on122094 117 PM on 081694 222 PM on 092794224 PM on 020195

4 The FOMC can also surprise the market by changingthe Fed funds target between FOMC meetings Because thisdoes not happen often in our sample (5 out of 62 times) andwe do not have the exact timing of such policy changes wedo not account for them Similarly data limitations preventus from investigating the effect of Fed open market opera-tions see Campbell R Harvey and Roger D Huang (2002)for a recent analysis involving the earlier 1982ndash1988 timeperiod

5 One could also attempt to infer expectations from Fedfunds futures prices as in Glenn D Rudebusch (1998) andKuttner (2001)

6 The design of the gure follows Chart 2 of Fleming andRemolona (1997)


TABLE 1mdashUS AND GERMAN NEWS ANNOUNCEMENTS

AnnouncementNumber of

observationsa Sourceb Datesc Announcement timed

US Announcements

Quarterly Announcements1 GDP advance 47 BEA 052287ndash103098 830 AM2 GDP preliminary 46 BEA 061787ndash122398 830 AM3 GDP nal 47 BEA 012287ndash112498 830 AM

Monthly AnnouncementsReal Activity

4 Nonfarm payroll employment 144 BLS 120586ndash120498e 830 AM5 Retail sales 145 BC 121186ndash121198 830 AM6 Industrial production 145 FRB 121586ndash121698 915 AM7 Capacity utilization 145 FRB 121586ndash121698 915 AM8 Personal income 142 BEA 121886ndash122498f 1000830 AMg

9 Consumer credit 129 FRB 040488ndash120798 300 PMh

Consumption

10 Personal consumption expenditures 143 BEA 121886ndash122498i 1000830 AMj

11 New home sales 117 BC 030289ndash120298 1000 AMInvestment

12 Durable goods orders 143 BC 122386ndash122398k 8309001000 AMl

13 Construction spending 128 BC 040188ndash120198m 1000 AM14 Factory orders 127 BC 033088ndash120498n 1000 AM15 Business inventories 129 BC 041488ndash121598 1000830 AMo

Government Purchases

16 Government budget de cit 124 FMS 042188ndash122198p 200 PMNet Exports

17 Trade balance 128 BEA 041488ndash121798 830 AMPrices

18 Producer price index 145 BLS 121286ndash121198 830 AM19 Consumer price index 145 BLS 121986ndash121598 830 AMForward-looking

20 Consumer con dence index 90 CB 073091ndash122998 1000 AM21 NAPM index 107 NAPM 020190ndash100198 1000 AM22 Housing starts 145 BC 123086ndash123098 830 AM23 Index of leading indicators 145 CB 123086ndash123098 830 AM

Six-Week AnnouncementsFOMC

24 Target federal funds rate 62 FRB 2592ndash122298 215 PMq

Weekly Announcements25 Initial unemployment claims 384 ETA 071891ndash123198 830 AM26 Money supply M1 628 FRB 120486ndash123198 430 PM27 Money supply M2 563 FRB 030388ndash123198 430 PM28 Money supply M3 563 FRB 030388ndash123198 430 PM

German Announcementsr

Quarterly Announcements29 GDP 24 GFSO 030993ndash120398 Varies


30 Employment 59 FLO 040693ndash120898 Varies31 Retail sales 59 GFSO 041493ndash121098 Varies32 Industrial production 63 GFSO 050493ndash120798 Varies


comparisons of responses of different exchangerates to different pieces of news Operationallywe estimate the responses by regressing assetreturns on news because sk is constant for anyindicator k the standardization affects neitherthe statistical signi cance of response estimatesnor the t of the regressions

Before proceeding we pause to discuss ingreater detail the possibility that the MMS fore-casts may not capture all information availableimmediately before the announcement Surelyinformation does not stop owing between the

time that the MMS forecast is produced and thetime that the macroeconomic indicator is real-ized hence the MMS forecasts may be ldquostalerdquoJust how stale they are however is an empiricalmatter This issue has been investigated alreadyin the context of news effects on interest ratesby Balduzzi et al (1998) who regress the actualannouncement Ai on the median forecast ofthe MMS survey Fi and the change in the(very announcement-sensitive) ten-year noteyield from the time of the survey to the time ofthe announcement Dy

TABLE 1mdashContinued



Investment

33 Manufacturing orders 62 GFSO 040693ndash120798 Varies34 Manufacturing output 64 GFSO 030293ndash120798 VariesNet Exports

35 Trade balance 61 GFSO 071393ndash121198 Varies36 Current account 61 BD 071393ndash121198 VariesPrices

37 Consumer price index 68 GFSO 030193ndash122398 Varies38 Producer prices 65 GFSO 031893ndash122298 Varies39 Wholesale price index 68 GFSO 031693ndash121698 Varies40 Import prices 68 GFSO 032693ndash122198 VariesMonetary

41 Money stock M3 66 BD 031893ndash121898 Varies

Notes We group the US monthly news announcements into seven groups Real activity the four components of GDP(consumption investment government purchases and net exports) prices and forward-looking Within each group we listUS news announcements in chronological order

a Total number of observations in the announcements sampleb Bureau of Labor Statistics (BLS) Bureau of the Census (BC) Bureau of Economic Analysis (BEA) Federal Reserve

Board (FRB) National Association of Purchasing Managers (NAPM) Conference Board (CB) Financial Management Of ce(FMO) Employment and Training Administration (ETA) German Federal Statistical Of ce (GFSO Statistisches BundesamtDeutschland) Federal Labor Of ce (FLO Bundesanstalt fur Arbeit) Bundesbank (BD)

c Starting and ending dates of the announcements sampled Eastern Standard Time Daylight saving time starts on the rst Sunday of April and ends on the last Sunday of Octobere 1098 is a missing observationf 1195 296 and 0397 are missing observationsg In 0194 the personal income announcement time moved from 1000 AM to 830 AMh Beginning in 0196 consumer credit was released regularly at 300 PM Prior to this date the release times variedi 1195 and 296 are missing observationsj In 1293 the personal consumption expenditures announcement time moved from 1000 AM to 830 AMk 0396 is a missing observationl Whenever GDP is released on the same day as durable goods orders the durable goods orders announcement is moved

to 1000 AM On 0796 the durable goods orders announcement was released at 900 AMm 0196 is a missing observationn 1098 is a missing observationo In 0197 the business inventory announcement was moved from 1000 AM to 830 AMp 0588 0688 1198 1289 and 0196 are missing observationsq Beginning in 32894 the Fed funds rate was released regularly at 215 PM Prior to this date the release times variedr Prior to 1994 the data refer only to West Germany Beginning in 1994 the data refer to the uni ed Germany The timing

of the German announcements is not regular but they usually occur between 200 AM and 800 AM Eastern StandardTime


A it 5 a0i 1 a1i F it 1 a2iDy t 1 e it

This particular regression facilitates the test-ing of several hypotheses First if there isinformation content in the MMS survey datathe coef cient estimates a1i should be posi-tive and signi cant Second if the surveyinformation is unbiased the a0i coef cientestimates should be insigni cant and theslope terms a1i should be insigni cantly dif-ferent from unity Finally if expectations arerevised between the survey and the announce-ment there should be a reaction in the bondprice at the time of the forecast revision andwe should see a relationship between thechange in yield and the announcement Asalready mentioned Balduzzi et al (2001) nd as have many others that most of theMMS forecasts contain information and are

unbiased7 More importantly for the issue athand however they also nd that for mostindicators the hypothesis that a2i 5 0 cannotbe rejected indicating that the MMS forecastsdo not appear signi cantly stale

II Exchange Rates and Fundamentals

We will specify and estimate a model ofhigh-frequency exchange-rate dynamics thatallows for the possibility of news affectingboth the conditional mean and the conditionalvariance Our goal is to determine whether

7 In addition to being unbiased Douglas K Pearce andV Vance Roley (1985) and Grant McQueen and Roley(1993) also nd that the MMS surveys are more accurate inthe sense of having lower mean squared errors than theforecasts from standard autoregressive time-series models

FIGURE 2 US MACROECONOMIC ANNOUNCEMENT RELEASE DATES DATA FOR MONTH X

Notes We show the sequence of announcement dates corresponding to data for month X for most of the economic indicatorsused in the paper For example March (month X) consumer credit data are announced between May (month X 1 2) 5 andMay 10 GDP data are special because they are released only quarterly Hence the GDP data released in a given month areeither advance preliminary or nal depending on whether the month is the rst second or third of the quarter For example rst quarter Q1 GDP advance data are announced between April (month X 1 1) 27 and May 4 rst quarter GDP preliminarydata are announced between May (month X 1 2) 27 and June 4 and rst quarter GDP nal data are announced between June(month X 1 3) 27 and July 4 The table is based on 2001 Schedule of Release Dates for Principal Federal EconomicIndicators produced by the US Of ce of Management and Budget and available at httpclinton4naragovtextonlyOMBpubpresspei2001html


high-frequency exchange-rate movements arelinked to fundamentals and if so how Ourmotivations are twofold The rst motivation isobviously the possibility of re ning our under-standing of the fundamental determinants ofexchange rates the central and still largely un-resolved question of exchange-rate economicsThe second motivation is the possibility of im-proved high-frequency volatility estimation viaallowance for jumps due to news as misspeci- cation of the conditional mean (for example byfailing to allow for jumps if jumps are in factpresent) will produce distorted volatility esti-mates in discrete time8

A Modeling the Response of Exchange Ratesto News

We model the 5-minute spot exchange rateRt as a linear function of I lagged values ofitself and J lags of news on each of K funda-mentals

(1) R t 5 b0 1 Oi 5 1

I

b i R t 2 i

1 Ok 5 1

K Oj 5 0

J

bkj Sk t 2 j 1 laquo t

t 5 1 T

As discussed earlier K 5 41 and T 5496512 We chose I 5 5 and J 5 2 based onthe Schwarz and Akaike information criteria9

We allow the disturbance term in the5-minute return model (1) to be heteroskedasticFollowing Andersen and Bollerslev (1998) weestimate the model using a two-step weightedleast-squares (WLS) procedure We rst esti-mate the conditional mean model (1) by ordi-nary least-squares regression and then weestimate the time-varying volatility of laquot fromthe regression residuals which we use to per-form a weighted least-squares estimation of (1)We approximate the disturbance volatility usingthe model

(2) zlaquo tz 5 c 1 csd~t

288

1 Ok 5 1

K Oj9 5 0

J9

bkj9zSkt 2 j9z

1 X Oq 5 1

Q X dqcosX q2pt

288 D1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J0

grj0 Drt 2 j0D 1 u t

The left-hand-side variable zlaquotz is the absolutevalue of the residual of equation (1) whichproxies for the volatility in the 5-minute intervalt As revealed by the right-hand side of equation(2) we model 5-minute volatility as drivenpartly by the volatility over the day containingthe 5-minute interval in question sd(t) partlyby news Skt and partly by a calendar-effectpattern consisting largely of intraday effects thatcapture the high-frequency rhythm of devia-tions of intraday volatility from the daily aver-age Speci cally we split the calendar effectsinto two parts The rst is a Fourier exibleform with trigonometric terms that obey a strict

8 In this paper we are primarily interested in exchange-rate volatility only insofar as it is relevant for inferenceregarding exchange-rate conditional mean dynamics Wehave reserved for future work a detailed analysis of vola-tility in relation to conditional mean and variance jumpsFor a discussion of the effects of conditional mean andvariance jumps on realized volatility see Andersen et al(2003)

9 We also tried allowing for negative J to account forannouncement leakage before the of cial time and moregenerally to account for the fact that the MMS forecastsmight not capture all information available immediatelybefore the announcement but doing so proved unnecessaryThis accords with the earlier-discussed nding of Balduzziet al (2001) that to a good approximation the MMSforecasts do capture all information available immediatelybefore the announcement Moreover if leakage is present(and introspection if not the empirics suggests that there is

likely some leakage however small) then our estimatednews response coef cients which correspond only to theimpact at the time of the of cial announcement are lowerbounds for the total news impact


periodicity of one day10 The second is a set ofdummy variables Drt capturing the Japaneselunch the Japanese open and the US lateafternoon during US daylight saving time

Let us explain in greater detail Consider rstthe daily volatility sd(t) which is the one-day-ahead volatility forecast for day d(t) (the daythat contains time t) from a simple daily con-ditionally Gaussian GARCH(1 1) model usingspot exchange-rate returns from January 2 1986through December 31 1998 Because sd(t) isintended to capture the ldquoaveragerdquo level of vol-atility on day d(t) it makes sense to construct itusing a GARCH(1 1) model which is routinelyfound to provide accurate approximations todaily asset return volatility dynamics11

Now consider the Fourier part for the calen-dar effects This is a very exible functionalform that may be given a semi-nonparametricinterpretation (A Ronald Gallant 1981) TheSchwarz and Akaike information criteria chosea rather low Q 5 4 for all currencies whichachieves parametric economy and promotessmoothness in the intraday seasonal pattern

Finally consider the news effects S and non-Fourier calendar effects D To promote tracta-bility while simultaneously maintaining exibilitywe impose polynomial structure on the responsepatterns associated with the bkj9 and grj0 param-eters12 For example if a particular news sur-prise affects volatility from time t0 to time t0 1J9 we can represent the impact over the eventwindow t 5 0 1 J9 by a polynomialspeci cation p(t) 5 c0 1 c1t 1 1 cPtPFor P 5 J9 this would imply the estimation ofJ9 1 1 polynomial coef cients and would not

constrain the response pattern in any way Useof a lower-ordered polynomial however con-strains the response in helpful ways it promotesparsimony and hence tractability retains exi-bility of approximation and facilitates the im-position of sensible constraints on the responsepattern For example we can enforce the re-quirement that the impact effect slowly fades tozero by imposing p( J9) 5 0

Polynomial speci cations ensure that the re-sponse patterns are completely determined bythe response horizon J9 the polynomial orderP and the endpoint constraint imposed onp( J9) For news effects S we take J9 5 12P 5 3 and p( J9) 5 013 The last conditionleads to a polynomial with one less parametersubstituting t 5 12 into p(t) we have p(t) 5c0[1 2 (t 12)3] 1 c1t[1 2 (t 12)2] 1c2t

2[1 2 (t 12)] We estimate each polyno-mial separately for all announcements and foreach exchange rate For example payroll em-ployment polynomial parameter estimates are(c0 c1 c2) 5 (0177175 200645 0008367) forthe DM$ (0163146 2005544 000704) for theCHF$ (0114488 2003795 000477) for thePound$ (0108867 2003186 0004003) for theEuro$ and (011717 2004619 0006289) forthe Yen$ Finally bkj9 5 gkpk( j9) where gk isthe coef cient estimate in equation (2) As forthe non-Fourier calendar-effect response pat-terns D for the Japanese market opening we useJ0 5 6 P 5 1 p(J0) 5 0 for the Japanese lunchhour we use J0 5 0 (ie a standard dummyvariable with no polynomial response) and forthe US late afternoon during US daylightsaving time we use J 0 5 60 P 5 2 andp(0) 5 p( J0) 5 014

In closing this subsection we note that wecould have handled the volatility dynamics dif-ferently In particular instead of estimating ex-plicit parametric models of volatility dynamicswe could have simply estimated equation (1)using heteroskedasticity- and serial-correlation

10 We also translate the Fourier terms leftward as appro-priate during US daylight saving time (Only North Amer-ica and Europe have daylight saving time)

11 For surveys of GARCH modeling in nancial envi-ronments see Bollerslev et al (1992) and Diebold and JoseLopez (1995) Other possibilities also explored with littlechange in qualitative results include use of daily realizedvolatilities as in Andersen et al (2001) and Andersen et al(2001a 2003)

12 This is particularly important in the case of condi-tional variance as opposed to conditional mean dynamicsbecause conditional variances turn out to adjust to shocksmore slowly than do conditional means thereby involvinglonger distributed lags as we will subsequently emphasizeHence although tractability did not require the impositionof polynomial shape on the conditional mean distributedlags it greatly enhances the accuracy of the conditionalvariance estimates

13 The ldquoconstraintrdquo that volatility news effects linger forat most an hour ( J9 5 12) is nonbinding Initial experi-mentation allowing for J9 5 36 revealed that one hour wasenough for full adjustment for all indicators and currencies

14 The Japanese opening is at 8 PM Eastern daylightsaving time the Japanese lunch hour is 11 PM through1230 AM Eastern daylight saving time the US lateafternoon during daylight saving time is de ned to start at 3PM Eastern daylight saving time


consistent (HAC) standard errors We nd thatapproach less attractive than the one we adoptedfor at least three reasons First we are interestednot only in performing heteroskedasticity-robustinference about the coef cients (done both byour WLS and by HAC estimation) in equation(1) but also in obtaining the most ef cientestimates of those coef cients Second al-though HAC estimation is asymptotically ro-bust to residual heteroskedasticity of unknownform its general robustness may come at theprice of inferior nite-sample performance rel-ative to the estimation of a well-speci ed para-metric volatility model15 Third despite the factthat they are not central to the analysis in thepresent paper both the intra- and interday vol-atility patterns are of intrinsic nancial eco-nomic interest and hence one may want estimatesof these in other situations Notwithstanding allof these a priori arguments against the use ofHAC estimation in the present context as acheck on the robustness of our results we alsoperformed all of the empirical work related tothe mean effects using HAC estimation with nochange in any of the qualitative results (al-though a number of the coef cients were nolonger statistically signi cant)

B News Effects I News AnnouncementsMatter and Quickly

The model (1)ndash(2) provides an accurate ap-proximation to both conditional mean and con-ditional variance dynamics Since the modelcontains so many variables and their lags itwould prove counterproductive to simply reportall of the parameter estimates Instead Figure 3shows the actual and tted average intradayvolatility patterns which obviously agree fairlyclosely Further in Figure 4 we present graph-ically the results for the most important indica-tors and we discuss those results (and someothers not shown in the gure) in what follows

Let us rst consider the effects of US mac-roeconomic news Throughout news exerts agenerally statistically signi cant in uence onexchange rates whereas expected announce-ments generally do not That is only unantici-pated shocks to fundamentals affect exchange rates in accordance with the predictions of ra-

tional expectations theory Many US indica-tors have statistically signi cant news effectsacross all currencies including payroll employ-ment durable goods orders trade balance ini-

15 See C Radhakrishna Rao (1970) and Andrew Chesherand Ian Jewitt (1987)

FIGURE 3 ACTUAL AND FITTED INTRADAY

VOLATILITY PATTERNS

Notes The solid line is the average intraday pattern of theabsolute residual return zlaquo tz over the 288 5-minute intervalswithin the day where laquot is the residual from the exchange-rate conditional mean model (1) in the text The dashed lineis the tted intraday pattern of zlaquo tz from the exchange-ratevolatility model (2) in the text To avoid contaminationfrom shifts in and out of daylight saving time we constructthe gures using only days corresponding to US daylightsaving time


tial unemployment claims NAPM index retailsales consumer con dence and advance GDP

The general pattern is one of very quick ex-change-rate conditional mean adjustment char-acterized by a jump immediately following theannouncement and little movement thereafterFavorable US ldquogrowth newsrdquo tends to producedollar appreciation and conversely This is con-sistent with a variety of models of exchange-ratedetermination from simple monetary models(eg Mark 1995) to more sophisticated frame-works involving a US central bank reactionfunction displaying a preference for low in a-

tion (eg John B Taylor 1993)16 One cansee from the center panel of the rst row ofFigure 4 for example that a one standarddeviation US payroll employment surprisetends to appreciate (if positive) or depreciate (ifnegative) the dollar against the DM by 016

16 For most of our macroeconomic indicators includingthose on which we primarily focus the sign of a ldquogoodshockrdquo is clear movements associated with increased realUS economic activity are good for the dollar Sometimeshowever it is not obvious which direction should be viewedas good as perhaps with consumer credit

FIGURE 4 EXCHANGE-RATE RESPONSES TO US NEWS

Notes We graph the three news response coef cients associated with the exchange-rate conditional mean regression (1)corresponding to responses at the announcement ve minutes after the announcement and ten minutes after the announce-ment We also show two standard error bands under the null hypothesis of a zero response obtained using the weightedleast-squares estimation method described in the text


percent17 This is a sizeable move from bothstatistical and economic perspectives On thestatistical side we note that only 07 percent ofour 5-minute returns show an appreciation ordepreciation bigger than 010 percent On theeconomic side we note that 016 percent is alsolarge relative to the average DM$ spreadwhich tends to be around 006 percent duringthe period we study (see Hendrik Bessem-binder 1994 and Joel Hasbrouck 1999 Table 1)

It is important to note that although closelytimed news events are highly correlated thecorrelation does not create a serious multicolin-earity problem except in a few speci c in-stances For example industrial production andcapacity utilization are released at the sametime and they are highly correlated (064) Ingeneral however the event that two announce-ments within the same category (eg real ac-tivity) are released simultaneously is rare

Now let us focus on the DM$ rate in somedetail It is of particular interest both because of itscentral role in the international nancial systemduring the period under study and because wehave news data on both US and German macro-economic indicators18 First consider the effects ofUS macroeconomic news on the DM$ rateNews announcements on a variety of US indica-tors signi cantly affect the DM$ rate includingpayroll employment durable goods orders tradebalance initial claims NAPM index retail salesconsumer con dence CPI PPI industrial produc-tion leading indicators housing starts construc-tion spending federal funds rate new home salesand GDP (advance preliminary and nal)

Now consider the effect of German macroeco-nomic news on the DM$ rate19 In sharp contrastto the large number of US macroeconomic indi-cators whose news affect the DM$ rate only veryfew of the German macroeconomic indicatorshave a signi cant effect (M3 and industrial pro-duction) We conjecture that the disparity may bedue to the fact detailed in Table 1 that the releasetimes of US macroeconomic indicators areknown exactly (day and time) but only inexactly

for Germany (day but not time) Uncertain releasetimes may result in less market liquidity (andtrading) around the announcement times henceresulting in smaller news effects around the an-nouncements ultimately producing a more grad-ual adjustment perhaps for a few hours after theannouncements Alternatively greater prean-nouncement leakage in Germany may result inadjustments taking place gradually in the daysprior to the actual announcement

Most of the explanatory power of the ex-change-rate conditional mean model (1) comesfrom the lagged values of the dependent vari-able and the contemporaneous news announce-ment Hence although 58 percent of the days inour sample contain a news announcement to agood approximation the news predicts only thedirection and magnitude of the exchange-ratemovement during the 5-minute post-release in-tervals which correspond to only two-tenths ofone percent of the sample observations To fo-cus on the importance of news during announce-ment periods we now estimate the model

(3) R t 5 bk S kt 1 laquo t

where Rt is the 5-minute return from time t totime t 1 1 and Skt is the standardized newscorresponding to announcement k (k 5 1 41) at time t and the estimates are based ononly those observations (Rt Skt) such that anannouncement was made at time t

We show the estimation results in Table 2which contains a number of noteworthy fea-tures First news on many of the fundamentalsexerts a signi cant in uence on exchange ratesThis is of course expected given our earlierestimation results for equation (1) as summa-rized in Figure 4 News from FOMC delibera-tions for example clearly in uences exchangerates the large and statistically signi cant co-ef cients and the high R2rsquos are striking Theirpositive signs indicate that as expected for ex-ample in a standard monetary model Fed tight-ening is associated with dollar appreciation20

17 We interpret a one-standard-deviation surprise as ldquotypicalrdquo18 German news is the only non-US news that is readily

available from MMS19 To a rst approximation German news is relevant

only for DM$ determination in contrast to US newswhich is relevant for the determination of all US dollarexchange rates

20 It would be interesting (with a longer sample of data) toexamine the stability of the response coef cient over differentstages of the business cycle see eg McQueen and Roley(1993) According to the standard US business-cycle chro-nology produced by the National Bureau of Economic Re-search the United States was in an expansion from March of1991 until March of 2001 hence our entire sample


TABLE 2mdashUS AND GERMAN CONTEMPORANEOUS NEWS RESPONSE COEFFICIENTS AND R2 VALUES

Announcement

Pound$ Yen$ DM$ CHF$ Euro$

bk R2 bk R2 bk R2 bk R2 bk R2

US Announcements

Quarterly Announcements1 GDP advance 0029 0098 0036 0102 008 0301 0079 0307 0061 04202 GDP preliminary 0038 0134 0022 0081 0055 0185 0057 0207 0017 00483 GDP nal 20004 0004 0019 0048 0017 0029 0010 0007 0006 0010


4 Nonfarm payroll employment 0098 0189 0084 0214 0161 0237 0144 0269 008 02325 Retail sales 0048 0225 0019 0066 0067 0241 0059 0170 0041 01936 Industrial production 0020 0105 0019 0078 0029 0131 0034 0147 0018 00867 Capacity utilization 0017 0061 0016 0055 0021 0046 0023 0058 0018 00418 Personal income 0007 0015 0001 0000 0006 0007 0003 0001 20005 00059 Consumer credit 0002 0002 0009 0019 0004 0012 0002 0002 20002 0004

Consumption

10 Personal consumption expenditures 20003 0003 0005 0006 20007 0010 20011 0012 0007 000811 New home sales 0002 0002 0011 0030 001 0015 20002 0001 0005 0003Investment

12 Durable goods orders 0055 0266 0027 0081 0088 0363 0085 0355 0043 023713 Construction spending 0019 0087 001 0026 0031 0091 0017 0034 0015 003014 Factory orders 0011 0024 0006 0006 0018 0038 0019 0041 0031 010215 Business inventories 20004 0008 001 0029 0009 0012 0002 0001 0007 0015Government Purchases

16 Government budget de cit 0007 0057 0008 0038 0002 0003 0010 0050 0003 0006Net Exports

17 Trade balance 0092 0529 0112 0370 0138 0585 0124 0480 0084 0414Prices

18 Producer price index 0005 0003 0000 0000 0019 0020 0017 0017 0018 004619 Consumer price index 0016 0048 0012 0033 0031 0101 0035 0104 0015 0027Forward-looking

20 Consumer con dence index 0037 0174 0022 0103 0058 0222 0054 0214 0035 018921 NAPM index 0028 0199 0012 0036 0039 0141 0036 0146 0025 007422 Housing starts 0006 0008 0005 0007 0017 0028 002 0033 0008 000923 Index of leading indicators 0012 0031 0009 0006 0012 0009 0011 0005 20005 0005

Six-Week Announcements24 Target federal funds rate 0048 0229 0050 0162 0072 0259 0072 0230 0032 0142

Weekly Announcements25 Initial unemployment claims 20014 0025 20012 0019 20022 0036 20026 0046 20019 005826 Money supply M1 0000 0000 0000 0000 0004 0020 0004 0019 0002 000927 Money supply M2 0000 0000 20001 0001 0004 0019 0005 0030 0002 001328 Money supply M3 0000 0000 0001 0002 0002 0004 0004 0023 0002 0011

German Announcements

Quarterly Announcements29 GDP 20004 0042 20002 0001 20007 0022 20011 0068 20004 0015


30 Employment 0000 0000 0002 0001 0000 0000 001 0045 0003 000331 Retail sales 0001 0001 0004 0008 20003 0004 20002 0003 2001 009132 Industrial production 20011 0059 20009 0036 20017 0172 20015 0105 20005 0015


Second unlike the R2 values for equation(1) which are typically very small the R2

values for equation (3) are often quite highNews announcements occur comparativelyrarely and have a nonnegligible but short-lived impact on exchange rates hence the R2

in an equation such as (3) must be low whencomputed across all 5-minute observations Incontrast one naturally expects higher R2 val-ues when computed using only announcementobservations although the precise size is ofcourse an empirical matter Table 2 reveals R2

values that are often around 03 and some-times approaching 06

Finally it is interesting to note that theresults of Yin-Wong Cheung and ClementYuk-Pang Wong (2000) and Cheung andMenzie David Chinn (2001) obtained by sur-veying traders cohere reassuringly with themodel-based results documented here In par-ticular Cheung and Chinn (2001) report thattraders believe that exchange rates adjust al-most instantaneously following news an-nouncements and that news regarding realvariables is more in uential than news re-garding nominal variables which is entirelyconsistent with the empirical results reportedin Table 2

C News Effects IIAnnouncement Timing Matters

One might wonder whether within the samegeneral category of macroeconomic indicatorsnews on those released earlier tend to havegreater impact than those released later Toevaluate this conjecture we grouped the USindicators into seven types real activity con-sumption investment government purchasesnet exports prices and forward-looking Withineach group we arranged the announcements inthe chronological order described in Figure 2The conjecture is generally veri ed In the es-timates of equation (3) within each indicatorgroup the announcements released earliest tendto have the most statistically signi cant coef -cients and the highest R2 values21 In Figure 5we plot the R2 of equation (3) within eachindicator group as a function of the announce-ment timing The clearly prevalent downwardslopes reveal that the early announcements doindeed have the greatest impact

The fact that ldquoannouncement timing mattersrdquo

21 One exception is the nominal group the consumerprice index seems more important than the producer priceindex despite its earlier release date


Announcement



Investment

33 Manufacturing orders 20007 0025 20008 0029 20011 0061 2001 0042 20002 000234 Manufacturing output 20001 0001 20017 0091 20007 0041 20009 0048 20007 0034Net Exports

35 Trade balance 20004 0018 0001 0000 0000 0000 0001 0001 20005 001936 Current account 20003 0009 0006 0019 20006 0035 20006 0033 20006 0031Prices

37 Consumer price index 20020 0159 20004 0016 0000 0000 0007 0016 20001 000138 Producer prices 20002 0003 0003 0012 20003 0003 20004 0011 20008 001539 Wholesale price index 0000 0000 0003 0003 20011 0039 20003 0005 0004 001240 Import prices 0007 0079 20009 0049 0003 0005 0006 0019 20003 0003Monetary

41 Money stock M3 2002 0215 0000 0000 20033 0181 2002 0113 20023 0161

Notes We estimate the contemporaneous exchange-rate news response model Rt 5 bkSkt 1 laquo t where Rt is the 5-minutereturn from time t to time t 1 1 and Skt is the standardized news corresponding to announcement k (k 5 1 41) madeat time t We estimate the regression using only those observations ( Rt Skt) such that an announcement was made at timet We report the bk and R2 values and we mark with an asterisk those coef cients that are statistically signi cant at the5-percent level using heteroskedasticity- and autocorrelation-consistent standard errors


helps with the interpretation of our earlier-reported empirical results in Table 2 whichindicate that only seven of the 40 announce-ments signi cantly impacted all the currency

speci cations The reason is that many of theannouncements are to some extent redundantand the market then only reacts to those releasedearlier Hence for example US durable goods

FIGURE 5 US NEWS EFFECTS AS A FUNCTION OF RELEASE TIME

Notes We estimate the contemporaneous exchange-rate news response model Rt 5 bkSkt 1 laquo t where Rt is the 5-minutereturn from time t to time t 1 1 and Skt is the standardized news corresponding to announcement k (k 5 1 17) madeat time t We estimate the regression using only those observations ( Rt Skt) such that an announcement was made at timet On the vertical axis we display the R2 values and on the horizontal axis we display macroeconomic news announcementsin the chronological order documented in Table 2 The ldquonews numbersrdquo are as follows

GDP Real Activity Investment Forward-Looking

1 GDP advance 4 Payroll employment 10 Durable goods orders 14 Consumer con dence2 GDP preliminary 5 Retail sales 11 Construction spending 15 NAPM index3 GDP nal 6 Industrial production 12 Factory orders 16 Housing starts

7 Capacity utilization 13 Business inventories 17 Index of leading indicators8 Personal income9 Consumer credit


orders matter for all currency pairs but USfactory orders which are released later do not

D News Effects III Volatility Adjusts toNews Gradually

As discussed previously and documented inFigure 4 exchange rates adjust to news imme-diately It is interesting to note however thatexchange-rate volatilities adjust only graduallywith complete adjustment occurring only afterJ9 5 12 5-minute periods or one hour

We provide details in Table 3 As alreadynoted and as shown again in the top panel ofthe table the contemporaneous return re-sponse coef cients are sizeable and statisti-cally signi cant and the full response occursimmediately In contrast the contemporane-ous volatility response coef cients althoughstatistically signi cant are smaller as shownin the middle panel of the table Importantlyhowever the complete response of volatilityto news occurs only after an hour or so and itis noticeably larger than either the contempo-

raneous volatility response or the contem-poraneous return response as shown in thebottom panel of the table

E News Effects IV Pure AnnouncementEffects are Present in Volatility

It is possible that the mere presence of anannouncement might boost volatility quiteapart from the size of the associated surpriseTo explore this possibility we add to the re-turns equation (1) J lags of announcementperiod dummies on each of K fundamentalsand we also add to the volatility equation (2)J9 lags of announcement period dummies oneach of K fundamentals As shown in Table4 the announcement dummies are generallyinsigni cant in the returns equation (1) butgenerally signi cant in the volatility equation(2) in line with earlier results for bond mar-kets such as Fleming and Remolona (19971999) News effects are still important how-ever in both conditional mean and variancedynamics

TABLE 3mdashRETURN AND VOLATILITY NEWS RESPONSE COEFFICIENTS

Announcement Pound$ Yen$ DM$ CHF$ Euro$

Contemporaneous Return Response bk0

Nonfarm payroll employment 0092 0072 0159 0115 0081Durable goods orders 0055 0029 0084 0083 0041Trade balance 0083 0115 0142 0131 0084Initial unemployment claims 20010 20009 20018 20024 20017

Contemporaneous Volatility Response bk0


Cumulative Volatility Response yenj95 0J9 bkj9


Notes We estimate the exchange-rate conditional mean model (1) Rt 5 b0 1 yeni5 1I biRt 2 i 1 yenk5 1

K yenj5 0J bkjSkt2 j 1 laquot

and we report estimates of the contemporaneous response of exchange-rate returns to news bk0 We also estimate thedisturbance volatility model (2)

zlaquo tz 5 c 1 csd ~t

2881 O

k 5 1

K Oj9 5 0

J9

bkj9zSk t 2 j9 z 1 X Oq 5 1

Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D 1 Or 5 1

R Oj0 5 0

J0

grj0 Dr t 2 j0D 1 ut

and we report estimates of the contemporaneous response of exchange-rate volatility to news bk0 5 gkpk(0) as describedin the text Finally we also report estimates of the cumulative volatility response yenj95 0

12 gkpk( j9) as described in the textAsterisks denote statistical signi cance at the 5-percent level


F News Effects V Announcement Effects areAsymmetricmdashResponses Vary with the

Sign of the News

We have seen that news about macroeconomicfundamentals signi cantly affect high-frequencyexchange rates Thus far we have allowed only forconstant news effects but it is natural to go fartherand ask whether the news effects vary with thesign of the surprise To address this issue wegeneralize equation (3) by allowing the impactresponse coef cient bk to be a linear function ofthe news surprise Skt allowing for a differentconstant and slope on each side of the origin

(4) bk 5 5 b0k 1 b1k S kt if S t 0b2k 1 b3k Skt if St 0

Inserting (4) into (3) yields the impact responsespeci cation

(5) R t 5 5 b0k Skt 1 b1k S kt2 1 laquo t if S t 0

b2k Skt 1 b3k Skt2 1 laquo t if St 0

Following Robert F Engle and Victor K Ng(1993) we call the union of b0kSkt 1 b1kSkt

2

to the left of the origin and b2kSkt 1 b3kSkt2 to the

TABLE 4mdashRETURN AND VOLATILITY NEWS RESPONSE COEFFICIENTS AND ANNOUNCEMENT DUMMY COEFFICIENTS


Contemporaneous Return ResponseNonfarm payroll employment

bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020

Durable goods ordersbk0 0052 0028 0082 0083 0039uk0 20023 20004 20023 20017 20010

Trade balancebk0 0086 0121 0144 0131 0085uk0 0013 0029 0013 0004 0012

Initial claimsbk0 20009 20009 20017 20023 20017uk0 0001 20010 20005 20002 20006

Contemporaneous Volatility ResponseNonfarm payroll employment

bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597




Notes We add to equation (1) J lags of announcement period dummies on each of K fundamentals Rt 5 b0 1 yeni5 1I

biRt2 i 1 yenk5 1K yenj5 0

J bkjSkt 2 j 1 yenk5 1K yenj5 0

J ukjDkt 2 j 1 laquot and we report estimates of the contemporaneous returnresponse to news and to announcement periods bk0 and uk0 respectively We also add to equation (2) J9 lags ofannouncement period dummies on each of K fundamentals

z laquo tz 5 c 1 csd~t

2881 O

k 5 1

K Oj9 5 0

J9

b kj9zSk t 2 j9z 1 Ok 5 1

K Oj9 5 0

J9

ukj9 D kt 2 j9 1 X Oq 5 1

Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0

g rj0 Drt 2 j0D 1 ut

and report estimates of the contemporaneous return response to news and to announcement periods bk0 and uk0 respectivelyAsterisks denote statistical signi cance at the 5-percent level


right of the origin the ldquonews impact curverdquo22 Inthe top row of Figure 6 we show the news impactcurves averaged across all macroeconomic funda-mentals k 5 1 41 It is clear that on average

the effect of macroeconomic news often varieswith its sign In particular negative surprises oftenhave greater impact than positive surprises23

It is interesting to see whether the sign effectprevails when we look separately at the most

22 Despite the super cial resemblance in terms of docu-menting asymmetric responses to news our work is verydifferent from that of Engle and Ng (1993) and manysubsequent related studies In particular the EnglendashNg newsimpact curve tracks the variance of equity returns condi-tional upon the sign and size of past returns (with noallowance for a time-varying conditional mean return)whereas our news impact curve tracks the mean of foreignexchange returns conditional upon the sign and size ofmacroeconomic news

23 To the best of our knowledge such sign effects havenot previously been documented for the foreign exchangemarket Evidence of asymmetric conditional-mean newseffects exists in other contexts however For exampleJennifer Conrad et al (2001) nd asymmetric effects ofearnings news on stock returns while recent concurrentwork by Hautsch and Hess (2001) details an asymmetricresponse to employment news in the T-bond futures market

FIGURE 6 US NEWS IMPACT CURVES

Notes In the top row we show the news impact curves averaged across all macroeconomic fundamentals k 5 1 41 Inthe remaining rows we show the news impact curves for payroll employment trade balance durable good orders and initialclaims See text for details


important news announcements In the remain-ing rows of Figure 6 we show the news impactcurves for payroll employment trade balancedurable goods orders and initial claims Thesign effect is generally maintained althoughthere is some variation across indicators andcurrencies Asymmetry in the Yen$ DM$ andCHF$ response to payroll employment andtrade balance news for example is very pro-nounced whereas it is largely absent in thePound$ and Euro$ response

In the next section we explore more deeplythe economics behind the asymmetric responseRecent theoretical models suggest that theasymmetry may be driven in part by the dy-namics of uncertainty regarding the underlyingstate of the economy It turns out that our MMSdata set contains not only expectations but alsoa measure of the cross-sectional dispersion ofexpectations the standard deviation Using thecross-sectional standard deviation as a proxy forstate uncertainty we can therefore directly as-sess a key mechanism thought to generateasymmetric response to which we now turn

III Asymmetric Response InformationProcessing and Price Discovery

Two strands of literature imply asymmetry inthe response of exchange rates to news In par-ticular they imply that bad news in ldquogoodtimesrdquo should have an unusually large impact aview that is also common in the practitionercommunity as emphasized by Conrad et al(2001) Note that our entire sample takes placein good timesmdash1992 through 1998 Hence thetheoretical prediction that ldquobad news in goodtimes should have unusually large effectsrdquo de-generates in our sample period to ldquobad newsshould have unusually large effectsrdquo which toa reasonably good approximation is what wefound earlier

The rst strand of the literature is ldquobehavior-alrdquo and focuses primarily on equities at the rmlevel Nicholas Barberis et al (1998) for exam-ple model investors as believing that rm earn-ings follow a two-state regime-switchingprocessmdasherroneously as earnings actually fol-low a random walkmdashwith mean-reverting earn-ings in state 0 and upward-trending earnings instate 1 Hence a series of positive earnings leadsinvestors to infer that state 1 holds with theconcomitant expectation of additional positive

earnings In such a situation bad news gener-ates a large negative response because it is asurprise whereas good news generates little re-sponse because it is anticipated

The second relevant strand of the literatureuses a rational-expectations equilibrium ap-proach and focuses more on the market level asopposed to the rm level as in Pietro Veronesi(1999) Timothy C Johnson (2001a b) andAlexander David and Veronesi (2001) Veronesi(1999) in particular models investors as (cor-rectly) believing that the economy follows atwo-state regime-switching process with ldquolowrdquoand ldquohighrdquo states corresponding to recessionsand expansions Agents solve a signal extrac-tion problem to determine the probability p(t)of being in the high state and equilibrium assetprices can be shown to be increasing and convexfunctions of p(t) The intuition for this keyresult is simple Suppose that p(t 2 1) rsquo 1ie investors believe that the high state almostsurely prevails Then if bad news arrives at timet two things happen rst expected future assetvalues decrease and second p(t) decreases(ie state risk increases) Risk-averse investorsrequire additional returns for bearing this addi-tional risk hence they require an additionaldiscount on the asset price which drops bymore than it would in a present-value modelConversely suppose investors are con dentthat the low state prevails ie p(t 2 1) rsquo 0Then if good news arrives at time t expectedfuture asset values increase but p(t) alsoincreases (ie state risk again increases) Asbefore investors require additional returns forbearing this additional risk hence they re-quire a discount on the asset price whichincreases by less than it would in a present-value model

For a number of reasons it is not our inten-tion here to explicitly test the practitioner claimthat prices respond most strongly to bad news ingood times or to directly implement Veronesirsquosmodel or to combine it with the Barberis-Shleifer-Vishy model First our data set is notwell-suited to that purpose as mentioned aboveit contains only the expansionary 1990rsquos Sec-ond Conrad et al (2001) have already madeadmirable progress in that regard nding gen-eral support for the assertion that (stock) pricesrespond most strongly to bad news in goodtimes Third the Barberis-Schleifer-Vishnymodel is not particularly well-suited to the forex


context relevant here as it focuses on the earn-ings stream for an individual rm

Instead we take as true the practitioner claimthat prices respond most strongly to bad news ingood times and we focus on the explanationembodied in Veronesirsquos model We use an in-teresting feature of our MMS expectations datato assess the key alleged mechanism throughwhich bad news in good times translates intolarge price moves increased uncertainty aboutthe state of the economy In particular we havedata not only on the median expectations ofmacroeconomic fundamentals but also on theassociated standard deviations across the indi-vidual forecasters Hence we can check directlywhether uncertainty about the state of the econ-omy as proxied by the standard deviation ofexpectations across the individual forecastersincreases following the arrival of bad news ingood times24 Before proceeding to examine theeffect of bad news arrivals on subsequent fore-cast dispersion however two issues arise

First it is not clear what timing in the datamatches the generic timing in the modelClearly bad news at time t 2 1 means thatexpectations for time t are formed in a bad newsenvironment but what if the news at t 2 2 wasbad and the news at t 2 1 was not Perhapsagents have a memory that lasts longer than oneannouncement period so that even the lattercase could be viewed as a bad news environ-ment In general we might say that we are in abad news environment if the news was bad atany of times t 2 1 t 2 2 t 2 d for somed Second to enhance our chances of detectingthe ldquoVeronesi effectrdquo if it exists we may notwant to track the arrival of all bad news butrather only bad news that exceeds some mini-mal threshold say the pth percentile of thedistribution of bad news where p like d mustbe chosen As a benchmark we simply set d 51 and p 5 50 percent (ie the median)

Figure 7 plots the corresponding standarddeviation of the MMS payroll employment du-rable goods orders and trade balance forecastsThe shaded areas indicate a bad news environ-ment using the criteria d 5 1 and p 5 50percent Analyst forecast dispersion is indeed

higher following bad news than at other timesspeci cally the uncertainty of payroll employ-ment is 30 percent higher the uncertainty ofdurable goods orders is 6 percent higher andthe uncertainty of the trade balance is 12 percenthigher These effects are robust to reasonablevariation in p and d

IV Concluding Remarks and Directions forFuture Research

The goal of the research on which this paperreports is to deepen our understandingof the links

24 Of course the notion of forecast uncertainty and theforecast dispersion across forecasters are not exactly equivalentconcepts Victor Zarnowitz and Louis A Lambros (1987)show however that they are generally positively correlated

FIGURE 7 FORECAST UNCERTAINTY

Notes We plot the time series of cross-sectional standarddeviations of the Money Market Services forecasts Theshaded areas denote ldquobad newsrdquo times See text for details


between exchange-rate movements and newsabout fundamentals To that end in this paper wehave documented important news effects withasymmetric response patterns Let us conclude byrelating our results to work on order ow anddrawing implications for future research

In recent innovative work Evans and Lyons(2002) show that signed order ow is a goodpredictor of subsequent exchange-rate move-ments This work is important in that it en-hances our understanding of the determinants ofhigh-frequency exchange-rate movements butless satisfying in that it remains ignorant aboutthe determinants of high-frequency order owWe in contrast have shown that news affectsexchange rates Combining our perspectives fo-cuses attention on the causal links among newsorder ow and forex movements which in ourview is a prime candidate for future research Itwill be of interest for example to determinewhether news affects exchange rates via order ow or instantaneously25 In work done subse-quently to the rst draft of this paper Evans andLyons (2001) and Kenneth A Froot and TarunRamadorai (2002) tackle precisely that issue

A second key direction for future research ispushing farther with the implications of Veronesi(1999) for the analysis of high-frequency newseffects Presently we have veri ed that the keymechanism that ampli es the effects of bad newsin good times in Veronesirsquos modelmdashincreasedstate uncertaintymdashis operative in the dataHowever one could potentially go farther andexploit the broader implications of Veronesirsquoswork for our approach namely that news effectsare in general a function of state uncertainty byincluding interactions of news with state uncer-tainty in both our conditional mean and condi-tional variance speci cations This would beparticularly interesting if data were available onexchange rates and fundamentals spanning badas well as good times but as of this writingsuch data remain elusive

Third it would be of interest to explore notonly the effects of regularly scheduled quanti-tative news on macroeconomic fundamentalsbut also the effects of irregularly scheduledqualitative ldquoheadline newsrdquo as prices and per-

haps order ow may reasonably be expected torespond to both26 It is not obvious howeverhow to do so in a compelling way both theconceptual and the practical complicationsseem daunting

Fourth it will be of interest to attempt ananalysis of structural stability as the marketmay change its view about which news is im-portant for exchange rates or about how tointerpret the sign of a surprise In some inter-pretations for example a positive US in ationsurprise would tend to produce dollar depreci-ation (eg when the US central bank reactionfunction assigns relatively low weight to thelevel of in ation) whereas in other interpreta-tions it would produce dollar appreciation (egwhen the US central bank reaction functionshows strong preference for low in ation as inTaylor 1993)

Finally we look forward to characterizing thejoint responses of the foreign exchange stockand bond markets to real-time news surprisesResponses have now been studied for each mar-ket in isolation Fleming and Remolona (1999)and Balduzzi et al (2001) study the bond mar-ket Mark J Flannery and Aris Protopapadakis(2002) study the stock market and this paper ofcourse studies the foreign exchange market Amultivariate framework however will facilitateanalysis of cross-market movements and inter-actions or lack thereof which may for exampleshed light on agentsrsquo views regarding centralbank reaction functions

REFERENCES

Almeida Alvaro Goodhart Charles A E andPayne Richard ldquoThe Effects of Macroeco-nomic News on High Frequency ExchangeRate Behaviorrdquo Journal of Financial andQuantitative Analysis September 199833(3) pp 383ndash408

Andersen Torben G and Bollerslev Tim ldquoDeut-sche Mark-Dollar Volatility Intraday Activ-ity Patterns Macroeconomic Announcementsand Longer Run Dependenciesrdquo Journal ofFinance February 1998 53(1) pp 219ndash65

Andersen Torben G Bollerslev Tim Diebold

25 For instance Kenneth R French and Richard Roll(1986) and Fleming and Remolona (1999) both argue thatpublicly available news may be incorporated in prices in-stantaneously even without trading

26 Indirect evidence is provided by Dirk Eddelbuttel andThomas H McCurdy (1998) who report signi cantly height-ened foreign exchange-rate volatility in response to the merefrequency of headline news items on the Reuters news screen


Francis X and Ebens Heiko ldquoThe Distribu-tion of Realized Stock Return VolatilityrdquoJournal of Financial Economics July 200161(1) pp 43ndash76

Andersen Torben G Bollerslev Tim DieboldFrancis X and Labys Paul ldquoThe Distributionof Realized Exchange Rate Volatilityrdquo Jour-nal of the American Statistical AssociationMarch 2001a 96(453) pp 42ndash55

ldquoModeling and Forecasting RealizedVolatilityrdquo Econometrica 2003 71 (forth-coming)

Balduzzi Pierluigi Elton Edwin J and Green TClifton ldquoEconomic News and the YieldCurve Evidence from the US TreasuryMarketrdquo Salomon Center Working PaperNo S-98-5 New York University 1998

ldquoEconomic News and Bond PricesEvidence from the US Treasury MarketrdquoJournal of Financial and Quantitative Anal-ysis December 2001 36(4) pp 523ndash43

Barberis Nicholas Shleifer Andrei and VishnyRobert ldquoA Model of Investor SentimentrdquoJournal of Financial Economics September1998 49(3) pp 307ndash43

Bessembinder Hendrik ldquoBidndashAsk Spreads inthe Interbank Foreign Exchange MarketsrdquoJournal of Financial Economics June 199435(3) pp 317ndash48

Bollerslev Tim Cai Jun and Song Frank MldquoIntraday Periodicity Long Memory Volatil-ity and Macroeconomic Announcement Ef-fects in the US Treasury Bond MarketrdquoJournal of Empirical Finance May 20007(1) pp 37ndash55

Bollerslev Tim Chou Ray Y and KronerKenneth F ldquoARCH Modeling in Finance AReview of the Theory and Empirical Evi-dencerdquo Journal of Econometrics AprilndashMay1992 52(1ndash2) pp 5ndash59

Bollerslev Tim and Domowitz Ian ldquoTrading Pat-terns and Prices in the Interbank Foreign Ex-change Marketrdquo Journal of FinanceSeptember 1993 48(4) pp 1421ndash43

Brandt Michael W Edelen R and KavajeczK A ldquoLiquidity in the US Treasury MarketAsymmetric Information and Inventory Ef-fectsrdquo Unpublished manuscript Universityof Pennsylvania 2001

Chesher Andrew and Jewitt Ian ldquoThe Bias of aHeteroskedasticity Consistent CovarianceMatrix Estimatorrdquo Econometrica September1987 55(5) pp 1217ndash22

Cheung Yin-Wong and Chinn Menzie DavidldquoCurrency Traders and Exchange Rate Dy-namics A Survey of the US Marketrdquo Jour-nal of International Money and FinanceAugust 2001 20(4) pp 439ndash71

Cheung Yin-Wong and Wong Clement Yuk-Pang ldquoA Survey of Market PractitionersrsquoViews on Exchange Rate Dynamicsrdquo Jour-nal of International Economics August2000 51(2) pp 401ndash19

Conrad Jennifer Cornell Bradford and LandsmanWayne ldquoWhen Is Bad News Really BadNewsrdquo Unpublished manuscript Universityof North Carolina-Chapel Hill 2001

Dacorogna Michel M Muller Ulrich A NaglerRobert J Olsen Richard B and Pictet OlivierV ldquoA Geographical Model for the Daily andWeekly Seasonal Volatility in the ForeignExchange Marketrdquo Journal of InternationalMoney and Finance August 1993 12(4) pp413ndash38

Danielsson Jon and Payne Richard ldquoReal Trad-ing Patterns and Prices in Spot Foreign Ex-change Marketsrdquo Unpublished manuscriptLondon School of Economics 1999

David Alexander and Veronesi Pietro ldquoIn ationand Earnings Uncertainty and the Volatilityof Asset Prices An Empirical InvestigationrdquoUnpublished manuscript University of Chi-cago 2001

Diebold Francis X and Lopez Jose ldquoModelingVolatility Dynamicsrdquo in Kevin Hoover edMacroeconometrics Developments tensionsand prospects Boston MA Kluwer Aca-demic Press 1995 pp 427ndash72

Dominguez Kathryn M ldquoThe Market Micro-structure of Central Bank Interventionrdquo Na-tional Bureau of Economic Research(Cambridge MA) Working Paper No 73371999

Eddelbuttel Dirk and McCurdy Thomas H ldquoTheImpact of News on Foreign Exchange RatesEvidence from High Frequency Datardquo Un-published manuscript University of Toronto1998

Ederington Louis H and Lee Jae Ha ldquoHowMarkets Process Information News Releasesand Volatilityrdquo Journal of Finance Septem-ber 1993 48(4) pp 1161ndash91

Engle Robert F and Ng Victor K ldquoMeasuringand Testing the Impact of News on Volatili-tyrdquo Journal of Finance December 199348(5) pp 1749ndash78


Evans Martin D D and Lyons Richard K ldquoWhyOrder Flow Explains Exchange Ratesrdquo Un-published manuscript University of Califor-nia-Berkeley 2001

ldquoOrder Flow and Exchange Rate Dy-namicsrdquo Journal of Political Economy Feb-ruary 2002 110(1) pp 170ndash80

Flannery Mark J and Protopapadakis ArisldquoMacroeconomic Factors Do In uence Ag-gregate Stock Returnsrdquo Review of FinancialStudies Summer 2002 15(3) pp 751ndash82

Fleming Michael J and Remolona Eli M ldquoWhatMoves the Bond Marketrdquo Economic PolicyReview Federal Reserve Bank of New YorkDecember 1997 3(4) pp 31ndash50

ldquoPrice Formation and Liquidity in theUS Treasury Market The Response to Pub-lic Informationrdquo Journal of Finance Octo-ber 1999 54(5) pp 1901ndash15

Frankel Jeffrey A and Rose Andrew K ldquoEm-pirical Research on Nominal ExchangeRatesrdquo in Gene Grossman and KennethRogoff eds Handbook of international eco-nomics Amsterdam North-Holland 1995pp 1689ndash1729

French Kenneth R and Roll Richard ldquoStockReturn Variances The Arrival of Informationand the Reaction of Tradersrdquo Journal of Fi-nancial Economics September 1986 17(1)pp 5ndash26

Froot Kenneth A and Ramadorai Tarun ldquoCur-rency Returns Institutional Investor Flowsand Exchange-Rate Fundamentalsrdquo NationalBureau of Economic Research (CambridgeMA) Working Paper No 9101 2002

Gallant A Ronald ldquoOn the Bias in FlexibleFunctional Forms and an Essentially Unbi-ased Form The Fourier Flexible FormrdquoJournal of Econometrics February 198115(2) pp 211ndash45

Goodhart Charles A E Hall Stephen GHenry S G Brian and Pesaran BahramldquoNew Effects in a High-Frequency Model ofthe Sterling-Dollar Exchange Raterdquo Journalof Applied Econometrics JanuaryndashMarch1993 8(1) pp 1ndash13

Goodhart Charles Ito Takatoshi and PayneRichard ldquoOne Day in June 1993 A Study ofthe Working of the Reuters 2000ndash2 Elec-tronic Foreign Exchange Trading Systemrdquo inJeffrey A Frankel Giampaolo Galli and Al-berto Giovannini eds The microstructure offoreign exchange markets Chicago Univer-

sity of Chicago Press for NBER 1996 pp107ndash79

Harvey Campbell R and Huang Roger D ldquoTheImpact of the Federal Reserve Bankrsquos OpenMarket Operationsrdquo Journal of FinancialMarkets April 2002 5(2) pp 223ndash57

Hasbrouck Joel ldquoSecurity BidAsk Dynamicswith Discreteness and Clustering SimpleStrategies for Modeling and EstimationrdquoJournal of Financial Markets February1999 2(1) pp 1ndash28

Hautsch Nikolaus and Hess Dieter ldquoWhatrsquos Sur-prising about Surprises The SimultaneousPrice and Volatility Impact of InformationArrivalrdquo Unpublished manuscript Univer-sity of Konstanz Germany 2001

Johnson Timothy C ldquoReturn Dynamics WhenPersistence Is Unobservablerdquo MathematicalFinance October 2001a 11(4) pp 415ndash45

ldquoVolatility Momentum and TimeVarying Skewness in Foreign Exchange Re-turnsrdquo Unpublished manuscript LondonBusiness School 2001b

Kuttner Kenneth N ldquoMonetary Policy Surprisesand Interest Rates Evidence from the FedFunds Futures Marketrdquo Journal of MonetaryEconomics June 2001 47(3) pp 523ndash44

Mark Nelson C ldquoExchange Rates and Funda-mentals Evidence on Long-Horizon Predict-abilityrdquo American Economic Review March1995 85(1) pp 201ndash18

Mark Nelson C and Sul Donggyu ldquoNominalExchange Rates and Monetary Fundamen-tals Evidence from a Small Post-BrettonWoods Panelrdquo Journal of International Eco-nomics February 2001 53(1) pp 29ndash52

McQueen Grant and Roley V Vance ldquoStockPrices News and Business Conditionsrdquo Re-view of Financial Studies Fall 1993 6(3) pp683ndash707

Meese Richard A and Rogoff Kenneth ldquoEmpir-ical Exchange Rate Models of the SeventiesDo They Fit Out of Samplerdquo Journal ofInternational Economics February 198314(1ndash2) pp 3ndash24

Muller Ulrich A Dacorogna Michel M OlsenRichard B Pictet Oliviet V Schwarz Mat-thias and Morgenegg Claude ldquoStatisticalStudy of Foreign Exchange Rates EmpiricalEvidence of a Price Change Scaling Law andIntraday Analysisrdquo Journal of Banking andFinance December 1990 14(6) pp 1189ndash208


Payne Richard ldquoAnnouncement Effects andSeasonality in the Intra-day Foreign Ex-change Marketrdquo Unpublished manuscriptLondon School of Economics 1996

Pearce Douglas K and Roley V Vance ldquoStockPrices and Economic Newsrdquo Journal ofBusiness January 1985 58(1) pp 49ndash67

Rao C Radhakrishna ldquoEstimation of Het-eroscedastic Variances in Linear ModelsrdquoJournal of the American Statistical Associa-tion March 1970 65(1) pp 161ndash72

Rudebusch Glenn D ldquoDo Measures of Mone-tary Policy in a VAR Make Senserdquo Interna-tional Economic Review November 199839(4) pp 907ndash31

Taylor John B ldquoDiscretion versus Policy Rules

in Practicerdquo Carnegie-Rochester ConferenceSeries on Public Policy December 199339(0) pp 195ndash214

Urich Thomas J and Wachtel Paul ldquoThe Ef-fects of In ation and Money Supply An-nouncements on Interest Ratesrdquo Journal ofFinance September 1984 39(4) pp 1177ndash88

Veronesi Pietro ldquoStock Market Overreaction toBad News in Good Times A Rational Ex-pectations Equilibrium Modelrdquo Review of Fi-nancial Studies Winter 1999 12(5) pp 975ndash1007

Zarnowitz Victor and Lambros Louis A ldquoCon-sensus and Uncertainty in Economic Predic-tionrdquo Journal of Political Economy June1987 95(3) pp 591ndash621


moreover that the adjustment response patternis characterized by a sign effect bad news hasgreater impact than good news Finally we re-late our results to recent theoretical and empir-ical work on asset return volatility and itsassociation with information processing andprice discovery

The paper relates to earlier work in intriguingways but at least three features differentiate our ndings from previous results along importantdimensions These include our focus on foreignexchange markets our focus on conditionalmean as opposed to conditional variance dy-namics and the length and breadth of our sampleof exchange-rate and announcement data Let usdiscuss them brie y in turn

First we focus on foreign exchange marketsas opposed to stock or bond markets and weaddress the central open issue in exchange-rateeconomicsmdashthe link between exchange ratesand fundamentals It is comforting howeverthat a number of recent papers focusing largelyon bond markets reach conclusions similar toours Pierluigi Balduzzi et al (2001) for exam-ple examine the effects of economic news onprices in the US interdealer government bondmarket nding strong news effects and quickincorporation of news into bond prices whileMichael J Fleming and Eli M Remolona(1997 1999) show that the largest bond pricemovements stem from the arrival of newsannouncements2

Second we focus primarily on exchange-rateconditional means as opposed to conditionalvariances That is we focus primarily on thedetermination of exchange rates themselves asopposed to their volatility We maintain thisfocus both because the conditional mean is ofintrinsic interest and because high-frequencydiscrete-time volatility cannot be extracted ac-curately unless the conditional mean is modeledadequately Hence our work differs in importantrespects from that of Louis H Ederington andJae Ha Lee (1993) Richard Payne (1996)Andersen and Bollerslev (1998) and Bollerslevet al (2000) for example who examine calen-dar and news effects in high-frequency asset-

return volatility but do not consider the effectsof news on returns themselves

Third we use a new data set which spans acomparatively long time period and includes abroad set of exchange rates and macroeconomicindicators

Notwithstanding the improvements obtainedthrough the above consideration our results arequite consistent with prior related work Indeedseveral studies have linked macroeconomicnews announcements to jumps in exchangerates and our ndings may be viewed as pro-viding con rmation and elaboration CharlesGoodhart et al (1993) for example examineone year of high-frequency DollarPound ex-change rates and two speci c news eventsmdashaUS trade gure announcement and a UK in-terest-rate changemdashand conclude in each casethat the news caused an exchange-rate jumpSimilarly Alvaro Almeida et al (1998) in theirstudy of three years of high-frequency DMDollar exchange rates and a larger set of newsannouncements document systematic short-livednews effects Finally Kathryn M Dominguez(1999) argues that most large exchange-ratechanges occur within ten seconds of a macroeco-nomic news announcement and that close timingof central bank interventions to news announce-ments increases their effectiveness

We proceed as follows In Section I we de-scribe our high-frequency exchange-rate andmacroeconomic expectations and announce-ments data In Section II we characterize thespeed and pattern of exchange-rate adjustmentto macroeconomic news and we documentamong other things the sign effects (ie alarger exchange-rate response to bad than goodnews) In Section III we relate the sign effects torecent theories of information processing andprice discovery We conclude in Section IV

I Real-Time Exchange Rates ExpectedFundamentals and Announced Fundamentals

Throughout the paper we use data on exchange-rate returns in conjunction with data on expecta-tions and announcements of macroeconomicfundamentals The data are novel in several re-spects such as the simultaneous high frequencyand long calendar span of the exchange-rate re-turns as well as the real-time nature of theexpectations and announcements of fundamen-tals Here we describe them in some detail

2 Also in concurrent related work for T-bond futuresNikolaus Hautsch and Dieter Hess (2001) report highlysigni cant but short-lived price and volatility impacts inresponse to new and revised employment gures
























S kt 5A kt 2 Ekt

sk











US Announcements





Consumption
























Investment
























(1) R t 5 b0 1 Oi 5 1

I

b i R t 2 i

1 Ok 5 1

K Oj 5 0

J


t 5 1 T




288

1 Ok 5 1

K Oj9 5 0

J9

bkj9zSkt 2 j9z

1 X Oq 5 1

Q X dqcosX q2pt

288 D1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J0




























VOLATILITY PATTERNS

























Announcement



US Announcements




Consumption
























Announcement



Investment




41 Money stock M3 2002 0215 0000 0000 20033 0181 2002 0113 20023 0161






























2881 O

k 5 1

K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D 1 Or 5 1

R Oj0 5 0

J0






Sign of the News







2





bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020





bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597









2881 O

k 5 1

K Oj9 5 0

J9


K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0







































REFERENCES























































































S kt 5A kt 2 Ekt

sk











US Announcements





Consumption
























Investment
























(1) R t 5 b0 1 Oi 5 1

I

b i R t 2 i

1 Ok 5 1

K Oj 5 0

J


t 5 1 T




288

1 Ok 5 1

K Oj9 5 0

J9

bkj9zSkt 2 j9z

1 X Oq 5 1

Q X dqcosX q2pt

288 D1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J0




























VOLATILITY PATTERNS

























Announcement



US Announcements




Consumption
























Announcement



Investment




41 Money stock M3 2002 0215 0000 0000 20033 0181 2002 0113 20023 0161






























2881 O

k 5 1

K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D 1 Or 5 1

R Oj0 5 0

J0






Sign of the News







2





bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020





bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597









2881 O

k 5 1

K Oj9 5 0

J9


K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0







































REFERENCES




































































US Announcements





Consumption
























Investment
























(1) R t 5 b0 1 Oi 5 1

I

b i R t 2 i

1 Ok 5 1

K Oj 5 0

J


t 5 1 T




288

1 Ok 5 1

K Oj9 5 0

J9

bkj9zSkt 2 j9z

1 X Oq 5 1

Q X dqcosX q2pt

288 D1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J0




























VOLATILITY PATTERNS

























Announcement



US Announcements




Consumption
























Announcement



Investment




41 Money stock M3 2002 0215 0000 0000 20033 0181 2002 0113 20023 0161






























2881 O

k 5 1

K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D 1 Or 5 1

R Oj0 5 0

J0






Sign of the News







2





bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020





bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597









2881 O

k 5 1

K Oj9 5 0

J9


K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0







































REFERENCES







































































Investment
























(1) R t 5 b0 1 Oi 5 1

I

b i R t 2 i

1 Ok 5 1

K Oj 5 0

J


t 5 1 T




288

1 Ok 5 1

K Oj9 5 0

J9

bkj9zSkt 2 j9z

1 X Oq 5 1

Q X dqcosX q2pt

288 D1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J0




























VOLATILITY PATTERNS

























Announcement



US Announcements




Consumption
























Announcement



Investment




41 Money stock M3 2002 0215 0000 0000 20033 0181 2002 0113 20023 0161






























2881 O

k 5 1

K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D 1 Or 5 1

R Oj0 5 0

J0






Sign of the News







2





bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020





bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597









2881 O

k 5 1

K Oj9 5 0

J9


K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0







































REFERENCES













































































(1) R t 5 b0 1 Oi 5 1

I

b i R t 2 i

1 Ok 5 1

K Oj 5 0

J


t 5 1 T




288

1 Ok 5 1

K Oj9 5 0

J9

bkj9zSkt 2 j9z

1 X Oq 5 1

Q X dqcosX q2pt

288 D1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J0




























VOLATILITY PATTERNS

























Announcement



US Announcements




Consumption
























Announcement



Investment




41 Money stock M3 2002 0215 0000 0000 20033 0181 2002 0113 20023 0161






























2881 O

k 5 1

K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D 1 Or 5 1

R Oj0 5 0

J0






Sign of the News







2





bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020





bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597









2881 O

k 5 1

K Oj9 5 0

J9


K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0







































REFERENCES






















































































VOLATILITY PATTERNS

























Announcement



US Announcements




Consumption
























Announcement



Investment




41 Money stock M3 2002 0215 0000 0000 20033 0181 2002 0113 20023 0161






























2881 O

k 5 1

K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D 1 Or 5 1

R Oj0 5 0

J0






Sign of the News







2





bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020





bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597









2881 O

k 5 1

K Oj9 5 0

J9


K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0







































REFERENCES























































































Announcement



US Announcements




Consumption
























Announcement



Investment




41 Money stock M3 2002 0215 0000 0000 20033 0181 2002 0113 20023 0161






























2881 O

k 5 1

K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D 1 Or 5 1

R Oj0 5 0

J0






Sign of the News







2





bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020





bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597









2881 O

k 5 1

K Oj9 5 0

J9


K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0







































REFERENCES











































































Announcement



Investment




41 Money stock M3 2002 0215 0000 0000 20033 0181 2002 0113 20023 0161






























2881 O

k 5 1

K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D 1 Or 5 1

R Oj0 5 0

J0






Sign of the News







2





bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020





bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597









2881 O

k 5 1

K Oj9 5 0

J9


K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0







































REFERENCES




























































































2881 O

k 5 1

K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D 1 Or 5 1

R Oj0 5 0

J0






Sign of the News







2





bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020





bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597









2881 O

k 5 1

K Oj9 5 0

J9


K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0







































REFERENCES


































































Sign of the News







2





bk0 0091 0071 0159 0115 0079uk0 0018 0008 0029 20002 0020





bk0 00173 00216 00215 00169 0015uk0 00566 00436 00873 00837 00597









2881 O

k 5 1

K Oj9 5 0

J9


K Oj9 5 0

J9


Q X dqcosX q2pt

288 D 1 fqsinX q2pt

288 D D1 O

r 5 1

R Oj0 5 0

J 0







































REFERENCES




































































































REFERENCES

































































Micro Effects of Macro Announcements: Real-Time …boller/Published_Papers/aer_03.pdfMicro Effects...

Documents

Transcript of Micro Effects of Macro Announcements: Real-Time …boller/Published_Papers/aer_03.pdfMicro Effects...