Leo M Broni 2021

JID: FINEC
ARTICLE IN PRESS [m3Gdc;May 30, 2021;23:52]

Journal of Financial Economics xxx (xxxx) xxx
Contents lists available at ScienceDirect
Journal of Financial Economics

journal homepage: www.elsevier.com/locate/jfec
Central bank communication and the yield curveR

Matteo Leombroni a, Andrea Vedolin b, Gyuri Venter c, Paul Whelan d,∗
a
Department of Economics, Stanford University, 579 Jane Stanford Way, Stanford, CA 94305, USA
b
NBER, CEPR, and Questrom School of Business, Boston University, 595 Commonwealth Avenue, Boston, MA 02118, USA
c
Warwick Business School, Warwick University, Coventry CV4 7AL, UK
d
Department of Finance, Copenhagen Business School, Solbjerg Plads 3, Frederiksberg 2000, Denmark
a r t i c l e i n f o a b s t r a c t
Article history: In this paper, we argue that monetary policy in the form of central bank communication
Received 16 April 2020 can shape long-term interest rates by changing risk premia. Using high-frequency move-
Revised 27 May 2020
ments of default-free rates and equity, we show that monetary policy communications by
Accepted 7 June 2020
the European Central Bank on regular announcement days led to a significant yield spread
Available online xxx
between peripheral and core countries during the European sovereign debt crisis by in-
JEL classification: creasing credit risk premia. We also show that central bank communication has a pow-
E43 erful impact on the yield curve outside regular monetary policy days. We interpret these
E58 findings through the lens of a model linking information embedded in central bank com-
G12 munication to sovereign yields.
Keywords: © 2021 Published by Elsevier B.V.

Interest rates
Monetary policy
Central bank communication
Eurozone
1. Introduction
The financial turmoil of 20 07–20 08 and the subsequent

R
European debt crisis have fueled a lively debate about
We thank the editor, Bill Schwert, and an anonymous referee for help-
ful comments and suggestions. We also thank Daniel Buncic, Stefania the role of central banks in controlling long-term interest
D’Amico, Paul Ehling, Jean-Sébastien Fontaine, Charles Goodhart, Robin rates. In this paper, we argue that monetary policy com-
Greenwood, Sven Klingler, Gábor Kőrösi, Lukas Kremens, David Lando, munication by central banks can have a dramatic impact
Anh Le, Wolfgang Lemke, David Lucca, Hanno Lustig, Aytek Malkhozov, on long-term interest rates via a risk premium channel. We
Charles Martineau, Felix Matthys, Leonardo Melosi, Lasse Nielsen, Ali
Ozdagli, Lasse Pedersen, Monika Piazzesi, Huw Pill, Gábor Pintér, Ricardo
establish this claim by showing that monetary policy com-
Reis, John Rogers, Dirk Schumacher, Ulf Söderström, Karlye Dilts Stedman, munications by the European Central Bank (ECB) led to
Alireza Tahbaz-Salehi, Jonathan Wright, Hongjun Yan, and seminar and a significant yield spread between peripheral (Italy and
conference participants at various universities and institutions for help- Spain) and core (Germany and France) countries during
ful comments and suggestions. We thank Now-casting Economics Ltd for
the European sovereign debt crisis by increasing credit
providing access to their eurozone now-casts. Gyuri Venter acknowledges
financial support from the Independent Research Fund Denmark (grant risk premia.
no. DFF-8019-00108B). Paul Whelan gratefully acknowledges support from Fig. 1 displays cumulative changes in ten-year core and
the FRIC Center for Financial Frictions (grant no. DNRF-102) and the Dan- peripheral sovereign yields between 2001 and 2015 on reg-
ish Finance Institute (DFI). ular ECB monetary policy meeting days, that is, days when
∗
Corresponding author.
the ECB sets the key interest rates for the euro area. The
E-mail addresses: leombm@stanford.edu (M. Leombroni),
avedolin@bu.edu (A. Vedolin), gyuri.venter@wbs.ac.uk (G. Venter),
plot shows that while core and peripheral yields moved
pawh.fi@cbs.dk (P. Whelan). one-for-one on these days before 2009, after the onset of
https://doi.org/10.1016/j.jfineco.2021.04.036
0304-405X/© 2021 Published by Elsevier B.V.
Please cite this article as: M. Leombroni, A. Vedolin, G. Venter et al., Central bank communication and the yield curve,
Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2021.04.036
JID: FINEC
M. Leombroni, A. Vedolin, G. Venter et al. Journal of Financial Economics xxx (xxxx) xxx
spreads in response to communication shocks must be due

to changes in risk premia as opposed to changes in expec-
tations of future short-term interest rates or term premia.
We start our analysis by developing a theoretical frame-
work that highlights how central bank communication af-
fects risk premia. We consider a currency union of multiple
countries, representing the eurozone. In the model, central
bank communication has two dimensions: one about the
intended future path of interest rates (forward guidance)
and the other about additional policies, such as asset pur-
chases, liquidity supports, or lending and refinancing op-
erations. The two shocks drive investors’ perceived prob-
ability of a credit event, such as a peripheral default or
the breakup of the eurozone, and hence impact the pre-
mia they demand on risky assets such as sovereign bonds
Fig. 1. European sovereign bond yield changes on ECB monetary policy
days This figure displays cumulative one-day changes in ten-year yields and equity. This mechanism is based on the premise that
for core (average of Germany and France) and peripheral (average of Italy market participants have imprecise knowledge about either
and Spain) bonds as well as the spread between peripheral and core the central bank’s reaction function, such as when it would
bonds only on European Central Bank meeting days. Data run from Jan- introduce unconventional policies, or about its private sig-
uary 2001 to December 2014.
nals, as in Romer and Romer (20 0 0) and Nakamura and
Steinsson (2018). Then, asset price movements around an-
nouncements are informative about market participants’
the European debt crisis yields diverged, leading to a sig-
reaction to the new information embedded in these an-
nificant spread. Importantly, this spread emerged during a
nouncements.
period when a series of unconventional measures were im-
The framework first formalizes how to identify inter-
plemented to reduce it.1
est rate communication shocks (also referred to as forward
Using high-frequency movements of default-free rates
guidance shocks) from changes in risk-free money market
and an equity index, we show that monetary policy com-
rates around communication events. Further, if equity is
munications conducted by the ECB on regular announce-
also expected to respond to a peripheral default, the equity
ment days were responsible for the pattern shown in
reaction that is orthogonal to interest rate shocks is infor-
Fig. 1 by increasing credit risk premia. These increases
mative about risk premia and provides an identification of
were economically sizable and, at the very least, amplified
pure risk premium shocks of monetary policy communica-
sovereign yield volatility, making it harder for the ECB to
tion.
succeed in reducing peripheral yields faster. However, we
The model also provides hypotheses about the impact
also show that speeches by the ECB president outside the
of these two types of shocks on sovereign yields. A neg-
regular monetary policy announcements significantly de-
ative forward guidance shock decreases bond yields uni-
creased the peripheral-core spread and, together with the
formly across all sovereigns by signaling lower future inter-
announcements of unconventional policies, led to a sizable
est rates than what the market expected but at the same
reduction in the yield spread. Taken together, our findings
time can also increase the required risk premium on all
provide novel evidence that monetary policy in the form of
sovereign debt, dampening the effect of the expectation
central bank communication can impact long-term interest
channel. A negative pure risk premium shock, on the other
rates by changing risk premia.
hand, increases credit-risky sovereign yields. Overall, inter-
Our empirical strategy exploits two key features of
est rate and risk premium shocks can help explain the dif-
monetary policy announcements in the eurozone. First, the
ference in peripheral and core yield reactions to monetary
ECB’s protocol for announcing monetary policy decisions
policy communication.
allows us to disentangle the component of the policy an-
To perform our empirical analysis, we extract the two
nouncement that contains new information about the fu-
monetary policy shocks on ECB announcement days using
ture path of interest rates or credit risk—what we refer
high-frequency data on money market rates and an eq-
to as communication shocks—from the announcement of
uity index. Because the timing of the press conferences is
the short-term interest rate. Second, the fact that (current
known precisely, we can identify surprises related to the
and future) short-term interest rates are common across
future path of short-term rates using changes in risk-free
all eurozone countries implies that any change in yield
interest rates with different maturities. Equity returns, also
sampled during ECB press conferences, allow us to extract
1
shocks that are informative about the probability of a fu-
On 8 August, 2011 and 10 May, 2010, the ECB announced di-
rect purchases of government debt through its Securities Markets Pro-
ture default in the euro area.
gramme, and on 6 September, 2012, it announced further purchases With the two communication shocks in hand, we test
via its Outright Monetary Transactions; Altavilla et al. (2014) and the model predictions and show a number of novel re-
Falagiarda and Reitz (2015), among others, show a significant reduction sults regarding eurozone yields. First, for our main result,
in the periphery-core spread due to these measures. In January 2015,
we split our sample into pre-sovereign debt crisis (January
the ECB launched its expanded Asset Purchase Programme; De Santis and
Holm-Hadulla (2017), among others, evaluate its effects on financial mar- 2001 to November 20 09, 10 0 observations) and sovereign
kets. debt crisis (December 2009 to December 2014, 61 ob-
2
JID: FINEC
servations) periods separately. We find that precrisis ECB These results relate to the broader notion that risk pre-
communication affected bond yields of euro-area countries mia due to monetary policy can also be earned outside
uniformly. However, we find that, during the crisis, pe- standard monetary policy announcement days; see, for ex-
ripheral yields’ response to interest rate shocks became ample, Neuhierl and Weber (2019). Moreover, we show
muted, while core yields continued to react strongly. Fur- that ECB president speeches led to a significant reduc-
ther, while the effect of risk premium shocks was negligi- tion in the peripheral-core spread, offsetting the increase
ble precrisis, they became the dominant force driving yield in spreads observed on regular ECB announcement days.
spreads afterwards. We find that interest rate and risk pre- This effect on the spread is further strengthened once we
mium shocks explain around 40% of changes in ten-year take into account announcements of unconventional poli-
yield spreads, with risk premium shocks being responsible cies: the overall effect—once we combine all announce-
for the majority of this variation. ment and speeches—was a sizable reduction of the yield
Second, using rolling regressions, we confirm that the spread. These results stress the relevance of taking into
effect of central bank communication about forward guid- account central bank communication outside regular an-
ance on peripheral bond yields declined during the cri- nouncement dates.
sis period, while risk premium shocks became increas- Fifth, we extend our analysis to the period after the
ingly important in driving up yield spreads. Taking into introduction of the ECB Quantitative Easing program.
account only scheduled announcement days, we find that To this end, we reestimate our baseline regressions for
central bank communication was responsible for a signifi- an extended sample period between 2015 and 2018
cant wedge that, at its peak around the end of 2013, repre- and include a Quantitative Easing (QE)-related policy
sented 25% of the total ten-year yield spread. This finding shock that we construct following Swanson (2018) and
is important since it coincides with a period when uncon- Altavilla et al. (2019). We find that the power of interest
ventional measures were implemented to reduce spreads. rate communication shocks returned to the precrisis level:
The dramatic difference between the effect of monetary estimated coefficients are highly statistically significant,
policy communications in the precrisis and crisis periods and these communication shocks can explain more than
is in line with two distinct regimes in our model and re- 60% of the variation in both core and peripheral yields. We
lates to the literature that links the European debt crisis to therefore conclude that the introduction of unconventional
self-fulfilling beliefs and multiple equilibria; see, for exam- monetary policies such as QE resurrected the power of
ple, Corsetti and Dedola (2016), Bocola and Dovis (2019), monetary policy communication about interest rates by
and Lorenzoni and Werning (2019). According to this in- reducing its risk premium effect.
terpretation, the period before late 2009 featured a small We perform a number of robustness checks to challenge
probability of a credit event and a low sensitivity of this our main finding by including macroeconomic announce-
probability to ECB communication shocks. In contrast, af- ments, changing the sampling frequencies of our left- and
ter December 2009, negative risk premia shocks, signaling right-hand variables, and considering alternative risk pre-
a lower probability of the introduction of necessary “save mium shocks. Taken together, our findings illustrate that
the euro” policies, increased agents’ perceived probability central bank communication can have significant effects on
of a credit event significantly, which in turn drove yield asset prices via a risk premium channel during and outside
spreads up even further. anticipated monetary policy announcements.
Third, we study the link between central bank commu- A large literature in macro-finance studies the effects
nication shocks and credit risk premia and show highly of the US Federal Reserve’s monetary policy on the cross-
significant effects on sovereign credit default swaps (CDSs) section of assets and market variables such as long-term
and, most importantly, their spread. This finding further real and nominal interest rates, equity returns, volatility,
corroborates our interpretation of risk premium shocks as and mortgage issuance; see, for example, Fama (2013),
being informative about the likelihood of a peripheral de- Hanson and Stein (2015), Boyarchenko et al. (2017),
fault. Hanson et al. (2020), and Neuhierl and Weber (2019).
Fourth, we investigate whether our results are exclu- While our approach is similar in spirit, we complement
sive to monetary policy announcements during press con- the above literature along at least two dimensions. First,
ferences or can be extended to central bank communi- we highlight the role of monetary policy to influence mar-
cation more generally—one of the most prominent being kets beyond the standard stance of conventional monetary
ECB President Draghi’s “whatever it takes” speech in 2012 policy and affect credit risk premia instead of term premia.
that immediately collapsed the peripheral-core spread and Second, the unique setting for the transmission of mone-
led to a rally in stock markets. To answer this ques- tary policy in the eurozone allows us to study central bank
tion and to go beyond anecdotal evidence, we construct communication separately from policy action.
interest rate and risk premium shocks during speeches An important literature studies the ECB’s action during
by ECB presidents akin to the procedure around mone- the European debt crisis. For example, Rogers et al. (2014),
tary policy press conferences. Using these shocks, we re- Fratzscher et al. (2016), Haitsma et al. (2016), Koijen et al.
run our main analysis and find patterns strikingly similar (2017, 2021), Krishnamurthy et al. (2018), and De San-
to standard monetary policy communication. While inter- tis (2019) all show that the unconventional policies of the
est rate and risk premium shocks have no significant ef- ECB successfully eased financial conditions in peripheral
fect on the yield spread before 2010, in the crisis period countries. In contrast to these papers, we study regular
both shocks explain around 35% of the variation of the monetary policy days, and our focus is on the different di-
yield spread on days when ECB presidents give speeches. mensions of central bank communication. Further, our long
3
JID: FINEC
time series enables us to show structural breaks in the ef- identification of communication shocks. We present our
fect these shocks have on the sovereign yield spread. We main empirical findings and perform various robustness
also extend our study from regular ECB monetary policy checks in Sections 5–7. An Online Appendix gathers addi-
meeting days to president speeches more generally. tional results omitted from the main paper.
The framework that guides our empirical approach
is also linked to a literature that explores belief- 2. Theoretical framework and main implications
driven equilibria around the European sovereign debt
crisis; see, for example, Corsetti and Dedola (2016), Our main premise is that monetary policy communi-
Bocola and Dovis (2019), Lorenzoni and Werning (2019), cation drives market participants’ beliefs about the future
and Bacchetta et al. (2020). We complement this theoret- path of interest rates as well as the implementation of ad-
ical literature by providing empirical evidence for a risk ditional policies, and we build a reduced-form model to
premium channel of monetary policy that arises in the study the cross-sectional impact of central bank commu-
“bad equilibria” of these models. nication on asset prices. Below we describe the model,
Our paper is also related to the literature that ex- highlight the mechanism we have in mind, describe how
plores the signaling channel of monetary policy: policy- to identify central bank communication shocks, and derive
makers’ actions reveal their private knowledge to mar- testable predictions. The formal model itself is delegated to
ket participants, which in turn can have real economic the Online Appendix.
effects; see, for example, Romer and Romer (20 0 0), While the main mechanism applies in general, to ac-
Campbell et al. (2016), and Nakamura and Steins- commodate our empirical application and provide testable
son (2018).2 We add to this literature by extracting two implications, we set up a modeling framework that can
distinct policy shocks that differentiate between standard represent the euro area. For this purpose, we consider a
interest rate shocks and news related to additional poli- currency union of multiple countries and think about a
cies that, in the eurozone setting, manifest as credit risk representative agent that trades default-free assets (e.g.
shocks. Different from this literature, we also argue that overnight indexed swap (OIS) rates), defaultable sovereign
our additional policy shocks (or risk premium shocks) can bonds in each country, and an aggregate equity index of
capture not only superior signals directly about macroe- the eurozone.
conomic variables but also information about the imple- The central bank (the ECB) has two roles in this
mentation of unconventional policies (or the lack thereof), economy: it sets the target short rate and communi-
which in turn naturally affect the macroeconomy. cates to market participants. We posit that central bank
Our identification of ECB communication shocks par- communication provides information about future short
tially follows Brand et al. (2010), who study the effect of rates (forward guidance) and additional policies. Our
monetary policy on eurozone money market rates; see also main interpretation of the latter type is signals about the
Altavilla et al. (2019). Our paper is different from theirs implementation of asset purchase programs or the lack
along several dimensions. First and foremost, we use not thereof.3 Market participants, in turn, update their beliefs
only money market rates but also equity returns to ex- about the probability of credit events that we think of as
tract two distinct channels of central bank communication, sovereign (mainly peripheral) defaults, or the breakup of
and we show that shocks driving credit risk premia have a the eurozone. In particular, we would expect credit risk
much more significant role in explaining sovereign yields to increase and future equity cash flows to decrease if
than the traditional interest rate shocks since 2009. Sec- the ECB signals lower future interest rates because the
ond, we study the cross-sectional differences in yield re- macroeconomy needs further stimulus, and if market
action to communication during the European sovereign participants find that either the probability or the scope
debt crisis, which is outside the sample period of Brand of future asset purchase programs is insufficient.
et al. (2010) and is not considered by Altavilla et al. (2019). In equilibrium, expected excess returns on all assets
Third, we show a more general link between central bank must compensate investors for the risk they bear: for
communication and asset prices that is also present when default-free bonds, this is only interest rate risk, whereas
ECB presidents give speeches. sovereign bonds and equity have risk premia that increase
The rest of the paper is organized as follows. in the probability of a credit event and the loss given a
Section 2 provides a simple theoretical framework to credit event (see, e.g., Duffie and Singleton, 1999). As a
study the impact of monetary policy communication on result, if and only if monetary policy communication is
sovereign yields. Section 3 presents the various data informative about the probability of the credit event and
sources used. Section 4 describes the institutional setting market participants consider peripheral countries (Greece,
of ECB monetary policy announcements and outlines the
3
This interpretation is consistent with the idea that monetary policy
2
Ellingsen and Soderstrom (2001), Woodford (2012), shocks are surprises about the central bank’s reaction to publicly avail-
Campbell et al. (2012), Gertler and Karádi (2015), Tang (2015), able information, as in Bauer and Swanson (2020). Alternatively, the stan-
Melosi (2016), Miranda-Agrippino and Ricco (2020), Ai and Bansal (2018), dard macro literature models central bank communication as revealing
Hansen et al. (2019), Laarits (2019), Andrade et al. (2019), and the bank’s private information about exogenous macro fundamentals such
Jarociński and Karádi (2020), among others, discuss further aspects as GDP growth, industrial production, or unemployment to the public;
of monetary policy signaling. See also Ehrmann and Fratzscher (2005), see, for example, Romer and Romer (20 0 0), Campbell et al. (2016), and
Lucca and Trebbi (2011), Schmeling and Wagner (2019), and Neuhierl and Nakamura and Steinsson (2018), among others. While both channels are
Weber (2019), who study the link between central bank tone and asset consistent with our formal model, we focus on the first interpretation due
returns. to the time period and the eurozone setting that we study.
4
JID: FINEC
Italy, Ireland, Portugal, and Spain) weaker/credit-riskier value can change across different regimes. In normal times,
than core countries such as Germany or France, the risk when the eurozone is in sound economic and financial
premium on the former are larger than on the latter; oth- condition, we would expect monetary policy communica-
erwise, there should be no difference. tion to have a small effect on credit risk, and as a result,
Consider now high-frequency intervals around commu- all sovereign bonds react to forward guidance shocks and
nication events such as ECB press conferences when all feature small risk premia. On the other hand, in more tur-
noncommunication shocks of the model are negligible. Our bulent (crisis) times, perceived credit risk is more sensi-
framework implies that one can identify shocks to the fu- tive to ECB communication. In turn, peripheral sovereign
ture path of interest rates from default-free rate changes in yields can stop reacting to conventional monetary policy,
these narrow intervals; we will denote these by IR. More- and negative additional policy shocks, which signal a lower
over, the impact of additional policy shocks can be identi- probability of the introduction of policies investors deem
fied by orthogonalizing high-frequency equity returns with necessary, drive up the perceived probability of a credit
respect to default-free yield changes and taking the resid- event, further raising yield spreads.
ual; we denote these by U. We summarize the above predictions in the follow-
The above setting has a series of implications about ing hypotheses: Hypothesis 1. In normal times, IR (forward
the effect of central bank communication shocks IR and guidance) communication shocks have a positive and uni-
U on the cross-section of sovereign yields. First, we show form impact on all sovereign yields. In crisis times, they
that sovereign yields of core countries react more to ECB have a positive effect on core yields, and a smaller or even
forward guidance shocks than peripherals. Sovereign bond negative impact on peripheral yields. Hypothesis 2. In nor-
yields are the average expected returns earned through the mal times, U (risk premium) communication shocks have
lifetime of bonds, which equal expected future risk-free a negligible effect on sovereign yields. In crisis times, they
rates and risk premia. Therefore, communication shocks have a negative impact on all sovereign yields, which is
about the future path of monetary policy can affect bond larger in absolute value for peripheral yields.
yields via two channels. While these predictions are intuitive, it is important to
On the one hand, forward guidance shocks provide in- show that they are consistent with a rational framework.
formation about future short rates, so a negative IR shock For this purpose, we build a reduced-form model of the
decreases all bond yields, and this effect is uniform across impact of central bank communication on asset prices in
all countries because they share the same short rate pro- the Online Appendix. In what follows, we perform empiri-
cess. On the other hand, innovations to the future path of cal tests suggested by Hypotheses 1 and 2.
interest rates also affect the perceived probability of the
credit event: an announcement that policy rates will be 3. Data
low for longer can increase the probability of default by
raising the market value of current liabilities and making it Interest Rates Swaps. From Reuters Datascope we col-
less profitable for bondholders to roll over sovereign debt, lect real-time quotes of overnight index swap rates with
and it can also be interpreted as a signal of weaker fu- maturities ranging between 1 and 12 months and swap
ture fundamentals (e.g., output or unemployment). These rates, written on the 6-month Euribor, with maturities
mechanisms increase the risk premia on credit-risky assets ranging between 2 and 10 years. Equity. Additionally, from
such as sovereign bonds. Reuters Datascope, we obtain high-frequency data on Eu-
Because the expectation channel is identical for all roStoxx 50 futures. We use futures data instead of the
countries and the risk premium channel counteracts it, cash index since futures markets are far more liquid than
core yields are overall more responsive to interest rate cash markets. Futures returns are computed on the most
shocks than peripheral yields. Intuitively, German and liquid (highest volume) contract, which is normally the
other core bonds, even in turbulent times, tend to fea- front month, or, in expiration months, the next to deliv-
ture small risk premia and thus interest rate shocks have ery. Sovereign bond yields. We use daily zero-coupon bond
an overall positive impact on their yields. In contrast, in yields of Germany, France, Italy, and Spain, with maturi-
stressful periods the risk premium channel on peripheral ties ranging between three months and ten years, available
bonds can be strong enough to dominate the expectation from Bloomberg. We focus on these four countries as both
channel and lead to negligible or even negative overall IR bond and CDS data coverage for these countries is reliable,
multipliers. and together they account for about 76% of the total GDP
A second result is that negative news about ECB poli- of the eurozone. We also use high-frequency bond yields of
cies, U < 0, increase the perceived probability of the credit the same set of countries available from Reuters Datascope.
event and hence the required risk premia; this raises Credit risk. To measure the credit risk of each country, we
sovereign yields, especially for peripheral countries. Since use US dollar-denominated credit default swaps sourced
these additional policy shocks have no expectation effect from Markit. News. For aggregate macroeconomic news
via influencing future short rates, we refer to them as pure about the eurozone, we rely on now-casts of current euro-
risk premium shocks in the rest of the paper. area GDP. Now-casts are based on a dynamic factor model
Notice that the described risk premium channel cru- (see, e.g., Giannone et al., 2008) to predict current and next
cially depends on the sensitivity of market participants’ quarter GDP growth and use a large and heterogeneous set
perceived probability to monetary policy shocks. While of predictors, including both “hard” and “soft” data, rang-
we take these parameters of the model as given, their ing from unemployment statistics to consumer surveys. We
5
JID: FINEC
use changes in the now-casting predictions between two Fig. 2 illustrates the timeline of events on days of the
ECB meetings to proxy for all relevant economic news re- meetings of the ECB Governing Council. At 13:45 CET, the
leased within this period. Announcement dates. Our main ECB publishes a press release announcing its policy rate
sample period runs from 1 January, 2001 to 31 Decem- decision, that is, the minimum bid rate for the main re-
ber, 2014. Since January 2015, the press release of the ECB financing operations of the Eurosystem. Then at 14:30 CET,
Governing Council policy decision refers to current and fu- the ECB president and vice president hold a press confer-
ture unconventional policy measures too; see the details in ence, during which they discuss the future path of mon-
Section 4. In addition, January 2015 also marks the begin- etary policy (forward guidance on interest rates) and the
ning of the ECB publishing its monetary policy delibera- state of the eurozone economy. As our focus is on the ef-
tions. Thus, our main period of interest ends in December fect of ECB communication on asset prices, to allow suffi-
2014 to keep our identification clean. We discuss the im- cient time for the market to reflect on rate decisions and
pact of the introduction of the Asset Purchase Programme information, we define our communication window start-
in January 2015 in Section 7. During the 2001–2015 pe- ing at 14:25 and ending at 15:30 CET, 40 minutes after the
riod, there is approximately one ECB meeting per month, press conference finishes.
except for in years 2001 and 2008, with 22 and 13 meet- The press conference begins with an introductory state-
ings, respectively. From the 179 announcement days, we ment, whose structure has remained the same since the
exclude 18 that were either not followed by a press con- inception of the ECB: it contains (i) a summary of the
ference or were unscheduled; these are summarized in the ECB’s monetary policy decision and balance of risks to
Online Appendix. Our final sample thus consists of 161 an- price stability and, since July 2013, an open-ended for-
nouncement days: 18 days when the main refinancing rate ward guidance; (ii) a discussion of both real and mone-
was cut, 11 days when the rate was raised, and 132 meet- tary developments in the euro area; and (iii) a conclu-
ings with no change. ECB president speeches. We combine sion with some considerations on fiscal policy and struc-
data on ECB president speeches from Bloomberg calendar, tural reforms. The press conference then continues with a
Bloomberg news, and the ECB website for the 2001 to 2015 question-and-answer session. Central bank communication
period. The Bloomberg economic calendar lists all speeches therefore not only reveals information about future inter-
performed by the ECB president together with the date est rates but also about the state of the economy. In the
of the speech. We then match the list of speeches pro- following, we draw on the joint dynamics of default-free
vided by Bloomberg with information from the ECB web- interest rates and equity during the 1-hour-and-45-minute
site, which provides the transcript for a set of speeches. For press conference window to capture the multidimensional
the purpose of our paper, we only use speeches that were nature of communication, as described by our theoretical
covered both by the Bloomberg calendar and by the ECB framework.
website. We filter out a small number of speeches such as We form a single composite forward guidance shock
award ceremonies, openings of museums, and book fairs from swap rates. Specifically, we measure changes in swap
that were clearly not discussing monetary policy-related rates with maturities ranging between one month and ten
issues. Finally, using the Bloomberg news database, we col- years over the press conference window and then esti-
lect the time stamp for the first news of the day that is re- mate latent factors via principal component analysis on
lated to the speech, focusing only on speeches held during the covariance matrix of the 161 (number of announce-
typical market trading hours, that is, between 09:00 and ments) × 21 (maturities) matrix of rate changes. We find
18:00 Central European Time (CET). This leaves us with 219 that the first principal component explains more than 86%,
ECB president speeches. and the first two principal components together explain
more than 93% of the total variation. To assess the eco-
4. ECB governing council meetings and policy shocks nomic significance of these factors, we regress zero-coupon
rate changes, bootstrapped from swap rate changes, on
A large empirical literature extracts monetary policy the first and second principal components. Our regressions
shocks from money market rates. We follow the approach reveal that almost all of the variation in bond yields is cap-
of Brand et al. (2010) based on high-frequency identifica- tured by the first principal component and the second fac-
tion, which exploits the fact that the ECB conducts the tar- tor has very little impact on yield changes during the com-
get rate announcement and the press conference at differ- munication window. Thus, we take PC1 as our proxy for the
ent points in time. This allows a simple yet clean separa- default-free interest rate communication shock, denoted by
tion of monetary policy action vis-à-vis communication. IR.
Fig. 2. Monetary policy decision window This figure illustrates the timeline of ECB monetary policy announcements. All times are in Central European
Time (CET).
6
JID: FINEC
Fig. 3. Time series of communication shocks This figure plots communication shocks extracted from interest rates and equity reactions in a tight window
around ECB press conferences. Data run from January 2001 to December 2014. Dashed blue lines indicate rate hikes, and bold red lines indicate rate cuts.
(For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
The equity response, EQ, is simply computed as the Table 1

Summary statistics of monetary policy communication shocks
log return of the most liquid EuroStoxx 50 futures con-
This table presents summary statistics for interest rate communication
tract during the same window used to estimate the for- shocks (IR) and pure risk premium shocks (U) in basis points (bps). IR
ward guidance shock. To disentangle the effect of shocks to is the first principal component from a principal component analysis ap-
risk premia that is independent of default-free interest rate plied to swap rate changes during the communication window with ma-
shocks, we then estimate an orthogonal component via or- turities ranging between one month and ten years. U is the residual when
regressing EuroStoxx 50 futures returns during the communication win-
dinary least squares (OLS):4
dow on IR. The communication window spans the ECB press conference
between 14:25 and 16:10 CET on ECB announcement days. The full sam-
EQt = a + b IRt + εt . (1) ple runs from January 2001 to December 2014 (161 announcements), pre-
crisis runs from January 2001 to November 2009 (100 announcements),
In our analysis, we orthogonalize equity shocks with re- and crisis runs from December 2009 to December 2014 (61 announce-
spect to the interest rate shock using the full sample pe- ments).
riod; however, our results remain the same if we orthogo- Mean Std Min Max Skew Kurt AR(1)
nalize with respect to the different periods. Thus, we ob- Full sample
tain pure risk premium shocks by
IR 0.20 3.19 13.43 14.34 0.11 8.01 0.24
U 0.00 72.73 247.32 180.24 0.31 4.20 0.09
Ut ≡ EQt − aˆ − bˆ IRt , (2)
Precrisis
where aˆ and bˆ are the OLS point estimates from (1). IR 0.19 3.29 13.43 14.34 0.03 7.78 0.26
Fig. 3 plots the time series of our estimated commu- U 2.76 64.71 187.37 173.79 0.04 4.15 0.15
nication shocks, and Table 1 presents summary statistics
Crisis
for the full sample and the two subsamples. For the full
sample, the interest rate shock is slightly negative at 0.20 IR 0.20 3.03 11.77 10.79 0.41 8.35 0.20
U 4.53 84.64 247.32 180.24 0.42 3.70 0.02
4
In the Online Appendix we present estimated coefficients for (1) and
a similar, multivariate specification that includes the first five principal basis points (bps), on average (U shocks are zero mean by
components of swap rates for three sample periods (full sample, precri- construction), and the volatility of risk premium shocks is
sis, and crisis). Our estimates show that for all sample periods, there is around 23 times larger than for interest rate shocks. Com-
a low correlation between equity returns and IR, and the maximum R2 is paring pre- and post-December 2009 summary statistics,
12%. Interestingly, unlike Bernanke and Kuttner (2005) for Federal Open
Market Committee (FOMC) meetings, we find that IR shocks have, on av-
we find that many characteristics are stable across subsam-
erage, a positive impact on equity returns around ECB press conferences ples. However, the risk premium shocks become more neg-
that increases over time, although all estimates are insignificant. ative as well as more volatile over time.
7
JID: FINEC
Fig. 4. Intraday asset price reaction to ECB communication This figure displays the response of two-year swap rates and the EuroStoxx index during the
2 August 2012, ECB press conference. The dashed lines mark the start of the target rate announcement (13:45 CET), and the start (14:30 CET), and end
(15:30 CET) of the press conference, respectively.
To motivate our approach, we discuss the events and 5. Central bank communication and sovereign yields
corresponding shocks on two particular days of our sam-
ple. On 4 August, 2011, the ECB Governing Council de- 5.1. Core versus peripheral yields
cided to keep interest rates on hold after a previous hike in
July, causing market participants to revise down their be- We regress daily changes of core and peripheral bond
liefs about the future path of the policy rate. This resulted yields on IR and U shocks for the pre-sovereign debt crisis
in a drop in interest rates, corresponding to a -11bp IR (January 2001 to November 20 09, 10 0 observations) and
shock, an approximately 3.5 standard deviation surprise— sovereign debt crisis (December 2009 to December 2014,
the largest dovish shock in the crisis period. 61 observations) periods separately.6
Fig. 4 shows the reaction of bond and stock markets Yields are defined as the arithmetic average of German
during the ECB press conference of 2 August, 2012, exactly and French yields, and peripheral yields are defined as the
one week after ECB President Draghi’s famous “whatever arithmetic average of Italian and Spanish yields; we report
it takes” speech. During the meeting, the ECB Governing individual country regressions in the Online Appendix. For-
Council decided that “it may undertake outright open mar- mally, as suggested by our theoretical framework, we run
ket operations of a size adequate to reach its objective.” As
a result, the spread between peripheral and core ten-year yτi,t = aτi + bτi IRt + ciτ Ut + i,t
τ , (3)
yields experienced the largest one-day increase on any day
between 2009 and 2015 (53bps), because after the speech where yτi,t are daily zero-coupon yield changes for i =
on 26 July, 2012, the market was expecting nothing short c, p (core and periphery), with maturities τ = 3, . . . , 120
of an announcement of QE.5 Fig. 4 shows that while the months, and we compare the obtained core and peripheral
two-year swap rate did not change significantly, EuroStoxx coefficients.
futures dropped by 2.66% during the first half of the press Fig. 5 visualizes our results. The left panels plot the ef-
conference. We measure the pure risk premium shock of fect of interest rate (upper left panel) and risk premium
this conference at 247bps, which corresponds to a three shocks (lower left panel) before December 2009. We find
standard deviation surprise—the largest negative U shock that before the European sovereign debt crisis, coefficients
in the sample. for the interest rate shock are statistically different from
Our proposed economic channel links the information zero for all maturities, and estimated coefficients for core
embedded in central bank communication to these swap, and peripheral countries are virtually the same, indicating
equity, and sovereign yield changes. In the following, we that monetary policy did not have a differential effect. For
study their relation more formally and use the above two example, for any negative 11bp forward guidance shock,
numerical examples, IR = -11bps and U = -247bps, to illus- there is an 18bp decrease in two-year bond yields and an
trate the economic significance of our results.
6
A formal analysis, following Bai and Perron (1998, 2003), identifies
three break points during the 2001–2018 period. The first is in December
5
The press headline that day read “ECB disappoints. The council is 2009, which was the first ECB meeting where Greek default was men-
clearly not in agreement on what can or will be deployed, and there tioned. The second occurs mid-2012, in the run-up to the “whatever it
are clearly a number of council members who are making further ECB takes” speech of ECB President Mario Draghi. The third break, in Decem-
action contingent on governments delivering on their side of the equa- ber 2014, marks the end of our “crisis” sample, after which the ECB (i)
tion and therefore whatever the ECB does will not be QE.” When asked introduced the Public Sector Purchase Program (PSPP) and (ii) changed its
during the question-and-answer session, President Draghi stated that the communication strategy by releasing some information about unconven-
move “was approved unanimously today with one exception and it was tional policies together with the monetary policy decision at 13:45 CET.
not me.” Bundesbank Chief Jens Weidmann allegedly voicing his reserva- Treating the 2009–2012 and 2012–2014 periods separately does not have
tions about bond buying caused uncertainty about future ECB monetary a qualitative impact on our results. Exhaustive estimation details are gath-
policy. ered in the Online Appendix.
8
JID: FINEC
IR loading: precrisis IR loading: crisis

2 2
Core
Periphery 1.5
Estimated coefficient
1.5
1
0.5
1
0
-0.5
0.5
-1
0 -1.5
3 6 12 24 36 48 60 72 84 96 108 120 3 6 12 24 36 48 60 72 84 96 108 120
Maturity (months) Maturity (months)
U loading: precrisis U loading: crisis

2 5
1 0
0 -5
-1 -10
-2 -15
3 6 12 24 36 48 60 72 84 96 108 120 3 6 12 24 36 48 60 72 84 96 108 120
Fig. 5. Core and peripheral yield responses before and during the crisis This figure plots the response of core and peripheral yields at different maturities
for IR and U shocks around ECB press conferences: yτi,t = aτi + bτi IRt + ciτ Ut + i,t
τ , τ = 3, . . . , 120 months. Data run from January 2001 to November 2009
on the left panels and from December 2009 to December 2014 on the right panels. Bands display 95% confidence intervals computed using HAC standard
errors with two lags.
8bp drop in ten-year yields for both core and peripheral core estimates but larger than the crisis peripheral effect.
countries. Pure risk premium shocks, on the other hand, Hanson and Stein (2015) report similar economic magni-
do not have a significant effect on bond yield changes as tudes for forward guidance shocks and their effects on real
estimated coefficients are insignificant at all maturities. yields.
The right panels present results from the crisis sub- Estimated coefficients on risk premium shocks for core
sample, the main focus of our paper. Interestingly, inter- countries are insignificant at all maturities except at the
est rate shocks have a differential effect on core versus shortest maturity. For peripheral yields, however, we find
peripheral countries in this period: for core countries we highly statistically significant estimates, which increase (in
find virtually the same hump-shaped pattern as in the first absolute value) with the maturity. To evaluate the effect
part of the sample, but peripheral countries are affected of risk premium shocks on peripheral yields, we refer to
much less; in fact, estimated coefficients beyond the one- the event shown in Fig. 4: a negative 247bp pure risk
year maturity are indistinguishable from zero. In particular, premium shock increases two-year peripheral yields by
we find that for any dovish 11bp surprise, two-year core 247 × 7.50/100 = 19bps and ten-year peripheral yields by
yields drop by 17bps, whereas the effect on a two-year pe- 247 × 9.17/100 = 23bps.
ripheral yield is a 2bp increase and statistically insignif- To highlight the effect of the shocks on the yield spread,
icant. This pattern extends to longer maturities: for ten- defined as the difference between peripheral and core
year bonds, the corresponding numbers are a 10bp core yields, we turn to Table 2, which presents estimated coef-
drop and a 4bp peripheral increase. We can compare these ficients for core and peripheral countries during the crisis.
numbers to those shown in the literature for US Treasury We find that interest rate shocks have a statistically sig-
bonds. For example, Nakamura and Steinsson (2018) find nificant effect on the spread for maturities ranging from
that any 100bp increase in their policy shock (which the two to ten years, with the largest effect for the interme-
authors interpret as a forward guidance shock) increases diate maturities around two years. For U shocks, we find
ten-year Treasury yields by 38bps. Since their largest (in that the estimated coefficients are again significant for ma-
absolute values) shock is a 13bp drop, this implies a 5bp turities ranging between two and ten years, and coeffi-
decrease in ten-year yields—close to our precrisis or crisis cients increase (in absolute value) with the maturity. The
9
JID: FINEC
Table 2
Core versus peripheral yield responses during the crisis
This table reports the results of multivariate regressions of zero-coupon one-day changes in core yields versus peripheral yields of
different maturities (months) on IR and U communication shocks:
yτi,t = aτi + bτi IRt + ciτ Ut + i,t

τ , τ = 3, . . . , 120 months.
Core yields are defined as the average of Germany and France and peripheral yields defined as the average of Italy and Spain. t-
statistics reported in parentheses are calculated using HAC standard errors. R2 is the change in the adjusted R2 when adding U
shocks to a univariate regression that uses only the IR shocks. Data run from December 2009 to December 2014.
3 6 12 24 36 48 60 72 84 96 108 120
Core
IR 0.64 0.99 1.23 1.52 1.52 1.47 1.42 1.31 1.19 1.08 0.99 0.94
(5.90) (6.14) (7.09) (7.55) (8.61) (8.18) (7.66) (7.39) (7.16) (6.75) (6.43) (6.18)
U (×10−2 ) 0.87 0.34 0.33 0.18 0.53 0.53 0.67 0.88 1.03 1.22 1.33 1.27
(2.04) (0.89) (1.04) (0.35) (0.87) (0.71) (0.78) (1.05) (1.24) (1.50) (1.63) (1.58)
2
R 17.40 27.16 63.07 60.08 59.00 54.40 48.09 46.95 44.63 41.91 38.74 36.30
R2 0.09 2.18 0.87 1.26 0.82 0.97 0.88 0.10 0.76 2.17 3.23 2.93
Periphery
IR 0.60 0.66 0.74 −0.21 −0.25 −0.29 −0.31 −0.31 −0.35 −0.33 −0.34 −0.34
(2.09) (1.54) (1.83) (0.50) (0.59) (0.75) (0.82) (0.87) (1.06) (1.04) (1.08) (1.08)
U (×10−2 ) 0.66 0.83 2.45 7.50 8.77 9.04 9.27 8.90 9.53 9.15 9.43 9.17
(0.88) (0.67) (1.74) (3.42) (4.08) (4.06) (4.08) (4.02) (3.65) (3.54) (3.35) (3.38)
2
R 15.93 12.48 5.60 16.43 20.92 23.65 25.14 25.49 29.25 28.91 30.22 29.79
R2 1.59 1.79 0.31 16.41 20.91 23.61 25.09 25.41 28.85 28.30 29.41 28.90
Periphery-core spread
IR 0.04 0.33 0.49 1.74 1.77 1.77 1.73 1.62 1.55 1.41 1.34 1.27
(0.18) (0.64) (1.20) (4.27) (4.46) (4.36) (4.14) (3.88) (3.88) (3.73) (3.55) (3.47)
U (×10−2 ) 0.21 0.49 2.78 7.68 9.29 9.57 9.95 9.78 10.55 10.37 10.76 10.44
(0.32) (0.41) (2.09) (3.85) (4.64) (4.29) (4.08) (4.02) (3.54) (3.47) (3.28) (3.30)
2
R 0.85 0.59 2.18 24.67 29.21 31.74 33.11 34.12 37.30 37.50 38.18 37.44
R2 3.23 3.09 0.84 15.12 20.59 22.90 24.82 26.30 30.31 31.28 32.81 32.29
last line of the table also reports the change in adjusted tors, for the omitted macro news variable.7 We find that
R2 when adding the risk premium shock to the regres- controlling for news does not affect our main result: re-
sion. We notice that the latter contributes the majority to gression coefficients are virtually the same as in Table 2,
the variation in bond yield changes, with its incremen- and the significance levels and regression R2 s are hardly
tal R2 s ranging between 1% and 32% for maturities above affected. This suggests that in the case of the eurozone,
one year. Repeating the same calculation as above, we find central bank communication still provides information rel-
that a -11bp IR shock increases the two-year (ten-year) evant for sovereign bond pricing beyond publicly available
yield spread by 19bps (14bps), whereas a 247bp U shock information.
increases the two-year (ten-year) yield spread by 19bps The above results indicate a regime change in terms of
(26bps). Therefore, while forward guidance and risk pre- central bank communication from the precrisis to crisis pe-
mium shocks have approximately the same effect on two- riod that led to significantly different patterns in sovereign
year yields, the latter have a twice as large impact on very yields’ reaction to monetary policy shocks. In particular,
long-term sovereign bond yields. This finding is also re- the precrisis regression coefficients are consistent with an
lated to Altavilla et al. (2019), who find that monetary pol- economy in which either there are no major differences
icy shocks extracted from default-free interest rates alone between core and peripheral countries’ economies or
have a small impact on long-term bonds during the crisis monetary policy communication does not contain signifi-
period—we show that in this period most of the variation cant new information about the state of the economy and
is risk premium related. hence credit risk. On the other hand, our results suggest
In a recent paper, Bauer and Swanson (2020) argue that that during the crisis, investors paid special attention to
because the Federal Reserve and market participants pay the health of the sovereign economies, with a particularly
attention to the same news, macro news are an omit-
ted variable in regressions similar to (3), and including
them drives out monetary policy shocks, questioning the 7
In the Online Appendix, we also present regression results from
“Fed information effect” shown in Nakamura and Steins- changes in expected output of core and peripheral countries on our mon-
son (2018). To address the concern that similar mecha- etary policy shocks and the news shock, similar to Bauer and Swan-
nisms are at work in the euro area too, in the Online Ap- son (2020). We find that estimated coefficients for U and IR shocks are
pendix we revisit the findings of Table 2 but also include positive for both countries. However, the coefficients are not precisely es-
timated: U shock coefficients, although larger for peripheral countries, are
changes in now-casts as a proxy, available in real time and statistically significant only for core countries, whereas IR are statistically
computed from a large panel of macroeconomic indica- significant only for peripheral countries.
10
JID: FINEC
IR: rolling coefficient and R2 IR: rolling coefficient and R2

40 1.2 35 1.2
35 R2 core R2 periphery
1.1 30 1
Coefficient core Coefficient periphery
30
1 25 0.8
25
0.9 20 0.6
R2
R2
20
0.8 15 0.4
15
0.7 10 0.2
10
5 0.6 5 0
0 0.5 0 -0.2
06 07 08 09 10 11 12 13 14 15 06 07 08 09 10 11 12 13 14 15
Date Date
U: rolling coefficient and R2 U: rolling coefficient and R2

10 2 40 5
9 R2 core 1.5 35 R2 periphery
8 Coefficient core 1 Coefficient periphery

30 0
7
0.5
25
6
0
R2
R2
5 20 -5
-0.5
4
15
-1
3
10 -10
-1.5
2
-2 5
1
0 -2.5 0 -15
06 07 08 09 10 11 12 13 14 15 06 07 08 09 10 11 12 13 14 15
Date Date
2
Fig. 6. Rolling regression estimates The upper panel plots the rolling betas and the rolling adjusted R s from regressions of core (left) and peripheral (right)
ten-year bond yields on the IR communication shocks in univariate regressions. The lower panel plots the rolling betas and the rolling adjusted R2 s from
regressions of core (left) and peripheral (right) ten-year bond yields on the U communication shocks in univariate regressions. The window size for the
rolling regression is set to 50 months.
sharp disconnect between core (e.g., Germany and France) signal a lower probability of the introduction of policies
and peripheral economies (e.g., Italy and Spain). It is also investors deem necessary, drive up the perceived probabil-
reasonable to assume that during this period, in case of a ity of a credit event, and yield spreads rise further and get
peripheral default or an eurozone breakup, bonds issued disconnected from fundamentals.
by peripheral countries would have been more exposed
to credit losses, potential redenomination, and liquidity 5.2. Communication effects in the time series
risks, that is, less valuable than bonds issued by core
countries. Thus, these results confirm the predictions of To get a better understanding of the time-series be-
Hypotheses 1–2. havior of the regime change noted earlier, Fig. 6 depicts
The regime change around the December 2009 ECB estimated coefficients and R2 s from rolling-window re-
meeting, the first one during which Greek default was gressions of ten-year bond yield changes of core and
mentioned, suggests that the failure of forward guidance peripheral countries on interest rate (upper panels) and
to impact peripheral yields and the dominance of U shocks risk premium shocks (lower panels). We find that the
might not be exclusively due to worsening fundamen- effect of IR shocks on core countries’ yields remains re-
tals. In fact, this dramatic change is consistent with the markably stable throughout the whole period as estimated
recent literature that links the European debt crisis to coefficients wiggle around 0.8. The effect on peripheral
self-fulfilling beliefs and multiple equilibria (see Corsetti yields, however, starts to weaken in 2011 and becomes
and Dedola, 2016; Bocola and Dovis, 2019; Lorenzoni and zero and insignificant in 2013. The effect of U shocks is
Werning, 2019; and Bacchetta et al., 2020, among oth- virtually the same for core and peripheral countries until
ers). In our framework, the precrises results correspond 2012, when the two start to diverge. While the effect on
to a “good” equilibrium in which all sovereign bonds re- core countries continues to be insignificant, the effect
act to forward guidance shocks and feature small risk pre- on peripheral yields strengthens as estimated coefficients
mia. The post-December 2009 results, on the other hand, become negative. A similar pattern emerges for the uni-
correspond to a “bad” equilibrium in which peripheral variate R2 s: for core yields, interest rate shocks explain,
sovereign yields stop reacting to conventional monetary on average, around 30% of the variation. For peripheral
policy and negative state-of-the-economy shocks, which countries, the R2 drops significantly in 2011 and converges
11
JID: FINEC
munication represented around a quarter of the total yield

spread.
5.4. Credit risk
Next, to study whether monetary policy communication

drives the yield spread through a credit risk channel, we
run the regressions
CDSi,t = ai + bi IRt + ci Ut + i,t ,

where CDSi,t is the change in the five-year CDS rate of
country i. Table 3 contains the results for the four individ-
ual countries, core and peripheral CDSs, and their spread.
We find that estimated coefficients for IR and U shocks
Fig. 7. Cumulative effect of communication This figure plots the cumu-
are significant and negative. In particular, a hypothetical
lative effect of IR and U communication shocks on the spread between negative 11bp IR shock increases the five-year peripheral-
ten-year peripheral and core bond yields. The cumulative effect is com- core CDS spread by 12bps, whereas a hypothetical nega-
puted from multivariate regression loadings estimated using a window tive 247bp U shock increases the difference in CDS rates
size set to 50 months, as in Fig. 6. The loadings are then multiplied by
by 23bps. Given that, on average, IR and U shocks are neg-
date t shocks and the overall effect computed by summing the fitted val-
ues over time. ative after December 2009, this implies that both shocks
significantly increase the credit risk premium spread be-
tween peripheral and core countries, and the majority of
to zero. Risk premium shocks, on the other hand, display this difference is driven by the U shock itself.8
exactly the opposite behavior: while the R2 is close to zero Our empirical results have two implications regarding
until the crisis for both core and peripheral yields, the the model described in Section 2. First, monetary policy
effect on the latter increases during the crisis, reaching communication seems not only to be an important driver
an R2 of 35% at the end of our sample. These results of investors’ beliefs about future interest rates but also
again suggest a radical change in how peripheral coun- about perceived credit risk, which supports our modeling
tries were perceived by market participants, even on ECB assumptions and interpretations. Second, the U shocks that
days. we back out from equity changes are essentially the main
drivers of credit risk premia. Since they have no effect
on expectations by construction, they can indeed be in-
5.3. Economic significance and the yield spread terpreted as sovereign credit risk premium shocks of ECB
communication.
Since the onset of the crisis in 2009, the ECB has
tried to ease distress in financial markets and to reduce 5.5. Are ECB days special?
sovereign spreads by (i) drastically lowering its target rate,
(ii) providing unprecedented amounts of liquidity support Our main results presented in Table 2 are based ex-
against a broader set of assets used as collateral, (iii) intro- clusively on days when the ECB makes its monetary pol-
ducing a series of unconventional measures such as its Se- icy announcement. It is natural to ask whether the rela-
curities Markets Programme and Outright Monetary Trans- tion between sovereign bond yields and shocks extracted
actions, and, since January 2015, (iv) introducing QE in the from risk-free interest rates and equity is different on ECB
form of its permanent Asset Purchase Programme. Our re- days relative to all other days. To study this question in
sults so far suggest that conventional monetary policy in more detail, we construct interest rate shocks by repeat-
the form of central bank communication is also a driver of ing the principal component analysis of risk-free interest
the yield spread. rates in high frequency during the communication window
To evaluate the realized effect and overall economic on all days from 2010 to 2015. Similarly, we construct risk
magnitude of this channel, we calculate the size and di- premium shocks from equity returns sampled in the same
rection of the spread implied by monetary policy shocks period on all days. Using these two shocks, we rerun our
and compare it to the time series of the yield spread. main regression augmented by a dummy, 1ECB,t , that takes
We compute the implied spread by multiplying realized
shocks with the difference in real-time policy loadings dis-
8
As an additional test for credit risk channel, we can look at corporate
played in Fig. 6 and add them up over time. The result-
credit spreads directly. To this end, we collect Markit iBoxx EUR price in-
ing spread is depicted in Fig. 7. Strikingly, we find that IR dices from Bloomberg and obtain the following estimates for the crisis
and U shocks had a consistently positive effect on the yield period:
spread starting at the onset of the crisis in 2010. Indeed,
(log BBBt − log AAAt ) = a + 3.42 IRt + 9.13 Ut + t , R̄2 = 40.17%.
the cumulative sum increases up to 50bps in late 2013 and (6.00 ) (4.58 )
has since then been declining. Economically, this effect is Since the left-hand side variables are price indices and not yields, more
large: at the end of 2013, the ten-year core-periphery yield negative shocks increase the corporate yield spread, in line with the re-
spread was 213bps, so at its peak, the spread due to com- sults of Table 3.
12
JID: FINEC
Table 3
Credit risk reaction during the crisis
This table reports estimated coefficients from the regression of changes in the five-year CDS rates on IR and U communication shocks:
CDSi,t = ai + bi IRt + ciUt + i,t ,

where CDSi,t is the change in the five-year CDS rate for country i. t-statistics reported in parentheses are calculated using HAC standard errors with two
lags. R2 is the change in the adjusted R2 when adding U shocks to a univariate regression on IR shocks. Data run from December 2009 to December
2014.
Germany France Italy Spain Core Periphery P-C
IR −0.28 −0.50 −1.55 −1.48 −0.39 −1.52 −1.12

(−2.32) (−3.85) (−4.42) (−3.69) (−3.73) (−4.16) (−3.61)
U (×10−2 ) −1.11 −2.59 −10.72 −11.23 −1.85 −10.97 −9.12
(−2.63) (−3.19) (−4.20) (−3.86) (−3.16) (−4.04) (−3.90)
2
R 16.06 28.36 36.01 36.48 25.91 36.87 36.29
R2 8.21 20.51 28.16 28.62 18.05 29.01 28.43
Table 4
Sovereign yield spreads on ECB versus non-ECB days
This table reports the results of multivariate regressions of zero-coupon one-day changes in peripheral minus core yields of different maturities (months)
on IR and U communication shocks as well as an ECB dummy variable that takes the value of one on days that the ECB announces its monetary policy and
zero otherwise, and an interaction term with each communication shock:

yτp,t − yτc,t = aτ1 + aτ2 × 1ECB,t + bτ1 IRt + c1τ Ut + bτ2 IRt × 1ECB,t + c2τ Ut × 1ECB,t + tτ , τ = 3, . . . , 120 months.
t-statistics reported in parentheses are calculated using HAC standard errors with two lags. Data run from December 2009 to December 2014.
3 6 12 24 36 48 60 72 84 96 108 120
IR −0.23 −0.49 −0.82 −1.27 −1.38 −1.41 −1.46 −1.47 −1.39 −1.32 −1.28 −1.23
(−0.70) (−1.48) (−2.45) (−2.87) (−3.21) (−3.36) (−3.63) (−3.86) (−3.67) (−3.62) (−3.61) (−3.56)
U (×10−2 ) −0.85 −1.13 −1.17 −3.65 −3.60 −3.53 −3.59 −3.34 −3.21 −3.15 −3.15 −3.15
(−1.28) (−1.41) (−1.52) (−3.93) (−4.07) (−4.31) (−4.67) (−4.54) (−4.56) (−4.77) (−4.85) (−4.99)
1ECB −1.43 −1.03 −0.99 −1.40 −1.53 −1.47 −1.66 −1.63 −1.34 −1.21 −1.22 −1.13
(−1.50) (−1.04) (−0.75) (−0.96) (−0.97) (−0.95) (−1.07) (−1.11) (−0.94) (−0.88) (−0.89) (−0.82)
IR × 1ECB 0.10 0.01 −0.11 −1.54 −1.77 −1.81 −1.82 −1.66 −1.82 −1.73 −1.76 −1.69
(0.24) (0.02) (−0.16) (−1.77) (−2.28) (−2.50) (−2.57) (−2.44) (−2.65) (−2.58) (−2.52) (−2.51)
U × 1ECB (×10−2 ) 0.64 0.68 −1.56 −3.85 −5.51 −5.86 −6.18 −6.27 −7.18 −7.07 −7.47 −7.16
(0.65) (0.43) (−0.77) (−1.28) (−2.07) (−2.40) (−2.62) (−2.76) (−3.04) (−3.04) (−2.97) (−2.96)
2
R 0.01 0.41 1.20 5.39 6.50 7.13 7.94 8.27 8.97 9.07 9.40 9.31
the value of one on days when the ECB makes its monetary Table 5
Credit risk on ECB versus non-ECB days
policy announcement and zero otherwise:
This table reports the results of multivariate regressions of changes in the

yτp,t − yτc,t = aτ1 + aτ2 1ECB,t + bτ1 IRt + c1τ Ut + bτ2 IRt five-year CDS rates on IR and U communication shocks as well as an ECB
dummy variable that takes the value of one on days that the ECB an-
× 1ECB,t + c2τ Ut × 1ECB,t + tτ . nounces its monetary policy and zero otherwise, and an interaction term
with each communication shock:
We present the results in Table 4 for sovereign yields and CDSi,t = a1,i + a2,i × 1ECB,t + b1,i IRt + c1,i Ut + b2,i IRt × 1ECB,t
in Table 5 for CDSs.
+c2,i Ut × 1ECB,t + i,t .
Except for the short end, changes in sovereign yields
are not significantly different on ECB days than other days t-statistics reported in parentheses are calculated using HAC standard er-
rors with two lags. Data run from December 2009 to December 2014.
as indicated by the insignificant estimates on the dummy
variable. The estimated coefficients on the IR and U shocks Core Periphery P-C
are negative and highly statistically significant, just as in IR −0.37 −1.25 −0.88
Table 2; the negative relation between yield spreads and (−3.46) (−3.18) (−2.71)
monetary policy shocks is significant in the crisis period, U (×10−2 ) −0.73 −3.48 −2.75
even on days when the ECB does not announce its mone- (−3.43) (−4.00) (−3.90)
1ECB −0.20 −2.13 −1.93
tary policy. However, estimated coefficients on the inter-
(−0.65) (−1.43) (−1.49)
action terms are also significant for long maturities, in- IR × 1ECB −0.28 −1.96 −1.68
dicating that the relation between communication shocks (−1.41) (−2.66) (−2.82)
and yield spreads is “special” on days when the ECB Gov- U × 1ECB (×10−2 ) −1.09 −7.34 −6.26
(−1.51) (−2.78) (−3.01)
erning Council hold their meetings. For a comparison of 2
R 5.96 8.17 7.54
the magnitudes, note that a hypothetical negative 11bp IR
shock increases the ten-year peripheral-core yields spread
by 14bps on normal days and by 32bps on ECB days, and by 25bps on ECB days—a three times larger effect. A
whereas a hypothetical negative 247bp U shock increases similar pattern emerges for CDSs: negative forward guid-
the peripheral-core yields spread by 8bps on normal days ance and risk premium shocks increase the credit risk of
13
JID: FINEC
Table 6
Alternative risk premium shock and yield spreads
This table reports the results of multivariate regressions of zero-coupon one-day changes in peripheral and core yield spreads of different maturities
(months) on IR and C communication shocks:

yτp,t − yτc,t = aτ + bτ IRt + cτ Ct + tτ , τ = 3, . . . , 120 months.
Core yields are defined as the average of Germany and France and peripheral yields defined as the average of Italy and Spain. t-statistics reported in
parentheses are calculated using HAC standard errors with two lags. R2 is the change in the adjusted R2 when adding C shocks to a univariate regression
on IR shocks. Data run from December 2009 to December 2014.
3 6 12 24 36 48 60 72 84 96 108 120
IRt −0.00 −0.36 −0.48 −1.62 −1.54 −1.54 −1.49 −1.40 −1.30 −1.15 −1.07 −1.00
(−0.00) (−0.69) (−1.10) (−4.76) (−4.13) (−3.52) (−3.18) (−3.04) (−2.98) (−2.81) (−2.65) (−2.55)
Ct 0.75 0.51 2.37 3.51 3.90 3.80 3.73 3.51 3.41 3.28 3.25 3.29
(1.99) (1.25) (3.41) (5.12) (4.99) (4.73) (4.50) (4.30) (4.13) (4.13) (3.94) (4.06)
2
R 13.01 3.23 26.19 42.97 45.10 45.42 43.21 40.61 37.74 36.62 34.89 36.21
R2 4.82 −0.77 24.45 32.64 36.70 36.51 34.86 32.78 30.73 30.55 29.64 31.29
peripheral countries relative to core countries, especially and results of these tests to the Online Appendix. First, we
on ECB days. study the effect of other macroeconomic announcements
on our results. Second, we explore the impact of varying
5.6. Alternative risk premium shocks the high-frequency window length to identify our mone-
tary shocks. Third, we use high-frequency changes in bond
In our framework, the nature of risk premium shocks yields instead of daily changes in our sovereign regres-
is informative about the implementation of unconventional sions. Fourth, we reconstruct our monetary policy commu-
policies (or the lack thereof) to the market. Our shocks nication shocks separately in the two relevant subsamples
are calculated from equity returns, but one might ar- and check whether they alter our results. Finally, we es-
gue that other asset prices capture changes in risk bet- timate our sovereign regression using bootstrapped stan-
ter around monetary policy announcements. For example, dard errors to take into account the extra sampling vari-
Rogers et al. (2014) use changes in the spread between ation due to the construction of our shocks. We find that
ten-year Italian and German bond yields to study the reac- our results are virtually unchanged in all the different ro-
tion of exchange rates and equity returns around ECB mon- bustness specifications.
etary policy announcements, and Bekaert et al. (2013) ar-
gue that equity-implied volatility is strongly related to the 6. ECB president speeches
stance of the US Federal Reserve’s monetary policy and in-
vestors’ risk aversion. One natural question is whether our results about press
In the following, we construct a composite measure conferences extend to other forms of central bank com-
from EuroStoxx returns and changes in two implied volatil- munication. This question is also related to a recent lit-
ities and study the impact of this alternative shock on erature that argues that a large fraction of risk premia
bond spreads during the crisis. To this end, we collect data earned on asset prices due to monetary policy occur out-
on the VSTOXX, an implied volatility index from options side of standard announcement days; see, for example,
written on the EuroStoxx, and extract an implied volatility Neuhierl and Weber (2019). The communication event that
measure from the cross-section of options written on the has gained most traction is undoubtedly the “whatever it
EUR/USD exchange rate. Using the first principal compo- takes” speech by ECB President Mario Draghi at an in-
nent of these implied volatilities and our equity returns, vestors’ conference in London on 26 July, 2012. The con-
and orthogonalizing the variable with respect to our IR sensus view in the literature is that the speech marked the
shock as in (2), we calculate an alternative risk premium beginning of the Outright Monetary Transaction program
shock, which we denote by C. intended to lower the high borrowing costs of peripheral
Table 6 summarizes estimated coefficients from re- countries; see, for example, Acharya et al. (2019). The up-
gressing peripheral-core yield spreads on the interest rate per panels of Fig. 8 illustrate the asset price reaction on
shock, IR, and the new risk premium shock, C. We notice that day for the two-year swap rate as well as for the Eu-
that all estimates for C are positive and highly statistically roStoxx index. While we notice an increase in the two-year
significant for long maturities; higher risk premium shocks swap rate, this was dwarfed by the sharp increase in the
lead to higher yield spreads. In terms of the IR coefficients equity index, with a daily return of almost 5%. The lower
and R2 s, we find the numbers to be very similar to those two panels, on the other hand, depict the well-known re-
reported in Table 2. Overall, we conclude that our results sult that during the days that followed the speech, neither
are robust to the set of assets we choose to construct risk German nor French yields moved much, while peripheral
premium shocks from. yields, as well as the spread, decreased significantly.
5.7. Robustness 6.1. Core versus peripheral yields
We perform a host of robustness checks to challenge In the following, we want to understand whether other
our main result; to save space, we defer exhaustive details central bank speeches command similar reactions in asset
14
JID: FINEC
Fig. 8. Intraday asset price reaction to “whatever it takes” on 26 July, 2012 The upper two panels depict the two-year swap rate and the EuroStoxx index
from 11:00 to 17:00 CET on 26 July, 2012. The dashed lines mark the beginning of ECB President Mario Draghi’s speech at the Global Investment Conference
in London. The lower two panels show the level and changes in yield spreads defined as the difference between the ten-year yield on peripheral and core
countries one day before the speech and two days after.
IR shock
5
WIT
Basis points
-5
01 02 03 04 05 06 07 08 09 10 11 12 13 14 15
Date
U shock
Percentage (%)
-2
01 02 03 04 05 06 07 08 09 10 11 12 13 14 15
Date
Fig. 9. Time series of president speech shocks This figure plots communication shocks extracted from interest rates and equity reactions in a tight window
around speeches by the ECB president. Data run from January 2001 to December 2014.
prices or whether 26 July, 2012 marked a special day. To shocks into the analysis. For example, the upper panel in-
this end, we collect data on ECB president speeches out- dicates that the forward guidance shock of 26 July, 2012
side the ECB announcement days as described in Section 3, does not signal a special event at all. In the lower panel,
and we apply the same identification to president speeches however, where we plot U shocks, this day clearly stands
to back out two communication shocks as described in out.
Section 4 for ECB press conferences. Fig. 9 plots the IR and With the president shocks at hand, we study the effect
U shocks that we obtain for president speeches, and it un- of these speeches on sovereign bond yields. To this end,
derscores nicely the importance of including risk premium we run the same regressions as in (3) but using the IR and
15
JID: FINEC
IR loading: precrisis IR loading: crisis

2.5 4
Core
Periphery
2 2
1.5 0
1 -2
0.5 -4
0 -6
-0.5 -8
3 6 12 24 36 48 60 72 84 96 108 120 3 6 12 24 36 48 60 72 84 96 108 120
U loading: precrisis U loading: crisis

3 5
2
0
1
0 -5
-1 -10
-2
-15
-3
-4 -20
3 6 12 24 36 48 60 72 84 96 108 120 3 6 12 24 36 48 60 72 84 96 108 120
Fig. 10. Core and peripheral yield responses to president speeches This figure plots the response of core and peripheral countries’ bond yields at different
maturities for IRt and Ut shocks around ECB president speeches: yτi,t = aτi + bτi IRt + ciτ Ut + i,t
τ , τ = 3, . . . , 120 months. Data run from January 2001 to
November 2009 on the left panels, and from December 2009 to December 2014 on the right panels. Bands display 95% confidence intervals computed
using HAC standard errors with two lags.
U shocks obtained around speeches. The estimated regres- yield spread due to the interest rate shock realization and
sion coefficients are plotted on Fig. 10, and Table 7 contains a 261 × 5.58/100 = 15bp drop due to the risk premium
estimated coefficients for the yield spreads. shock realization. On that day, the spread between periph-
We notice a strikingly similar pattern compared to our eral and core yields decreased by 40bps. Forward guidance
baseline results presented on Fig. 5: interest rate commu- and risk premium shocks thus contributed to more than
nication shocks have a significantly positive hump-shaped half of the overall reduction.
effect on all sovereign yields before December 2009 and on While we measure IR and U shocks on regular mon-
core yields during the crisis period, but the coefficients are etary days and during president speeches the same way,
insignificant for the periphery during the crisis. Moreover, these events can contain different types of information, so
loadings on risk premium shocks are insignificant for all one should expect a different impact on sovereign yields
countries precrisis and for core countries during the crisis too. Comparing our estimates of Tables 2 and 7, we find
but are negative and large in absolute value for peripherals that the two shocks contribute approximately the same to
during the crisis. the overall variation in the ten-year yield spread, at around
The results in Table 7 indicate that while precrisis there 35%. However, while on regular monetary policy days most
was almost no significant reaction of the yield spread to of the explanatory power comes from risk premia shocks,
either IR or U shocks, during the crisis, estimated coeffi- forward guidance shocks contribute a bigger fraction to
cients are significant. To compare the economic size of the the overall R2 during president speeches (10% compared to
estimates, we can again use the largest shocks in the sam- 32%).
ple to study the effects of forward guidance and risk pre- We also study the relation between credit risk and
mia shocks on yields. For both shocks, the largest shocks President speeches in Table 8. The results are somewhat
occurred during the “whatever it takes” speech in July similar to Table 3. We find that when the ECB president
2012: we find an IR shock of 2.63bps and a U shock of gives a speech, an IR shock of 2.63bps leads to a 2.63 ×
261bps (these correspond to a 2 standard deviation and a 2.15 = 6bp drop in the CDS peripheral-core spread. On the
4.69 standard deviation surprise, respectively). As a result, other hand, a hypothetical 261bp pure risk premium de-
we should see a 2.63 × 3.29 = 9bp drop in the ten-year creases the spread by 261 × 5.78/100 = 15bps. Overall, our
16
JID: FINEC
Table 7
President speeches and yield spreads
This table reports the results of multivariate regressions of zero-coupon one-day changes in peripheral and core yield spreads of different maturities
(months) on IR and U communication shocks during ECB president speeches that are not standard monetary policy announcements:

yτp,t − yτc,t = aτ + bτ IRt + cτ Ut + tτ , τ = 3, . . . , 120 months.
Core yields are defined as the average of Germany and France, and peripheral yields are defined as the average of Italy and Spain. t-statistics reported in
parentheses are calculated using HAC standard errors with two lags. R2 is the change in the adjusted R2 when adding U shocks to a univariate regression
on IR shocks. Precrisis runs from January 2001 to November 2009. Crisis runs from December 2009 to December 2014.
3 6 12 24 36 48 60 72 84 96 108 120
Precrisis
IR 0.09 −0.10 −0.37 −0.09 −0.14 −0.08 −0.04 −0.07 −0.07 −0.08 −0.14 −0.10
(0.49) (−0.63) (−1.93) (−1.12) (−1.68) (−1.02) (−0.59) (−1.05) (−1.02) (−0.90) (−1.66) (−0.97)
U (×10−2 ) −0.20 −0.53 0.04 −0.69 −0.74 −0.58 −0.62 −0.44 −0.19 −0.20 −0.18 −0.14
(−0.63) (−1.60) (0.12) (−2.51) (−2.41) (−2.01) (−2.21) (−1.72) (−0.82) (−0.82) (−0.61) (−0.43)
2
R −0.55 2.62 4.21 6.51 5.41 3.60 3.06 2.09 −0.23 0.01 1.53 0.16
R2 −1.12 1.37 −1.32 4.36 3.13 1.61 1.59 0.36 −1.06 −0.95 −1.04 −1.20
Crisis
IR −0.93 −1.55 −1.27 −4.29 −4.33 −4.01 −3.93 −3.79 −3.58 −3.48 −3.33 −3.29
(−0.90) (−1.95) (−1.33) (−2.94) (−2.82) (−2.76) (−3.04) (−3.31) (−3.45) (−3.75) (−3.79) (−3.79)
U (×10−2 ) −6.31 −6.08 −7.76 −11.14 −9.57 −9.20 −8.62 −7.81 −6.80 −6.18 −5.80 −5.58
(−2.69) (−2.08) (−1.75) (−3.21) (−2.78) (−3.09) (−3.17) (−3.14) (−3.22) (−3.06) (−2.85) (−2.88)
2
R 10.40 14.26 21.89 40.05 36.13 38.69 39.61 40.38 39.27 37.36 36.73 35.98
R2 5.49 7.05 13.39 16.67 12.57 14.18 13.80 13.42 12.14 10.72 10.21 9.71
Table 8
President speeches and credit risk
This table reports estimated coefficients from the regression of changes in the five-year CDS rates on IR and U communication shocks sampled during ECB
president speeches:
τ = aτ + bτ IR + cτ U + τ ,
CDSi,t τ = 60, . . . , 120 months,
i i t i t i,t
τ is the change in the two-year (top panel) or five-year (bottom panel) CDS rate for country i. t-statistics reported in parentheses are calculated
where CDSi,t
using HAC standard errors with two lags. R2 is the change in the adjusted R2 when adding U shocks to a univariate regression on IR shocks. Data run
from December 2009 to December 2014.
Germany France Italy Spain Core Periphery P-C
IR −0.37 −0.85 −2.93 −2.58 −0.61 −2.76 −2.15

(−3.58) (−2.87) (−2.64) (−2.60) (−3.33) (−2.80) (−2.58)
U (×10 −2
) −1.26 −1.43 −6.74 −7.51 −1.34 −7.13 −5.78
(−5.65) (−2.58) (−3.64) (−4.54) (−4.07) (−4.27) (−4.04)
2
R 36.01 15.90 26.02 31.00 23.32 30.12 28.92
R2 18.44 −1.67 8.45 13.43 5.75 12.55 11.35
results indicate that central bank communication has a sig- creased, by the end of the sample, the communication ef-
nificant effect on asset prices not just around monetary fects of regular announcement days were completely offset
policy announcements but also during other ECB president by the reduction due to ECB president speeches.
speeches. Further, the lower panel of Fig. 11 adds days when un-
conventional monetary policies were announced. As shown
6.2. Economic significance by Krishnamurthy et al. (2018), the Securities Markets Pro-
gramme announcements led to significant drops in the pe-
In Section 5, we argued that ECB communication on its ripheral bond yields and hence the periphery-core spreads.
regular monetary policy announcement days contributed Once we combine all the announcements days, we find
to an increasing yield spread between core and peripheral that the ECB successfully narrowed the spread between pe-
countries, and this spread emerged during a period when ripheral and core countries. Nevertheless, our results show
a series of unconventional measures were implemented to that ECB communication on regular announcement days
reduce it. Therefore, we also want to study the combined partially offset some of these effects. The temporary in-
effect of ECB regular announcement days with ECB presi- creases in default risk premia and peripheral yields due to
dent speeches and unconventional announcement days. the ECB’s communication on regular monetary policy days,
The upper panel of Fig. 11 extends the cumulative both between 2009 and mid-2010 and throughout 2013,
changes in yields and the periphery-core spread of Fig. 1 to were economically sizable and at the very least increased
include days when the ECB president gives speeches. While sovereign yield volatility and made it harder for the ECB to
we find that between 2010 and 2012 the yield spread in- succeed in bringing down peripheral yields quicker.
17
JID: FINEC
100
WIT
50
0
Basis points
-50
-100
-150 Peripheral (P)

Core (C)
Yield spread (P-C)
-200
01 02 03 04 05 06 07 08 09 10 11 12 13 14 15
Date
100
SMP
SMP
WIT
50
-50
Basis points
-100
-150
-200
-250
Peripheral (P)
-300 Core (C)
Yield spread (P-C)
-350
01 02 03 04 05 06 07 08 09 10 11 12 13 14 15
Date
Fig. 11. Cumulative yield spreads on ECB days, president speeches, and unconventional monetary policy The upper panel displays cumulative one-day
changes in ten-year yields for core (average of Germany and France) and peripheral (average of Italy and Spain) bonds as well as the spread between
peripheral and core bonds on European Central Bank meeting days and days when the ECB president gives speeches. The lower panel adds days when
unconventional monetary policies were announced.
7. Quantitative easing and reconnecting monetary To this end, we extend our analysis to the 2015–
policy 2018 period, and we augment the set of our mon-
etary policy shocks following Swanson (2018) and
Our previous results indicate that during the 2009–2014 Altavilla et al. (2019) to construct a QE-related policy
period, even around the time of the announced unconven- shock. These authors argue that with the introduction
tional monetary policy measures, communication on the of QE, there exists a third dimension to monetary policy
ECB monetary policy meeting days significantly increased communication that is independent of the “standard” tar-
yield spreads, which at the very least made it harder for get and forward guidance shocks. We follow the authors’
the ECB to succeed in bringing down peripheral yields. Fol- procedure, and we extract three principal components
lowing the end of our main sample, in December 2014, from the cross-section of high-frequency changes in the
the ECB announced the decision to launch its permanent communication window and impose the following factor
QE, called the Asset Purchase Programme, and one natural rotation: (i) the second and third (when the third factor
question is whether and how this affected the transmission is present) factors do not load on the one-month OIS
of monetary policy communication to asset prices. and (ii) the rotation is such that the third factor has
18
JID: FINEC
Table 9
Core versus peripheral yield responses post 2014 with QE shocks
This table reports the results of multivariate regressions of zero-coupon one-day changes in core yields versus peripheral yields of different maturities
(months) on IR and U communication shocks as well as QE shocks:
yτi,t = aτi + bτi IRt + ciτ Ut + diτ QEt + i,t

τ , τ = 3, . . . , 120 months.
Core yields are defined as the average of Germany and France, and peripheral yields are defined as the average of Italy and Spain. t-statistics reported in
parentheses are calculated using HAC standard errors with two lags. R2 is the change in the adjusted R2 when adding QE shocks to the regression that
only uses IR and U shocks. Data run from January 2015 to September 2018.
3 6 12 24 36 48 60 72 84 96 108 120
Core
IR 0.35 0.54 0.74 1.38 1.72 2.06 2.31 2.49 2.68 2.81 2.90 2.92
(2.73) (4.28) (4.59) (8.93) (11.90) (13.89) (16.27) (17.23) (18.17) (15.49) (13.70) (12.58)
U (×10−2 ) −1.21 −1.53 −1.96 −2.59 −2.73 −2.82 −2.69 −2.56 −2.36 −2.11 −1.96 −1.78
(−3.06) (−4.61) (−4.81) (−9.23) (−8.40) (−7.32) (−6.36) (−5.98) (−5.03) (−4.18) (−3.47) (−3.11)
QE −0.36 −0.39 −0.35 −0.47 −0.37 −0.24 −0.05 0.19 0.45 0.69 0.93 1.00
(−3.10) (−3.46) (−2.86) (−2.55) (−1.65) (−0.92) (−0.16) (0.61) (1.39) (1.84) (2.16) (2.23)
2
R 46.94 65.01 64.51 70.74 71.80 73.50 74.52 75.79 77.60 76.10 74.11 73.09
R2 7.65 7.06 2.57 1.91 0.29 −0.55 −0.93 −0.70 0.18 1.27 2.63 3.07
Periphery
IR 0.36 0.43 0.66 1.51 1.98 2.39 2.63 2.77 2.91 3.09 3.31 3.30
(4.34) (4.58) (5.08) (7.08) (7.98) (7.85) (7.47) (7.55) (7.38) (7.58) (7.82) (7.36)
U (×10−2 ) −0.08 −0.35 −0.98 −1.62 −1.98 −2.28 −2.58 −2.54 −2.34 −2.64 −2.89 −2.70
(−0.59) (−1.92) (−1.81) (−2.58) (−2.76) (−2.99) (−3.04) (−2.86) (−2.49) (−2.70) (−2.87) (−2.60)
QE 0.10 0.08 0.27 0.24 0.18 0.32 0.44 0.64 0.76 0.85 0.97 0.99
(0.94) (0.46) (1.11) (0.84) (0.55) (0.81) (0.99) (1.54) (1.71) (2.03) (2.40) (2.25)
2
R 41.25 49.70 33.99 62.72 64.17 65.13 63.42 62.76 61.75 62.50 64.13 61.75
R2 −0.76 −1.22 −0.47 −0.77 −1.11 −0.80 −0.62 0.01 0.40 0.66 1.06 1.03
Periphery-core spread
IR 0.01 −0.11 −0.08 0.14 0.26 0.33 0.32 0.28 0.23 0.27 0.41 0.38
(0.10) (−0.75) (−0.46) (0.65) (0.96) (0.97) (0.80) (0.66) (0.54) (0.60) (0.86) (0.74)
U (×10−2 ) 1.13 1.18 0.99 0.97 0.74 0.54 0.11 0.02 0.02 −0.53 −0.93 −0.92
(2.76) (2.61) (1.26) (1.55) (1.17) (0.88) (0.16) (0.02) (0.03) (−0.66) (−1.14) (−1.07)
QE 0.46 0.47 0.61 0.71 0.55 0.57 0.49 0.45 0.31 0.16 0.04 −0.02
(3.34) (2.51) (2.62) (2.05) (1.27) (1.10) (0.87) (0.87) (0.67) (0.37) (0.08) (−0.03)
2
R 39.67 34.39 11.62 27.71 21.39 24.44 20.20 19.91 19.13 18.82 19.07 14.94
R2 17.78 12.58 8.17 9.92 2.72 1.81 −0.06 −0.66 −1.84 −2.70 −2.98 −3.15
the smallest variance in the precrisis period. The latter the bottom panel shows that our communication shocks
enforces the factor unimportant in the precrisis period. as well as the QE factor are not significant at any maturity.
As Altavilla et al. (2019) note, this factor should only Overall, these findings also suggest that the muted sen-
contribute to the movements in the long end of the yield sitivities of peripheral bond yields to forward guidance and
curve and only be active post-2014, leading to the “QE the extreme sensitivity to U shocks during the crisis should
factor” label. be ascribed to the risk premium effect of monetary policy
Table 9 collects the results of regressions of core and communication in that period rather than to measurement
peripheral yields as well as their spread in the post-2014 errors or other confounding effects. Further, in reference to
period. Comparing estimated coefficients to those esti- our model and interpretation, the introduction of the As-
mated during the crisis period (Table 2), we find that the set Purchase Programme can be seen as a commitment by
effect of monetary policy communication returned to al- the ECB strong enough to eliminate the “bad” equilibrium
most precrisis levels: regression coefficients of both types of the crisis period and collapse yield spreads.
of shocks are of the same magnitude for core and pe-
ripheral yields, and IR shocks feature extremely high t- 8. Conclusion
statistics. In terms of R2 , our communication shocks ex-
plain, on average, more than 60% of the variation in pe- The ECB press conference meeting of 12 March, 2020
ripheral yields, while the QE shock explains an additional was a stark reminder that monetary policy communication
∼2% and only at the long end, in line with the interpre- matters. When ECB President Christine Lagarde mentioned
tation in Altavilla et al. (2019).9 Regarding yield spreads, that it is not the ECB’s job to “close the spread” between
bonds of different member states, asset markets reacted
promptly: the ten-year yield on Italian sovereigns jumped
9
Note that our IR communication shocks are not orthogonal to
Altavilla et al. (2019)’s QE factor. Therefore, our findings do not imply that
QE shocks are not relevant in this sample. Rather, they suggest that our provide a more formal comparison between our monetary policy shocks
interest rate and risk premium shocks capture most of the QE effects. We and those of the authors in the Online Appendix.
19
JID: FINEC
from 117bps to 174bps, whereas for Germany it remained Bauer, M. D., Swanson, E. T., 2020. The Fed’s Response to Economic News
at -74bps, implying an 35% increase in the spread. At the Explains the ‘Fed Information Effect’. Unpublished working paper.
Federal Reserve Bank of San Francisco.
same time, the EuroStoxx index fell by 13%. In this paper, Bekaert, G., Hoerova, M., Lo Duca, M., 2013. Risk, uncertainty and mone-
we offer a formal treatment of how central bank communi- tary policy. J Monet Econ 60, 771–788.
cation affects the cross-section of asset prices and provide Bernanke, B.S., Kuttner, K.N., 2005. What explains the stock market’s re-
action to federal reserve policy? Journal of Finance 60 (3), 1221–1257.
evidence for the presence of a central bank communication Bocola, L., Dovis, A., 2019. Self-fulfilling debt crises: a quantitative analy-
risk premium channel. sis. American Economic Review 109 (12), 4343–4377.
We make four novel contributions. First, drawing on the Boyarchenko, N., Haddad, V., Plosser, M.C., 2017. Market Confidence and
Monetary Policy. Unpublished working paper, Federal Reserve Bank of
joint dynamics between interest rates and an equity index
New York
sampled during a narrow window around ECB press con- Brand, C., Buncic, D., Turunen, J., 2010. The impact of ECB monetary de-
ferences, we construct monetary policy shocks related to cisions and communication on the yield curve. J. Eur. Econ. Assoc. 8,
1266–1298.
two distinct channels of central bank communication: the
Campbell, J.R., Evans, C.L., Fisher, J.D.M., Justiniano, A., 2012. Macroeco-
path of interest rates (forward guidance) and credit risk nomic effects of Federal Reserve forward guidance. Brook. Pap. Econ.
premia in the eurozone. Act. 44 (1), 1–80.
Using these shocks, we show that in the precrisis pe- Campbell, J.R., Fisher, J.D.M., Justiniano, A., Melosi, L., 2016. Forward guid-
ance and macroeconomic outcomes since the financial crisis. Eichen-
riod (January 2001 to November 2009), forward guidance baum, M., Parker, J. A. (Eds.), NBER Macroeconomics Annual 2016, Vol-
shocks were the important communication instrument and ume 31. Chicago University Press, Chicago, pp. 283–357..
had a uniform effect on core and peripheral bond yields. In Corsetti, G., Dedola, L., 2016. The mystery of the printing press: mone-
tary policy and self-fulfilling debt crises. J. Eur. Econ. Assoc. 14 (6),
contrast, during the sovereign debt crisis period (Decem- 1329–1371.
ber 2009 to December 2014), core bonds only reacted to De Santis, R.A., 2019. Redenomination risk. J. Money Credit Bank. 51 (8),
forward guidance shocks and peripheral yields were driven 2173–2206.
De Santis, R. A., Holm-Hadulla, F., 2017. Flow effects of central bank asset
almost exclusively by credit risk premium shocks, leading purchases on euro area sovereign bond yields: evidence from a natu-
to a significant wedge between core and peripheral yields. ral experiment. Unpublished working paper. European Central Bank.
We also show that our results are not exclusive to Duffie, D., Singleton, K.J., 1999. Modeling term structures of defaultable
bonds. Rev. Financ. Stud. 12, 687–720.
ECB press conferences during monthly monetary policy
Ehrmann, M., Fratzscher, M., 2005. How Should Central Banks Communi-
announcements but more generally whenever ECB presi- cate? Unpublished working paper. European Central Bank.
dents give speeches, and we demonstrate that president Ellingsen, T., Soderstrom, U., 2001. Monetary policy and market interest
rates. Am. Econ. Rev. 91 (5), 1594–1607.
speeches and unconventional policy announcements man-
Falagiarda, M., Reitz, S., 2015. Announcements of ECB unconventional pro-
aged to be effective and overcome the negative effect grams: implications for the sovereign spreads of stressed euro area
of ECB standard monetary policy announcements on the countries. J. Int. Money Financ. 53, 276–295.
yield spread. Finally, we show that the introduction of un- Fama, E.F., 2013. Does the Fed control interest rates? Rev. Asset Pric. Stud.
3, 180–199.
conventional monetary policy in 2015 restored the trans- Fratzscher, M., Lo Duca, M., Straub, R., 2016. ECB unconventional monetary
mission of monetary policy communication on sovereign policy actions: market impact, international spillovers and transmis-
yields as core and peripheral bonds again reacted homo- sion channels. IMF Econ. Rev. 64 (1), 36–74.
Gertler, M., Karádi, P., 2015. Monetary policy surprises, credit costs, and
geneously to communication shocks. economic activity. Am. Econ. J. 7 (1), 44–76.
The recent events of 2020 have again highlighted the Giannone, D., Reichlin, L., Small, D., 2008. Nowcasting: the real-time in-
importance of studying monetary policy beyond setting formational content of macroeconomic data. J. Monet. Econ. 55 (4),
665–676.
risk-free interest rates, as is standard in the literature. For Haitsma, R., Unalmis, D., de Haan, J., 2016. The impact of the ECB’s con-
example, while the forward guidance shock does not in- ventional and unconventional monetary policies on stock markets. J.
dicate a significant move on 12 March, 2020, we observe Macroecon. 48, 101–116.
Hansen, S., McMahon, M., Tong, M., 2019. The long-run information effect
a -215bp (almost three standard deviation) risk premium
of central bank communication. J. Monet. Econ. 108, 203–210.
shock. To facilitate research in this exciting area, we will Hanson, S.G., Lucca, D.O., Wright, J.H., 2020. Rate-Amplifying Investor De-
keep an up-to-date time series of our monetary policy mand and the Excess Sensitivity of Long-Term Interest Rates. Unpub-
lished working paper. Harvard Business School.
shocks on our webpages.
Hanson, S.G., Stein, J.C., 2015. Monetary policy and long-term real rates. J.
Financ. Econ. 115 (3), 429–448.
References
Jarociński, M., Karádi, P., 2020. Deconstructing monetary policy surprises:
the role of information shocks. Am. Econ. J. 12 (2), 1–43.
Acharya, V.V., Eisert, T., Eufinger, C., Hirsch, C., 2019. Whatever it takes: Koijen, R.S.J., Koulischer, F., Nguyen, B., Yogo, M., 2017. Euro-area quanti-
the real effects of unconventional monetary policy. Rev. Financ. Stud. tative easing and portfolio rebalancing. Am. Econ. Rev. Pap. Proc. 107,
32, 3366–3411. 621–627.
Ai, H., Bansal, R., 2018. Risk preferences and the macroeconomic an- Koijen, R.S.J., Koulischer, F., Nguyen, B., Yogo, M., 2021. Inspecting the
nouncement premium. Econometrica 86, 1383–1430. mechanism of quantitative easing in the euro area. J. Financ. Econ.
Altavilla, C., Brugnolini, L., Gürkaynak, R.S., Motto, R., Ragusa, G., 2019. 140, 1–20.
Measuring euro area monetary policy. J. Monet. Econ. 108, 162–179. Krishnamurthy, A., Nagel, S., Vissing-Jorgensen, A., 2018. ECB policies in-
Altavilla, C., Giannone, D., Lenza, M., 2014. The Financial and Macroeco- volving government bond purchases: impact and channels. Rev. Fi-
nomic Effects of OMT Announcements. Unpublished working paper. nanc. 22 (1), 1–44.
European Central Bank. Laarits, T., 2019. Pre-Announcement Risk. Unpublished working paper.
Andrade, P., Gaballo, G., Mengus, E., Mojon, B., 2019. Forward guidance New York University.
and heterogeneous beliefs. Am. Econ. J. 11, 1–29. Lorenzoni, G., Werning, I., 2019. Slow moving debt crises. Am. Econ. Rev.
Bacchetta, P., Perazzi, E., van Wincoop, E., 2020. Self-fulfilling Debt Crises: 109 (9), 3229–3263.
Can Monetary Policy Really Help? Unpublished working paper. Na- Lucca, D.O., Trebbi, F., 2011. Measuring Central Bank Communication: an
tional Bureau of Economic Research. Automated Approach With Application to FOMC Statements. Unpub-
Bai, J., Perron, P., 1998. Estimating and testing linear models with multiple lished working paper. Federal Reserve Bank of New York.
structural changes. Econometrica 66, 47–78. Melosi, L., 2016. Signalling effects of monetary policy. Rev. Econ. Stud. 84
Bai, J., Perron, P., 2003. Computation and analysis of multiple structural (2), 853–884.
change models. J. Appl. Econ. 18 (1), 1–22.
20
JID: FINEC
Miranda-Agrippino, S., Ricco, G., 2020. The transmission of monetary pol- Schmeling, M., Wagner, C., 2019. Does Central Bank Tone Move Asset
icy shocks. Am. Econ. J.. forthcoming Prices? Unpublished working paper. Goethe University Frankfurt.
Nakamura, E., Steinsson, J., 2018. High frequency identification of mone- Swanson, E.T., 2018. Measuring the Effects of Federal Reserve Forward
tary non-neutrality. Q. J. Econ. 133 (3), 1283–1330. Guidance and Asset Purchases on Financial Markets. Unpublished
Neuhierl, A., Weber, M., 2019. Monetary policy communication, policy working paper. University of California.
slope, and the stock market. J. Monet. Econ. 108, 140–155. Tang, J., 2015. Uncertainty and the Signaling Channel of Monetary Policy.
Rogers, J.H., Scotti, C., Wright, J.H., 2014. Evaluating asset market effects of Unpublished working paper. Federal Reserve Bank of Boston.
uncoventional monetary policy: a multi-country review. Econ. Policy Woodford, M., 2012. Methods of policy accommodation at the inter-
29 (80), 749–799. est-rate lower bound. In: Proceedings from the Federal Reserve Bank
Romer, C.D., Romer, D.H., 20 0 0. Federal reserve information and the be- of Kansas City Economic Policy Symposium, Jackson Hole, Wyoming,
havior of interest rates. Am. Econ. Rev. 90 (3), 429–457 . pp. 185–288..
21

Leo M Broni 2021

Uploaded by

Copyright:

Available Formats

Leo M Broni 2021

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Leo M Broni 2021

Uploaded by

Copyright:

Available Formats

JID: FINEC

ARTICLE IN PRESS [m3Gdc;May 30, 2021;23:52]

Contents lists available at ScienceDirect

Journal of Financial Economics

Central bank communication and the yield curveR

Keywords: © 2021 Published by Elsevier B.V.

The ﬁnancial turmoil of 20 07–20 08 and the subsequent

spreads in response to communication shocks must be due

The equity response, EQ, is simply computed as the Table 1

IR loading: precrisis IR loading: crisis

U loading: precrisis U loading: crisis

yτi,t = aτi + bτi IRt + ciτ Ut + i,t

IR: rolling coefficient and R2 IR: rolling coefficient and R2

U: rolling coefficient and R2 U: rolling coefficient and R2

9 R2 core 1.5 35 R2 periphery

8 Coefficient core 1 Coefficient periphery

munication represented around a quarter of the total yield

5.4. Credit risk

Next, to study whether monetary policy communication

CDSi,t = ai + bi IRt + ci Ut + i,t ,

CDSi,t = ai + bi IRt + ciUt + i,t ,

Germany France Italy Spain Core Periphery P-C

IR −0.28 −0.50 −1.55 −1.48 −0.39 −1.52 −1.12

5.7. Robustness 6.1. Core versus peripheral yields

IR loading: precrisis IR loading: crisis

U loading: precrisis U loading: crisis

Germany France Italy Spain Core Periphery P-C

IR −0.37 −0.85 −2.93 −2.58 −0.61 −2.76 −2.15

-150 Peripheral (P)

yτi,t = aτi + bτi IRt + ciτ Ut + diτ QEt + i,t

You might also like

yτi,t = aτi + bτi IRt + ciτ Ut + i,t

CDSi,t = ai + bi IRt + ci Ut + i,t ,

CDSi,t = ai + bi IRt + ciUt + i,t ,

yτi,t = aτi + bτi IRt + ciτ Ut + diτ QEt + i,t