The News Content of Macroeconomic Announcements: What ...Vol. 8 No. 3 News Content of Macroeconomic Announcements 3 its information becomes stale over time. Third, the price response

The News Content of MacroeconomicAnnouncements: What if Central Bank

Communication Becomes Stale?∗

Michael Ehrmann and David SondermannEuropean Central Bank

How do financial markets incorporate news? This paperargues that one piece of news not only has direct effects onasset prices and market volatility, but it can also alter the rel-ative importance of other news. Studying the reaction of UKshort-term interest rates to the Bank of England’s InflationReport and to macroeconomic announcements, this paper findssupport for the notion of interdependent news effects. Withtime elapsing since the latest release of an Inflation Report,market volatility increases, suggesting that market uncertaintyrises until the central bank updates its communication. Atthe same time, the price response to other macroeconomicannouncements becomes more pronounced, and they play amore important role in reducing uncertainty.

JEL Codes: D83, E43, E58, G12, G14.

1. Introduction

How do financial markets incorporate news? If a particular newsitem contains relevant information, it might exert immediate effectson asset prices and market volatility. Furthermore, it might alsoalter the importance that gets attached to other, subsequent, pieces

∗Copyright c© 2012 European Central Bank. We would like to thank MartinT. Bohl, Georgios Chortareas, Philippine Cour-Thimann, Reint Gropp, BerndWilfling, two anonymous referees, and seminar participants at the ECB, the Euro-pean Business School, Marburg University, Norges Bank, Oxford University, theNational Bank of Poland conference on “Publishing Central Bank Forecasts inTheory and Practice” and the 2010 ASSA meetings in Atlanta for helpful com-ments and suggestions. This paper presents the authors’ personal opinions anddoes not necessarily reflect the views of the European Central Bank. Authore-mails: [email protected], [email protected].

1

2 International Journal of Central Banking September 2012

of news (for instance, by making them less relevant). This createsan interdependence of news, making the information content of oneitem dependent on the relative importance of another item. The aimof this paper is to verify the interdependence of news and to doc-ument the corresponding reactions of financial markets. The mainhypothesis of the paper is that shortly after the release of a “domi-nant” piece of news, other news leads to relatively small reactions infinancial markets. By contrast, with larger distance to this dominantpiece, the relative information content in other news increases, thusalso triggering more substantial effects on asset prices.

To study this hypothesis, we will use the Bank of England’s Infla-tion Report as an example for potentially dominant news and lookat the reaction of British short-term interest rates to the InflationReports themselves. In addition, we incorporate other macroeco-nomic news and model the reaction to this news as dependent onthe time that has elapsed since the last release of an Inflation Report.This choice rests on the assumption that central bank communica-tions are important drivers of short-term interest rates, and exploitsthe fact that the Bank of England’s Inflation Report is releasedat a quarterly frequency. This implies that in the interim periodbetween two Inflation Reports, financial market participants needto update their beliefs about the course of the economy and thelikely setting of monetary policy based on other information, suchas the releases of macroeconomic data. Shortly after the release ofan Inflation Report, markets should have a relatively clear pictureabout the future course of interest rates, but with increasing dis-tance to this communication, this picture becomes blurred giventhat the economy evolves, yet the central bank does not provide anupdate of its own assessment. Accordingly, other pieces of informa-tion (as given by macroeconomic announcements) should becomemore newsworthy and therefore receive more attention in financialmarkets.

The evidence provided in this paper is in line with these hypothe-ses. There are three main findings. First, the release of the InflationReport leads to a marked decline in market volatility, suggestingthat it triggers a reduction in uncertainty among financial marketparticipants about the future course of monetary policy. Second,market volatility increases with the time that has elapsed since thelatest release of an Inflation Report, which supports the notion that

Vol. 8 No. 3 News Content of Macroeconomic Announcements 3

its information becomes stale over time. Third, the price response toother macroeconomic announcements does indeed become more pro-nounced with increasing distance to the latest Inflation Report, andthey play a more important role in reducing uncertainty. Finally,the paper finds the same patterns with regard to other news, asuncertainty also increases in between two announcements of infla-tion data. The findings therefore apply more generally and are notrestricted to central bank communications.

These findings have a number of important implications. In therealm of monetary policy, central banks routinely study the marketresponse to macroeconomic news. To appropriately judge the mag-nitude of such market movements, it is important to compare theseagainst the correct benchmark, which is not the average responsebut rather the average response conditional on the freshness of cen-tral bank communication. An above-average response to the releaseof inflation data shortly prior to the publication of a major cen-tral bank release, for instance, is to be expected, and should beassessed accordingly. Beyond a pure monetary policy perspective,these findings might help explain why it is at times difficult toestablish a link between asset prices and macroeconomic fundamen-tals (see, e.g., the discussion on this issue in Andersen et al. 2003).What the current paper suggests is that a missing (or a muted)responsiveness need not automatically imply a broken link. Rather,muted responsiveness can very well arise if the information con-tained in another source (that could have entered the information setalready at an earlier stage) is sufficiently precise to dominate the newsignal.

The paper relates to the extensive literature on how macroeco-nomic announcements (such as the release of data for consumerprice inflation) are incorporated into asset prices. Early contribu-tions have shown the responsiveness of stock prices (McQueen andRoley 1993), of money and bond markets (Fleming and Remolona1999b; Thornton 1998), and exchange rates (Andersen et al. 2003;Faust et al. 2007). Further work has subsequently provided evi-dence for asymmetries and time variations in the effect of macro-economic announcements. Andersen et al. (2003), for instance, showthat bad news has a greater impact on exchange rates than goodnews, and that more timely news releases lead to larger market reac-tions. Andersen et al. (2007) furthermore show that the news effects


are dependent on the business cycle. Finally, a recent contributionby Gilbert et al. (2010) finds a somewhat smaller responsivenessto macro announcements that are subject to larger revisions, and inparticular a substantially larger reaction to news with more informa-tion content about future monetary policy decisions by the FederalOpen Market Committee.

This suggests that interest rates react to macroeconomicannouncements because their release provides signals which allowfinancial market participants to update their expectations aboutthe future course of monetary policy. If this is the case, then sig-nals which are provided by the central bank itself should receiveeven more attention in financial markets. As a matter of fact, cen-tral bank communications have been found to be among the mostimportant market movers, especially for interest rates (Guthrie andWright 2000; Kohn and Sack 2004; Andersson, Dillén, and Sellin2006; Ehrmann and Fratzscher 2007). Reinhart and Sack (2006) aswell as Reeves and Sawicki (2007) show that, in particular, commu-nication on behalf of the entire policy-making committee is a strongmarket mover; for the case of the Bank of England, the InflationReport has been shown to be particularly important.1

The current paper is also closely related to Gropp and Kadareja(2012), who look at the aging of the information content contained incommercial banks’ annual reports. The idea underlyng that paperis that with increasing distance to an annual report, the qualityof the public information about the bank deteriorates. If a signalarrives shortly after the release of the annual report, it is much eas-ier for analysts to determine the implications of this signal for thestock price of the individual bank, whereas over time, uncertaintyincreases. In line with this hypothesis, the paper finds that withincreasing distance to the annual report, the volatility response ofstock prices to monetary policy shocks becomes larger and morepersistent. Our paper complements theirs by looking at a different

1Whereas earlier studies of the market reactions to Bank of England commu-nication (Clare and Courtenay 2001, Lasaosa 2007) had concluded that the newscomponent contained in interest rate decisions increased with the independenceof the Bank of England, Bell (2005) argues that in more recent years, the newscontent of the decisions has declined, due to a more central role of the InflationReport in the decision-making process.


informational setup (the central bank informing about the state ofthe economy rather than a commercial bank providing informationabout itself), different financial markets (interest rates versus stockmarkets), different data frequency (daily as opposed to intradaily),and a different volatility measure (conditional versus realized). Thefact that both papers come to similar conclusions suggests a widerapplicability of the phenomenon.

The remainder of this paper is structured as follows. It developssome hypotheses about the reception of news in financial marketsand describes the data and the methodology underlying our empiri-cal analysis in section 2. Section 3 provides the empirical results, firstlooking at the unconditional reception of news and subsequently ask-ing how this changes if central bank communication becomes stale.Section 4 concludes.

2. Modeling the Reception of News in Financial Markets

The purpose of this section is to develop a few hypotheses about theeffect of the release of Inflation Reports and of macroeconomic newson financial markets and their interaction, as well as to present thesetup of our subsequent empirical analysis.

2.1 Some Hypotheses about the Reception of News inFinancial Markets

The impact of macroeconomic announcements on financial marketshas typically been studied in the context of Bayesian learning mod-els (Kim and Verrecchia 1991; Veronesi 2000; Gilbert et al. 2010;Hess and Niessen 2010), where market participants need to forman expectation about some economic variable, but they receive onlynoisy signals about this economic fundamental. Let us assume thatprior to the reception of such a signal, market participants haveformed an expectation, with μF denoting the mean expectation andρF its precision.2 If they receive a noisy signal μA about the economicfundamental, with precision ρA, they will update their expectationby calculating a weighted average of the prior expectation and the

2This follows the notation of Hess and Niessen (2010).


signal, with the weights being related to the relative quality of each.The updated belief is given as

μP = μFρF

ρF + ρA+ μA

ρAρF + ρA

(1)

and has precision

ρP = ρF + ρA. (2)

This very simple setup allows drawing a number of important con-clusions. First, an update of the mean expectations will only takeplace if μA �= μF , i.e., if there is some “news” in the signal. In con-trast, the precision of the expectation will change as long as ρA �= 0,regardless of whether the signal contains some news content. Sec-ond, the updating of the mean expectation will crucially depend onthe precision of the prior expectation: the more precise the prior,the less value added is implied in receiving a signal, and the lessupdating will take place in response to it.

Accordingly, we can formulate the following hypotheses:

1. The release of public signals should lead to an update ofagents’ expectations about the economic fundamental.

2. The release of public signals should lead to an increased pre-cision of agents’ expectations.

3. With passage of time after the release of a signal, its infor-mation becomes stale, which should ceteris paribus lead to adecreasing precision of expectations.

4. The more time has elapsed since the release of a signal, themore pronounced the update of expectations in response toother signals.

5. The more time has elapsed since the release of a signal,the stronger the increase in the precision of expectations inresponse to another signal.

2.2 The Data

We would like to put these hypotheses to an empirical test. The prin-ciple should apply generally, for any financial market, and for anypiece of news. However, as argued above, the example of central bankcommunications and short-term interest rates might be promising,


given that they satisfy the main requirements: (i) there is known tobe a tight link between the two, (ii) some central bank communica-tions are infrequent, implying that there is sufficient time in betweentwo releases to study time variations in the response to other news,and (iii) some central bank communications have a regular releaseschedule, which makes them a convenient object of study.

Accordingly, for the financial market we chose British zero-coupon government bond yields for different maturities ranging fromtwo to twelve months. Our dependent variable is defined as the firstdifference of the daily yields, as is common practice in the announce-ment literature. These data have been provided by the Bank of Eng-land. A possibly preferable alternative would have been to use inter-bank rates, as these are arguably more tightly linked to monetarypolicy; unfortunately, the available series typically do not vary atthe daily frequency, especially at the beginning of our sample, suchthat we had to use government bond yields instead.3 The same datahave been fruitfully employed in related studies; Gürkaynak, Levin,and Swanson (2010), for instance, use them to investigate whetherthe UK monetary policy has succeeded in anchoring inflation expec-tations. The sample spans all trading days from March 1997 untilDecember 2008. While the start of the sample is due to the unavail-ability of earlier zero-coupon yields data, it coincides roughly withthe independence of the Bank of England and, as such, constitutesa meaningful choice. The starting point is also just prior to the 1998Bank of England Act, which requires the Monetary Policy Commit-tee (MPC), i.e., the Bank of England’s rate-setting body, to sign offon the Bank’s Inflation Report.4

3In order to check the robustness of our results, we also used available inter-bank rates. Although these rates are only available for a considerably shortertime sample (starting in 2001, thus reducing the sample by around one-third) andexhibit consecutive periods of non-fluctuating rates, we find our results broadlyunchanged. In particular, we still observe an increase in volatility with the timeelapsed since the last Inflation Report.

4This need not imply that the MPC, as a group, endorses the entire content ofthe report; however, the forecasts of inflation and GDP and the associated rangesare typically endorsed by the MPC. Judging from the media (and especiallynewswire) reporting about the Inflation Report, where especially the forecastsreceive a lot of attention, we assume that this part of the reports is central to thefinancial market participants. For a detailed discussion on the forecasting processand its importance, see Goodhart (2001).


As to the central bank communication variable, we focus on theBank of England’s Inflation Report, despite the fact that there area large number of other important communication events. Monetarypolicy decisions and the accompanying statements or the subse-quent release of the minutes of MPC meetings are obvious exam-ples, as are speeches by MPC members. However, in contrast tothese other publications, Inflation Reports have a regular publica-tion schedule (compared with speeches), a relatively low publicationfrequency (compared with minutes and policy statements), and theyare clearly a publication of major importance: Inflation Reports con-tain an in-depth analysis of the Bank of England’s assessment of theeconomy (taking the length of the documents as a proxy, InflationReports are at around fifty to sixty pages, more than five times aslong as minutes, which typically contain around ten pages). Impor-tantly, the Inflation Report also contains the forecasts of inflationand GDP, and as such an important forward-looking element ofthe Bank’s economic assessment, which directly feeds into policydeliberations. According to the Bank of England (2009), the Infla-tion Report “serves two purposes. First, its preparation provides acomprehensive and forward-looking framework for discussion amongMPC members as an aid to our decision making. Second, its publica-tion allows us to share our thinking and explain the reasons for ourdecisions to those whom they affect.” According to Lomax (2005),the Inflation Report plays a “central role” in the MPC’s communica-tion. As further indications of its prominence, note that the InflationReport is the first on the list of the Bank’s “Main Publications”on its web site (www.bankofengland.co.uk), and that its publica-tion is accompanied by a regular press conference. As a matter offact, Reeves and Sawicki (2007) have found the Inflation Report tobe a particularly strong market mover, and the reports have beenjudged to be of very high quality (Fracasso, Genberg, and Wyplosz2003).

We have collected the release dates of the Bank of England’sInflation Report from the Bank of England web site, and createa dummy variable that is equal to one on release dates and zerootherwise. Our sample covers forty-six Inflation Reports. Note thatthe release schedule of the Inflation Reports is clearly exogenousto market events: the Bank of England publishes the exact releasedates several issues ahead. The release pattern is furthermore highly


regular: 55 percent of reports are released sixty-five days after thepreceding report, another 13 percent deviate from this by one day,another 24 percent by four days, and the remaining 6 percent bysix days. The time elapsed since the last report is measured bycounting the days since the preceding release and dividing this num-ber by the total number of days between the preceding and thesubsequent release. To facilitate the interpretation of coefficients,we have normalized the resulting variable by subtracting its mean.The variable therefore ranges from −0.5 (on the first day after therelease of an Inflation Report) to +0.5 (on the day of the releaseof the subsequent Inflation Report). Note that this normalizationdoes not affect any of our results qualitatively. Finally, given thatasset prices, at the time of a news release, react only to the sur-prise component contained therein (Kuttner 2001), we need to con-struct a proxy for the news content of each Inflation Report. Wedo so by calculating the change in the central tendency of the infla-tion forecast at the policy-relevant two-year horizon. Of course, thisproxy is rather crude, as it neither does justice to the breadth anddepth of the report nor takes into account the fact that changes inthe inflation forecast might have been partially expected. However,a cross-check with media reporting about the Bank of England’sInflation Reports shows a very strong emphasis on precisely thechange in the two-year inflation forecast, such that we consider thisproxy as capturing the most relevant news content in the InflationReports.

As to other news, we follow a large literature and look at themarket response to the release of macroeconomic data. To constructthe surprise component, we follow the standard in the announce-ment literature and deduct the expectation of the announcementfrom the actual announcement value of the variable. To allow a com-parison of the magnitude of the financial market responses, we fur-thermore standardize the surprises by their own standard deviation.This standardization removes differences in the unit of measurementacross variables, and regression coefficients for each series can thenbe interpreted as a response per one-standard-deviation surprise.We obtained data on financial market expectations of the variousmacroeconomic data releases from two sources, namely Money Mar-ket Services (MMS) and Bloomberg Financial Services, and use themedian response of the respective polls as our measure of market


expectations.5 The macroeconomic announcements contained in ourdata set relate to the Consumer Price Index (month on month), theManufacturing Purchasing Managers Index, retail sales (month onmonth), unemployment, and the trade balance. By opting for nomi-nal and real indicators, we aim to cover a relatively broad spectrumof the economy; by choosing forward-looking ones such as the PMI,we intend to capture indicators that might help markets infer newsabout the future course of monetary policy. As an alternative toincluding the CPI, one might consider using the RPIX, given thatthe Bank of England’s inflation target was formulated in terms ofthis price index until 2003. While having opted for the CPI, we canconfirm that using the RPIX does not affect our results.

Table 1 reports summary statistics for all variables and alsoshows how much time has, on average, elapsed between the release ofan Inflation Report and the respective macroeconomic data releases.It is important to note that all releases are rather regular, such thatthere is relatively little variability in average numbers over time.

Given that the bulk of the announcement literature has studiedmarket responses to U.S. releases, we furthermore collected U.S.announcement data, again covering nominal and real aspects aswell as leading indicators, namely the Conference Board’s compos-ite index of leading indicators, industrial production, core CPI, andnon-farm payrolls. The U.S. data will provide a basis for testing therobustness of our results. Note that we can be agnostic on why U.S.surprises move British yields, given that we are only interested inhow financial markets react to public signals conditional on the timeelapsed since the preceding Inflation Report, regardless of the under-lying perceived transmission mechanism from the U.S. to the Britisheconomy.

2.3 The Econometric Model

As mentioned above, we are interested in how expectations aboutthe fundamental, as well as their precision, evolve. Our empirical

5Ehrmann et al. (2011) note that the information content of the MMS andBloomberg series is very similar. In particular, when both surveys co-exist, therelease data are identical, and the expectations agree almost perfectly. The qual-ity of these data as measures of expectations has been verified by, e.g., Balduzzi,Elton, and Green (2001) and Andersen et al. (2003) for MMS, and by Ehrmannand Fratzscher (2005) for Bloomberg.


Tab

le1.

Sum

mar

ySta

tist

ics

Avg.

Tim

eEla

pse

dSin

ceLas

tIn

flat

ion

Rep

ort

No.

of1s

t2n

d3r

d

Obs.

Min

.M

ax.

Mea

nR

elea

seR

elea

seR

elea

se

Rel

ease

Dum

my

2943

0.00

1.00

0.02

——

—In

flat

ion

Rep

ort

Cha

nge

of29

43−

1.15

0.81

0.00

——

—V

aria

ble

sIn

flati

onFo

reca

st(T

wo

Yea

rsA

head

)

Min

utes

Dum

my

2943

0.00

1.00

0.05

——

—B

oEV

aria

ble

sSp

eech

es29

43−

1.00

1.00

0.00

——

—In

tere

stR

ate

2943

−0.

330.

250.

00—

——

Surp

rise

s

Tw

oM

onth

2943

1.40

7.44

5.03

——

—G

over

nm

ent

Thr

eeM

onth

2943

1.26

7.44

5.02

——

—B

ond

Yie

lds

Six

Mon

th29

431.

137.

415.

01—

——

Tw

elve

Mon

th29

431.

307.

275.

00—

——

(con

tinu

ed)


Tab

le1.

(Con

tinued

)

Avg.

Tim

eEla

pse

dSin

ceLas

tIn

flat

ion

Rep

ort

No.

of1s

t2n

d3r

d

Obs.

Min

.M

ax.

Mea

nR

elea

seR

elea

seR

elea

se

UK

CP

I29

43−

2.59

1.95

0.00

1635

57U

KP

MI

2943

−2.

871.

520.

0016

3655

UK

UK

Ret

ailSa

les

2943

−3.

025.

440.

0110

3151

Mac

roec

onom

icU

K29

43−

1.33

2.67

0.00

1740

60A

nnou

nce

men

tsU

nem

ploy

men

tU

KTra

de29

43−

3.39

2.42

−0.

0114

3652

Bal

ance

US

Com

posi

te29

43−

1.53

2.15

0.00

931

52In

dex

US

US

Indu

stri

al29

43−

6.31

4.50

0.00

930

44M

acro

econ

omic

Pro

duct

ion

Annou

nce

men

tsU

SC

PI

2943

−2.

581.

72−

0.01

1031

47U

SN

on-F

arm

2943

−2.

971.

84−

0.01

1739

60Pay

rolls

Note

s:T

he

table

show

ssu

mm

ary

stat

isti

csfo

rth

eva

riab

les

use

din

the

empir

ical

anal

ysis

.T

he

last

thre

eco

lum

ns

repor

thow

man

yday

s,on

aver

age,

hav

eel

apse

dbet

wee

nth

epre

cedin

gre

leas

eof

anIn

flat

ion

Rep

ort

and

the

mac

roec

onom

ican

nou

nce

men

t.G

iven

that

the

mac

ro-

econ

omic

annou

nce

men

tsar

em

onth

ly,th

ere

are

typic

ally

thre

ere

leas

esin

bet

wee

ntw

oIn

flat

ion

Rep

orts

.In

afe

win

stan

ces,

ther

ehas

also

bee

na

fourt

hre

leas

e,nu

mber

sfo

rw

hic

har

enot

repor

ted

for

bre

vity

.


counterparts to these two concepts are the response of interest ratesand their volatility, respectively. It is important to note, however,that we are interested in the volatility response conditional on theresponse of interest rate levels. As shown in equations (1) and (2),the arrival of news can have two effects: first, a change in the expec-tation about the future path of interest rates. This will be reflected inthe price response. Second, also the precision of expectations mightbe affected. Once the market has settled to reflect the new meanexpectation, we would expect to see a change in volatility comparedwith the time prior to the arrival of the news. The “immediate”price response to the signal should therefore not be considered partof the volatility response. We therefore need to estimate an econo-metric model that allows testing for the effect of news on boththe conditional mean and the conditional variance of asset prices.We estimate an exponential GARCH (EGARCH) model, followingNelson (1991).6 An EGARCH(1,1) model is sufficient to address thenon-normality of the data, in particular the serial correlation andheteroskedasticity of the daily interest rate series. We will use differ-ent variants of the econometric model. In its most extended version,the conditional mean equation is formulated as

rt = c1 + δIRsIRt +∑

k

δkskt +∑

k

φkskt gt + ωgt + ψrt−1

+∑

l

νldlt + μt, (3)

with rt as the change in the daily UK zero-coupon rates, rt−1 asthe lagged change, and dlt as a set of controls containing day-of-the-week effects. skt denotes the standardized surprise componentcontained in macroeconomic announcement k, and sIRt denotes theproxy for the news component in the Inflation Report. Finally, gtstands for the time that has elapsed since the preceding releaseof the Inflation Report. Conditioned on the information set of last

6In order to test for asymmetries in volatility, we apply the Engle and Ng(1993) sign and size bias test. It examines whether sign and size of the shocksimpact differently upon the conditional variance. The joint hypothesis of noasymmetry is rejected for all maturities, suggesting the use of Nelson’s (1991)EGARCH. The results are not displayed but are available from the authors uponrequest.


period (It−1), we assume the distribution of the disturbance to beμt|It−1 ∼ (0, ht). Hence, we express the conditional variance of UKinterest rate changes, ht, as

log(ht) = c2 + κ1

(∣∣∣∣∣ μt−1√ht−1∣∣∣∣∣ −

√2/π

)+ κ2

(μt−1√ht−1

)+ κ3 log(ht−1)

+ λIRaIRt +∑

k

λkakt +∑

k

ρkakt gt + ζgt +∑

l

τ ldlt. (4)

akt and aIRt are announcement dummies that take the value one on

all days a macroeconomic announcement or an Inflation Report isreleased, and zero otherwise.7 All other variables are as describedin equation (3). The model is estimated via maximum likelihood,using the BHHH algorithm for optimization. Note that the model isestimated for all business days in the sample, i.e., also for days whenneither an Inflation Report is issued nor a macroeconomic announce-ment is made. The corresponding variables are equal to zero on suchdays.

In the context of this econometric model, the hypotheses devel-oped in section 2.1 translate into the following:

1. The publication of an Inflation Report and the release ofmacro news should lead to an update of agents’ expectationsabout the future course of monetary policy, and thus to aresponse of interest rates: H0 : δk �= 0; δIR > 0.8

2. The publication of an Inflation Report and the release ofmacro news should lead to an increased precision of agents’expectations and thus to lower conditional volatility: H0 :λk < 0; λIR < 0.

3. With passage of time, the information contained in a givenInflation Report becomes stale, which should lead to a

7Introducing the announcement dummies lagged by one day does not affectour results.

8The expected sign depends on the type of news. An increased inflation fore-cast in the Inflation Report or an unexpectedly high inflation release or releaseabout real variables should lead to increasing interest rates, whereas unexpectedlyhigh unemployment numbers should lower interest rates.


decreasing precision of expectations and thus to an increase inconditional volatility—until the subsequent Inflation Report isreleased: H0 : ζ > 0.9

4. The update of expectations in response to macro news shouldbe more pronounced the more time has elapsed since thepreceding release of an Inflation Report: H0 : sign(δk) =sign(φk).

5. The increase in the precision of expectations in response tomacro news should be stronger the more time has elapsed sincethe preceding release of an Inflation Report: H0 : ρk < 0.

Hypotheses 2, 3, and 5 are illustrated in figure 1. The solid lineshows the evolution of market volatility that is predicted by thevarious hypotheses. Importantly, the figure also contains a dottedline, which shows that the estimate of the volatility increase overtime depends crucially on the inclusion of macro announcementsin the regression model: without these, the slope of the line getsunderestimated. This is important to bear in mind when interpret-ing our results: while our model controls for a few macroeconomicannouncements, it will never be able to include all relevant news,such that we would expect to find a relatively subdued estimatefor ζ.

It is important to note that much of the literature has discussedthat macro announcements increase market volatility. Ederingtonand Lee (1993) and Fleming and Remolona (1999a) have identifiedtwo distinct adjustment processes: In the first stage, prices adjustnearly instantaneously to news, thus confirming the conjecture ofFrench and Roll (1986) that public information affects prices beforeanyone can trade on it. In the second stage, which is somewhatmore persistent, trading volume surges and price volatility persists,as residual disagreements among the traders about the interpreta-tion of the news triggers trading. Note, however, that these twoperiods span a couple of hours, i.e., are typically concluded within

9We are interested in the direct effect of the aging process of the InflationReport and not the overall effect (which furthermore takes into account thatwith stale central bank communication, macroeconomic announcements will leadto a stronger volatility reduction). Accordingly, we formulate this hypothesis asH0 : ζ > 0 rather than H0 : ζ +

∑

k

ρkak > 0.


Figure 1. The Predicted Evolution of Market Volatility

inflation report

macro news

macro news

macro news

Market volatility

inflation report

t

inflation report

macro news

macro news

macro news

Market volatility

inflation report

t

Notes: This stylized figure shows the predicted evolution of market volatility.Volatility falls in response to news such as the Inflation Report. Subsequently,it increases with the distance to the last Inflation Report. The reduction involatility in response to macro news increases with the distance to the last Infla-tion Report. The dotted line shows the slope for this volatility increase thatwould be estimated if macro news were mistakenly omitted from the econometricmodel.

the course of a trading day. Accordingly, we might expect volatil-ity to increase in the very short run yet still be muted over longerhorizons.

Another distinction that is important in that respect is whetherthe increase or decrease is related to a realized or a conditionalvolatility. Realized volatility is likely to increase (due to a response ofthe conditional mean equation), whereas conditional volatility couldalso fall in response to news. A number of papers in this litera-ture estimate conditional mean and conditional variance equations:Andersen et al. (2003, 2007) and Ehrmann and Fratzscher (2005)using a weighted least-squares approach; Jones, Lamont, and Lums-daine (1998) or Flannery and Protopapadakis (2002) using GARCHmodels; and Goodhart et al. (1993) even using GARCH in meanmodels. The key insight from these studies is that the effects onthe conditional mean might very well differ from the effects onthe conditional variance. Flannery and Protopapadakis (2002), forinstance, find that inflation news affects only the conditional mean


of stock returns, whereas some real factors trigger responses in theconditional variance without affecting the conditional mean.

Still, positive reaction coefficients are also found in studies ofconditional volatility. How does this result square with equation (2),which states that any additional piece of information will increasethe precision of the expectation?10 This simplistic model, of course,assumes that the economic fundamental is static in nature, suchthat the new information helps in updating and narrowing downthe prior belief. In a dynamic setting, the economic fundamentalcan move over time, however, such that a large surprise (i.e., a casewhere μA is substantially different from μF ) might make the priorexpectation look unreliable or biased with regard to the new stateof the economic fundamental. In this case, the piece of news servestwo purposes: its incidence should lower uncertainty, as there is anew piece of information about the fundamental, but depending onthe magnitude of the surprise, it might increase uncertainty, as itmakes the prior expectation less valuable. Which effect dominatesremains an empirical question. In our analysis, we will therefore dis-till the pure effect on volatility arising from the incidence of the pub-lic news by controlling for the effect of the surprise component in thesignal.

3. Financial Market Reactions to the Arrival of News

This section presents the empirical results. We will first ask howfinancial markets respond to the arrival of news in general, lookingat the response to Inflation Reports as well as to macro releases. Sub-sequently, we will test what happens over time following the releaseof an Inflation Report, with the information becoming stale.

3.1 The Unconditional Reception of News

Our first test relates to the reaction of yields to the release of pub-lic signals, i.e., hypotheses 1 and 2. Table 2 shows the results ofa model that contains, in the variance equation, dummies that are

10This issue is touched upon in DeGennaro and Shrieves (1997), Kim and Sheen(2000), and Kim, McKenzie, and Faff (2004).


Tab

le2.

The

Ave

rage

Effec

tof

Annou

nce

men

tson

Yie

lds

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

UK

CP

I0.

004∗

∗∗0.0

020.

012∗

∗∗0.0

020.

021∗

∗∗0.0

020.

028∗

∗∗0.0

03U

KP

MI

0.00

5∗∗∗

0.0

020.

006∗

∗0.0

030.

010∗

∗∗0.0

040.

019∗

∗∗0.0

07U

KR

etai

lSa

les

0.00

3∗∗∗

0.0

010.

005∗

∗∗0.0

010.

014∗

∗∗0.0

020.

018∗

∗∗0.0

03U

K−

0.00

10.0

01−

0.00

20.0

01−

0.00

10.0

02−

0.00

40.0

04U

nem

ploy

men

tU

KTra

de−

0.00

3∗∗∗

0.0

01−

0.00

4∗∗∗

0.0

01−

0.00

5∗∗∗

0.0

02−

0.00

20.0

02B

alan

ceΔ

IRIn

flati

on0.

008

0.0

100.

010

0.0

120.

018

0.0

140.

069∗

∗∗0.0

19Fo

reca

st

(con

tinu

ed)


Tab

le2.

(Con

tinued

)

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

Var

ianc

eEqu

atio

n

UK

CP

I−

0.16

90.1

170.

046

0.1

140.

105

0.1

040.

018

0.1

00(D

umm

y)U

KP

MI

−0.

555∗

∗∗0.0

89−

0.62

7∗∗∗

0.0

80−

0.41

2∗∗∗

0.0

80−

0.16

5∗∗

0.0

73(D

umm

y)U

KR

etai

lSa

les

−0.

861∗

∗∗0.0

64−

0.87

9∗∗∗

0.0

61−

0.59

8∗∗∗

0.0

60−

0.36

1∗∗∗

0.0

69(D

umm

y)U

K−

0.80

8∗∗∗

0.0

99−

0.73

2∗∗∗

0.0

98−

0.54

1∗∗∗

0.0

89−

0.16

7∗∗

0.0

77U

nem

ploy

men

t(D

umm

y)U

KTra

de0.

112∗

0.0

600.

126∗

∗0.0

580.

155∗

∗∗0.0

570.

100∗

0.0

57B

alan

ce(D

umm

y)In

flati

onR

epor

t−

0.93

3∗∗∗

0.0

82−

0.85

9∗∗∗

0.0

82−

0.51

7∗∗∗

0.0

85−

0.27

0∗∗∗

0.0

86(D

umm

y)

Note

s:T

he

table

show

s,fo

rth

ediff

eren

tm

aturi

ties

(in

the

vari

ous

colu

mns)

,th

ere

acti

onco

effici

ent

ofU

Kze

ro-c

oupon

yiel

ds

tom

acro

eco-

nom

ican

nou

nce

men

tsan

dth

ere

leas

eof

the

Inflat

ion

Rep

ort.

Num

ber

sin

ital

ics

den

ote

the

stan

dar

der

rors

.**

*,**

,an

d*

indic

ate

stat

isti

cal

sign

ifica

nce

atth

e1%

,5%

,an

d10

%le

vel,

resp

ecti

vely

.T

he

mea

neq

uat

ion

issp

ecifi

edas

rt

=c1

+δIR

sIR t

+∑ k

δks

k t+

ψr

t−

1+

∑ lν

ld

l t+

μt,

the

vari

ance

equat

ion

aslo

g(h

t)

=c2

+κ1(|

μt−

1/√

ht−

1|−

√2/

π)+

κ2(μ

t−

1/√

ht−

1)+

κ3lo

g(h

t−

1)+

λIR

aIR t

+∑ k

λka

k t+

∑ lτd

l t.


equal to one on release dates for Inflation Reports and the UK macroannouncements, and that controls, in the mean equation, for theeffect of macroeconomic releases and the Inflation Report.11 As tohypothesis 1, we do indeed find that interest rates are responsive toa number of macroeconomic releases and (although not at all matu-rities) to our proxy of the news content of the Inflation Reports.The direction of these effects is as one would expect, with interestrates rising in response to higher than expected inflation and realdevelopments, and with interest rates falling in response to higherthan expected unemployment (although this effect is not statisticallysignificant).

The important thing to note in table 2 is that conditional volatil-ity is reduced in response to nearly all announcements, i.e., providingsupport for hypothesis 2. The importance of the Inflation Reportto market participants is confirmed by the fact that the volatility-reducing effect in response to its release is one of the strongest acrossthe different announcements contained in this model. Another inter-esting fact is the consistency of these results across maturities. Atall maturities, the announcements reduce volatility, with the effectbecoming smaller the longer the horizon. The single exception in thetable relates to the UK trade balance. Its release seems to heightenrather than dampen market uncertainty.

As argued above, the precision-enhancing effect of macro newsmight be difficult to detect if one does not control for the possibil-ity that large surprises in a given piece of news lead to a revisionof the prior expectation about the economic fundamental. In orderto take this into account, we have expanded the current regressionmodel by furthermore controlling for the absolute surprise in theconditional volatility equation. Table 3 reports the corresponding

11For brevity, the EGARCH coefficients are not displayed in table 2 and thefollowing tables. Throughout all estimations, the GARCH term is strongly signif-icant and close to but below one, indicating a high persistence of the conditionalvolatility. Both EGARCH parameters are equivalently significant in all estima-tions. The sign asymmetry parameter exhibits a negative sign, although small insize. The estimates confirm the asymmetry assumption proposed by the Engleand Ng (1993) test results and can be read as a negative innovation increasingvolatility more than a positive innovation of equal magnitude. The results areavailable from the authors upon request.


Tab

le3.

The

Ave

rage

Effec

tof

Annou

nce

men

tson

Yie

lds:

Inci

den

cevs.

Surp

rise

Com

pon

ent

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

UK

CP

I0.

006∗

∗∗0.0

020.

013∗

∗∗0.0

020.

022∗

∗∗0.0

030.

028∗

∗∗0.0

03U

KP

MI

0.00

8∗∗

0.0

030.

007∗

0.0

040.

011∗

0.0

050.

019∗

∗0.0

08U

KR

etai

lSa

les

0.00

3∗∗∗

0.0

010.

005∗

∗∗0.0

010.

014∗

∗∗0.0

020.

019∗

∗∗0.0

03U

K−

0.00

10.0

01−

0.00

10.0

02−

0.00

20.0

02−

0.00

40.0

04U

nem

ploy

men

tU

KTra

de−

0.00

10.0

02−

0.00

30.0

02−

0.00

30.0

03−

0.00

20.0

03B

alan

ceΔ

IRIn

flati

on−

0.00

90.0

110.

013

0.0

120.

036∗

∗0.0

160.

082∗

∗∗0.0

23Fo

reca

st

Var

ianc

eEqu

atio

n

UK

CP

I−

0.53

9∗∗∗

0.1

75−

0.47

3∗∗∗

0.1

65−

0.19

50.1

47−

0.04

50.1

45(D

umm

y)U

KP

MI

−1.

009∗

∗∗0.1

25−

1.15

7∗∗∗

0.1

07−

0.79

4∗∗∗

0.1

06−

0.36

9∗∗∗

0.0

96(D

umm

y)U

KR

etai

lSa

les

−1.

038∗

∗∗0.0

83−

1.03

2∗∗∗

0.0

81−

0.57

4∗∗∗

0.0

77−

0.36

2∗∗∗

0.0

79(D

umm

y)U

K−

1.10

1∗∗∗

0.1

50−

1.14

7∗∗∗

0.1

49−

0.91

3∗∗∗

0.1

27−

0.32

8∗∗∗

0.1

09U

nem

ploy

men

t(D

umm

y)U

KTra

de−

0.43

8∗∗∗

0.0

89−

0.41

9∗∗∗

0.0

69−

0.34

6∗∗∗

0.0

72−

0.10

50.0

76B

alan

ce(D

umm

y)

(con

tinu

ed)

22 International Journal of Central Banking September 2012Tab

le3.

(Con

tinued

)

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

Var

ianc

eEqu

atio

n

UK

CP

I(A

bs.

0.33

6∗∗

0.1

310.

508∗

∗∗0.1

210.

365∗

∗∗0.1

210.

118

0.1

23Su

rpri

se)

UK

PM

I(A

bs.

0.83

1∗∗∗

0.1

750.

874∗

∗∗0.1

540.

617∗

∗∗0.1

550.

276∗

∗∗0.1

05Su

rpri

se)

UK

Ret

ailSa

les

0.30

6∗∗∗

0.0

540.

308∗

∗∗0.0

500.

108∗

∗0.0

530.

066

0.0

53(A

bs.Su

rpri

se)

UK

0.12

50.1

070.

238∗

∗0.1

000.

245∗

∗∗0.0

880.

040

0.0

75U

nem

ploy

men

t(A

bs.Su

rpri

se)

UK

Tra

de0.

644∗

∗∗0.0

600.

629∗

∗∗0.0

540.

561∗

∗∗0.0

540.

246∗

∗∗0.0

49B

alan

ce(A

bs.

Surp

rise

)In

flati

onR

epor

t−

0.96

0∗∗∗

0.1

11−

0.97

2∗∗∗

0.1

11−

0.63

0∗∗∗

0.1

11−

0.43

1∗∗∗

0.1

13(D

umm

y)In

flati

onR

epor

t0.

562

0.7

461.

079∗

0.6

350.

890

0.6

360.

953∗

0.5

25(A

bs.Su

rpri

se)

Note

s:T

he

table

show

s,fo

rth

ediff

eren

tm

aturi

ties

(in

the

vari

ous

colu

mns)

,th

ere

acti

onco

effici

ent

ofU

Kze

ro-c

oupon

yiel

ds

tom

acro

eco-

nom

ican

nou

nce

men

tsan

dth

ere

leas

eof

the

Inflat

ion

Rep

ort,

separ

atin

gth

eeff

ect

ofth

ean

nou

nce

men

tin

ciden

cean

dth

eab

solu

tesu

rpri

sein

the

vari

ance

equat

ion.

Num

ber

sin

ital

ics

den

ote

the

stan

dar

der

rors

.**

*,**

,an

d*

indic

ate

stat

isti

cal

sign

ifica

nce

atth

e1%

,5%

,an

d10

%le

vel,

resp

ecti

vely

.T

he

mea

neq

uat

ion

issp

ecifi

edas

rt

=c1

+δIR

sIR t

+∑ k

δks

k t+

ψr

t−

1+

∑ lν

ld

l t+

μt,

the

vari

ance

equat

ion

as

log(

ht)

=c2

+κ1(|

μt−

1/√

ht−

1|−

√2/

π)+

κ2(μ

t−

1/√

ht−

1)+

κ3lo

g(h

t−

1)+

λIR

aIR t

+∑ k

λka

k t+

∑ k

k|s

k t|+

∑ lτd

l t.


results and very convincingly supports the conjecture. First, volatil-ity increases in response to the surprise component of virtually allannouncements (including our proxy for the surprise component inthe Inflation Report, albeit at low levels of statistical significance).Second, the results with regard to the announcement dummies (nowcleanly measuring the effect of the incidence of a release) improvein several ways: the coefficient estimate with regard to the tradebalance dummy switches sign and turns negative in line with allother coefficients; at the two- and three-month maturities, the pre-viously insignificant coefficient on the CPI dummy is now also sig-nificant and negative; all coefficients are substantially larger thantheir correspondents in table 2.

Based on this evidence, we conclude that the arrival of newsleads to an update in expectations. Their incidence tends to enhancethe precision of expectations, but this effect can possibly be over-shadowed by the volatility-enhancing effect of the announcement’ssurprise component. However, in the majority of cases, the volatility-reducing factor turns out to be dominating. Taken together,these results provide strong evidence in favor of hypotheses 1and 2.

3.2 What if Central Bank Communication Becomes Stale?

To test hypotheses 3–5, we will now turn to measuring the marketreaction conditional on the freshness of central bank communication.For that purpose, we expand the econometric model by including thevariable that measures how much time has elapsed since the preced-ing release of an Inflation Report, and by interacting this variablewith the macroeconomic release data. The results are provided intable 4.

Looking at the direct effect of the elapsed time, it is apparent thatvolatility does indeed increase with the distance from the precedingInflation Report release. This effect is strongest at the two-monthmaturity but is also present and statistically significant up to a hori-zon of six months. Despite its statistical significance, it has to benoted that the magnitude of the effect is small. However, figure 1has shown that the omission of macroeconomic announcements willtend to depress estimates of ζ and, as we will see later on, this


le4.

The

Effec

tof

Annou

nce

men

tson

Yie

lds:

Dis

tance

toth

eP

rece

din

gIn

flat

ion

Rep

ort

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

UK

CP

I0.

004∗

∗0.0

020.

009∗

∗∗0.0

020.

018∗

∗∗0.0

020.

026∗

∗∗0.0

03U

KP

MI

0.00

30.0

020.

005∗

0.0

030.

010∗

∗0.0

040.

020∗

∗∗0.0

07U

KR

etai

lSa

les

0.00

4∗∗∗

0.0

010.

006∗

∗∗0.0

010.

014∗

∗∗0.0

020.

017∗

∗∗0.0

03U

K−

0.00

20.0

01−

0.00

20.0

02−

0.00

20.0

02−

0.00

30.0

04U

nem

ploy

men

tU

KTra

de−

0.00

20.0

01−

0.00

5∗∗∗

0.0

02−

0.00

4∗0.0

020.

000

0.0

02B

alan

ceU

KC

PI

*T

ime

0.01

4∗∗∗

0.0

050.

018∗

∗∗0.0

050.

023∗

∗∗0.0

070.

027∗

∗∗0.0

10E

laps

edU

KP

MI

*T

ime

0.01

4∗0.0

070.

007

0.0

090.

001

0.0

13−

0.00

80.0

24E

laps

edU

KR

etai

lSa

les

*0.

006

0.0

040.

004

0.0

040.

000

0.0

06−

0.00

50.0

09T

ime

Ela

psed

UK

−0.

006∗

∗0.0

03−

0.00

40.0

05−

0.00

70.0

050.

003

0.0

10U

nem

ploy

men

t*

Tim

eE

laps

edU

KTra

de−

0.00

50.0

06−

0.00

50.0

06−

0.00

70.0

06−

0.01

2∗0.0

07B

alan

ce*

Tim

eE

laps

edT

ime

Ela

psed

−0.

001

0.0

010.

001

0.0

010.

001

0.0

020.

001

0.0

02Δ

IRIn

flati

on0.

006

0.0

100.

002

0.0

110.

017

0.0

150.

061∗

∗∗0.0

19Fo

reca

st

(con

tinu

ed)

Vol. 8 No. 3 News Content of Macroeconomic Announcements 25Tab

le4.

(Con

tinued

)

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

Var

ianc

eEqu

atio

n

UK

CP

I−

0.23

1∗0.1

38−

0.02

40.1

300.

072

0.1

10−

0.02

60.0

99(D

umm

y)U

KP

MI

−0.

549∗

∗∗0.0

99−

0.58

9∗∗∗

0.0

85−

0.43

6∗∗∗

0.0

86−

0.18

2∗∗

0.0

74(D

umm

y)U

KR

etai

lSa

les

−0.

857∗

∗∗0.0

74−

0.90

3∗∗∗

0.0

71−

0.61

1∗∗∗

0.0

71−

0.39

7∗∗∗

0.0

73(D

umm

y)U

K−

0.75

8∗∗∗

0.1

04−

0.66

4∗∗∗

0.1

03−

0.57

4∗∗∗

0.0

96−

0.10

30.0

79U

nem

ploy

men

t(D

umm

y)U

KTra

de0.

140∗

∗0.0

650.

135∗

∗0.0

620.

154∗

∗∗0.0

600.

080

0.0

58B

alan

ce(D

umm

y)U

KC

PI

−1.

129∗

∗∗0.2

83−

0.93

9∗∗∗

0.2

64−

0.33

00.2

340.

339

0.2

49(D

umm

y)*

Tim

eE

laps

edU

KP

MI

−0.

332

0.2

79−

0.24

40.2

58−

0.18

10.2

800.

289

0.2

97(D

umm

y)*

Tim

eE

laps

edU

KR

etai

lSa

les

−0.

428∗

0.2

42−

0.46

0∗0.2

37−

0.63

8∗∗∗

0.2

47−

0.42

8∗0.2

26(D

umm

y)*

Tim

eE

laps

ed

(con

tinu

ed)


Tab

le4.

(Con

tinued

)

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

Var

ianc

eEqu

atio

n

UK

0.50

6∗0.2

800.

368

0.2

750.

202

0.2

96−

0.35

90.2

47U

nem

ploy

men

t(D

umm

y)*

Tim

eE

laps

edU

KTra

de−

0.44

3∗∗

0.2

25−

0.07

90.2

490.

146

0.2

260.

136

0.1

82B

alan

ce(D

umm

y)*

Tim

eE

laps

edT

ime

Ela

psed

0.07

4∗∗∗

0.1

160.

031∗

0.0

170.

040∗

∗0.0

17−

0.00

70.0

17In

flati

onR

epor

t−

1.09

1∗∗∗

0.1

10−

1.00

4∗∗∗

0.1

06−

0.70

4∗∗∗

0.1

210.

335∗

∗∗0.0

95(D

umm

y)

Note

s:T

he

table

show

s,fo

rth

ediff

eren

tm

aturi

ties

(in

the

vari

ous

colu

mns)

,th

ere

acti

onco

effici

ent

ofU

Kze

ro-c

oupon

yiel

ds

tom

acro

eco-

nom

ican

nou

nce

men

tsan

dth

ere

leas

eof

the

Inflat

ion

Rep

ort

inge

ner

alan

ddep

endin

gon

the

tim

eel

apse

dsi

nce

the

rele

ase

ofth

epre

cedin

gIn

flat

ion

Rep

ort.

Num

ber

sin

ital

ics

den

ote

the

stan

dar

der

rors

.**

*,**

,an

d*

indic

ate

stat

isti

calsi

gnifi

cance

atth

e1%

,5%

,an

d10

%le

vel,

resp

ecti

vely

.T

he

mea

neq

uat

ion

issp

ecifi

edas

rt

=c1

+δIR

sIR t

+∑ k

δks

k t+

∑ kφ

ks

k tg

t+

ωg

t+

ψr

t−

1+

∑ lν

ld

l t+

μt,th

eva

rian

ceeq

uat

ion

as

log(

ht)

=c2

+κ1(|

μt−

1/√

ht−

1|−

√2/

π)+

κ2(μ

t−

1/√

ht−

1)+

κ3lo

g(h

t−

1)+

λIR

aIR t

+∑ k

λka

k t+

∑ kρ

ka

k tg

t+

ζg

t+

∑ lτd

l t.


conjecture is confirmed by the inclusion of four U.S. announcementsin a subsequent robustness test, with the estimates of ζ doublingor even tripling in magnitude. The volatility-increasing effect disap-pears at the one-year horizon, and consistently so in all further mod-els that we will analyze. At the same time, as should be expected,the elapsed time does not matter for the mean of yields, as shownby the very small and statistically highly insignificant coefficients forthis variable in the mean equation. This result is in clear support ofhypothesis 3.

What about the volatility reduction in response to macroeco-nomic releases? Hypothesis 5 stated that this effect should be morepronounced the more time has elapsed since the preceding releaseof an Inflation Report. As a matter of fact, with very few excep-tions, the interaction terms are typically negative, supporting thehypothesis. Although a number of these are not estimated to besignificant, the picture that emerges is fairly consistent: with oneexception, all significant coefficients are negative. Furthermore, theeffects are sizable. Take the example of the CPI release: its effecton average is estimated to be −0.23 for the two-month maturity.However, the first CPI release after a given Inflation Report has vir-tually no effect on volatility, with its coefficient estimated at +0.063,while the third CPI release has a coefficient estimate of −0.591.12As before, the effects are particularly pronounced at shortermaturities.

The results strengthen by imposing the split of announcementeffects into the announcement incidence and the absolute surprisecomponent, as shown in table 5. In this specification, all statisticallysignificant coefficients on the non-interacted dummies are negative,as are all the significant coefficients on the interaction terms. Atthe same time, all significant coefficients on the absolute surprise aswell as on the respective interaction terms are positive. Controllingfor the surprise component, in the example of CPI the differences

12These numbers are calculated as follows: the average value of the “time-elapsed” variable gt is calculated for the first and the third CPI releases.These values amount to −0.260 and 0.319, respectively. The average effect ofthe first CPI release on volatility is then given by λCPI + (−0.260)ρCPI =−0.231 + 0.260 ∗ 1.129 = 0.063; the average effect for the third CPI release isλCPI + 0.319ρCPI = −0.231 − 0.319 ∗ 1.129 = −0.591.


Tab

le5.

The

Effec

tof

Annou

nce

men

tson

Yie

lds:

Dis

tance

toIn

flat

ion

Rep

ort

and

Inci

den

cevs.

Abso

lute

Surp

rise

,M

ean

Equat

ion

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

UK

CP

I0.

004∗

0.0

020.

009∗

∗∗0.0

030.

018∗

∗∗0.0

030.

026∗

∗∗0.0

03U

KP

MI

0.00

30.0

040.

007

0.0

040.

011∗

0.0

060.

020∗

∗0.0

08U

KR

etai

lSa

les

0.00

4∗∗∗

0.0

020.

006∗

∗∗0.0

020.

014∗

∗∗0.0

020.

018∗

∗∗0.0

03U

KU

nem

ploy

men

t−

0.00

10.0

01−

0.00

20.0

02−

0.00

20.0

02−

0.00

30.0

04U

KTra

deB

alan

ce0.

000

0.0

02−

0.00

20.0

03−

0.00

30.0

040.

000

0.0

03U

KC

PI

*T

ime

0.01

4∗∗

0.0

060.

016∗

∗0.0

070.

021∗

∗∗0.0

080.

025∗

∗∗0.0

09E

laps

edU

KP

MI

*T

ime

0.01

10.0

120.

002

0.0

14−

0.00

50.0

18−

0.00

90.0

27E

laps

edU

KR

etai

lSa

les

*0.

005

0.0

050.

003

0.0

060.

002

0.0

06−

0.00

30.0

09T

ime

Ela

psed

UK

Une

mpl

oym

ent

−0.

004

0.0

03−

0.00

20.0

04−

0.00

40.0

070.

004

0.0

12*

Tim

eE

laps

edU

KTra

deB

alan

ce−

0.00

70.0

08−

0.00

70.0

09−

0.00

60.0

11−

0.01

10.0

08*

Tim

eE

laps

edT

ime

Ela

psed

−0.

001

0.0

010.

000

0.0

010.

000

0.0

010.

000

0.0

02Δ

IRIn

flati

on0.

010

0.0

100.

000

0.0

120.

039∗

∗0.0

170.

078∗

∗∗0.0

24Fo

reca

st

(con

tinu

ed)


le5.

(Con

tinued

)

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

Var

ianc

eEqu

atio

n

UK

CP

I(D

umm

y)−

0.57

3∗∗∗

0.2

09−

0.45

4∗∗

0.2

02−

0.16

00.1

76−

0.05

10.1

44U

KP

MI

(Dum

my)

−0.

989∗

∗∗0.1

40−

1.12

7∗∗∗

0.1

22−

0.79

4∗∗∗

0.1

13−

0.36

0∗∗∗

0.0

97U

KR

etai

lSa

les

−1.

048∗

∗∗0.0

91−

1.10

9∗∗∗

0.0

89−

0.64

3∗∗∗

0.0

88−

0.46

0∗∗∗

0.0

87(D

umm

y)U

KU

nem

ploy

men

t−

0.92

2∗∗∗

0.1

79−

1.02

7∗∗∗

0.1

78−

0.87

3∗∗∗

0.1

45−

0.11

70.1

09(D

umm

y)U

KTra

deB

alan

ce−

0.49

3∗∗∗

0.0

94−

0.48

8∗∗∗

0.0

80−

0.35

7∗∗∗

0.0

81−

0.10

90.0

85(D

umm

y)U

KC

PI

(Abs

.0.

294

0.1

900.

414∗

∗0.1

690.

284∗

0.1

670.

073

0.1

27Su

rpri

se)

UK

PM

I(A

bs.

0.76

9∗∗∗

0.2

040.

765∗

∗∗0.1

850.

599∗

∗∗0.1

650.

260∗

∗∗0.1

01Su

rpri

se)

UK

Ret

ailSa

les

0.37

7∗∗∗

0.0

690.

412∗

∗∗0.0

670.

201∗

∗∗0.0

590.

107∗

0.0

55(A

bs.Su

rpri

se)

UK

Une

mpl

oym

ent

−0.

022

0.1

250.

094

0.1

250.

179∗

0.1

02−

0.03

00.0

75(A

bs.Su

rpri

se)

UK

Tra

deB

alan

ce0.

735∗

∗∗0.0

710.

755∗

∗∗0.0

660.

609∗

∗∗0.0

660.

211∗

∗∗0.0

57(A

bs.Su

rpri

se)

UK

CP

I(D

umm

y)*

−1.

080∗

∗0.5

08−

1.03

1∗0.5

43−

0.81

60.5

34−

0.00

30.4

07T

ime

Ela

psed

UK

PM

I(D

umm

y)*

0.12

40.4

120.

016

0.3

940.

100

0.4

090.

357

0.3

81T

ime

Ela

psed

(con

tinu

ed)


le5.

(Con

tinued

)

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

Var

ianc

eEqu

atio

n

UK

Ret

ailSa

les

−0.

828∗

∗∗0.3

22−

0.94

2∗∗∗

0.3

29−

1.48

8∗∗∗

0.3

33−

1.10

2∗∗∗

0.3

10(D

umm

y)*

Tim

eE

laps

edU

KU

nem

ploy

men

t−

0.20

00.5

77−

0.26

80.5

64−

0.28

90.4

59−

1.14

4∗∗∗

0.3

74(D

umm

y)*

Tim

eE

laps

edU

KTra

deB

alan

ce−

0.53

1∗0.3

19−

0.46

30.3

39−

0.39

20.2

97−

0.25

50.2

94(D

umm

y)*

Tim

eE

laps

edU

KC

PI

(Abs

.0.

159

0.5

510.

321

0.4

880.

730

0.4

980.

775∗

∗0.3

71Su

rpri

se)

*T

ime

Ela

psed

UK

PM

I(A

bs.

−0.

289

0.7

04−

0.77

70.6

87−

0.70

90.6

96−

0.61

60.5

75Su

rpri

se)

*T

ime

Ela

psed

UK

Ret

ailSa

les

0.40

90.3

420.

383

0.3

480.

925∗

∗∗0.3

110.

607∗

∗0.2

56(A

bs.Su

rpri

se)

*T

ime

Ela

psed

(con

tinu

ed)


Tab

le5.

(Con

tinued

)

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

Var

ianc

eEqu

atio

n

UK

Une

mpl

oym

ent

1.30

2∗∗∗

0.4

051.

472∗

∗∗0.4

031.

102∗

∗∗0.3

230.

807∗

∗∗0.2

89(A

bs.Su

rpri

se)

*T

ime

Ela

psed

UK

Tra

deB

alan

ce−

0.58

7∗0.3

11−

0.35

30.3

14−

0.07

00.3

250.

178

0.2

38(A

bs.Su

rpri

se)

*T

ime

Ela

psed

Tim

eE

laps

ed0.

085∗

∗∗0.0

180.

061∗

∗∗0.0

180.

063∗

∗∗0.0

200.

012

0.0

20In

flati

onR

epor

t−

1.23

1∗∗∗

0.1

45−

1.20

7∗∗∗

0.1

45−

0.88

5∗∗∗

0.1

55−

0.52

0∗∗∗

0.1

21(D

umm

y)In

flati

onR

epor

t1.

503∗

0.8

461.

670∗

0.8

641.

348

0.8

381.

518∗

∗∗0.5

17(A

bs.Su

rpri

se)

Note

s:T

he

table

show

s,fo

rth

ediff

eren

tm

aturi

ties

(in

the

vari

ous

colu

mns)

,th

ere

acti

onco

effici

ent

ofU

Kze

ro-c

oupon

yiel

ds

tom

acro

-ec

onom

ican

nou

nce

men

tsan

dth

ere

leas

eof

the

Inflat

ion

Rep

ort

inge

ner

alan

ddep

endin

gon

the

tim

eel

apse

dsi

nce

the

rele

ase

ofth

epre

cedin

gIn

flat

ion

Rep

ort,

separ

atin

gth

eeff

ect

ofth

ean

nou

nce

men

tin

ciden

cean

dth

eab

solu

tesu

rpri

sein

the

vari

ance

equat

ion.

Num

-ber

sin

ital

ics

den

ote

the

stan

dar

der

rors

.**

*,**

,an

d*

indic

ate

stat

isti

cal

sign

ifica

nce

atth

e1%

,5%

,an

d10

%le

vel,

resp

ecti

vely

.T

he

mea

neq

uat

ion

issp

ecifi

edas

rt

=c1

+δIR

sIR t

+∑ k

δks

k t+

∑ kφ

ks

k tg

t+

ωg

t+

ψr

t−

1+

∑ lν

ld

l t+

μt,

the

vari

ance

equat

ion

aslo

g(h

t)

=

c2

+κ1(|

μt−

1/√

ht−

1|−

√2/

π)+

κ2(μ

t−

1/√

ht−

1)+

κ3lo

g(h

t−

1)+

λIR

aIR t

+∑ k

λka

k t+

∑ kρ

ka

k tg

t+

∑ k

k|s

k t|+

∑ kιk

|sk t|g

t+

ζg

t+

∑ lτd

l t.


between the first and the last announcement are even more striking,with an effect of −0.292 for the first and of −0.918 for the last,suggesting that the volatility reduction of the last announcement ismore than three times as large as for the first announcement. Theevidence is therefore clearly in support of hypothesis 5.

The last hypothesis to be tested, number 4, stated that the reac-tion of asset prices to macroeconomic news should be stronger themore time has elapsed since the preceding release of an InflationReport. To see whether this is the case, we need to check the coeffi-cients in the mean equation. As mentioned above, the non-interactedcoefficients in the mean equation show the sign that should beexpected according to economic reasoning. The question to be set-tled is whether the coefficients on the interacted variables show thesame sign and are statistically significant. While it is generally truethat the coefficients of the interacted variables have the same signas those of the non-interacted variables, their statistical significance(against zero) is rather weak. Only in the case of the CPI announce-ment do we find such a pattern consistently across the differentmaturities.

One interesting observation in that context relates to the evo-lution of coefficients across the maturity spectrum. Contrary towhat was observed for the volatility equation, effects are now largerwith increasing maturities. This finding is in line with Fleming andRemolona (1999b), who had also found an increasing magnitudeover the maturities studied in the current paper. At the one-yearhorizon, the effects are relatively stark, though: the response of one-year yields to a one-standard-deviation surprise in the first CPIrelease following an Inflation Report is estimated at 1.8 basis points,whereas the response to the last CPI release is substantially largerat 3.3 basis points.

3.3 Robustness and Extensions

We have tested the robustness of these results in a number ofways. First, we are interested in whether the inclusion of otherBank of England communication events makes a difference, giventhat these provide an occasion for financial market participants toupdate their information set. This might apply in particular to the


communications in the context of interest rate changes,13 to therelease of the minutes of the MPC meetings, and possibly also tospeeches by MPC members. For interest rate decisions, it is possibleto control for the surprise component by means of a Reuters surveyamong financial market participants; for minutes, our analysis justcomprises the inclusion of a dummy variable that is equal to one onthe days of the release of the minutes. With regard to speeches andinterviews by MPC members, we have used an update of the dataset by Ehrmann and Fratzscher (2007). This data set collects allspeeches by MPC members containing some forward-looking refer-ence to monetary policy inclinations. The communications are codedas +1 if they indicate a tendency towards tightening, as 0 for neutralstatements, and as −1 for statements suggesting an easing of mon-etary policy. As in Ehrmann and Fratzscher (2007), we will enterthis variable into the mean equation and a dummy variable for theincidence of such speeches in the variance equation. As shown intable 6, our findings remain qualitatively robust.

The additional communication variables provide interestingresults themselves. The surprise component in policy rate changesleads to significant responses of market interest rates, with the effectsdecreasing for longer maturities. Also the speeches affect interestrates consistently, and in the expected direction: relatively hawkishstatements raise interest rates, while relatively dovish statementslower them. The magnitude is very much in line with the results ofEhrmann and Fratzscher (2007), with 0.004 vs. 0.005 at the three-month maturity and 0.009 in both studies at the one-year maturity.Turning to the variance equation, we find—in line with the resultsfor macroeconomic announcements—the surprise component in pol-icy rate changes raises conditional volatility, whereas the incidence

13A broader measure would be to include all monetary policy decisions, i.e.,also the no-change decisions. Decisions not to change interest rates (especiallywhen they had not been fully anticipated, i.e., some market participants hadindeed expected a change to occur at a given MPC meeting) typically raise ques-tions as to the timing of the next policy rate change, an uncertainty that, bydefinition, is less imminent in the case of policy rate changes. Using the broadervariable in our regressions does not affect our results, with the exception of theresponse of conditional volatility to the incidence of a decision, where we finda reduction for change decisions and an increase for the broader measure. Wesuspect that this relates to the uncertainty about the timing of the next moveand should vanish at longer maturities.


le6.

The

Effec

tof

Annou

nce

men

tson

Yie

lds:

Dis

tance

toIn

flat

ion

Rep

ort

and

Inci

den

cevs.

Abso

lute

Surp

rise

and

Oth

erB

oEC

omm

unic

atio

ns,

Mea

nEquat

ion

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

UK

CP

I0.

005∗

∗0.0

020.

009∗

∗∗0.0

020.

018∗

∗∗0.0

030.

026∗

∗∗0.0

03U

KP

MI

0.00

30.0

030.

004

0.0

040.

010∗

0.0

050.

019∗

∗∗0.0

07U

KR

etai

lSa

les

0.00

4∗∗∗

0.0

010.

006∗

∗∗0.0

020.

014∗

∗∗0.0

020.

017∗

∗∗0.0

03U

K−

0.00

10.0

01−

0.00

10.0

02−

0.00

20.0

02−

0.00

40.0

04U

nem

ploy

men

tU

KTra

de−

0.00

10.0

03−

0.00

40.0

04−

0.00

40.0

040.

000

0.0

03B

alan

ceU

KC

PI

*T

ime

0.00

90.0

060.

015∗

∗0.0

070.

021∗

∗∗0.0

080.

023∗

∗∗0.0

08E

laps

edU

KP

MI

*T

ime

0.00

70.0

10−

0.00

10.0

12−

0.00

90.0

16−

0.01

10.0

23E

laps

edU

KR

etai

lSa

les

0.00

40.0

050.

002

0.0

05−

0.00

10.0

06−

0.00

50.0

09*

Tim

eE

laps

edU

K−

0.00

50.0

03−

0.00

30.0

05−

0.00

40.0

070.

004

0.0

12U

nem

ploy

men

t*

Tim

eE

laps

edU

KTra

de−

0.00

40.0

09−

0.00

50.0

10−

0.00

30.0

11−

0.01

10.0

08B

alan

ce*

Tim

eE

laps

edIn

tere

stR

ate

0.50

0∗∗∗

0.0

210.

482∗

∗∗0.0

200.

464∗

∗∗0.0

340.

418∗

∗∗0.0

34C

hang

esSp

eech

esby

0.00

20.0

010.

004∗

∗∗0.0

020.

005∗

∗0.0

020.

009∗

∗∗0.0

03M

PC

Mem

bers

(con

tinu

ed)


le6.

(Con

tinued

)

Tw

oM

onth

sT

hre

eM

onth

sSix

Mon

ths

Tw

elve

Mon

ths

Std

.Std

.Std

.Std

.C

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

orC

oeff

.Err

or

Tim

eE

laps

ed−

0.00

10.0

010.

000

0.0

010.

000

0.0

01−

0.00

10.0

02Δ

IRIn

flati

on−

0.01

30.0

10−

0.00

40.0

120.

037∗

∗0.0

170.

068∗

∗∗0.0

24Fo

reca

st

Var

ianc

eEqu

atio

n

UK

CP

Documents

The News Content of Macroeconomic Announcements: What ...Vol. 8 No. 3 News Content of Macroeconomic Announcements 3 its information becomes stale over time. Third, the price response