Does Central Bank Tone Move Asset Prices?

Abstract

This article shows that changes in the tone of central bank communication have a significant effect on asset prices. Tone captures how the central bank frames economic fundamentals and its monetary policy. A positive tone surprise is associated with increases in stock prices and interest rates, whereas credit spreads and volatility risk premia decrease. These tone effects are robust to controlling for policy actions as well as for conventional measures of monetary policy shocks. Our results suggest that communication tone is a powerful instrument of monetary policy, which affects risk premia embedded in asset prices.

Full text

JOURNAL OF FINANCIAL AND QUANTITATIVE ANALYSIS
© The Author(s), 2024. Published by Cambridge University Press on behalf of the Michael G. Foster
School of Business, University of Washington. This is an Open Access article, distributed under the terms
of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0), which permits
unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
doi:10.1017/S0022109024000073
Does Central Bank Tone Move Asset Prices?
Maik Schmeling
Goethe University Frankfurt and Centre for Economic Policy Research (CEPR)
schmeling@finance.uni-frankfurt.de
Christian Wagner
WU Vienna University of Economics and Business, Vienna Graduate School of Finance (VGSF) and Centre for
Economic Policy Research (CEPR)
christian.wagner@wu.ac.at (corresponding author)
Abstract
This article shows that changes in the tone of central bank communication have a significant
effect on asset prices. Tone captures how the central bank frames economic fundamentals and
its monetary policy. A positive tone surprise is associated with increases in stock prices and
interest rates, whereas credit spreads and volatility risk premia decrease. These tone effects
are robust to controlling for policy actions as well as for conventional measures of monetary
policy shocks. Our results suggest that communication tone is a powerful instrument of
monetary policy, which affects risk premia embedded in asset prices.
As I had often remarked, monetary policy is 98 percent talk and 2 percent
action. (Ben Bernanke, 2016, p. 498)
I don’t think I’m stepping up my rhetoric on inflation, Draghi said […].
Financial market analysts nonetheless detected a shift in tone if not in substance
of monetary policy. (Reuters, Apr. 4, 2012)
All eyes will be on the ECB this afternoon. If the tone is clearly dovish, then it
could maybe stop the bleeding on the market. (Reuters, Aug. 7, 2014)
I. Introduction
Monetary policy strongly affects asset prices, a prime example being the effect
of monetary policy announcements on stock prices (e.g., Bernanke and Kuttner
(2005), Lucca and Moench (2015), Cieslak, Morse, and Vissing-Jorgensen (2019),
We thank two anonymous referees, Alessandro Beber, Oliver Boguth, Jennifer Conrad (the editor),
Gino Cenedese, Zhi Da, Marco Di Maggio, Chris Downing, Michael Ehrmann, Rainer Haselmann,
Tarek Hassan, Marcin Kacperczyk, Mark Kamstra, Anil Kashyap, Ralph Koijen, Holger Kraft, Tim
Kroencke, David Lando, Christian Laux, Michael Lemmon, Tim Loughran, Mamdouh Medhat, Michael
Melvin, Alex Michaelides, Menno Middeldorp, Silvia Miranda-Agrippino, Philippe Mueller, Thomas
Nagel, Florian Nagler, Evgenia Passari, Lasse Pedersen, Tarun Ramadorai, Jesper Rangvid, Lucio
Sarno, Christian Schlag, Andreas Schrimpf, Vania Stavrakeva, Andrea Tamoni, Pietro Veronesi, Anette
Vissing-Jorgensen, Desi Volker, Paul Whelan, Fredrik Willumsen, and participants at the 2017
1
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

and Neuhierl and Weber (2021)). A large part of the information released on
announcement days comes in the form of verbal communication, rather than
quantitative releases, and central banks (CBs) use such communication to explain
their policy decisions and the economic outlook and to shape market expectations.
CB communication is thus closely followed by market participants, extensively
covered by the financial press, and CBs evaluate the media coverage of their
statements to gauge the effectiveness of their communication.
1
Importantly, market
participants do not only pay attention to the content but also, as the above quotes
illustrate, to the tone of CB statements (i.e., how the central bank frames its policy
decisions and the economic outlook). Hence, a natural question is “Does commu-
nication matter for asset prices beyond policy actions?”Ben Bernanke’s view that
“monetary policy is 98 percent talk and 2 percent action”suggests that it should.
The contribution of our article is to answer this question by showing that the
tone of CB communication matters for asset prices. A positive tone surprise is
associated with higher equity market returns, lower volatility risk premia (a proxy
for risk aversion implied by equity options), and lower credit spreads (in particular
for financial institutions). At the same time, a positive tone surprise is associated
with higher risk-free interest rates. Our results suggest that policy tone affects risk
premia embedded in asset prices and that these effects are very similar to those of
policy actions on stocks (e.g., Bernanke and Kuttner (2005)), variance risk premia
(e.g., Bekaert, Hoerova, and Lo Duca (2013)), and credit spreads (e.g., Gertler and
Karadi (2015)). Given that our analysis controls for policy actions, our findings
imply that communication tone is an additional policy tool that supplements other
instruments of monetary policy.
In the empirical analysis, we measure the tone of the European Central Bank
(ECB) president in press conferences (PCs) held after policy meetings, which offers
an ideal setup for our analysis.
2
The ECB holds scheduled monetary policy meetings
on Thursdays and announces its interest rate decision at 13:45 CET. The policy
statement issued at that time contains little to no information other than the actual
interest rate decision. At 14:30, the PC starts. Since PCs take place during trading
hours, financial markets can react to new information instantaneously, and the stag-
gered timing of rate announcement and PC allows to disentangle market reactions to
news about policy rates and communication (e.g., Ehrmann and Fratzscher (2009)).
American Finance Association (AFA) Meetings, 2016 Western Finance Association (WFA) Meetings,
the 2016 SFS Cavalcade, the 2016 INQUIRE UK Conference, the 2015 London Empirical Asset Pricing
(LEAP) Meeting, and the 2015 European Finance Association (EFA) Meetings, as well as seminar
participants at Aarhus University, Aalto University, the Bank for International Settlements, Bank of
England, the Board of Governors of the Federal Reserve System, BlackRock, Copenhagen Business
School, the German Institute for Economic Research (DIW, Berlin), Norges Bank, Norges Bank
Investment Managers, Sveriges Riksbank, Goethe University Frankfurt, and the Vienna Graduate
School of Finance (VGSF) for helpful comments and suggestions. Wagner acknowledges support by
the Danish National Research Foundation (DNRF102). Schmeling gratefully acknowledges financial
support by the German Science Foundation (DFG).
1
For an overview of the literature on CB communication, see, e.g., Woodford (2005) and Blinder,
Ehrmann, Fratzscher, De Haan, and Jansen (2008). Berger, Ehrmann, and Fratzscher (2011) discuss how
the ECB evaluates communication effectiveness via media reception.
2
The ECB was the first major central bank to use press conferences to inform the public about the
rationale behind its decisions and to provide an outlook, but recently, other central banks (including the
Fed) have started to adopt similar communication strategies.
2 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

To quantify tone, we use the financial dictionary developed by Loughran and
McDonald (2011) to identify negative words and evaluate each statement’s tone by
assessing the prevalence of negative words. We verify that tone indeed captures
how the ECB frames macroeconomic fundamentals by showing that phrases such
as “global imbalances,”“disorderly correction,”“excessive deficit,”and discus-
sions about fundamentals that, for example, “remain weak,”are among the most
important drivers of tone.
Turning to the relation between CB communication and asset prices, we first
study how equity markets respond to changes in tone. Figure 1 illustrates our results
by plotting the average cumulative returns of the EuroStoxx 50 (a European large-
cap stock index) in a 48-h window around policy rate announcements of the ECB.
The middle line (in blue) of Figure 1 shows the average cumulative return
across all 241 PCs in our sample. There is a preannouncement drift before the policy
rate announcement at 13:45 CET (indicated by the solid vertical line labeled “RA”),
akin to the findings in Lucca and Moench (2015) for FOMC meetings. Contrary to
the FOMC preannouncement drift, however, these returns are completely reversed
in the 24 h after the announcement. The other 2 lines show average cumulative
returns over the same time window but separately for PCs with a more positive tone
(upper line in green) and PCs with a more negative tone (lower line in red) compared
to the previous PC. Three effects stand out from Figure 1. First, PCs with a more
positive tone are associated with higher returns than PCs with a more negative tone.
Second, tone-conditional returns co-move until the beginning of the PC and then
start to diverge. Third, unlike the preannouncement drift, the return spread between
PCs with positive and negative tone changes is not reversed.
The link between tone changes and equity markets is statistically significant
for intraday returns measured from the beginning of the PC as well as for returns
FIGURE 1
Stock Returns in the 48 Hours Around ECB Policy Rate Announcements
Figure 1 shows the cumulative returns of the EuroStoxx 50 index in the 48 hours around ECB policy rate announcements. The
ECB announces its rate decision at 13:45 (CET) and then holds a press conference, which starts at 14:30 CET. The time
window shown is from 13:45 on the day before until 13:45 on the day after the announcement. The dashed vertical lines
indicate the end of a trading day, whereas the 2 solid lines indicate the time of the policy rate announcement (“RA”) and the
start of the press conference (“PC”), respectively. The three lines correspond to all press conference days (middle line, blue),
the subset of days with positive tone changes (upper line, green), or negative tone changes (lower line, red).
Time (Hour CET)
Cumulative Return (basis points)
−40
−30
−20
−10
0
10
20
30
40
14 16 10 12 14 16 10 12
RA PC
All press conferences
PCs with more positive tone
PCs with more negative tone
Schmeling and Wagner 3
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

measured over the full PC day, for Eurozone indices as well as for country indices.
Our key finding is that the effect of tone changes on returns is robust to controlling
for market-based proxies for financial conditions leading up to the policy meeting,
the ECB’s policy rate and unconventional policy announcements, interest rate-
based measures of monetary policy shocks associated with the rate announcement,
and the PC (e.g., Altavilla, Brugnolini, Gürkaynak, Motto, and Ragusa (2019)), as
well as past tone changes and other textual characteristics. Accounting for all these
control variables, we can interpret our results in terms of tone surprises that move
stock prices, and we validate this interpretation using the residuals from autore-
gressive (AR) models of ECB tone as well.
Our results imply that surprises in ECB tone convey new information for stock
markets, which raises the question of why and how tone matters for asset prices. To
shed light on this question, we explore tone effects in risk-free interest rates and
other classes of risky assets.
We start by documenting that a positive tone surprise is associated with higher
(risk-free) interest rates, which implies that tone does not move stock prices through
a simple discount rate effect. Instead, we show that tone surprises have a large effect
on risk premia embedded in asset prices.
When ECB tone becomes more positive, the VSTOXX volatility index
(similar to the VIX in the U.S.) decreases, which implies that volatility insurance
becomes cheaper. At the same time, realized volatility is essentially unrelated to
tone changes. As a consequence, changes in the price of volatility insurance are
primarily driven by lowered risk premia required by investors in excess of expected
volatility. This, in turn, implies that positive tone surprises are associated with
market participants lowering their risk aversion. Thus, our finding represents a
communication-based analog to Bekaert et al. (2013), who find that monetary
easing decreases risk aversion as measured by variance risk premia.
As an alternative proxy for risk premia, we analyze the response of credit
spreads to tone changes. We find that a positive tone surprise is associated with a
decrease in credit spreads (i.e., the yield differential of BBB- and AAA-rated
corporate bonds), and this result is most pronounced for the credit spreads of
financial institutions. These tone responses of credit spreads as well as the responses
of stocks and interest rates are qualitatively the same as the joint asset price
responses due to changes in the risk aversion of the financial sector in Gilchrist
and Zakrajšek (2012).
Since positive tone shocks are associated with higher stock prices and higher
interest rates, they bear a resemblance to CB information effects (e.g., Nakamura
and Steinsson (2018), Jaroci
nski (2020)). To explore this further, we expand our
analysis of tone surprises to account for information effects and find that controlling
for information shocks does not crowd out the effect of tone shocks on asset prices.
These empirical results cannot distinguish whether the significance of tone stems
from being a novel, text-based proxy for information effects or whether tone matters
through a separate channel. However, they clearly show the CB tone moves for asset
prices because it conveys news not captured by empirical measures of policy and
information shocks.
At the beginning of the introduction, we asked the question whether a CB’s
communication matters for asset prices beyond policy actions. We find that it does.
4 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

Our results suggest that CB tone affects the risk-taking of market participants and
the risk premia they require, which implies that communication tone is an important
instrument in the monetary policy toolkit.
Related literature. Our work relates to previous research that analyzes the
effects of monetary policy on the prices of stocks and other assets as well as to the
literature on CB communication.
Bernanke and Kuttner (2005) are among the first to show that Fed policy
decisions have a strong effect on stock prices. Other studies of equity returns around
policy meetings provide evidence for a preannouncement drift leading up to FOMC
meetings (Lucca and Moench (2015)) and weekly return patterns over FOMC
cycles (Cieslak et al. (2019)). Neuhierl and Weber (2019) show that the expected
path of monetary policy, measured from Fed Fund futures, predicts stock returns.
There is ample evidence that monetary policy affects prices of other assets as well
(e.g., Rigobon and Sack (2004), Campbell, Pflueger, and Viceira (2020)), and our
work is most closely related to those who document risk premium effects, such as in
term premia (e.g., Shiller, Campbell, and Schoenholtz (1983), Gertler and Karadi
(2013), and Hanson and Stein (2015)), equities (e.g., Bernanke and Kuttner (2005)),
credit spreads (e.g., Gilchrist and Zakrajšek (2012), Gertler and Karadi (2015)), and
options-implied measures of risk premia (e.g., Bekaert et al. (2013)).
A related literature focuses on quantifying monetary policy surprises from
changes in asset prices in short windows around policy announcements (e.g.,
Kohn and Sack (2004), Guerkaynak, Sack, and Swanson (2005), Nakamura and
Steinsson (2018), Cieslak and Schrimpf (2019), Ferrari, Kearns, and Schrimpf
(2021), Leombroni, Vedolin, Venter, and Whelan (2021), and Swanson (2021)).
In our empirical analysis, we use the interest rate shocks proposed by Altavilla et al.
(2019) and Jaroci
nski (2020) as well as the policy and information shocks identified
by Jaroci
nski and Karadi (2020) from the joint market reactions of interest rates and
equities.
Since we measure tone from CB statements, our work relates to the large
literature on CB communication (e.g., Woodford (2005), Blinder et al. (2008), for
a comprehensive survey). Early workincludes Romer and Romer(2004) who apply a
narrative approach to identify monetary policy shocks from CB documents. Lucca
and Trebbi (2009) analyze the content of FOMC statements by semantic orientation
scores estimated from a large information set obtained through search engines.
Jegadeesh and Wu (2017) assess how the market responds to different topics dis-
cussed in FOMC minutes. Hansen, McMahon, and Prat (2017)) investigate how
transparency affects deliberation of FOMC members, and Hansen and McMahon
(2016) study how FOMC communication about economic conditions and forward
guidance affect economic and financial variables. More recently, Ehrmann and Talmi
(2020) use a human scoring approach to investigate how (small) changes in CB
communication affect financial markets. Picault and Renault (2017) develop a lex-
icon to quantify ECB communication and show that it is helpful in explaining future
monetary policy outcomes. Other articles that analyze different communication
characteristics (such as content, tone, similarity, readability, etc.) include Bligh and
Hess (2007), Rosa and Verga (2007), Rosa (2011), and Amaya and Filbien (2015).
Our contribution to these branches of research is to show that policy commu-
nication matters for asset prices through a risk-based channel, beyond policy
Schmeling and Wagner 5
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

actions, because tone surprise conveys news not captured by monetary and infor-
mation shocks.
II. Measuring CB Tone
Our empirical analysis focuses on the ECB. Throughout our sample period
from 1999 to 2021, the ECB has held its monetary policy meetings on Thursdays
(scheduled well in advance), announced its interest rate decision at 13:45 CET, and
held a PC at 14:30.
The announcements and PCs take place during European trading hours and are
closely followed by market participants who can react to new information instan-
taneously. The staggered timing of rate announcement and PC provides an ideal
setup for disentangling market reactions to news about policy rates and communi-
cation tone. Our focus is on the tone surprise revealed during the PC, controlling for
any information released with the rate announcement 45 min earlier.
3
The ECB was the first major CB to adopt this form of communication and thus
offers the longest history to study the impact of CB tone on asset prices. Impor-
tantly, other CBs have recently followed the ECB’s example and started to hold PCs
after their policy meetings. For example, the Federal Reserve has started to hold
PCs very similar to the ECB’s setup in Apr. 2011, but only after every other FOMC
meeting. Boguth, Gregoire, and Martineau (2018) provide the first evidence that
markets pay higher attention and respond more strongly to FOMC meetings with
PCs than without PCs. In 2018, chairman Jay Powell announced that the Fed would
hold PCs after every FOMC meeting from 2019, emphasizing that increasing the
number of PCs is no indication about future policy actions but only about improving
communication.
4
With more and more CBs seeking to improve communication
with the public by holding PCs after policy meetings, our results should be a useful
benchmark for assessing the likely effects of PCs on financial markets as CBs adopt
this form of communication as well.
5
In total, our sample covers 241 ECB PCs from Jan. 1999 (the introduction of
the Euro) to Dec. 2021. For these PCs, we obtain transcripts of the ECB president’s
opening statements, which are carefully drafted in advance with a 2-fold purpose: to
inform the general public about the rationale underlying the interest rate decision
made by the Governing Council and to provide a general outlook.
3
For most of our sample (i.e., 1999 to 2016), the statement issued at 13:45 contained little to no
information other than the ECB’s interest rate decision. From Mar. 2016 onward, the press release reports
all monetary policy decisions, including unconventional monetary policy actions which had previously
been announced during press conferences. Since July 2016, the ECB has also included rate guidance in
the press release. With the onset of the COVID-19 crisis, the length of the press releases has increased
considerably, mostly due to the ECB announcing pandemic-related policy decisions.
4
In his PC on June 13, 2018 (link), Chairman Powell states, “As Chairman, I hope to foster a public
conversation about what the Fed is doing to support a strong and resilient economy. And one practical
step in doing so is to have a press conference like this after every one of our scheduled FOMC meetings.
[…] I want to point out that having twice as many press conferences does not signal anything about the
timing or pace of future interest rate changes. This change is only about improving communications.”
5
Other central banks include the Bank of England, who started to hold press conferences after
inflation reports in 2015, but also other central banks (e.g., New Zealand, Norway, Sweden, and
Switzerland).
6 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

Below, we discuss how we measure tone, present summary statistics for ECB
tone, and provide evidence that the ECB uses its tone to frame its judgment about
economic conditions and to adumbrate its future actions.
A. Measuring Tone from ECB PC Statements
The objective of our article is to quantify how changes in CB tone matter for
asset prices. For our analysis, we deliberately choose a simple dictionary-based
measure of tone that we quantify from ECB statements as described below. Addi-
tionally, we use the transcripts to compute other text-based measures proposed by
previous research to capture changes in the statements’wording, complexity, and
lexical diversity. We discuss the processing of transcripts and the measurement of
these textual characteristics in detail in Appendix A.
We use the transcripts of the ECB PCs to construct a proxy for CB tone based
on the financial dictionary developed by Loughran and McDonald (LM) (2011).
More specifically, we use this dictionary to identify words that can be classified as
negative in financial contexts. In each transcript, we count the number of negative
words (N) as well as the total number of words (T) and define CB tone (τ)as
τ=1N=T,(1)
such that lower values reflect a more negative CB tone and higher values imply a
less negative tone. Our empirical analysis focuses on changes in tone,Δτ, measured
as the first difference in τbetween two subsequent PCs. Accordingly, we interpret
increases in τas tone becoming more positive and decreases in τas tone becoming
more negative.
Our choice to measure CB tone based on negative words listed in the LM
dictionary is driven by our objective to use a simple, transparent approach that does
not require any form of subjective judgment and thereby minimizes concerns
related to data mining. More specifically, our considerations are as follows.
First, we only use negative words because the usefulness of positive words for
measuring tone is very limited. On the one hand, positive words are frequently
negated (whereas negative words are not) and the framing of bad news often
involves positive words (e.g., Loughran and McDonald (2011), (2016)).
6
Second, by relying on the well-established LM dictionary, we avoid the need
for a subjective classification of words as being negative or not. The LM dictionary
is explicitly designed to be informative for financial documents (in contrast to, e.g.,
the widely used Harvard Dictionary), and while it was originally applied to 10-K
filings, it has proven useful in other financial contexts as well (see, e.g., Gurun and
Butler (2012), Hillert, Jacobs and Müller (2014), and the surveys of Loughran and
6
For example, Loughran and McDonald (2016) note (p. 1217) that “The framing of negative
information is so frequently padded with positive words that the measured positive sentiment is
ambiguous”such that ultimately there typically is “little incremental information in positive word
lists.”One could attempt to account for negations by training an algorithm to ignore or reinterpret
positive words in the vicinity of negations; however, doing so raises data mining concerns and does not
increase the accuracy of tone measurement. On the other hand, previous research suggests that market
participants tend to focus on negative words while paying less attention to positive words (e.g., Loughran
and McDonald (2020)).
Schmeling and Wagner 7
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

McDonald (2016), (2020)). Alternatively, we could build our own dictionary of CB
language, either by labeling words as negative based on common sense or based on
a statistical procedure that classifies certain words as negative based on the market’s
reaction to the occurrence of these words. However, defining such a list ourselves
would essentially mean that we have control over the resulting time series of tone
and, thus, the outcome of our empirical analysis later in the article, which could
make our results susceptible to p-hacking concerns (e.g., Loughran and McDonald
(2020)). Using a statistical procedure to generate a word list would either require to
reserve some of the data for training the model (which limits the sample available
for the economic analysis) or to use the data twice, first to build the dictionary and
subsequently to analyze the effect of tone on asset prices (which creates hindsight
bias).
Finally, and again to avoid data mining concerns, we choose to measure tone
by means of simple word counts rather than more elaborate techniques. Approaches
such as term weighting or topic modeling use the full sample, which implies
hindsight bias. Hence, to avoid all these potential biases, we choose simplicity
and transparency over more elaborate alternatives in our empirical tests.
7
The downside of our approach, as for any other method of textual analysis, is
that there can be misclassifications (i.e., cases where a phrase is identified as being
negative even though it is not). In Sections II.B and II.C, we document the useful-
ness of our tone measure by providing excerpts from PC statements and showing
which words and phrases drive ECB tone.
B. Descriptive Statistics for ECB Tone
Table 1 presents some descriptive statistics for ECB PCs. The first column
shows that PCs take place regularly but not at equidistant intervals. The average PC
cycle is around 23 trading days, with 9 and 49 days for the shortest and longest
intervals, respectively. The second column summarizes statistics for the ratio of the
number of negative words to the number of total words (N=T), which we use to
compute the tone measure defined in equation (1). The average N=Tis around 2.5%
and is associated with substantial variability within the range of 0.4% and 5.7%.
The third column shows that tone changes (Δτ) are close to 0 on average and at the
median but exhibit substantial variation in the range from 2.4% to +2.0% as well
as a significant first-order autocorrelation. Of the 240 ECB tone changes in our
sample, we find that tone increases at 128 PCs and decreases in 112 cases. Figure 2
plots the time series of ECB tone and changes in ECB tone. The gray vertical lines
mark the dates of the ECB PCs. Graph A shows that ECB tone reached its minimum
at the end of 2008/beginning of 2009 during the financial crisis, and Graph B
illustrates that the volatility of tone changes over time.
7
For the same reason, we do not ask human readers to evaluate CB statements. For instance, while a
potential advantage of that approach may be that human readers are better in processing certain nuances
of texts, a disadvantage is that human judgment cannot be avoided in the scoring process, thereby neither
guaranteeing an avoidance of misclassification nor “reader-fixed effects”in tone measures (e.g.,
Ehrmann and Fratzscher (2007)). Moreover, it would be difficult to set up a proper out-of-sample
analysis of how CB tone matters for asset prices, as multiple readers would have to be trained on a
large body of statements.
8 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

C. Which Words Drive ECB Tone?
To provide evidence that tone indeed captures how the ECB frames macro-
economic fundamentals, we present summary statistics for the most frequently used
negative words that drive our tone measure as well as for bigrams and trigrams
(i.e., sequences of 2 and 3 adjacent words) in which they appear. Table 2 shows that
the most frequently used negative words are “weak,”“decline,”and “imbalances.”
8
The most common bigrams and trigrams involving negative words include, for
instance, “global imbalances,”“weaker (than) expected,”“disorderly correction,”
“financial market volatility,”and “high level (of) unemployment.”This suggests
that our simple, dictionary-based measure correctly captures negative phrases
commonly used by the ECB. With this first evidence for tone picking up how the
ECB interprets and judges economic developments, we provide several PC excerpts
to illustrate the broader context in which tone is measured.
Table 3 presents excerpts from the PC held on Jan. 15, 2009, which our
measure identifies to exhibit the most negative tone during our sample period. In
these excerpts, we highlight word sequences involving negative words that we have
identified in multiple statements (in red italic font) and mark the negative words
with asterisks (*). From this statement, the sentence having the largest impact on our
tone measure is from the discussion of economic risks, stating that
They relate mainly to the potential for a stronger impact on the real economy of
the *turmoil* in financial markets, as well as to *concerns* about the emer-
gence and intensification of protectionist pressures and to possible *adverse*
developments in the world economy stemming from a *disorderly* *correc-
tion* of global *imbalances*.
TABLE 1
The Tone of ECB Press Conference Statements
Table 1 reports descriptive statistics for the 241 ECB press conferences between Jan. 7, 1999 and Dec. 16, 2021. The first
column reports the number of business days between press conferences (PCs). N=Treports the ratio of the number of
negative words (N) divided by the total number of words (T) in the president’s opening statement at the PC (in percentage
points). Δτmeasures the change in tone τcompared to the tone at the previous PC, where τ¼1N=Tas defined in equation
(1); reported numbers are the changes in percentage points. For the 240 realizations of Δτ, we also report the coefficient of an
AR(1) regression and the associated t-statistic. “No. of obs. Δτ>0”denotes the number of tone changes when tone becomes
more positive and “No. of obs. Δτ<0”counts the observations when tone becomes more negative.
Days Between PCs N=T[in %] Δτ[in %]
Mean 23.2 2.545 0.006
Std. dev. 6.9 0.994 0.745
Min 9.0 0.361 2.409
Median 20.0 2.503 0.028
Max 49.0 5.651 2.015
AR(1) 0.403
t-stat. [6.78]
No. of obs. Δτ>0128
No. of obs. Δτ<0112
8
These counts are based on aggregating words by their word stem (e.g., the 467 occurrences we
summarize for “weak”are the sum of occurrences for “weak”(194), “weaken”(6), “weakened”(22),
“weakening”(58), “weaker”(121), “weakness”(60), and “weaknesses”(6)).
Schmeling and Wagner 9
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

In general, reading through these paragraphs, we find support for the view that
our tone measure picks up the ECB’s framing of economic and financial conditions
as well as the economic outlook. To provide a broader picture of what our tone
measure captures, we present additional excerpts in Section IA.A of the Supple-
mentary Material.
III. CB Tone and Equity Returns
In this section, we document a strong link between stock prices and the tone
of ECB PC statements. A more positive (negative) tone compared to the previous
PC is associated with higher (lower) equity market returns. These results are
robust to controlling for market-based proxies for financial conditions leading
up to the policy meeting, the ECB’s policy actions, and interest rate-based
measures of monetary policy shocks associated with the rate announcement
and the PC.
FIGURE 2
The Tone of ECB Press Conference Statements
Figure 2 plots the time series of ECB tone, τ, and changes in ECB tone, Δτ, in Graphs A and B, respectively. Tone is defined as
τ¼1N=T, see equation (1), where Nand Tdenote the number of negative words and the total number of words in a press
conference statement. Δτis measured as the difference in τbetween two consecutively held press conferences. Tone is
measured from the ECB president’s opening statements at the 241 ECB press conferences between Jan. 7, 1999, and Dec.
16, 2021. The vertical lines mark these 241 press conferences.
Graph A. Tone Level (τ)
Graph B. Changes in ECB Tone (Δτ)
ECB Tone
0.94 0.95 0.96 0.97 0.98 0.99 1.00
Jan/99 Jan/01 Jan/03 Jan/05 Jan/07 Jan/09 Jan/11 Jan/13 Jan/15 Jan/17 Jan/19 Jan/21
Changes in ECB Tone
–0.02 –0.01 0.00 0.01 0.02
Jan/99 Jan/01 Jan/03 Jan/05 Jan/07 Jan/09 Jan/11 Jan/13 Jan/15 Jan/17 Jan/19 Jan/21
10 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

A. Equity Returns Around ECB PCs
Akin to the literature that quantifies monetary policy shocks from changes in
market prices in short windows around policy announcements, we start by studying
the impact of tone changes on asset prices in daily data. The high-frequency results,
TABLE 3
Excerpts from the ECB President’s Statement on Jan. 15, 2009
Table 3 presents excerpts of the ECB president’s introductory statement, given at the press conference on Jan. 15, 2009. Our
measure of CB tone identifies this statement to exhibit the most negative tone of all statements in our sample. From this
statement, we present the three paragraphs that have the largest impact on our tone measure, i.e., the three paragraphs with
the highest ratio of negative words to total words. Words highlighted in italic font and marked by asterisks (*) are negative
words identified by the dictionary we employ. Other words highlighted in italic font are common word sequences involving
negative words that we have identified in multiple statements.
•Looking further ahead, on the basis of our current analysis and assessment, we continue to see global economic *weak-
ness* and very *sluggish* domestic demand *persisting* in the coming quarters as the impact of the financial tensions on
activity continues. At the same time, we expect the fall in commodity prices to support real disposable income in the period
ahead. Furthermore, the euro area should over time reap the full benefit from the effects of policy measures announced over
recent weeks.
•In the view of the Governing Council, this outlook for the economy remains surrounded by an exceptionally high degree of
uncertainty. Overall, risks to economic growth remain clearly on the downside. They relate mainly to the potential for a
stronger impact on the real economy of the *turmoil* in financial markets, as well as to *concerns* about the emergence and
intensification of protectionist pressures and to possible *adverse* developments in the world economy stemming from a
*disorderly* *correction* of global *imbalances*.
•Risks to price stab ility over the medium term are broadly balanced .* Unexpected* further *dec lines* in commodity prices
or a stronger than exp ected slowdown in the economy could pu t *downward* pressure on infla tion, while upside risks to
price stabilit y could materia lize, particul arly if the rece nt fall in commod ity prices were to reverse or if d omestic price
pressures turn o ut to be stronger tha n assumed. It is ther efore *crucial* that price and wag e-setters fully live up to their
responsibili ties.
TABLE 2
Which Words Drive Tone?
Table 2 presents descriptive statistics for the “negative”words (as classified by the dictionary of Loughran and McDonald
(2011)) that are most prevalent in ECB press conference statements. Column 1 reports the 20 most frequently used negative
words, ordered by the number of their occurrence across all ECB press conference statements; these counts are based on
aggregating words by their word stem. Columns 2 and 3 show the context in which negative words are most frequently used by
the ECB by presenting counts for bigrams and trigrams (i.e., sequences of 2 and 3 adjacent words), respectively. The analysis
is based on 241 ECB press conference statements between Jan. 7, 1999, and Dec. 16, 2021.
Words No. Bigrams No. Trigrams No.
Weak 467 Global imbalances 86 Correction global imbalances 38
Decline 397 Weaker expected 56 Global imbalances regard 36
Imbalances 233 Structural unemployment 50 Imbalances regard price 36
Concerns 210 Fiscal imbalances 39 Disorderly correction global 36
Negative 202 Correction global 38 Lagged relationship business 33
Slow 176 Imbalances regard 37 Financial market volatility 32
Unemployment 175 Disorderly correction 36 Reduce structural unemployment 29
Crucial 163 Possibility disorderly 35 Reflect lagged relationship 29
Volatility 161 Market volatility 34 Possibility disorderly correction 26
Dampened 150 Prolonged period 34 Pressures possibility disorderly 25
Deficit 149 Lagged relationship 33 Prolonged period low 25
Downward 131 Excessive deficit 30 Continue reflect lagged 24
Challenges 113 Reflect lagged 29 Structural unemployment boost 24
Adverse 98 Level unemployment 28 High-level unemployment 22
Lagging 81 Disorderly developments 25 Financial market turmoil 21
Correction 80 Remain weak 25 Reducing vulnerabilities implementation 21
Disorderly 67 Revised downward 25 Vulnerabilities implementation structural 21
Restructuring 67 Excessive deficits 24 Disorderly developments owing 20
Excessive 63 High unemployment 24 Owing global imbalances 20
Turmoil 61 Negative impact 24 Vulnerabilities emerging markets 20
Schmeling and Wagner 11
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

shown in Figure 1, suggest that the effect of ECB tone changes on EuroStoxx
50 prices over the full trading day is very similar to that arising during the
PC. Accordingly, we should find similar PC effects in daily data when we compute
returns from the closing prices on the day preceding the PC and the day on which the
PC is held.
To study the effect of changes in ECB tone on Eurozone equity returns, we
obtain daily equity data for i) the EuroStoxx 50 (ESX50), which covers the 50 larg-
est firms in the Eurozone, from STOXX; ii) the MSCI EMU Index, a broad
Eurozone index, from Datastream; iii) 10 MSCI country indices, for EMU countries
with data from 1999 through 2021 (Austria, Belgium, Finland, France, Germany,
Ireland, Italy, Netherlands, Portugal, and Spain), from Datastream as well. The data
cover the period from the first to the last PC in our sample (i.e., Jan. 7, 1999, to Dec.
16, 2021) with 5,825 daily observations, of which 240 are PC days (with tone
changes) and 5,585 are non-PC days. Table IA.3 in the Supplementary Material
reports summary statistics for equity index returns over the full sample as well as
separately for non-PC days and PC days.
Table 4 provides such evidence for the ESX50 as well as the broad MSCI EMU
index and 10 EMU country indices. In Columns 1 and 2 of Table 4, we report results
from regressions of daily returns on PC day dummies and find that not a single
coefficient is significantly different from 0. Hence, there is no general premium on
PC days, unlike the FOMC premium for the U.S. as documented in Lucca and
Moench (2015). Columns 3–6 present results for regressing returns on separate
dummies for PCs with positive tone changes and negative tone changes, respec-
tively and testing whether the estimated coefficients are equal. All dummies for
positive tone changes carry a positive slope coefficient and all dummies for negative
tone changes have a negative coefficient estimate; many of the estimates for
positive and/or negative tone change dummies are significantly different from
0. Moreover, we can reject equality of coefficients (based on an Ftest) at the 5%
level for both EMU market indices and all 10 countries.
B. Regressions of Equity Returns on ECB Tone Changes
The above results suggest that there is no PC day premium in EMU equity
markets but that stocks react differently when the ECB’s tone change is positive or
negative. We now provide evidence that tone changes convey new information for
stock returns that is not subsumed by control variables that account for policy
actions, market conditions, measures of monetary policy shocks proposed in the
literature, and other textual characteristics of the PC statements. We provide rele-
vant details on the construction of the monetary policy shock variables in Appendix
Band present summary statistics for all control variables in Table IA.4 in the
Supplementary Material.
Table 5 presents regression results for the ESX50. Specification 1 regresses PC
day returns only on tone changes to provide a benchmark estimate. We find a
significantly positive effect of tone changes on returns with a coefficient estimate
of 0.44. In economic terms, a 1 Std. Dev. increase (decrease) in tone changes, where
σΔτðÞ¼0:00745, translates into a positive (negative) return of around 33 basis
points on a PC day. With 8 to 12 PCs per year, this translates into 2.6% to 3.3% p.a.,
12 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

which seems sizeable given that the average annualized return of the ESX50 during
our sample is of a similar magnitude.
Specification 2 adds lagged tone changes (to control for autocorrelation in tone
changes) and various measures of market conditions prior to the PC day (i.e., stock
returns, return volatility, implied volatility (VSTOXX), the level of interest rates
(German 2-year yield), and the term spread (German 10- minus 1-year yields)).
These controls are measured from the previous PC to the day before the current PC,
to control for the possibility that the ECB might adjust its tone to recent market
conditions (e.g., Cieslak and Vissing-Jorgensen (2021), provide such evidence for
the Federal Reserve). In essence, by controlling for lagged tone and financial
market developments prior to the PC, we are testing whether tone surprises move
stock prices.
9
These controls hardly affect the estimate and significance of the
coefficient on tone changes.
In specification 3, we also control for other textual characteristics of PC
statements, discussed in more detail in Appendix A. First, we add a proxy for the
TABLE 4
ECB Press Conferences, Tone Changes, and Equity Returns
Table 4 presents results for the role of ECB press conferences (PCs) for daily equity returns of Eurozone market and country
indices. In Panel A, we report results from regressing returns on a constant and a dummy, 1(PC), that is, 1 on days with PCs,
and 0 otherwise. In Panel B, we report results from regressing returns on a constant and separate dummies for PC days with
positive tone changes (Δτ>0) and negative tone changes (Δτ<0). Numbers in brackets are t-statistics based on White (1980)
standard errors. Additionally, we report the pvalue of an Ftest that the coefficient estimates for both dummies are equal. The
data cover the period from the first to the last PC in our sample, that is, Jan. 7, 1999, to Dec. 16, 2021, with 5,825 daily
observations including 241 PCs, that is, we have 240 PC days with tone changes.
Panel A. PC Days Panel B. PC Days with Positive Versus Negative Tone Changes
Const 1(PC) Const 1Δτ>0ðÞ 1Δτ<0ðÞ p[F]
EMU Market Indices
Eurostoxx 50 1.44 3.63 1.44 24.13 35.36 0.01
[0.85] [0.32] [0.85] [1.64] [2.11]
MSCI EMU 1.67 4.65 1.67 21.30 34.30 0.01
[1.04] [0.44] [1.04] [1.57] [2.14]
Country Indices
Austria 2.09 0.16 2.09 19.95 23.14 0.05
[1.03] [0.01] [1.03] [1.52] [1.32]
Belgium 0.45 0.58 0.45 32.92 38.86 0.00
[0.25] [0.05] [0.25] [2.74] [2.08]
Finland 2.25 6.91 2.25 40.02 30.93 0.03
[0.93] [0.42] [0.93] [2.12] [1.14]
France 2.32 6.88 2.32 20.90 38.63 0.01
[1.44] [0.64] [1.44] [1.49] [2.37]
Germany 2.18 8.75 2.18 15.49 36.44 0.02
[1.22] [0.78] [1.22] [1.01] [2.24]
Ireland 0.10 4.41 0.10 29.75 24.54 0.02
[0.05] [0.36] [0.05] [1.71] [1.54]
Italy 0.64 7.62 0.64 20.65 39.94 0.02
[0.35] [0.61] [0.35] [1.40] [1.97]
Netherlands 2.47 1.43 2.47 19.84 25.73 0.02
[1.50] [0.15] [1.50] [1.55] [1.84]
Portugal 0.48 2.69 0.48 20.92 29.67 0.00
[0.29] [0.30] [0.29] [1.88] [2.10]
Spain 0.96 0.85 0.96 27.54 29.66 0.02
[0.53] [0.07] [0.53] [1.87] [1.59]
9
In robustness checks, we repeat the empirical analysis with tone surprises which we obtain as the
residuals from autoregressive (AR) models for the level of tone, as we discuss in Section V.B.
Schmeling and Wagner 13
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

distance (DISt) of statements, which captures how much the wording of a statement
differs from that of the previous statement. DIStmight matter for asset prices if
changes in communication reflect changes in the monetary policy stance or eco-
nomic environment (also see, e.g., Ehrmann and Talmi (2020)). Second, we add
proxies for changes in readability, as measured by the FOG index (ΔFOGt), and
TABLE 5
Equity Market Returns and Changes in ECB Tone
Table 5 presents results on the link between EuroStoxx 50 returns and changes in the ECB’s communication tone. On each
ECB press conference (PC) day, we compute the change in tone (Δτt) compared to the previous PC and the equity return from
the closing prices on the day preceding the PC and the day on which the PC is held. Our sample includes a total of 240 returns
and tone changes, computed from the 241 PCs between Jan. 7, 1999 and Dec. 16, 2021. We regress returns on tone changes
and the following control variables. To control for autocorrelation in tone changes, we add lagged tone changes (Δτt1). To
control for ECB policy actions, ΔMROtdenotes the change in the policy rate announced at the PC at time tand UMPtis a
dummy that takes the value 1 for PCs at which unconventional monetary policy actions are announced, and 0 otherwise. To
control for monetary policy surprises, we use high-frequency interest rate data, either the first principal component (PC1) of
short-term interest rate changes around the press release announcing the policy rates and around the press conference or the
four factors proposed by Altavilla et al. (2019). To control for communication features other than tone, we include the distance
in the wording (DISt), change in complexity measured by the FOG index (ΔFOGt), and change in lexical diversity measuredby
the type-token ratio (ΔTTRt) of the current compared to the previous PC statement. Finally, to account for the information set of
market participants prior to the PC day, we control for the stock market return and volatility, changes in the VSTOXX, and
interest rates (level and term spread) since the previous PC. We report coefficient estimates, t-statistics based on White (1980)
standard errors in brackets, the regressions’adjusted R2, and the number of observations.
123456
Const 0.00 0.00 0.00 0.00 0.00 0.00
[0.18] [1.00] [0.63] [0.67] [0.18] [0.18]
ECB Tone
Δτt0.44 0.44 0.46 0.47 0.48 0.54
[2.34] [2.65] [2.64] [2.71] [2.93] [2.82]
Δτt10.03 0.02 0.08 0.13 0.15
[0.16] [0.12] [0.47] [0.85] [0.86]
ECB Actions
ΔMROt0.69 1.98 1.54
[0.69] [1.62] [1.04]
UMPt0.01 0.01 0.01
[1.61] [1.55] [1.63]
Monetary Policy Shocks
PC1 –press release 0.26
[2.58]
PC1 –press conference 0.07
[2.45]
Target 0.16
[1.50]
Timing 0.12
[2.51]
FG 0.00
[0.08]
QE 0.02
[0.36]
Text Controls
DISt0.00 0.00 0.00 0.00
[0.10] [0.22] [0.11] [0.12]
ΔFOGt0.05 0.04 0.03 0.06
[0.52] [0.39] [0.25] [0.50]
ΔTTRt0.04 0.04 0.05 0.08
[1.30] [1.35] [1.89] [2.59]
Pre-PC Market Controls
Market return 0.00 0.01 0.00 0.01 0.00
[0.03] [0.13] [0.03] [0.12] [0.03]
Market volatility 0.07 0.07 0.07 0.01 0.01
[0.93] [0.88] [0.89] [0.16] [0.18]
VSTOXX 0.01 0.01 0.01 0.00 0.01
[0.37] [0.33] [0.46] [0.29] [0.36]
Interest rate level 1.05 0.95 0.87 0.38 0.20
[1.85] [1.68] [1.54] [0.61] [0.30]
Term spread 0.45 0.58 0.50 0.50 0.22
[0.62] [0.81] [0.69] [0.71] [0.30]
Adj. R2(%) 3.20 7.29 6.71 7.71 14.43 11.49
No. of obs. 240 239 239 239 239 206
14 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

lexical diversity, which we measure by the type-token ratio (ΔTTRt). More complex
and lexically diverse statements are potentially harder to interpret, might increase
uncertainty, and could thus matter for asset prices. However, these three additional
characteristics turn out to be insignificant and they also do not affect the signifi-
cance of tone changes. Hence, we can rule out that tone changes matter for stocks
because they capture features of other textual characteristics.
The results of specification 4 show that policy actions taken by the ECB hardly
affect the coefficient on tone changes by controlling for changes in policy rates and
for unconventional monetary policy announcements. More specifically, we com-
pute changes in the rate on main refinancing operations (ΔMRO)
10
and construct a
dummy variable (UMPt) which takes a value of 1 when an unconventional policy
action (according to Cieslak and Schrimpf (2019)) is announced during a PC, and
0 otherwise. Neither of these policy actions are associated with regression coeffi-
cients different from 0, whereas the coefficient estimate for tone changes remains
unchanged and significantly positive.
Finally, we rule out the possibility that changes in tone capture the same
information as monetary policy shocks measured from high-frequency changes
in interest rates; we discuss the shock construction in detail in Appendix B.In
specification 5, we follow Jaroci
nski (2020) and measure shocks as the first
principal component (PC1) of changes in short-term interest rates. In the regression,
we control for shocks associated with the press release announcing policy rates and
shocks associated with the PC. In specification 6, we use the term structure-based
shock factors proposed by Altavilla et al. (2019) (i.e., their target factor for the rate
announcement as well as the timing, forward guidance (FG), and quantitative
easing (QE) factors for the PC). For both regressions, we find that coefficient
estimates for shocks reflecting responses in short-term rates during the PC are
significantly positive, but the coefficient on tone changes remains largely
unchanged and significantly positive as well.
These results show that changes in ECB tone convey new information for
EMU equity markets, which is not subsumed by policy actions, not due to market
conditions prior to PCs, and not captured by measures of monetary policy shocks. In
the Supplementary Material, we report additional results that corroborate our
findings. Repeating the regressions with ESX50 high-frequency returns measured
over different time periods of the PC day, Table IA.5 in the Supplementary Material
confirms that the significance of tone changes only arises during the PC (i.e., in the
time window from 14:30 to 17:30 CET) and not before, as already suggested by
Figure 1.
11
The results in Table 5 are also very similar to those for the broader MSCI
10
The MRO rate is the main policy rate, but using the rates of the deposit facility or the marginal
lending facility does not change the results as all three rates are highly correlated. All ECB-related data
can be obtained from the statistics section of the ECB website (https://www.ecb.europa.eu/stats/).
11
Figure 1 shows that returns on days with more positive versus negative tone start to diverge at the
beginning of the press conference. There appears to be some preannouncement effect on the day of the
press conference; however, most of this can be traced back to the COVID-19-related stock market crash
in Mar. 2020. The scheduled ECB press conference on Mar. 12 happened to coincide with the Eurostoxx
50 losing more than 10% in a single day and much of this right after the morning opening. Removing this
one outlier day from our sample substantially reduces the spread between the red and green line prior to
the PC.
Schmeling and Wagner 15
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

EMU index (Table IA.6 in the Supplementary Material) as well as for country
indices, where we find that Ireland is the only case in which equity returns are not
significantly related to tone changes in all specifications (see Table IA.7 in the
Supplementary Material).
IV. Why Does Tone Matter?
Our finding that changes in ECB tone significantly move stock markets raises
the question of why and how tone matters for asset prices. To shed light on this
question, we now study how the prices of other assets respond to changes in
ECB tone.
Our results suggest that the relation between tone changes and stock returns
cannot be explained by movements in risk-free rates but must be driven by how
tone matters for risk premia embedded in asset prices. A more positive tone is
associated with a lower option-implied risk aversion and with lower credit
spreads, in particular for financial institutions. We discuss how the tone effects
on asset prices are consistent with a risk-based channel of monetary policy and
more specifically with the linkages between credit spreads, interest rates, and
stock returns documented by Gilchrist and Zakrajšek (2012). We also show that
these tone effects are robust to controlling for policy and information shocks, as
in Jaroci
nski and Karadi (2020), which supports our conclusion that tone sur-
prises primarily affect risk premia.
A. ECB Tone and Risk-Free Government Bond Yields
A natural starting point for our further analysis of market responses to CB
communication is to consider interest rates. Using German government bonds,
Figure 3 presents results for the term structure of yield changes on ECB PC
days.
12
Graph A of Figure 3 shows that, on average across all PC days (dashed blue
line), yields of all maturities increase and more so for longer as compared to shorter
maturities. When we separate PC days with positive (green) and negative (red) tone
changes, we see a similar slope effect for both, but the level of yield changes is
significantly different across all maturities: When ECB tone becomes more posi-
tive, all yields increase and more so for longer maturities. When ECB tone becomes
more negative, yields of shorter maturities decrease whereas yields of longer
maturities increase on average. Graph B presents results from regressing yield
changes on tone changes on PC days as well as our standard control variables for
other textual characteristics, policy actions, market conditions, and monetary policy
shocks. We plot the tone coefficient estimates along with 95% confidence intervals
and find that estimates are positive for all maturities with the link being statistically
significant for maturities up to 5 years.
12
We use Bundesbank data for the term structure of German government bond yields. These data are
available over our full sample period, whereas European yield data available from the ECB only start in
2004. Over the joint sample period, the German yield curve is highly correlated with the ECB AAAyield
curve.
16 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

These results show that a more positive ECB tone is not only associated with
higher stock prices but also with higher interest rates. Finding that interest rates and
equity prices respond to tone changes in the same direction is interesting for two
reasons. On the one hand, this finding suggests that tone does not move stock prices
through a simple risk-free rate discount effect. Instead, stock returns in response to
changes in ECB tone appear to reflect news about risk premia, and we provide
evidence supporting this notion below. On the other hand, the positive co-movement
of interest rates and equity prices appears similar to that underlying the identification
of “central bank information shocks”(e.g., Nakamura and Steinsson (2018), Jaro-
ci
nski and Karadi (2020)) as we discuss in more detail in Section IV.D.
FIGURE 3
Government Yield Changes on ECB Press Conference Days
Figure 3 presents results on changes in the German government bond yield curve (for maturities ranging from 1 to 20 years,
x-axis) in response to changes in ECB tone. Graph A presents average PC day yield changes in basis points, for all PC days
(dashed lined in blue) as well as conditional on the tone changes at the most recent PC having been positive (green triangles)
or negative (red bullets). Graph B presents results from regressing PC day yield changes on changes in ECB tone (Δτ) as well
as our standard control variables for other textual characteristics, policy actions, market conditions, and monetary policy
shocks. We plot the slope coefficients for tone changes, along with 95% confidence bands (based on White (1980) standard
errors). The sample spans a total of 240 tone changes from 241 ECB press conferences between Jan. 1999 and Dec. 2021.
Graph A. Positive Versus Negative Tone Changes on PC Days
0.0
0.5
1.0
1.5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Δτ>0
Δτ<0
Maturity in Years
Change (in bp)
Graph B. Regressions of PC Day Yield Changes on Tone Changes
0.000
0.005
0.010
0.015
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Maturit
y
in Years
Coefcient Estimate
Schmeling and Wagner 17
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

B. Does Tone Matter for Risk Premia? Evidence from Options
Our findings in Section III suggest that investors adjust their expectations for the
stock market return in response to changes in ECB tone. Conceptually, such adjust-
ments may be driven by changes in the quantity of risk that investors face or the
premium they require per unit of risk. To analyze these different dimensions, we assess
the realized volatility of ESX50 returns, changes in index options-implied volatility,
and the link between realized volatility and changes in implied volatility.
13
We follow
Bekaert et al. (2013) and Bekaert, Engström, and Xu (2022), who propose to measure
time variation in risk aversion via variance risk premia implied by equity options.
Bekaert et al. (2013) show that unexpected monetary policy easing is associated with a
decrease in variance risk premia, which implies a lower risk aversion by market
participants. Similarly, we find that a more positive CB tone is associated with a
significant decrease in options-implied volatility as well as in volatility risk premia.
1. Realized Volatility, Implied Volatility, and Risk Premia
First, we use high-frequency data to compute the realized volatility (RV) of the
ESX50 for each trading day in our sample, following the approach of Bollerslev,
Hood, Huss, and Pedersen (2018).
14
For each day, we also compute the realized
volatility from 14:30 to 17:30 (RVPC), which captures the time window of the PC on
ECB announcement days. Using both estimates, we check whether realized vola-
tility is different on PC and non-PC days and whether realized volatility is different
on PC days with positive compared to negative tone changes.
Panel A in Table 6 reports the results from regressing RV or RVPC on PC and
PC tone change dummies. We find that realized volatility is significantly higher on
PC days compared to non-PC days by about 15 basis points over the full trading day
and by about 17 basis points in the time period from 14:30 to 17:30. However, the
sign of ECB tone changes does not appear to matter for realized volatility, as we are
far from rejecting the null hypothesis of equal coefficients when we regress RV and
RVPC on separate dummies for PCs with positive and negative tone changes; the p
values of the Ftests are 0.30 for RV and 0.56 for RVPC.
Next, we compute changes in index options-implied volatility, measured by
the VSTOXX, which is a volatility index computed from options on the ESX50,
similar to the VIX based on S&P 500 options in the U.S.
15
The VSTOXX can be
interpreted as a price of volatility insurance, since VSTOXX is the fixed leg in a
volatility swap that pays the difference in implied volatility and future realized
volatility of the ESX50. To analyze whether ECB tone matters for the pricing of
insurance against future volatility, we compute log changes in VSTOXX from the
13
For summary statistics of all volatility quantities, see Table IA.8 in the Supplementary Material.
14
For each day in our sample, i) we compute 5 daily series of squared 5-min log returns, starting at the
first five unique 1-min marks, respectively; ii) we compute the sum of squared returns for each of the five
series; iii) we obtain that day’s estimate of realized variance as the average of the five sums; iv) we take
the square root to obtain our estimate of realized volatility. Bollerslev et al. (2018) provide a discussion
that this procedure provides an efficient estimate of realized volatility.
15
The VSTOXX is designed to make pure volatility tradable and to be replicable by options
portfolios that do not react to ESX50 price changes but only to volatility changes. The VSTOXX is
computed from maturity-specific subindices, which themselves are computed from ESX50 options in
predefined maturity buckets and across moneyness levels. For details, see the STOXX (2018).
18 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

TABLE 6
Realized Versus Implied Volatility and Changes in ECB Tone
Table 6 presents results on the link between stock market volatility and changes in the ECB’s communication tone. For each
day in our sample, we measure the realized volatility of the Eurostoxx 50 from intraday data over the full day (RV) and over the
time window from 14:30 to 17:30 (RVPC). We measure changes in implied volatility as daily log changes in the VSTOXX,
Δlog VSTOXXðÞ. Finally, as a proxy for changes in the VRP, we compute the ratio of changes in implied volatility to realized
volatility. On each ECB press conference (PC) day, we compute the change in tone (Δτt) compared to the previous PC. The
data cover the period from the first to the last PC in our sample, that is, Jan. 7, 1999, to Dec. 16, 2021. Panel A reports results
from using all days in our sample. In columns 1 and 2 of Panel A, we report results from regressing the volatility quantities on a
constant and a dummy, 1(PC), that is 1 on days with PCs, and 0 otherwise. In columns 3–6, we report results for regressions on
a constant and separate dummies for PC days with positive tone changes (Δτ>0) and negative tone changes (Δτ<0).
Numbers in brackets are t-statistics based on White (1980) standard errors. Additionally, we report the pvalue of an Ftest that
the coefficient estimates for both dummies are equal. Panel B reports results for PC day regressions of changes in implied
volatility and changes in volatility risk premia on tone changes and a set of control variables; for detailed variable descriptions,
we refer to Table 5.
Panel A. ECB Press Conferences and Tone Changes
PC Days PC Days with Positive Versus Negative Tone Changes
Const 1(PC) Const 1Δτ>0ðÞ 1Δτ<0ðÞ p[F]
Realized Volatility
Trading day RV 97.67 15.27 97.67 11.33 19.78 0.30
[44.98] [4.26] [44.98] [2.44] [3.19]
From 14:30 to 17:30RVPC 61.15 17.48 61.15 15.84 19.35 0.56
[43.76] [6.48] [43.76] [4.34] [4.40]
Changes in Implied Volatility
Δlog VSTOXXðÞ 4.23 121.72 4.23 200.14 32.10 0.05
[0.59] [2.81] [0.59] [3.40] [0.52]
Proxies for Volatility Risk Premia
Δlog VSTOXXðÞ=RV 0.34 1.52 0.34 2.17 0.78 0.09
[4.06] [3.50] [4.06] [3.45] [1.39]
Δlog VSTOXXðÞ=RVPC 0.34 1.52 0.34 2.17 0.78 0.09
[4.06] [3.50] [4.06] [3.45] [1.39]
Panel B. Regressions on ECB Tone Changes
Δlog VSTOXXðÞ Δlog VSTOXXðÞ=RV Δlog VSTOXXðÞ=RVPC
123123123
Const 0.02 0.01 0.01 0.77 0.35 0.28 1.70 0.90 0.93
[0.91] [0.59] [0.49] [0.32] [0.15] [0.11] [0.47] [0.25] [0.24]
ECB Tone
Δτt1.48 1.49 2.08 134.49 135.30 179.38 204.94 206.59 273.39
[2.74] [2.84] [3.17] [2.69] [2.72] [3.01] [2.77] [2.86] [3.10]
Δτt10.34 0.47 0.50 11.34 16.49 37.91 18.69 21.05 55.99
[0.55] [0.79] [0.69] [0.20] [0.27] [0.53] [0.22] [0.23] [0.52]
ECB Actions
ΔMROt0.37 3.93 3.84 170.51 2.32 154.60 295.39 6.98 259.02
[0.14] [1.19] [0.99] [0.81] [0.01] [0.47] [0.91] [0.02] [0.53]
UMPt0.05 0.04 0.04 1.75 1.68 1.19 2.15 2.04 1.36
[2.39] [2.29] [1.99] [1.25] [1.18] [0.74] [1.20] [1.11] [0.64]
Monetary Policy Shocks
PC1 –press release 0.73 35.98 65.59
[2.52] [2.40] [3.08]
PC1 –press conference 0.19 7.90 7.45
[1.61] [0.83] [0.51]
Target 0.39 21.18 33.45
[1.42] [1.36] [1.52]
Timing 0.42 31.32 39.67
[1.97] [1.81] [1.60]
FG 0.02 10.51 18.92
[0.18] [0.89] [0.99]
QE 0.46 58.24 91.81
[2.00] [2.72] [2.69]
Text Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
Pre-PC Market Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj. R2(%) 3.20 6.33 4.12 2.16 2.21 3.66 1.21 1.48 2.28
No. of obs. 239 239 206 239 239 206 239 239 206
Schmeling and Wagner 19
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

close on the day before the PC to the close on the PC day (i.e., the timing is exactly
the same as in our analysis of stock returns above (e.g., Table 5)).
The results in Panel A of Table 6 show that implied volatility significantly
decreases on PC days by about 1.2%.
16
However, once we distinguish between
PCs with positive and negative tone changes, we find that implied volatility
significantly decreases only on days with positive tone changes (by 2.00%),
whereas it is not different from 0 on PC days with negative tone changes; accord-
ingly, we can reject the hypothesis of equal dummy coefficients with a pvalue of
0.05. Hence, our results suggest that volatility insurance becomes cheaper when
ECB tone becomes more positive.
The above findings are intriguing, because they suggest that ECB tone matters
for the volatility risk premium (VRP) and hence for investors’risk aversion. Changes
in implied volatility are either due to changes in expected future volatility or changes
in the VRP that investors are willing to pay on top of expected volatility. Given that
realized volatility is not significantly different on PC days with positive and negative
tone changes, it seems unlikely that ECB tone affects expectations about future
realized volatility, and we provide more evidence for this view below. Instead,
ECB tone appears to affect the VSTOXX through changes in volatility risk premia.
To assess changes in the VRP, we compute log changes in the VSTOXX
relative to realized volatility, using both RV and RVPC.
17
Similar to the VSTOXX,
we find that VRPs decrease on PC days and that once we control for the sign of ECB
tone changes, this is mostly due to PCs with positive tone changes.
2. Regressions on ECB Tone Changes
To provide further evidence for a link of implied volatility and volatility risk
premia to ECB tone, we run regressions of changes in VSTOXX and VRPs on tone
changes and the set of control variables that we have also used in our analysis of
stock returns above. The results in Panel B of Table 6 show that the coefficient
estimate for tone changes is significantly negative in all specifications, which
implies that a positive tone surprise is associated with lower volatility risk premia
and a decline in the pricing of volatility insurance.
18
Beyond these significant tone effects, we also find (mostly significant) neg-
ative coefficients for UMPtin the VSTOXX regressions, which suggests that
announcements of unconventional policy actions reduce options-implied volatility,
in line with, for example, Hattori, Schrimpf, and Sushko (2016). Moreover, we find
16
This finding is similar to the negative VIX changes on FOMC announcement days documented by
Boguth et al. (2018). To the extent that such decreases in VIX reflect a reduction in uncertainty, one can
rationalize announcement premia in the theoretical framework of Ai and Bansal (2018). Recall, however,
from Section III that we do not find significant ECB announcement day effects in the ESX50.
17
Our goal is to track changes in VRP at high frequency. Ideally, we would like to measure VRP from
a 1-day volatility swap that pays the difference between 1-day VSTOXX (fixed leg) and realized
volatility over the PC day (floating leg), but unfortunately such contracts do not exist. To assess whether
VRP increases or decreases, we compare the 1-day change in the VSTOXX relative to realized volatility.
To rule out the hypothetical case that tone changes may not affect RV and RVPC but realized volatility
going forward, we verify that there are no tone-related patterns in realized volatility over the next week,
month, and 3 months; see Table IA.9 in the Supplementary Material.
18
Table IA.10 in the Supplementary Material additionally reports regression estimates for all control
variables.
20 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

that changes in implied volatility and volatility risk premia are positively related to
interest rate-based measures of monetary policy shocks, either to the “PC1 –press
release”shock in specification 2 or the “QE factor”shock in specification 3. Put
differently, unexpected monetary policy tightening is associated with higher
implied volatility and risk premia. Additionally, we repeat the regression analysis
for different VSTOXX maturities, ranging from 1 month to 2 years. Figure 4
illustrates that the estimated coefficients are significantly negative and monotoni-
cally increase with maturity, except for a small twist at the 1-year horizon. These
results suggest that communication tone has a stronger impact on short-term
compared to longer-term risk premia.
Hence, akin to the finding of Bekaert et al. (2013) that monetary easing
decreases variance risk premia, we find that a more positive communication tone
is associated with a significant decrease in volatility risk premia. Considering
that we control for policy actions, changes in market conditions since the last PC
day, and yield-based monetary policy shocks, our results suggests that changes in
ECB tone affect risk premia embedded in asset prices. In other words, ECB tone
matters for asset prices through a risk-based channel by affecting investor risk
aversion.
C. ECB Tone and Corporate Credit Spreads
The results above show that there is a link between CB tone and economic
fundamentals and that tone matters for asset prices through risk premia. To better
understand this combination of results, we now study the relation between ECB
tone and credit spreads, motivated by previous evidence that changes in credit
spreads are driven by risk premia and reflect the risk-bearing capacity of financial
intermediaries.
Gilchrist and Zakrajšek (2012) study the interrelations between credit
spreads, economic activity, and monetary policy. First, they show that the pre-
dictive relation between credit spreads and economic activity is driven by the
spreads’embedded risk premia, which also account for most of the spreads’
variation. Second, they argue that increases in credit spreads reflect a reduction
in the effective risk-bearing capacity of the financial sector, which in turn leads to
a reduction in credit supply, a contraction in economic activity, a decline in
interest rates, and a fall in stock markets. Third, they provide evidence that shocks
to credit spreads are linked to the deterioration in the profitability and creditwor-
thiness of broker–dealers, who are the marginal investors in corporate debt
markets. The results of Gilchrist and Zakrajšek (2012) are consistent with earlier
evidence that changes in monetary policy that affect the risk-bearing capacity of
intermediaries will directly matter for asset prices, such that looser policy leads to
a lower price of risk (see, e.g., Adrian and Shin (2008) and Adrian, Moench, and
Shin (2010)). For a recent survey of this “risk-taking channel”of monetary policy,
see Adrian and Liang (2018). More generally, the idea that financial intermedi-
aries are the marginal investors in asset markets and therefore play a crucial role
for the pricing of assets is central to the recent literature on intermediary asset
pricing (see, e.g., He and Krishnamurthy (2013), Adrian, Etula, and Muir (2014),
and He, Kelly and Manela (2017)).
Schmeling and Wagner 21
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

To analyze whether changes in ECB tone matter for EMU credit spreads, we
obtain data on IBOXX credit indices to compute corporate yield spread differ-
entials between BBB- and AAA-rated firms.
19
Tab le 7 presents results for broad
credit indices and for indices covering either financial or nonfinancial firms.
Panel A shows that credit spreads tend to decrease on PC days, but the only
significant effect we find is for financial firms (1.39 basis points, t-statistics
of 2.36). When we test for differences in PC day effects conditional on tone
becoming more positive or negative, we find a significant difference for financial
firms (pvalue 0.05), where a more positive tone is associated with a spread
decrease of 2.4 basis points. Using the same dummy regressions, we find
FIGURE 4
Changes in ECB Tone and Term Structures of Volatility Changes
Figure 4 shows the slope coefficient estimates from regressions of changes in implied volatility and proxies for volatility risk
premia on changes in ECB tone and our standard set of control variables. Our sample covers the 241 press conferences(PCs)
held by the ECB between Jan. 7, 1999, to Dec. 16, 2021, from which we compute 240 changes in ECB tone. On each PC day,
we measure the change in implied volatility as the daily log change in the VSTOXX, Δlog VSTOXXðÞ, from the closing values of
the day preceding the PC and the day on which the PC is held. As proxies for changes in volatility risk premia, we scale
changes in implied volatility by the realized volatility, computed from high-frequency data either over the full PC day (ΔVRP) or
over the time window from 14:30 to 17:30 (ΔVRPPC). We compute changes in implied volatility and volatility risk premia using
VSTOXX indices with maturities between 1 month and 24 months and present coefficient estimates (solid line with bullets)
along with 95% confidence bands (dashed lines, based on White (1980) standard errors).
–3.5
–3.0
–2.5
–2.0
–1.5
–1.0
–0.5
0.0
1m 2m 3m 6m
Graph A. Tone Sensitivities of Changes in
Implied Volatility (VSTOXX)
Δlog(VSTOXX)
Graph C. Tone Sensitivities of Changes in
Volatility Risk Premia (ΔVRPPC)
Δlog(VSTOXX)/RVPC
9m 12m 18m 24m
Graph B. Tone Sensitivities of Changes in
Volatility Risk Premia (ΔVRP)
Δlog(VSTOXX)/RV
–300
–200
–100
0
1m 2m 3m 6m 9m 12m 18m 24m
–500
–400
–300
–200
–100
0
1m 2m 3m 6m 9m 12m 18m 24m
19
For summary statistics of changes in credit spreads, see Table IA.11 in the Supplementary Material.
For most of their empirical analysis, Gilchrist and Zakrajšek (2012) use the excess bond return of their
self-constructed credit index, because it is the best predictor of future economic activity in their
U.S. sample. For the BBB–AAA spread, they find that the predictive ability is less significant but
qualitatively the same. We use the BBB–AAA spread because Krylova (2016) finds that the BBB–AAA
spread mostly dominates alternative corporate spread measures as leading indicator for the Eurozone.
More recently, Gilchrist and Mojon (2018) provide credit risk indices for the euro area.
22 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

weaker results for the credit spreads of all firms (pvalue of 0.11) and no PC
effects for nonfinancial firms (pvalue of 0.30).
Turning to the regression analysis in Panel B of Table 7, we obtain a similar
picture but with more pronounced results.
20
There is a negative relation between
TABLE 7
Corporate Credit Spreads and Changes in ECB Tone
Table 7 presents results on the link between changes in corporate credit spreads and changes in the ECB’s communication
tone. For each day in our sample, we compute changes in credit spreads, defined as the yield differentials of BBB- and AAA-
rated bonds of all corporates and separately for financials and nonfinancials. On each ECB press conference (PC) day, we
compute the change in tone (Δτt) compared to the previous PC. Daily data on credit spreads of corporates and financials are
available from Apr. 1999 and for nonfinancials from Aug. 1999 to Dec. 2021. Panel A reports results from using all days in our
sample. In Columns 1 and 2 of Panel A, we report results from regressing changes in credit spreads on a constant and a
dummy, 1(PC), that is 1 on days with PCs, and 0 otherwise. In columns 3–6, we report results for regressions on a constant and
separate dummies for PC days with positive tone changes (Δτ>0) and negative tone changes (Δτ<0). Numbers in brackets
are t-statistics based on White (1980) standard errors. Additionally, we report the pvalue of an Ftest that the coefficient
estimates for both dummies are equal. Panel B reports results for PC day regressions of changes in credit spreads on tone
changes and a set of control variables; for detailed variable descriptions, we refer to Table 5.
Panel A. ECB Press Conferences and Tone Changes
PC Days PC Days with Positive Versus Negative Tone Changes
Const 1(PC) Const 1Δτ>0ðÞ 1Δτ<0ðÞ p[F]
All corporates 0.02 0.60 0.02 1.13 0.01 0.11
[0.31] [1.64] [0.31] [1.74] [0.04]
Financials 0.06 1.39 0.06 2.44 0.18 0.05
[0.27] [2.36] [0.27] [2.48] [0.30]
Nonfinancials 0.01 0.27 0.01 0.67 0.18 0.30
[0.09] [0.62] [0.09] [0.86] [0.75]
Panel B. Regressions on ECB Tone Changes
All Corporates Financials Nonfinancials
123123123
Const 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
[1.18] [1.32] [1.01] [0.50] [0.25] [0.70] [1.17] [1.08] [1.09]
ECB Tone
Δτt0.01 0.01 0.01 0.02 0.02 0.02 0.01 0.01 0.00
[2.48] [2.51] [1.87] [2.75] [2.78] [2.47] [1.81] [1.91] [0.86]
Δτt10.01 0.01 0.01 0.02 0.02 0.00 0.01 0.01 0.01
[1.89] [1.95] [1.46] [1.81] [2.11] [0.34] [2.22] [2.05] [1.85]
ECB Actions
ΔMROt0.09 0.10 0.13 0.16 0.18 0.29 0.01 0.00 0.01
[1.50] [1.39] [1.40] [1.89] [1.72] [2.21] [0.68] [0.10] [0.26]
UMPt0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
[1.83] [1.92] [2.11] [1.22] [1.34] [1.27] [2.28] [2.06] [2.82]
Monetary Policy Shocks
PC1 –press release 0.00 0.00 0.00
[0.78] [0.84] [1.51]
PC1 –press conference 0.00 0.00 0.00
[0.15] [0.80] [0.38]
Target 0.00 0.01 0.00
[1.14] [2.34] [0.05]
Timing 0.00 0.00 0.00
[0.28] [0.98] [0.39]
FG 0.00 0.00 0.00
[0.79] [0.68] [0.88]
QE 0.01 0.01 0.01
[1.80] [1.33] [1.49]
Text Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
Pre-PC Market Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj. R2(%) 4.14 3.70 5.05 12.63 12.94 19.69 2.49 1.88 1.28
No. of obs. 239 239 206 239 239 206 235 235 206
20
Table IA.12 in the Supplementary Material additionally reports regression estimates for all control
variables.
Schmeling and Wagner 23
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

changes in credit spreads and changes in ECB tone, with the link being most
significant for spreads of financial firms.
21
Among the control variables, we note
that the ECB’s policy actions have a significant impact on credit spreads as well.
UMP announcements significantly lower credit spreads (in line with, e.g.,
Chodorow-Reich (2014)) in the sample of all corporates, mostly driven by the
impact on the spreads of nonfinancial firms. For financials, we find in our most
comprehensive specification that changes in credit spreads are positively related to
changes in the policy rate and negatively related to target shocks. Controlling for
these and other effects, the coefficient estimate on tone changes is significantly
negative in all specifications for financial firms (t-stats between 2.47 and 2.78),
mostly significant for the set of all firms (t-stats between 1.87 and 2.78) but less
so for nonfinancial firms (t-stats between 0.86 and 1.91).
Taken together, the confluence of our results suggests that the answer to the
question how and why tone matters for asset prices is a risk-based channel. We find
that tone affects risk premia very similarly to policy actions, as shown by, for
example, Bernanke and Kuttner (2005) for stocks, Bekaert et al. (2013) for variance
risk premia, and Gertler and Karadi (2015) for credit spreads. Put differently, CB
tone moves asset prices because it seems to affect the risk aversion of market
participants. More specifically, the ECB tone-related linkages we document
between stock returns, interest rates, and credit spreads are qualitatively the same
as those that arise in Gilchrist and Zakrajšek (2012) due to shocks to intermediary
risk-bearing capacities. Our finding that the results are more pronounced for the
credit spreads of financial institutions than for nonfinancial corporations provides
further evidence for a risk-based channel of CB tone and suggests a link between
CB communication, intermediaries, and asset prices.
D. CB Tone and Information Shocks
In our empirical analysis, we find that changes in CB tone move stock prices
and interest rates in the same direction. We now connect this finding to recent
research on CB information effects, which has proposed to use the co-movement
between stocks and interest rates around CB announcements to distinguish between
“policy shocks”and “information shocks”(e.g., Nakamura and Steinsson (2018),
Jaroci
nski and Karadi (2020)).
We use the (updated) shock series of Jaroci
nski and Karadi (JK) (2020) who
identify policy and information shocks from the high-frequency co-movement of
interest rates and stock prices via sign restrictions; for details, see Appendix B. The
intuition is as follows: A monetary policy shock (i.e., an unexpected tightening or
easing of the monetary policy stance) should move stock prices and interest rates in
opposite directions. That is, a tightening shock should increase discount rates and,
as a consequence, decrease stock prices. By contrast, interest rates and stock prices
co-move in the same direction in case of an information shock, that is, if an
announcement reveals unexpectedly good (bad) news about economic conditions,
this will drive up (down) both stocks and interest rates.
21
We note that lagged tone changes are significant in some specifications as well and account for this
in robustness checks that we present in Section V.B..
24 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

Our finding that tone surprises move stock prices and interest rates in the same
direction raises the question whether these tone effects are similar to the JK
information shocks. To address this question for stocks and interest rates as well
as for the tone effects in volatility and credit markets, we extend the regression
analysis by including the JK policy and information shocks as additional variables.
Table 8 reports the main results across asset classes using the Eurostoxx 50 (as in
Table 5), the 2-year German government bond yield (as in Figure 3), options-
implied volatility and volatility risk premia (as in Table 6), and the credit spreads
of financials (as in Table 7). These results show that the coefficient estimates for
tone changes remain significant in all asset classes.
For interest rates and credit spreads, we find that the results are virtually
unchanged when including JK policy and information shocks. For returns on the
Eurostoxx 50 as well as changes in the VSTOXX and in volatility risk premia, we
find that including policy and information shocks substantially increases the
adjusted R2values and somewhat reduces the magnitude of the tone coefficients,
but the tone effects remain statistically significant. The coefficient estimates for,
both, policy and information shocks are statistically significant as well, with the
estimates for information shocks having the same sign as the estimates for the tone
surprises, whereas the policy shock coefficients take the opposite sign.
TABLE 8
ECB Tone and Policy Shocks Versus Information Shocks
Table 8 presents results on the link between asset price responses and changes in ECB tone, controlling for “policy shocks”
and “information shocks”as proposed by Jaroci
nski and Karadi (2020). Each column refers to a different asset class: ESX50
refers to returns in the Eurostoxx50 equity index, 2Y to changes in the German 2-year government bond yield, VSTOXX to log
changes in the VSTOXX volatility index, VRP and VRP-PC to changes in the proxies for variance risk premia (i.e., log changes in
the VSTOXX scaled by realized volatility measured over the full day or from 14:30 to 17:00), and Credit-Fin to changes in the
credit spread of financial institutions. On each press conference, we regress the asset price responses on changes in ECB
tone compared to the previous press conference, a large set of control variables (for detailed variable descriptions we refer to
Table 5), as well as policy and information shocks. We report coefficient estimates, t-statistics based on White (1980) standard
errors in brackets, the regressions’adjusted R2, and the number of observations.
ESX50 2Y VSTOXX VRP VRP-PC Credit-Fin
Const 0.00 0.00 0.02 0.97 1.86 0.00
[1.06] [1.01] [1.08] [0.40] [0.51] [0.30]
ECB Tone
Δτt0.27 0.01 0.96 94.80 151.91 0.02
[2.27] [2.75] [2.05] [1.96] [2.13] [2.68]
Δτt10.02 0.00 0.08 6.95 11.56 0.02
[0.20] [0.40] [0.15] [0.12] [0.14] [2.08]
ECB Actions
ΔMROt0.25 0.04 1.97 360.63 512.42 0.18
[0.22] [1.81] [0.64] [1.51] [1.41] [2.04]
UMPt0.00 0.00 0.03 0.86 0.91 0.00
[1.07] [2.32] [2.01] [0.73] [0.58] [1.21]
Monetary Policy Shocks
Policy shock 0.23 0.00 0.58 43.10 64.43 0.00
[5.12] [0.07] [3.40] [4.33] [4.48] [0.50]
Information shock 0.25 0.00 0.61 49.07 59.52 0.00
[6.05] [0.66] [3.39] [3.97] [2.96] [1.83]
Text Controls Yes Yes Yes Yes Yes Yes
Pre-PC Market Controls Yes Yes Yes Yes Yes Yes
Adj. R2(%) 38.30 5.64 16.23 9.59 6.88 13.56
No. of obs. 239 239 239 239 239 239
Schmeling and Wagner 25
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

From a conceptual perspective, it is difficult to say whether tone shocks are a
new, text-based proxy for CB information effects, capturing different aspects than
the JK information shocks, or whether tone surprises affect asset prices through a
separate channel. In other words, it could be that tone and information shocks are
(imperfectly correlated) proxies for the same underlying effect.
22
Drawing a clear
distinction between these two possibilities presents a challenge, and we defer this
task to future research since it is not necessary for the purpose of our study. What we
show is that CB tone indeed matters for asset prices, also when controlling for the
policy and information shock measures proposed by Jaroci
nski and Karadi (2020).
V. Additional Results and Robustness Tests
This section summarizes additional results and robustness checks, which we
present in the Supplementary Material.
A. Robustness over Subsample Periods
To show that our results are not driven by a particular period in our sample
(e.g., the financial crisis), we repeat the empirical analysis for 18 6-year subsamples.
Figure IA.1 in the Supplementary Material shows that there is a positive spread in
stock market returns on days with positive compared to negative tone changes in
17 of the 18 subsamples. For interest rates, we find that positive (negative) tone
changes are typically associated with increases (decreases) in the 2-year yield of
German government bonds or, at least, less of a decrease (less of an increase).
Moreover, we find that a more positive tone is usually associated with a decrease in
the VSTOXX, whereas a negative tone change is associated with an increase or a
smaller decrease. The inverse relation between tone changes and credit spreads of
financial firms appears to have started in 2009, that is, after the onset of the financial
crisis when investors became particularly concerned with the health of financial
institutions and have become less important in recent years. Taken together, these
results show how tone effects vary over time and corroborate our conclusion that
tone conveys information that matters for asset prices through a risk-based channel.
B. Surprises Based on AR-Models of CB Tone
In our main analysis, we have studied surprises in ECB tone via regressions of
asset price responses on tone changes, lagged tone changes, and proxies for the
information set of market participants prior to the PC. The advantage of using tone
changes and control variables is that all variables are observable in real time and no
separate estimation is required to obtain tone surprises, thereby avoiding generated
regressor issues. An alternative approach is to model ECB tone as an AR process in
a first step, potentially including other variables as well, and to obtain tone surprises
as the residuals from the AR model. We then use these residuals as tone surprises in
a second step, in which we relate changes in asset prices to these tone surprises.
22
The sample correlation of tone changes with policy shocks is 0.032 and with information shocks
is 0.167.
26 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

We now repeat the empirical analysis using AR(1), AR(3), and AR(5) models
for the level of ECB tone in two specifications. In the first specification, we only
include lags of ECB tone on the right-hand side of the regression. In the second
specification, we extend the AR models to also include other information available
to market participants prior to the PC, that is, the stock market and interest rate
quantities that we have used as control variables for market conditions in our main
analysis. We then regress asset price changes on the AR model tone surprises as well
as the control variables we have included in our main analysis (except for the
variables already incorporated in the respective AR models). We report the results
in Tables IA.13 and IA.14 in the Supplementary Material, which show that the tone
coefficient estimates have very similar magnitudes and levels of significance as in
the regressions reported in Sections III and IV.
C. Tone Surprises, Policy Actions, and Fundamentals
Our results suggest that CB tone moves asset prices through a risk-based
channel, with tone surprises affecting risk premia required by market participants.
In additional analyses, we check whether tone could also matter for asset prices by
signaling news about the future path of monetary policy or economic fundamentals.
We discuss our findings below but delegate details of the econometric setup to
Section IA.B of the Supplementary Material.
First, we show that (lagged) tone changes have some predictive power for
future policy rate changes over and above the information contained in lagged
MRO changes: A more positive (negative) tone predicts future increases
(decreases) in policy rates. This finding is generally consistent with CB tone
surprises being informative about the future policy stance, but we have already
shown above that tone-related risk premium effects dominate risk-free discount rate
effects.
Second, we use predictive regressions to study whether tone changes are
informative about future fundamentals. The signs of the estimated coefficients
support the intuition that a more positive tone is associated with better economic
conditions, which is in line with our finding that tone surprises share some simi-
larities with CB information effects. However, we only find some degree of sig-
nificance for growth in (real) industrial production and, somewhat more
pronounced, for business confidence, which may be indicative of either cash flow
or risk premium effects; all other estimates are insignificant. Hence, it appears
unlikely that tone surprises matter for asset prices mostly due to predictive power
for future fundamentals.
D. Sovereign Yield Spreads
In an additional empirical exercise, we examine the effect of tone surprises on
the term structures of sovereign yield spreads on Italian and Spanish government
bonds compared to German government bonds. Repeating the analysis of the term
structure of German government bonds (from Section IV.A), we now present
analogous results for sovereign yield spread curves in Figure IA.2 in the Supple-
mentary Material. To have full sample coverage from 1999 to 2021, we focus on
maturities of 1–10 years. Economically, our findings are consistent with a risk-
Schmeling and Wagner 27
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

based channel of tone, that is, we find in Panel A that a more positive tone is, on
average, associated with lower sovereign yield spreads. Panel B shows the results
from regressing sovereign yield spread changes on changes in ECB tone (Δτ)as
well as our standard control variables for other textual characteristics, policy
actions, market conditions, and monetary policy shocks. With these control vari-
ables, all estimates for maturities of 2 years or longer are negative, but most are not
significantly different from 0 at the 5% level.
VI. Conclusion
We propose to assess market-relevant news in CB announcements directly
from the verbal communication of policy makers. Using a systematic approach to
measure the tone of the ECB president in PCs held after policy meetings, we find
that a positive tone surprise is associated with significantly higher stock prices,
higher interest rates, lower volatility risk premia, and lower credit spreads. These
tone effects are robust to controlling for policy actions and standard measures of
monetary policy shocks, which implies that tone surprises convey price-relevant
news to the market.
Our findings contribute to the debate on effective CB communication. We
show that a simple measure of CB tone conveys news for asset prices through a risk-
based channel, similarly to the risk premium effects of policy actions. Thus, from
the perspective of policy makers, our results imply that communication tone is an
important instrument in the monetary policy toolkit. By tilting their communication
to have a more positive or negative tone, central bankers can affect the risk appetite
of market participants and, thereby, financial conditions and the pricing of risky
assets. This may prove particularly useful in times of high uncertainty or when the
scope for policy actions is limited.
Appendix A. ECB Statements and Textual Characteristics
The transcripts of the ECB press conferences are publicly available on the ECB
website (https://www.ecb.europa.eu/press/pressconf/).
After obtaining the transcripts, we follow standard procedures of the textual
analysis literature in preparing the transcripts for further analysis: We i) convert all
words to lowercase, ii) remove numbers, iii) remove punctuation, iv) remove English
stop words (e.g., for, very, and, of, are, etc.), and v) strip excessive whitespace.
The objective of our article is to assess how surprises in ECB affect asset prices.
Throughout our empirical analysis, we control for other textual characteristics of ECB
press conference statements to rule out that tone changes matter for asset prices because
they capture features of these other characteristics, which are described in more detail
below.
First, we compute the “distance”(DISt) between two consecutive PC statements.
DIStis based on the Euclidean distance between two vectors (one for each PC) where
each vector counts the number of occurrences of each word (each word is represented
by 1 row in the 2 vectors). Thus, larger values imply larger differences in the wording
28 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

used in the two PCs.
23
We control for DIStbecause Bholat, Santos, and Schonhardt-
Bailey (2015) and Ehrmann and Talmi (2020) suggest that CB communication might
affect markets differently depending on how much it deviates from the previous
communication.
Second, we use the FOG index to quantify the complexity/readability of ECB
statements and measure changes in the index (ΔFOGt) between consecutive PCs. The
FOG index aims at measuring the number of years of education needed to understand a
text on first reading and is computed from the text’s average number of words per
sentence and its percentage of complex words (defined as words with more than
2 syllables). For more details about the FOG index, its application in financial research,
and alternative readability measures (that in our application yield very similar results),
see the survey of Loughran and McDonald (2016). Third, and somewhat related, we
control for changes in the lexical diversity of ECB statements. Specifically, we compute
the type-token ratio, that is, the ratio of unique words (types) to total words (tokens) and
its changes (ΔTTRt) between consecutive PCs.
We con t r o l f or ΔFOGtand ΔTTRtto account for the possibility that tone changes
may be correlated with changes in complexity and lexical diversity. Complex and lexi-
cally diverse statements may have a worse “signal-to-noise”ratio than simpler statements,
which could impact on the effectiveness of central bank (CB) communication to markets
(e.g., Woodford (2005), Blinder et al. (2008)). By controlling for ΔFOGtand ΔTTRt,we
can rule out that tone effects on asset prices may simply reflect that market participants
respond differently to complex and lexically diverse compared to simple statements
because these are harder to interpret and lead to more uncertainty.
Appendix B. Measures of Monetary Policy Shocks
In our empirical analysis, we control for three sets of monetary policy shock
measures. We obtain the monetary policy shock data from Altavilla et al. (2019)as
well as the high-frequency asset price data from the (updated) euro area monetary policy
database (EA-MPD), which was established by these authors. Additionally, we obtain
the updated series of the policy and information shocks proposed by Jaroci
nski and
Karadi (2020) from Marek Jarocinski’s website. We describe the shock measures below
and present summary statistics in Table IA.4 in the Supplementary Material.
First, we use the four factors proposed by Altavilla et al. (2019) as control vari-
ables, using the replication data available for their article. Using high-frequency data on
the term structure Overnight Index Swaps (OIS) and German yields, they provide four
factors, estimated as rotated factors from principal component analyses. The “Target”
factor summarizes the interest rate response to the rate announcement, measured in a
narrow time window around the press release (using median rates between 13:25–13:35
and 14:00–14:10). Using interest rate changes during a narrow time window around the
press conference (using median rates between 14:15–14:25 and 15:40–15:50), they
rotate the first three principal components into the “timing,”“forward guidance,”and
“quantitative easing”factors. The interpretation provided by Altavilla et al. (2019)
suggests that the “timing”-factor captures information relevant for the short run, the
23
We choose Euclidean distance for its simplicity. We have also experimented with alternative
distance measures such as cosine similarity, which are immune to mechanical effects due to variation
in text lengths across documents, and obtained similar results. Moreover, computing distance metrics
based on bigrams (e.g., Tetlock (2011), Amaya and Fibien (2015)) leads to very similar results.
Schmeling and Wagner 29
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

“FG”-factor for monetary policy over the medium term, and the “QE”-factor contains
information relevant for long-term yields. For more details on the econometric proce-
dure, see their article. The factor data are available from the beginning of the year 2002
until the end of our sample period.
Second, we follow Jaroci
nski (2020) in measuring surprises in euro area short-
term interest rates. To do so, we use high-frequency EA-MPD data and extract the first
principal component of changes in the 1-, 3-, and 6-month, and 1-year OIS rates. We
apply the same time windows as Altavilla et al. (2019) to measure shocks associated
with the rate announcement in the press release (“PC1 –Press release”) as well as shocks
associated with the press conference (“PC1 –Press conference”).
Third, we use the updated shock series of Jaroci
nski and Karadi (2020) to assess
the importance of controlling for “policy shocks”and “information shocks”when
testing for the effect of tone surprises on asset price. Jaroci
nski and Karadi (2020)
use sign restrictions on the co-movement of short-term interest rates and stock returns to
identify “policy”and “information”shocks. The key idea can be summarized as
follows: Conventional monetary policy shocks should lead to a negative co-movement
of interest rates and stocks, because unexpectedly higher rates should depress asset
valuations through stronger discounting of future cash flows. One can think of such
shocks as, for example, news about the CB reaction function. Hence, in a structural VAR
framework, such policy shocks can be identified by imposing that shocks to stock prices
and short-term rates move in opposite directions. However, news released during a
monetary policy event can also refer to information about the CB’s view of the state of
the business cycle and/or near-term economic growth. Such news would imply an
increase in short-term rates as well as higher stock prices (due to higher cash flows).
Imposing this positive co-movement of shocks allows for identifying information
shocks. We refer to Jaroci
nski and Karadi (2020) for details on the computation of
these shocks.
Supplementary Material
To view supplementary material for this article, please visit http://doi.org/
10.1017/S0022109024000073.
References
Adrian, T.; E. Etula; and T. Muir. “Financial Intermediaries and the Cross-Section of Asset Returns.”
Journal of Finance, 69 (2014), 2557–2596.
Adrian, T., and N. Liang. “Monetary Policy, Financial Conditions, and Financial Stability.”Interna-
tional Journal of Central Banking, 14 (2018), 73–132.
Adrian, T.; E. Moench; and H. S. Shin. “Financial Intermediation, Asset Prices and Macroeconomic
Dynamics.”FRB of New York Staff Report (2010).
Adrian, T., and H. S. Shin. “Financial Intermediaries, Financial Stability, and Monetary Policy.”In
Jackson Hole Economic Symposium Proceedings 2008, Federal Reserve Bank of Kansas City
(2008), 287–334.
Ai, H., and R. Bansal. “Risk Preferences and the Macroeconomic Announcement Premium.”Econo-
metrica, 86 (2018), 1383–1430.
Altavilla, C.; L. Brugnolini; R. S. Gürkaynak; R. Motto; and G. Ragusa. “Measuring Euro Area
Monetary Policy.”Journal of Monetary Economics, 108 (2019), 162–179.
Amaya, D., and J.-Y. Filbien. “The Similarity of ECB Communications.”Finance Research Letters,13
(2015), 234–242.
30 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

Bekaert, G.; E. Engström; and N. Xu. “The Time Variation in Risk Appetite and Uncertainty.”Man-
agement Science, 68 (2022), 3975–4004.
Bekaert, G.; M. Hoerova; and M. Lo Duca. “Risk, Uncertainty, and Monetary Policy.”Journal of
Monetary Economics, 60 (2013), 771–788.
Berger, H.; M. Ehrmann; and M. Fratzscher. “Monetary Policy in the Media.”Journal of Money, Credit
and Banking, 43 (2011), 689–709.
Bernanke, B. The Courage to Act: A Memoir of a Crisis and its Aftermath. New York: W. W. Norton &
Company (2016).
Bernanke, B. S., and K. N. Kuttner. “What Explains the Stock Market’s Reaction to Federal Reserve
Policy?”Journal of Finance, 60 (2005), 1221–1257.
Bholat, D.; S. Hansen; P. Santos; and C. Schonhardt-Bailey. “Text Mining for Central Banks.”CCBS
Handbook No. 33, London: Bank of England (2015).
Bligh, M. C., and G. D. Hess. “The Power of Leading Subtly: Alan Greenspan, Rhetorical Leadership,
and Monetary Policy.”Leadership Quarterly, 18 (2007), 87–104.
Blinder, A. S.; M. Ehrmann; M. Fratzscher; J. De Haan; and D.-J. Jansen. “Central Bank Communication
and Monetary Policy: A Survey of Theory and Evidence.”Journal of Economic Literature,46
(2008), 910–945.
Boguth, O.; V. Gregoire; and C. Martineau. “Coordinating Attention: The Unintended Consequences of
FOMC Press Conferences.”Journal of Financial and Quantitative Analysis, 54 (2018), 2327–2353.
Bollerslev, T.; B. Hood; J. Huss; and L. H. Pedersen. “Risk Everywhere: Modeling and Managing
Volatility.”Review of Financial Studies, 31 (2018), 2729–2773.
Campbell, J.; C. Pflueger; and L. Viceira. “Macroeconomic Drivers of Bond and Equity Risks.”Journal
of Political Economy, 128 (2020), 3148–3185.
Chodorow-Reich, G. “Effects of Unconventional Monetary Policy on Financial Institutions.”Brookings
Papers on Economic Activity, Spring (2014), 155–227.
Cieslak, A.; A. Morse; and A. Vissing-Jorgensen. “Stock Returns over the FOMC Cycle.”Journal of
Finance, 74 (2019), 2201–2248.
Cieslak, A., and A. Schrimpf. “Non-Monetary News in Central Bank Communication.”Journal of
International Economics, 118 (2019), 293–315.
Cieslak, A., and A. Vissing-Jorgensen. “The Economics of the Fed Put.”Review of Financial Studies,34
(2021), 4045–4089.
Ehrmann, M., and M. Fratzscher. “Communication by Central Bank Committee Members: Different
Strategies, Same Effectiveness?”Journal of Money, Credit and Banking, 39 (2007), 509–541.
Ehrmann, M., and M. Fratzscher. “Explaining Monetary Policy in Press Conferences.”International
Journal of Central Banking, 5 (2009), 41–84.
Ehrmann, M., and J. Talmi. “Starting from a Blank Page? Semantic Similarity in Central Bank Com-
munication and Market Volatility.”Journal of Monetary Economics, 111 (2020), 48–62.
Ferrari, M.; J. Kearns; and A. Schrimpf. “Monetary Policy’s Rising FX Impact in the Era of Ultra-Low
Rates.”Journal of Banking and Finance, 129 (2021), 106–142.
Gertler, M., and P. Karadi. “QE 1 vs. 2 vs. 3…: A Framework for Analyzing Large-Scale Asset Purchases
as a Monetary Policy Tool.”International Journal of Central Banking, 9 (2013), 5–53.
Gertler, M., and P. Karadi. “Monetary Policy Surprises, Credit Costs, and Economic Activity.”American
Economic Journal: Macroeconomics, 7 (2015), 44–76.
Gilchrist, S., and B. Mojon. “Credit Risk in the Euro Area.”Economic Journal, 128 (2018), 118–158.
Gilchrist, S., and E. Zakrajšek. “Credit Spreads and Business Cycle Fluctuations.”American Economic
Review, 102 (2012), 1692–1720.
Guerkaynak, R.; B. Sack; and E. Swanson. “Do Actions Speak Louder than Words? The Response of
Asset Prices to Monetary Policy Actions and Statements.”International Journal of Central Banking,
1 (2005), 55–93.
Gurun, U. G., and A. W. Butler. “Don’t Believe the Hype: Local Media Slant, Local Advertising, and
Firm Value.”Journal of Finance, 67 (2012), 561–598.
Hansen, S., and M. McMahon. “Shocking Language: Understanding the Macroeconomic Effects of
Central Bank Communication.”Journal of International Economics, 99 (2016), S114–S133.
Hansen, S.; M. McMahon; and A. Prat. “Transparency and Deliberation Within the FOMC: A Compu-
tational Linguistics Approach.”Quarterly Journal of Economics, 133 (2017), 801–870.
Hanson, S., and J. C. Stein. “Monetary Policy and Long-Term Real Rates.”Journal of Financial
Economics, 115 (2015), 429–448.
Hattori, M.; A. Schrimpf; and V. Sushko. “The Response of Tail Risk Perceptions to Unconventional
Monetary Policy.”American Economic Journal: Macroeconomics, 8 (2016), 111–136.
He, Z.; B. Kelly; and A. Manela. “Intermediary Asset Pricing: New Evidence from Many Asset Classes.”
Journal of Financial Economics, 126 (2017), 1–35.
Schmeling and Wagner 31
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.

He, Z., and A. Krishnamurthy. “Intermediary Asset Pricing.”American Economic Review, 103 (2013),
732–770.
Hillert, A.; H. Jacobs; and S. Müller. “Media Makes Momentum.”Review of Financial Studies,27
(2014), 3467–3501.
Jaroci
nski, M. “Central Bank Information Effects and Transatlantic Spillovers.”ECB Working Paper
No. 2482 (2020).
Jaroci
nski, M., and P. Karadi. “Deconstructing Monetary Policy Surprises –The Role of Information
Shocks.”American Economic Journal: Macroeconomics, 12 (2020), 1–43.
Jegadeesh, N., and D. Wu. “Deciphering Fedspeak: The Information Content of FOMC Meetings.”
Working Paper, Emory University (2017).
Kohn, D. L., and B. P.Sack. Central Bank Talk: Does It Matter and Why?”“Macroeconomics, Monetary
Policy, and Financial Stability. Ottawa: Bank of Canada (2004), 175–206.
Krylova, E. “Leading Indicator Properties of Corporate Bond Spreads, Excess Bond Premia and Lending
Spreads in the Euro Area.”ECB Working Paper No. 1911 (2016).
Leombroni, M.; A. Vedolin; G. Venter; and P. Whelan. “Central Bank Communication and the Yield
Curve.”Journal of Financial Economics, 141 (2021), 860–880.
Loughran, T., and B. McDonald. “When is a Liability Not a Liability? Textual Analysis, Dictionaries,
and 10-Ks.”Journal of Finance, 66 (2011), 35–65.
Loughran, T., and B. McDonald. “Textual Analysis in Finance and Accounting: A Survey.”Journal of
Accounting Research, 54 (2016), 1187–1230.
Loughran, T., and B. McDonald. “Textual Analysis in Finance.”Annual Review of Financial Economics,
12 (2020), 357–375.
Lucca, D. O., and E. Moench. “The Pre-FOMC Announcement Drift.”Journal of Finance, 70 (2015),
329–371.
Lucca, D. O., and F. Trebbi. “Measuring Central Bank Communication: An Automated Approach with
Application to FOMC Statements.”NBER Working Paper No. 15367 (2009).
Nakamura, E., and J. Steinsson. “High Frequency Identification of Monetary Non-Neutrality: The
Information Effect.”Quarterly Journal of Economics, 133 (2018), 1283–1330.
Neuhierl, A., and M. Weber. “Monetary Policy Communication, Policy Slope and the Stock Market.”
Journal of Monetary Economics, 108 (2019), 140–155.
Neuhierl, A., and M. Weber. “Time Series Momentum Around FOMC Meetings.”Working Paper,
Chicago Booth (2021).
Picault, M., and T. Renault. “Words Are Not All Created Equal: A New Measure of ECB
Communication.”Journal of International Money and Finance, 79 (2017), 136–156.
Rigobon, R., and B. Sack. “The Impact of Monetary Policy on Asset Prices.”Journal of Monetary
Economics, 51 (2004), 1553–1575.
Romer, C. D., and D. H. Romer. “A New Measure of Monetary Shocks: Derivation and Implications.”
American Economic Review, 94 (2004), 1055–1084.
Rosa, C. “Words that Shake Traders: The Stock Market’s Reaction to Central Bank Communication in
Real Time.”Journal of Empirical Finance, 18 (2011), 915–934.
Rosa, C., and G. Verga. “On the Consistency and Effectiveness of Central Bank Communication:
Evidence from the ECB.”European Journal of Political Economy, 23 (2007), 146–175.
Shiller, R.; J. Campbell; and K. Schoenholtz. “Forward Rates and Future Policy: Interpreting the Term
Structure of Interest Rates.”Brookings Papers on Economic Activity (1983), 173–223.
STOXX. “STOXX Strategy Index Guide.”(2018).
Swanson, E. T. “Measuring the Effects of Federal Reserve Forward Guidance and Asset Purchases on
Financial Markets.”Journal of Monetary Economics, 118 (2021), 32–53.
Tetlock, P. C. “All the News That’s Fit to Reprint: Do Investors React to Stale Information?”Review of
Financial Studies, 24 (2011), 1481–1512.
White, H. “A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for
Heteroscedasticity.”Econometrica, 48 (1980), 817–838.
Woodford, M. “Central Bank Communication and Policy Effectiveness.”NBER Working Paper
No. 11898 (2005).
32 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109024000073
Downloaded from https://www.cambridge.org/core. IP address: 94.176.62.97, on 17 Dec 2024 at 19:03:11, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.