PreprintPDF Available
Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

Have the content, sentiment, and timing of the Federal Reserve (Fed) communications changed across communication types during the COVID-19 pandemic? Did similar changes occur during the global financial and dot-com crises? We compile dictionaries specific to COVID-19 and unconventional monetary policy (UMP) and utilize sentiment analysis and topic modeling to study the Fed’s communications and answer the above questions. We show that the Fed’s communications regarding the COVID-19 pandemic concern matters of financial volatility, contextual uncertainty, and financial stability, and that they emphasize health, social welfare, and UMP. We also show that the Fed’s communication policy changes drastically during the COVID-19 pandemic compared to the GFC and dot-com crisis in terms of content, sentiment, and timing. Specifically, we find that during the past two decades, a decrease in the financial stability sentiment conveyed by the Fed’s interest rate announcements and minutes precedes a decrease in the Fed’s interest rate.
Content may be subject to copyright.
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Covid Economics Issue 79, 27 May 2021
Copyright: Jonathan Benchimol, Sophia Kazinnik and Yossi Saadon
Federal Reserve communication
and the COVID-19 pandemic1
Jonathan Benchimol,2 Sophia Kazinnik3 and Yossi Saadon4
Date submitted: 19 May 2021; Date accepted: 25 May 2021
Have the content, sentiment, and timing of the Federal Reserve (Fed)
communications changed across communication types during the
COVID-19 pandemic? Did similar changes occur during the global
financial and dot-com crises? We compile dictionaries specific to
COVID-19 and unconventional monetary policy (UMP) and utilize
sentiment analysis and topic modeling to study the Fed’s communications
and answer the above questions. We show that the Fed’s communications
regarding the COVID-19 pandemic concern maers of financial
volatility, contextual uncertainty, and financial stability, and that
they emphasize health, social welfare, and UMP. We also show that the
Fed’s communication policy changes drastically during the COVID-19
pandemic compared to the GFC and dot-com crisis in terms of content,
sentiment, and timing. Specifically, we find that during the past two
decades, a decrease in the financial stability sentiment conveyed by the
Fed’s interest rate announcements and minutes precedes a decrease in
the Fed’s interest rate.
1 This paper does not necessarily reflectthe views of the Bank of Israel, the Federal Reserve Bank of Richmond,
or the Federal Reserve System. We thank Itamar Caspi (discussant) and participants at the 133rd American
Economic Association (ASSA) annual meeting and the Bank of Israel research seminars for their useful
comments.
2 Bank of Israel, Research Department.
3 Federal Reserve Bank of Richmond, Quantitative Supervision & Research (QSR).
4 Bank of Israel, Research Department.
218
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
1. Introduction
At the outbreak of COVID-19, most central banks declared that they would take “all
necessary steps” to mitigate the impact of the pandemic on their economies and
decreased their interest rates to the zero-lower bound. Although conventional
monetary policy tools have proven almost ineffective in the ensuing unprecedented
economic crisis, unconventional monetary policy (UMP) tools have drastically
changed both within and between central banks. At the same time, the content,
sentiment, and timing of the central banks’ communications regarding their UMP
measures and sentiments have changed as well.
However, there are no studies on how the content, sentiment, and timing of the
communications of the Federal Reserve (Fed) change. In this paper, we study such
changes across three different communication types (namely, Fed fund rate
announcements, Federal Open Market Committee minutes, and Fed chairman
speeches) during the past two decades and three different economic crises (namely,
the GFC, dot-com, and COVID-19 crises). We pay particular attention to the Fed’s
communications regarding financial stability and conventional and unconventional
monetary policies.
Using state-of-the-art text-mining methodologies, we first analyze the Fed’s
communications throughout 2020 to show they were uncertain1 and heterogeneous
during the COVID-19 crisis over time and across communications types. We then
analyze how the Fed’s communications relate to the outbreak of the pandemic and
subsequently derive potential policy implications from this analysis. Finally, we show
that the Fed’s communications and actions have been more reactive to the COVID-19
crisis than to the global financial crisis (GFC) and the dot-com crisis. Indeed, some of
the Fed’s communications and UMP-related actions appear to have anticipated the
spread of COVID-19. Taken together, our findings show that the Fed’s communication
policy has been drastically different during the COVID-19 pandemic than during the
GFC and dot-com crises.
Central banks communicate on a variety of topics, through different channels,
and with well-defined objectives (Hansen et al., 2019; Benchimol et al., 2020a). Central
bank communication aims to inform (e.g., current and future policy objectives and
decisions), explain (e.g., past, current, and future economic outlook and decisions),
and influence (e.g., current and future uncertainty and financial decisions). These
communications are usually published and stored as text (Haldane and McMahon,
2018).
The COVID-19 pandemic has affected all sectors of the global economy (Chetty
et al., 2020). In particular, the effect of the pandemic on financial markets and social
welfare has led to changes in monetary policy and threatened financial stability (Daly,
2020; Craig et al., 2021). Naturally, central banks have played an influential role during
the COVID-19 crisis, and have adapted their communication policies to the current
global economy. As in the GFC, central banks have managed the COVID-19 crisis
using UMP tools (e.g., forward guidance, quantitative easing, funding and lending
1 Communications are “uncertain” in the sense that they use nonspecific word such as “approximate,”
“contingent,” “indefinite,” and “uncertain.”
219
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
facilities, adjustments to market operations, negative or dual interest rates, etc.). Their
objective has been to decrease uncertainty and increase financial stability.
In this paper we examine the Fed’s communications on conventional and
unconventional monetary policies during the COVID-19 crisis by focusing on the most
significant types of communications. Specifically, we look at Fed fund rate (FFR)
announcements, Federal Open Market Committee (FOMC) minutes (policy decision
discussions and deliberations), and speeches given by the Fed chairman. Like most
central banks whose primary policy instrument (the nominal interest rate) has a zero
or negative lower bound, the Fed’s policy stance and subsequent expectation shaping
is implemented through other channels, including communications, quantitative
easing (QE), balance sheet policies, lending facilities, fiscal and money drops, forward
guidance, and other market operations.2 Although these decisions are undeniably
related to global economic stability, which may involve monetary policy, credibility,
and independence risks, the COVID-19 pandemic has led to unprecedented central
bank monetary policy decisions. We examine the Fed’s communications related to
these decisions over the past two decades. This analysis allows us to study whether
the Fed successfully implemented clear and transparent communications to support
UMP measures addressing the economic challenges caused by the COVID-19
pandemic. Our descriptive analysis shows that Fed’s communication was used in a
timed and targeted way, showcasing Fed’s increasing experience in crisis-specific
communication management. The methodologies and technical tools used in this
paper, such as R functions and applications to central bank texts, are available in
Benchimol et al. (2020b).
In a highly uncertain economic environment such as that of the COVID-19 crisis,
the standard for effective central bank communications would normally consist of
straightforward and timely updates about current and near-term policy actions.
Accordingly, financial stability updates have prevailed in the Fed’s monetary policy
and financial market-related communications. To proxy for the degree of financial
stability conveyed in a central bank communication, we calculate a financial stability
score for each relevant communication based on a word count of the terms that can
also be found in the financial stability dictionary (Correa et al., 2021).
We study how monetary policy and financial stability are considered in central
bank communications over time, from the dot-com crisis to the COVID-19 pandemic.
Specifically, we analyze how the Fed’s communication policy during the COVID-19
pandemic compares to that of the dot-com and GFC crises. The correlation of the
sentiment and uncertainty of the Fed’s communications to economic and financial
variables, unconventional monetary policies, and the COVID-19 pandemic is also
investigated. Finally, we focus on how the Fed’s UMP-related communications have
evolved during the past two decades.
We find that the content, timing, and sentiment of the Fed’s communications
exhibit noteworthy differences conditional on the crisis. Since the GFC,
communications regarding UMP have become the “new normal,” as reflected in all
three communication types, namely, FFR announcements, FOMC minutes, and Fed
chairman speeches. COVID-19 appears to have caused structural changes in the Fed’s
2 See, e.g., Bianchi et al. (2020) and Guerrieri et al. (2020).
220
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
communication content. We also find evidence for a link between conventional
monetary policy and financial stability sentiment.
The remainder of the paper is organized as follows. Section 2 describes the data.
Section 3 presents the results of the text analysis. Section 4 presents the results of the
sentiment analysis. Section 5 presents the results of the topic modeling of the past and
current economic situations. Section 6 examines the Fed’s communications on
unconventional monetary policy. Section 7 discusses the Fed’s early communications
on the pandemic. Section 8 compares the Fed’s conventional monetary policy to its
financial stability sentiment over the past two decades. Section 9 derives some policy
implications, and Section 10 concludes. Section 12 presents the dictionaries used to
analyze UMP and COVID-19 and additional results. The methodologies used in this
paper are presented in Benchimol et al. (2020b).
2. Data
2.1 Text Data
Our study focuses on the most significant types of Fed’s communications intended for
public consumption. We gathered 776 of these communications for the period 2000
2020. The sample contains formal communications detailing monetary policy
discussions (FFR announcements and FOMC minutes) as well as less formal
communications (Fed chairman speeches).
Our dataset is summarized in Table 1. In addition to the texts, we use dictionaries
for our text analysis purposes: namely, a well-known finance dictionary (Loughran
and McDonald, 2011), a financial stability dictionary (Correa et al., 2021), an UMP
dictionary (Christensen and Rising, 2017; Henry, 2008), and our own COVID-19
dictionary.3
Loughran and McDonald (2011) is a dictionary developed to measure the
sentiment of financial texts better than general dictionaries. This dictionary is widely
used in text analyses in the finance and economics literatures (Loughran and
McDonald, 2016; Benchimol et al., 2020a). Loughran and McDonald (2011) also
developed a dictionary to measure the uncertainty conveyed in financial texts.
Table 1. Descriptive Statistics: Federal Reserve Texts
No. Texts
No. Words (average)
Sample
FFR Announcements
181
400
2000–2020
FOMC Minutes
170
6809
2000–2020
Chairman Speeches
425
2931
2000–2020
Total
776
3213
2000–2020
Notes: “Total” refers to the sum of all three communication types.
Sources: The Federal Reserve Board of Governors and FederalReserve.gov archives.
3 Our UMP and COVID-19 dictionaries are presented in the Appendix.
221
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Correa et al. (2021) construct a dictionary explicitly tailored to financial stability
contexts. This dictionary classifies words as positive or negative based on the
sentiment they convey in financial stability reports.
For the UMP dictionary, we rely on a dictionary that translates central bank
communications about future monetary policy into groups of positive and negative
words (Christensen and Rising, 2017), and merge this dictionary with a more market-
related dictionary (Henry, 2008). These two dictionaries are analyzed and compared
in Erasmus and Hollander (2020).
Finally, we construct a COVID-19 dictionary by compiling relevant keywords
that relate to the pandemic. We use this dictionary to capture the frequency (or
“intensity”) of words associated with the COVID-19 pandemic in the Fed’s
communications in order to identify virus-related content in those communications.
We describe how we apply these dictionaries to our sample in Section 3. See the
Appendix for more details about the UMP and COVID-19 dictionaries.
2.2 COVID-19 Data
The database for the COVID-19 statistics is the COVID-19 Data Repository4
maintained by the Johns Hopkins University Center for Systems Science and
Engineering (CSSE), with the support of the ESRI Living Atlas Team and the Johns
Hopkins University Applied Physics Lab (APL). In our analysis, we essentially utilize
the daily number of new COVID-19 cases.
2.3 Financial Data
The daily financial dataset used in this paper is collected through Bloomberg. It
includes the SP500 equity index, the CBOE VIX, the nominal effective exchange rate
(broad), and the nominal interest rate (FFR).
3. Methodology
This study aims to capture the change and impact of the Fed’s communications during
the COVID-19 pandemic.5 To this end, we build text-based measures of uncertainty
and sentiment in the Fed’s communications by utilizing an array of custom
dictionaries, as described in the previous section.
We use three text-mining techniques: word counting, sentiment scoring, and
topic modeling. First, we use simple word-counting procedures. Specifically, we count
the terms related to UMP and COVID-19 that appear in the Fed’s communications.
Second, we use sentiment scoring. This supervised machine-learning method allows
us to measure sentiments conveyed by the Fed’s communications. Specifically, we use
the Loughran and McDonald’s (2011) dictionary to proxy for sentiment and
uncertainty in the Fed’s communications and build several sentiment scores and
polarity indicators based on general (NRC, SentiWords, Hu&Liu, Jockers) and
specialized (financial stability, UMP) dictionaries. Third, we use topic modeling. This
unsupervised machine-learning method allows us to extract and examine the thematic
4 The full dataset can be downloaded from https://github.com/CSSEGISandData/COVID-19
5 Central bank communications are defined as significant communications such as FFR announcements,
FOMC minutes, and Fed chairman speeches.
222
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
content of the Fed’s communications. Specifically, we use the latent Dirichlet
allocation (LDA) algorithm to compare the content of Fed’s communications to
economic and financial developments.
The results presented below are obtained with the statistical software R. The R
functions and packages used in this paper are described in Benchimol et al. (2020b).
3.1 Word Counting
To estimate the amount of coverage of the COVID-19 pandemic in each of the Fed’s
communications, we construct a COVID-19-specific dictionary, presented in the
Appendix. By counting the number of COVID-19-related words in each Fed
communication, we can estimate how the Fed perceived the severity of the pandemic
at that time.
In addition, we construct another dictionary that captures communications
regarding the Fed’s UMP measures by merging two existing dictionaries, those of
Erasmus and Hollander (2020) and Henry (2008). We then calculate the overall
sentiment related to UMP measures using this merged dictionary.
Finally, we explore whether the onset of the COVID-19 pandemic impacted
communication clarity.
3.2 Sentiment Scoring
To measure the sentiment of the Fed’s communications, we implement several
methods of capturing text sentiment. The general methodology is based on counting
positive and negative words according to a specific dictionary. The sentiment score is
calculated by dividing the difference between the number of positive and negative
words by the total amount of words.
To this end, we use Correa et al.’s (2021) dictionary, which is specifically tailored
to capture financial stability sentiment. Correa et al. (2021) explain movements in
financial cycle indicators related to credit, asset prices, systemic risk, and monetary
policy rates and classify words (positive/negative) based on the sentiment conveyed
in financial stability reports. The second dictionary focuses on forward-guidance and
quantitative measures (Christensen and Rising, 2017) that we merge with another
dictionary that focuses more on the regulatory context, structural attributes, and dual
informational-promotional role of earnings press releases (Henry, 2008).
We use another set of dictionaries to capture different dimensions of the
sentiment expressed in the text, such as the Loughran and McDonald (2011) sentiment
dictionary, as well as a set of commonly used sentiment dictionaries in the text-mining
literature, such as the Jockers, NRC, and Hu&Liu dictionaries. We use these
dictionaries in conjunction with the so-called valence shifters (i.e., negators,
amplifiers/intensifiers, de-amplifiers/downtoners) to capture nuances in the
sentiment of the relevant text.
Last but not least, we construct two types of sentiment indicators based on the
Loughran and McDonald (2011) dictionary. One is the standard score measure
described above. The other is a polarity measure that includes the possibility of
neutral, positive, negative, very positive, or very negative sentiment, according to the
sentiment of the words immediately preceding and following the considered word.
223
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Section 4 presents sentiment scores produced from these dictionaries for FFR
announcements, FOMC minutes, and Fed chairman speeches. For these three
communication types, we note a sharp decrease in sentiments in the first quarter of
2020, as well as a spike in uncertainty-related words during that same period. This
finding suggests that the Fed’s communications reflect its willingness to address the
ongoing developments during the COVID-19 pandemic proactively.
3.3 Topic Modeling
Our goal in this section is to identify underlying themes that drive the Fed’s
communications and to capture theme prevalence over time. For this purpose, we rely
on the so-called topic modeling approach. This algorithm allows us to identify a small
number of verbal themes that best explain thematic variation over time, and we use it
to capture the Fed’s assessments of economic and financial risks in real-time.
Topic modeling is an unsupervised machine-learning technique that does not
require any training or dictionary-based analysis. Here, we use LDA, which works in
the following way. The LDA algorithm views each document as a mixture of topics
that are present in the body of the text. The algorithm scans a set of relevant
documents (FFR announcements, FOMC minutes, and Fed chairman speeches),
detects words and phrases within them, and automatically clusters word groups (i.e.,
topics) that best characterize a set of documents. In essence, the algorithm identifies
the different topics represented in the document, and calculates the prevalence of each
of them (Blei et al., 2003).
In the following Section 4, we present the results of the word-counting and
sentiment-scoring analysis and interpret them by providing a descriptive picture of
the sentiments the Fed’s communications conveyed. Section 5 presents the results of
the topic modeling analysis. Overall, we find that at the onset of the COVID-19
pandemic, the topics related to policy intervention gained prominence, at the expense
of other topics. We relate UMP to our various indicators and the Fed’s communication
types in Section 6 and control for the severity of the outbreak by including the number
of COVID-19 cases at the time of the communication in Section 7. We also describe
and relate our text-mining indicators to financial stability and interest rate decisions
over the past two decades in Section 8.
4. Sentiment Analysis
Figure 1 presents several sentiment indicators extracted from the text analysis of FFR
announcements. The sentiments sharply degraded following the COVID-19 outbreak
in China in January 2020 and the US in March 2020. This date range corresponds to an
increase in the contextual uncertainty indicator (the number of words reflecting
uncertainty, scaled by text length).
Based on the Loughran and McDonald (2011) dictionary, our new polarity
indicator decreases but displays a more optimistic sentiment for 2020Q3. FFR
announcements summarize the current state of the economy and monetary policy
decisions. Figure 1 shows that the shock that occurred in January 2020 lasts up until
April 2020.
224
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
The sharp decrease in sentiments in 2020Q1-Q2, and the increase in the
contextual uncertainty in 2020Q1-Q3, correspond to the beginning of the COVID-19
crisis.
Figure 1. Sentiment Scores in FFR Announcements
Notes: Solid black lines represent sentiment score values. Red arrows represent trends related to the
COVID-19 outbreak.
Figure 1 also shows that the SentiWords, Jockers, and NRC polarity indexes are
(apparently) less informative, while the Hu&Liu polarity index displays similar
dynamics to our Loughran and McDonald-based polarity index. SentiWords, a high-
coverage polarity index, captures an interesting increase in the sentiment of FFR
announcements from the beginning of the COVID-19 crisis. A potential explanation
for this increase is that the Fed used a different communication strategy in this crisis
than in the GFC and dot-com crisis.6 The financial stability sentiment has been
decreasing into negative territory since 2019Q4, which means that more negative
financial stability-related words were present in the FFR announcements than positive
ones.
Interestingly, the UMP sentiment decreased mainly because this indicator
includes forward-guidance sentiment but also words found in standard dictionaries.
6 SentiWords polarity index sharply decreases for the GFC and dot-com crisis. The results of the full
sample are available in the Appendix (Figure A1).
225
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Note that these results are based on FFR announcements, which are the most
supervised and controlled7 communication type.
Figure 2 presents selected sentiment indicators over the full sample. As proxied
by our sentiment measures, the Fed’s communications differ significantly in sentiment
during the COVID-19 crisis, compared to the GFC and the dot-com crisis.
Figure 2. A Tale of Three Crises: Sentiment.
Notes: The gray shaded area represents the NBER recessions periods. Solid lines represent sentiment
scores computed from FFR announcements.
The financial stability sentiment sharply deteriorated before the GFC and dot-
com crisis, whereas it is significantly positive before the COVID-19 pandemic. The
FFR announcements appear to be a significant predictor of future conventional
monetary policy (see Section 8).
The Hu and Liu sentiment polarity index improved from the GFC until the
COVID-19 crisis. It decreased less in the COVID-19 crisis than in the other crises.
While the Loughran and McDonald sentiment index did not improve between the
crises, it decreased less in the COVID-19 crisis than in the other crises. Also, the
volatility of these indicators was less pronounced in the COVID-19 crisis than in the
GFC and dot-com crisis. Although Figure 2 presents the tale of three crises, it also
clearly shows a tale of three communication policies.
7 By the spokesperson, and other Fed’s departments or officials.
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
-5,0
-3,0
-1,0
1,0
3,0
5,0
7,0
9,0
Financial Stability Sentiment UMP Sentiment Loughran (score) Hu & Liu
226
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Figure 3 presents the same indicators for the FOMC minutes. The dynamics for
almost all sentiment indicators display a sharp deterioration in 2020Q2, which is much
more pronounced for the FOMC minutes than for the FFR announcements.
Figure 3. Sentiment Scores for the Fed’s FOMC Minutes
Notes: Solid black lines represent sentiment score values. Red arrows represent trends related to the
COVID-19 outbreak.
This probably reflects the gap between the description of the current economic
situation (FFR announcements) and the discussions and tentative solutions to the
COVID-19 crisis (discussed in the minutes). The Loughran and McDonald score and
polarity indexes showed a sharp decrease in sentiment related to financial uncertainty
from January to April 2020.
The UMP sentiment score decreases until 2020Q2 and then sharply increases
until it becomes positive. This phenomenon corresponds to the more positive
language adopted in FOMC minutes regarding UMP steps taken during the COVID-
19 crisis. Figure 3 also shows that according to FOMC minutes (i.e., according to
policymakers during monetary policy committee discussions), financial stability was
perceived to be at risk in 2020Q2. This was effectively the case in reality but less so
than during the GFC, as explained in the Appendix (Figure A5).
The high coverage of the SentiWords index captures an interesting pattern of
increasing sentiment from the beginning of the COVID-19 crisis. This may result from
the Fed’s communication strategy to calm and reassure economic agents with the use
227
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
of more positive words. This was not the case during the GFC, where the SentiWords
indicator sharply declined to historically low levels. 8
Figure 4 presents the sentiment indicators for the official speeches of the
chairman of the Federal Reserve.
Figure 4. Sentiment Scores for Fed Chairman Speeches
Notes: Solid black lines represent sentiment score values. Red arrows represent trends related to the
COVID-19 outbreak. Green arrows represent trends that remain constant before and after the COVID-
19 outbreak.
The sentiment conveyed by Fed chairman speeches decreases less in comparison
to the minutes and announcements. A potential explanation is that the speeches might
be aimed at managing expectations more than the minutes and announcements are.
Although the economic situation worsened and sentiments degraded from February
2020 onward, contextual uncertainty decreased.
The sentiment conveyed by the Fed chairman speeches is generally more volatile
than the sentiment conveyed by FFR announcements and minutes (Benchimol et al.,
2020a). However, the small sample makes the sentiment indicators for the COVID-19
crisis presented in Figure 4 less volatile than those for the GFC. 9
8 The results of the full sample are available in the Appendix (Figure A2).
9 The results of the full sample are available in the Appendix (Figure A3).
228
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
As for the FFR announcements and minutes, the large scope of the SentiWords
dictionary captures this specific Fed’s communication policy held during the COVID-
19, which seems to have been in force at least up until 2020Q4. The Loughran and
McDonald dictionary also shows a decrease in the use of uncertainty-related words in
the Fed chairman speeches since the COVID-19 outbreak in China, which may also
result from a specific communication policy.
Figure 5 presents aggregated sentiment indicators based on FFR
announcements, FOMC minutes, and Fed chairman speeches.
Figure 5. Sentiment Scoring of Main Fed Communications
Notes: Solid black lines represent sentiment score values. Red arrows represent trends related to the
COVID-19 outbreak. Green arrows represent trends that remain constant before and after the COVID-
19 outbreak.
We provide a global picture of the Fed’s communications by aggregating all
communication types in our dataset. Figure 5 shows a sharp decrease in the
uncertainty sentiment, which is mainly driven by minutes and speeches. The financial
stability sentiment sharply decreases from January to April 2020, which may be
correlated with volatility measures such as the CBOE VIX (see Section 7).
The results presented in Figures 15 point to the same conclusion. Namely, it is
highly likely that the Fed has globally implemented the same communication policy
across all communication types (FFR announcements, FOMC minutes, Fed chairman
229
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
speeches) during the COVID-19 crisis. These results also demonstrate the different
sentiments involved in the COVID-19 crisis compared to the GFC.10
This section covers the special sentiment deterioration that occurred from
January to April of 2020. The recovery of sentiment into positive territory took place
following this deterioration, with some sentiment measures rising above pre-crisis
levels. Looking at the pronounced differences in sentiment over time, we find
evidence that the Fed’s communications were used to shape the narrative and manage
expectations.
5. Topic Modeling
In this section, we focus on the topics extracted from our sample of texts. It is
important to note that the topic modeling methodology does not use any
predetermined dictionaries. In contrast to sentiment analysis, it is a more structural
and unsupervised approach to interpreting word-topic linkages in texts.
Figure 6. A Tale of Three Crises: Topics.
Note: Bars represent the topic probability computed from FFR announcements. For clarity and
robustness, we restrict attention to the six most frequently discussed topics.
Source: Benchimol et al. (2020a).
Figure 6 presents the six topics extracted from FFR announcements over the past
two decades. It shows that discussion of policy interventions was more pronounced
during the COVID-19 than during the other crises. Interestingly, Figure 6 shows that
10 The results of the full sample are available in the Appendix (Figure A4).
230
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
the topic of inflation expectations decreased in importance while the topic of economic
growth, which includes economic growth considerations and concerns, increased.
Another interesting observation from Figure 6 is related to each crisis’s relative
influence on the Fed’s communications. While the dot-com crisis had almost no effect
on the topics conveyed to the public in the FFR announcements, the GFC and COVID-
19 crises strongly shaped the topics conveyed in these announcements.
Figure 7 shows that the probability that policy intervention was discussed in FFR
announcements significantly increased after the COVID-19 outbreak in China.
However, this topic had begun to increase earlier, which indicates that it may be at
least partly related to previous concerns unrelated to COVID-19.
Figure 7. Topic Analysis of FFR Announcements
Notes: For clarity and robustness, we restrict attention to the six most frequently discussed topics.
A sharp decrease in the topic probability of inflation expectations coincides with
the COVID-19 outbreak, which is in line with monetary policy considerations at that
time, when the focus shifted to policy intervention.
To a lesser extent, inflation is less discussed, and economic growth more
discussed, in FFR announcements during the COVID-19 crisis.
The increase in the topic probability of policy intervention in FFR
announcements decreases the topic probability of inflation expectations and, to a
lesser extent, inflation. Although the Fed’s main objective is to stabilize prices, this
231
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
finding demonstrates that its FFR announcements were less related to inflation
concerns after the COVID-19 outbreak.
The topic of economic growth slightly increased after the COVID-19 outbreak in
China. This finding indicates the Fed’s concern that a pandemic would pose a threat
to economic growth.
Figure 8 presents the topic analysis of FOMC minutes. It shows that the COVID-
19 outbreak significantly shaped the discussions of the Federal Open Market
Committee.
Figure 8. Topic Analysis of FOMC Minutes
Notes: For clarity and robustness, we restrict attention to the six most frequently discussed topics.
Interestingly, like the FFR announcements, the FOMC minutes are also
influenced by the topic of policy intervention, even though the interest rate is the most
prominent topic. The probability of discussion of inflation also decreased for the
FOMC minutes due to the COVID-19 outbreak, while discussions of policy
intervention and financial markets increased.
The topics conveyed by the Fed’s FOMC minutes reflect a sharp increase in
coverage of policy intervention and foreign economy. However, the coverage of
foreign economy had begun to increase even before the COVID-19 outbreak while
inflation and interest rate topics had begun to decrease.
Figure 9 presents the topic analysis of Fed chairman speeches. These speeches
focused on social welfare concerns after the COVID-19 outbreak, similar to the pre-
232
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
crisis concerns about education and inequality in the US. This finding shows that Fed
chairman speeches are often devoted to issues unrelated to its primary objective of
stabilizing prices, such as education, health, and development economics, including
family and labor markets.11
Figure 9. Topic Analysis of Fed Chairman Speeches
Notes: For clarity and robustness, we restrict attention to the six most frequently discussed topics.
The fact that the speeches are less supervised and held to a broader audience
than traditional FFR announcements and FOMC minutes, which are more focused on
inflation and output growth, may explain the increase in discussion of social welfare
issues. We also observe an increase in discussions of economic policy after June 2020.
This may be due to COVID-19 spillovers, but we cannot reject the US election effect.
Before the COVID-19 outbreak, economic policy considerations occupied the attention
of most Fed chairman speeches.
Figure 10 presents the topics discussed in the Fed’s communications in the
aggregate. We provide a global picture of the topic modeling of the Fed’s
communications by aggregating all communication types in our dataset, namely, FFR
announcements, FOMC minutes, and Fed chairman speeches. Because of the
considerable number of texts analyzed, and their respective, often different,
characteristics, we were constrained to include a larger number of topics for the global
11 The most frequently used words and word fragments (root words) in the context of the topic of social
welfare are communiti, economi, educ, work, develop, research, busi, job, peopl, mani, help, opportun, can,
import, family.
233
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
topic modeling. Taken together, these topics include most of the topics described in
Figures 5 to 7.
Figure 10. Topic Analysis of Main Fed Communications
Notes: For clarity and robustness, we restrict attention to the six most frequently discussed topics.
Note that the monetary policy topic contains some references to UMP and
unemployment, which may be attributed to the Fed’s dual mandate.
The topic of inflation expectations continues to attract most of the Fed’s attention
overall, which the COVID-19 crisis and long-term interest rate concerns may have
reinforced.
Figure 10 shows three prevalent topics that emerged in quick succession. First,
the Fed’s communications regarding the foreign economy increased as the COVID-19
pandemic spread from China to the rest of the world (1). Second, the Fed’s
communications regarding financial stability increased as fears about the financial
system due to the impact of the COVID-19 crisis increased (2). Third, the Fed’s
communications regarding social welfare increased as the potential need of
Americans for additional relief plans from the government and the Fed increased (3).
The Fed’s communications regarding conventional monetary policy (especially
average inflation targeting) and UMP decreased at the beginning of the COVID-19
pandemic but increased thereafter (4).
4
2
3
1
234
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Overall, the content and timing of the Fed’s communications exhibit differences
across the three crises discussed above. Unlike the GFC and dot-com crises, the
COVID-19 crisis has caused the focus of the Fed’s communications to shift away from
discussions of inflation expectations to discussions of policy intervention.
Interestingly, the topic of policy intervention is much more prevalent in the Fed’s
communications during the Covid-19 crisis, compared to GFC and dot-com crises. It
seems that the policymakers not only implement policy interventions but discuss
these interventions differently across the crises, and that the content, sentiment, and
timing of these communications are conditioned on the crisis.
6. Unconventional Monetary Policy
This section analyzes the link between the Fed’s communications, its actions, UMP,
and COVID-19. Figure 11 compares UMP terms with UMP measures as reflected by
the Fed’s balance sheet.
Figure 11. Unconventional Monetary Policy in FFR Announcements
Notes: The gray shaded area represents NBER recession periods. The red shaded area represents total
assets (minus eliminations from consolidation) in the Fed’s balance sheet in millions of US dollars. The
blue shaded area represents the word-counting indicator based on our UMP dictionary presented in
the Appendix. The red arrows indicate UMP terms’ intensity from the onset of the GFC to the outbreak
of COVID-19. The green arrows indicate the timing of the Fed’s UMP communications.
Source: Board of Governors of the Federal Reserve System (US).
0
1000000
2000000
3000000
4000000
5000000
6000000
7000000
0,0
1,0
2,0
3,0
4,0
5,0
6,0
7,0
UMP Terms Fed Balance Sheet (Total Assets)
235
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Figure 11 shows that the Fed communicated more extensively about UMP
during the GFC than during the COVID-19 crisis. However, it is important to note the
timing. The Fed communicated about and acted against the GFC after a delay of nine
months, whereas it hastened to do so during the COVID-19 crisis and the
communications and actions were more clearly coordinated.
Figure 11 shows that the Fed’s communications regarding UMP during the
COVID-19 crisis (according to a word count of the terms listed in our UMP dictionary
in the Appendix) correspond to effective UMP measures that led to Fed balance sheet
changes with several lags.
It is notable that whereas actions were implemented before they were
communicated during the GFC, they were implemented after they were
communicated after the GFC and during the COVID-19 crisis.
Another feature of our dictionary is that it can be used to identify other periods
beyond the above-noted crises when substantial UMP measures were taken to support
the US economy. Indeed, as can be seen in Figure 11, each communication peak related
to a UMP measure influenced the Fed’s balance sheet shortly after the communication
shock. The communications about UMP in 20132014 were devoted to conveying the
message that these expansionary policies would cease (and effectively did so
according to the Fed’s balance sheet), thus proving that our UMP dictionary captures
tapering communication policies.
The Fed implemented this gradual reversal of QE easing policies to mitigate
economic growth expectations. The “tapering” effectively started in 2013 when Ben
Bernanke, the Fed chairman at the time, commented that the Fed would lower the
amount of purchased assets each month if economic conditions, such as inflation and
unemployment, continued to be favorable.12
Figure 12 shows that FOMC minutes discuss UMP actions relatively earlier for
the COVID-19 crisis than for the GFC. The quantity of these UMP discussions is
similar to tapering discussions held by the FOMC in 2013, and the overall level is
higher in late 2020 compared to the GFC period.
It is worth noting the difference between the FFR announcements and the FOMC
minutes with respect to UMP. Although FOMC minutes are less supervised and
longer than FFR announcements, pre-COVID-19 crisis UMP FOMC discussions were
more intense than pre-GFC or even during the first half of the GFC, a behavior
confirmed by comparing Figures 11 and 12.
In summarizing the discussions held between monetary policy committee
members, FOMC minutes typically contain more UMP terms (discussions or
controversies about potential solutions or policy implementations) than FFR
announcements do. It is interesting to note that such terms have remained in frequent
use since the GFC.
12 “Tapering” refers to gradually reducing the Fed’s asset purchases, not altogether eliminating them.
236
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Figure 12. Unconventional Monetary Policy and Minutes
Notes: The gray shaded area represents NBER recession periods. The red shaded area represents total
assets (minus eliminations from consolidation) in the Fed’s balance sheet in millions of US dollars. The
blue shaded area represents the word-counting indicator based on our UMP dictionary presented in
the Appendix. The red arrows indicate UMP terms’ intensity from the onset of the GFC to the outbreak
of COVID-19. The green arrows indicate the timing of the Fed’s UMP communications. The blue arrow
indicates that UMP discussions in the Fed’s minutes during the pre-tapering period were almost as
high as during the COVID-19 crisis.
Source: Board of Governors of the Federal Reserve System (US).
Figure 13 presents the word counts related to UMP terms in Fed chairman
speeches.
Figure 13 shows that relevant communication shocks, like during the GFC or the
third round of monthly purchases of Treasury securities and mortgage-backed
securities (MBS) in September 2012 (third QE), are strongly related to the UMP content
of the Fed chairman speeches. Indeed, during or after each communication peak, the
dynamics of the Fed’s balance sheet changed. Most of the peaks that did not influence
the Fed’s balance sheet were related to forward-guidance communications.
The Fed chairman speeches appear to be the privileged platform for mentioning
UMP terms. It is interesting to compare the frequency of UMP terms in the speeches
delivered before and after each crisis. As the Figure 13 shows, the frequency of the
UMP terms in the speeches increases in the wake of the GFC, dot-com, and COVID-
19 crises. However, the frequency is higher, and earlier, in the COVID-19 crisis,
compared to the other crises.
0
1000000
2000000
3000000
4000000
5000000
6000000
7000000
0,0
0,5
1,0
1,5
2,0
2,5
3,0
3,5
4,0
4,5
5,0
UMP Terms Fed Balance Sheet (Total Assets)
237
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Figure 13. Unconventional Monetary Policy and Fed Chairman Speeches
Notes: The gray shaded area represents NBER recession periods. The red shaded area represents total
assets (minus eliminations from consolidation) in the Fed’s balance sheet in millions of US dollars. The
blue shaded area represents the word-counting indicator based on the dictionary presented in the
Appendix.
Source: Board of Governors of the Federal Reserve System (US).
Figure 14 presents an aggregated UMP indicator for main Fed communications.
The figure shows differences in the timing of UMP communications and actions for
the COVID-19 and GFC crises
Notable is the post-GFC “new normal,” where UMP communications and
actions are more frequent than in the pre-GFC period. This continuous need for UMP
tools may eventually transform their unconventional character into a more
conventional or regular one.
Overall, communicating about QE and forward-guidance (UMP) actions became
the “new normal” for the Fed since the GFC (Bernanke, 2020), while the frequency of
UMP terms remains higher for the COVID-19 crisis than for previous crises.
0
1000000
2000000
3000000
4000000
5000000
6000000
7000000
0
1
2
3
4
5
6
UMP Terms Fed Balance Sheet (Total Assets)
238
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Figure 14. Unconventional Monetary Policy in Main Fed Communications
Notes: The gray shaded area represents NBER recession periods. The red shaded area represents total
assets (minus eliminations from consolidation) in the Fed’s balance sheet in millions of US dollars. The
blue shaded area represents the word-counting indicator based on the dictionary presented in the
Appendix. The arrows indicate the time periods between the beginning of the GFC crisis and the first
UMP communications (red) and the first UMP measures (green) influencing the Fed’s balance sheet.
Source: Board of Governors of the Federal Reserve System (US).
7. COVID-19
In this section we examine the use of COVID-19 terms with our dictionary presented
in the Appendix. We compare these terms with UMP and contextual uncertainty
terms, and financial volatility and new COVID-19 cases.
Figure 15 presents the repartition of COVID-19-related terms used in the main
Fed’s communications in 2020.
The figure shows that the Fed chairman speeches anticipated the waves of new
COVID-19 cases. One has to consider this result cautiously since the first tests started
later in the US compared to other countries. Nevertheless, the speeches anticipated
the spillovers of the virus from China, focusing on the US economy.
It is worth noting that the Fed chairman speeches provide a more timely and
flexible communication vehicle than FOMC minutes and FFR announcements. They
are disseminated quickly and informally compared to the other communication types,
and allow health, political or foreign considerations that are less discussed in the other
communication types.
0
1000000
2000000
3000000
4000000
5000000
6000000
7000000
0,0
0,5
1,0
1,5
2,0
2,5
3,0
3,5
4,0
4,5
5,0
UMP Terms Fed Balance Sheet (Total Assets)
239
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Interestingly, Figure 15 presents two types of COVID-19 waves: the first type of
wave plots new cases of COVID-19 based on medical statistics from the COVID-19
Data Repository, and the second type of wave plots the intensity of COVID-19-related
terms in the Fed’s communications based on our COVID-19 dictionary presented in
the Appendix. It is visually apparent that the Fed communication waves precede the
virus waves.13
The magnitude and severity of the COVID-19 virus were rapidly understood and
communicated to the public by the Fed via its FFR announcements and chairman
speeches. The FFR announcements used more COVID-19-related terms than the other
communication types and contributed better to the first communication wave than the
speeches, but they lagged a few weeks behind the first Fed chairman speeches
mentioning COVID-19-related terms.
Figure 15. COVID-19 and Main Fed Communications
Notes: The shaded areas represent the word-counting indicator for each communication type based on
our COVID-19 dictionary presented in the Appendix. The dashed line represents the number of new
COVID-19 cases in the US (right axis).
Source: Johns Hopkins University Center for Systems Science and Engineering (CSSE).
The decrease in the intensity of COVID-19 terms in the Fed chairman speeches
in the second quarter of 2020 is directly correlated to the decrease in positive sentiment
reported in Figure 5 for the same period. It is also correlated to the increase in the
13 Granger causality tests also confirm this finding, but given the few observations available, the results
are not reported.
0
20000
40000
60000
80000
100000
120000
0
0,5
1
1,5
2
2,5
3
3,5
4
4,5
Announcements Minutes Speeches New Cases
240
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
topicality of social welfare in the Fed chairman speeches during this period, as
reported in Figure 9. Consequently, both the topics and the sentiments of the Fed
chairman speeches were affected by the COVID-19 outbreak.
Interestingly, the increase in the SentiWords sentiment exposed in Section 4
seems to be confirmed by the decrease in the frequency of COVID-19-related terms
after 2020Q3 despite the increase in the number of new cases during that quarter. In
other words, the Fed’s communications conveyed a more positive message than the
reality of the pandemic and its economic spillovers would warrant. This seems to be
the result of a crisis-specific communication strategy.
Overall, the waves of Fed communications about the COVID-19 crisis
anticipated the waves of new COVID-19 cases. The Fed chairman speeches
communicated about the first wave of COVID-19 earlier than the other
communication types (FFR announcements and FOMC minutes). This result again
confirms that speeches are less supervised than announcements and minutes and thus
allow the Fed to communicate in a timelier manner. It also indicates that the Fed had
an early understanding of the severity and magnitude of the COVID-19 pandemic and
its economic spillovers.
Figure 16 compares new COVID-19 cases in the US and word-counting
indicators based on the dictionaries of UMP and COVID-19 terms. This figure also
includes the Chicago Board Options Exchange (CBOE) Volatility Index (VIX) to
compare market volatility and potential financial uncertainty underlying the virus
outbreak with UMP and COVID-19 terms.
Figure 16 shows that the COVID-19-related terms in the Fed’s communications
in January 2020 about the virus outbreak in China considerably upset financial
markets in the US. However, the UMP-related terms in the Fed’s subsequent
communications, about the UMP actions taken by the Fed in response to the
pandemic, helped to decrease this financial volatility.
The CBOE VIX dramatically increased with the COVID-19 outbreak in China and
several other countries, including the US. COVID-19 mentions in the Fed’s
communications preceded UMP considerations and waves of new cases in the US.
Between May and July of 2020, the Fed extensively communicated about
unconventional monetary policies. During this period, although new cases of COVID-
19 significantly increased, the Fed’s communications and actions slightly decreased
financial volatility.
Following this period, the increase in the frequency of UMP-related terms in the
Fed’s communications as the pandemic continued to worsen may have stabilized the
volatility of financial markets.
241
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Figure 16. COVID-19 and UMP Terms in Main Fed Communications
Notes: The dark blue line represents the word-counting indicator based on our dictionary of COVID-19
terms presented in the Appendix. The dashed line represents the number of new COVID-19 cases in
the US (right axis). The green arrow indicates the CBOE VIX reversal due to UMP communications (and
actions) highlighted with the blue arrow. The blue circle highlights the other relations between
increases in UMP communications and decreases in the VIX.
Sources: Bloomberg and Johns Hopkins University Center for Systems Science and Engineering (CSSE).
Figure 17 presents our COVID-19 and UMP word-counting indicators based on
Loughran and McDonald’s (2011) dictionary of contextual uncertainty terms.
Uncertainty in the Fed’s communication is related to the number of UMP terms
found in those communications, what we interpret as the “uncertainty effect”.
COVID-19, UMP, and uncertainty comove in the Fed’s communications, especially
during the second half of 2020. This is not necessarily the case at the beginning of the
COVID-19 sample period, mainly because the suddenness of the outbreak of the virus
took everyone by surprise and increased the frequency of the uncertainty-related
terms before the others. The “uncertainty effect” appears during crisis periods
necessitating UMP to mitigate market and economic uncertainty.
Figure 17 also demonstrates the anticipatory effects of uncertainty- and UMP-
related terms in the Fed’s communication regarding COVID-19. The increase in the
use of uncertainty terms appears to precede increases in new cases of the virus.
The correlation between the contextual uncertainty from Loughran and
McDonald’s (2011) dictionary and the UMP-related terms from the UMP dictionary
0
20000
40000
60000
80000
100000
120000
0
0,2
0,4
0,6
0,8
1
1,2
1,4
1,6
1,8
2
COVID-19 Terms UMP Terms VIX New Cases
242
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
presented in the Appendix is significantly positive at 0.44 for weekly average
communications between 2000 and 2020 (i.e., 1090 observations).
Figure 17. COVID-19, UMP, and Uncertainty in Main Fed Communications
Notes: The dashed line represents the number of new COVID-19 cases in the US (right axis). The
contextual uncertainty indicator is the number of uncertainty terms according to Loughran and
McDonald (2011).
Source: Johns Hopkins University Center for Systems Science and Engineering (CSSE).
Figure 18 presents our COVID-19 and UMP word-counting indicators together
with the financial stability sentiment (Correa et al., 2021) and the number of new
COVID-19 cases in the US.
Except at the beginning of the COVID-19 crisis, increases in the sentiment
associated with financial stability are correlated to increases in the number of new
virus cases. The end-of-sample decrease in the financial stability sentiment is partly
driven by the lack of chairman’s speeches during this period.
Figure 18 shows that the decrease in sentiment associated with financial stability
lags a few weeks behind the increases in both COVID-19- and UMP-related terms in
the Fed’s communications. This result is not surprising given that discussions and
decisions related to financial stability generally occur after shocks to financial stability.
The several deteriorations in financial stability sentiment that precedes the increase in
the number of new COVID-19 cases may confirm the anticipatory effect of the Fed’s
discussions of their stabilization policies.
0
20000
40000
60000
80000
100000
120000
0
10
20
30
40
50
60
70
COVID-19 Terms UMP Terms Uncertainty Terms New Cases
243
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Figure 18. COVID-19, UMP, and Financial Stability in Main Fed Communications
Notes: The blue line represents the word-counting indicator based on our COVID-19 dictionary
presented in the Appendix. The dashed line represents the number of new COVID-19 cases in the US
(right axis). The financial stability index is rescaled to match scale constraints. The blue arrows indicate
similar trends between the number of new COVID-19 cases and the financial stability index.
Sources: Johns Hopkins University Center for Systems Science and Engineering (CSSE).
Following the GFC, the Fed’s communications anticipated effective UMP
implementations (actions). The timing and magnitude of these implementations were
dramatically different between crises. It was shown in Figure 18 that the Fed’s
communications about the COVID-19 crisis also anticipated waves of new COVID-19
cases. The UMP implementations were indeed aimed at reducing market volatility
together with COVID-19 spillovers.
We have also shown that the contextual uncertainty in the Fed’s communications
well anticipated COVID-19 waves. Finally, it appears that the decrease in sentiment
associated with the Fed’s communications about financial stability generally
anticipated increases in the number of new COVID-19 cases. The anticipatory effects
of contextual uncertainty in the Fed’s communications seem to confirm its early
understanding of the COVID-19 spreading and its economic spillovers.
8. Financial Stability
UMP- and uncertainty-related terms are associated with financial stability and
volatility. This section focuses on the financial stability sentiment and contextual
0
20000
40000
60000
80000
100000
120000
-2
-1
0
1
2
3
4
COVID-19 Terms UMP Terms Financial Stability Sentiment New Cases
244
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
uncertainty and their respective dynamics relative to the conventional monetary
policy instrument (the FFR) and financial market volatility (CBOE VIX).
Figure 19 compares the financial stability sentiment, UMP- and uncertainty-
related terms from FFR announcements and the CBOE VIX.
Figure 19. Financial Stability and Monetary Policy in FFR Announcements
Notes: The gray shaded area represents NBER recession periods. The right axis indicates the VIX level.
Figure 19 shows the Fed used fewer uncertainty-related words in its FFR
announcements, and so voluntarily limited contextual uncertainty, during the
COVID-19 crisis than in previous years and previous crises. This suggests a
communication strategy that aimed to decrease market uncertainty (volatility) when
the number of new virus cases sharply increased. Before the dot-com and GFC crises,
a strong decrease in the positive sentiment associated with financial stability
sentiment occurred, but this was not the case in the COVID-19 crisis. This is mainly
due to the unpredictable characteristics of this crisis as well as its rapid spillovers on
market uncertainty rather than the banking system.
The UMP-related terms played a significant role in decreasing market volatility
during the dot-com and GFC crises but also during the COVID-19 crisis. In all of these
crises, FFR announcements about the implementation of UMP measures reduced the
VIX.
The financial stability sentiment is closely related to the FFR level. A decrease in
the financial stability sentiment generally corresponds to a decrease in the FFR.
10
20
30
40
50
60
70
-3,0
-1,0
1,0
3,0
5,0
7,0
9,0
Financial Stability Sentiment UMP Terms Uncertainty Terms FFR VIX
245
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Figure 20 is the same as Figure 19 except that FOMC minutes are considered
instead of FFR announcements.
Figure 20. Financial Stability and Monetary Policy in FOMC Minutes
Notes: The gray shaded area represents NBER recession periods. The right axis indicates the VIX level.
Interestingly, although each crisis was preceded by an increase in the use of
uncertainty-related terms in FOMC minutes, a clear tendency to reduce uncertainty-
related words during crises is observed in Figure 20, similar to Figure 19. The VIX is
negatively correlated to financial stability sentiment, and both precede UMP terms,
which generally leads to UMP actions being taken to stabilize the markets and
financial stability fears and financial volatility (VIX).
As FOMC minutes provide detailed information on the monetary policy
committee views about the suitable and near-term policy stance and the US economic
outlook, they convey financial stability sentiments and UMP terms earlier than FFR
announcements.
The FFR increases correspond to high financial stability sentiment levels, except
during 20122015 where the financial stability sentiment was driven up by UMP
communications and actions.
Figure 21 presents Fed chairman speeches together with the FFR announcements
and financial volatility.
The financial stability sentiment present in the speeches is less indicative of the
future FFR compared to the announcements and minutes. Chairman speeches may
convey a negative financial stability sentiment even as the FFR increases, which on
0
10
20
30
40
50
60
70
-3,0
-1,0
1,0
3,0
5,0
7,0
Financial Stability Sentiment UMP Terms Uncertainty Terms FFR VIX
246
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
average does not happen in FFR announcements and FOMC minutes (see Figures 19
and 20).
Figure 21. Financial Stability and Monetary Policy in Fed Chairman Speeches
Notes: The gray shaded area represents NBER recession periods. The right axis indicates the VIX level.
Comparing the period between the dot-com and GFC crises (P1) with the period
between the GFC and the COVID-19 crises (P2) is informative. While few Fed
chairman speeches contained UMP-related terms during P1, the “new normal” is on
its way to being established during P2. More interestingly, the UMP- and uncertainty-
related terms are relatively correlated during P2, whereas this correlation is
nonexistent during P1. This comparison is, to a lesser extent, also valid for the VIX-
and uncertainty-related terms, which are less correlated during P1 compared to P2.
Figure 21 shows that uncertainty-related terms in Fed chairman speeches were
fewer during the COVID-19 crisis than in the previous years and previous crises.
However, the contrast is less stark for the announcements and the minutes. The strong
instability in the speeches is due to the wide-ranging fields and objectives they cover,
coupled with the fact that speeches are generally less supervised than announcements
and minutes.
The communication related to UMP occurred after the volatility peaks during
the GFC and COVID-19 crises. Figures 20 and 21 highlight the Fed interventionism
policy, which most central banks of developed countries employ: after each FFR
decreases, UMP communication, usually followed by actions, compensates for the
0
10
20
30
40
50
60
70
-3,0
-2,0
-1,0
0,0
1,0
2,0
3,0
4,0
5,0
6,0
7,0
Financial Stability Sentiment UMP Terms Uncertainty Terms FFR VIX
247
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
inability of the central bank to use the nominal interest rate, their main policy
instrument, stuck at the ZLB.
Figure 22 aggregates the Fed’s three communication types to present a global
picture of the Fed’s communications.
Our previous finding that the Fed had a crisis-specific communication strategy
is confirmed. Indeed, Figure 22 supports the finding that, during crises, the Fed
decreased the sentiment associated with UMP measures around the same time that it
decreased the frequency of uncertainty-related terms in its communications.
Decreases in financial stability sentiment generally precede VIX increases, except
in the COVID-19 crisis. A potential explanation for this finding is that the COVID-19
crisis was less predictable than the dot-com and GFC crises.
Figure 22. Financial Stability and Monetary Policy in Main Fed Communications
Notes: The gray shaded area represents NBER recession periods. FSS stands for “financial stability
sentiment.” The right axis indicates the VIX and uncertainty terms levels.
Following the GFC, a new normal was established in which the Fed’s
communications came to be increasingly fed by discussions of UMP tools, including
forward-guidance measures. Typically, these UMP discussions were characterized by
a high-level contextual uncertainty. This new normal was partly upset by the COVID-
19 crisis, where the Fed adopted a communication strategy to use fewer uncertainty-
related terms in their UMP communications.
10
20
30
40
50
60
70
-3,0
-2,0
-1,0
0,0
1,0
2,0
3,0
4,0
5,0
6,0
7,0
FSS UMP Terms UMP Sentiment FFR VIX
Uncertainty Terms
248
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
The previous figures were focused on financial stability, but the Fed’s
communications also addressed economic stability. Figure 23 relates the word-
counting indicators discussed above to the nominal effective exchange rate (NEER).
Figure 23. Financial Stability, NEER, and FFR Announcements
Notes: The gray shaded area represents NBER recession periods. The NEER corresponds to the amount
of US dollars needed to purchase foreign currency (right axis). The financial stability sentiment, UMP,
and uncertainty terms are related to FFR announcements.
Figure 23 shows that increases in the NEER are correlated to decreases in the
frequency of uncertainty-related words in the Fed’s FFR announcements. It also shows
UMP communications (and actions) generally decrease the NEER except during the
tapering period where the NEER increased. High NEER levels also correspond to low
levels of uncertainty-related terms in the Fed’s communications.
Figure 24 relates the sentiment indicators discussed above to another measure of
economic stability: the unemployment rate.
Figure 24 shows that sentiment and the unemployment rate are almost always
inversely related. When the aggregated sentiment is positive, the unemployment rate
tends to decrease. A switch from a positive to a negative aggregated sentiment usually
coincides with an increase in the unemployment rate. The unpredictable nature of the
COVID-19 crisis makes this statement debatable but not necessarily wrong. Unlike
previous crises, the COVID-19 crisis increased the unemployment rate in the US from
3.5% to 13% in a short period of time, between January and May 2020. The
unemployment rate decreased to lower levels after May 2020, around 6%. The sharp
0,8
0,85
0,9
0,95
1
1,05
1,1
1,15
1,2
-3,0
-1,0
1,0
3,0
5,0
7,0
9,0
Financial Stability Sentiment UMP Terms Uncertainty Terms NEER
249
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
and short-time shock on the aggregate sentiment was even shorter than the one on the
unemployment rate, which may indicate a crisis-specific communication strategy
toward communication optimism during crises.
Figure 24. Sentiment and Unemployment
Notes: The gray shaded area represents NBER recession periods. The right axis indicates the
unemployment rate levels. The average sentiment aggregate contains an equally weighted average of
sentiments according to the Loughran and McDonald (score and polarity), Hu&Liu (polarity), Jockers
(polarity), NRC (polarity), SentiWords (polarity), UMP (score), and financial stability (score)
sentiments. To achieve a balanced aggregated indicator for each communication type, we weight this
average sentiment aggregate for FFR announcements more than for FOMC minutes, which in turn are
weighted more than for chairman speeches.
Figure 24 confirms this intuition again that the Fed adopted a communication
strategy to convey positive sentiments during the COVID-19 crisis.
To summarize, we show that the financial stability sentiment relates to FFR
decisions in FFR announcements and FOMC minutes but not significantly enough in
speeches. Increases in conventional monetary policy are often preceded by increases
in financial stability sentiment, except during the period where UMP terms are used
by Fed’s communications, which generally involve actions improving financial
stability.
We have also shown that the positive aggregated sentiment in the main Fed’s
communications correlates with decreasing unemployment. Except in times of
2
4
6
8
10
12
14
-3,0
-2,0
-1,0
0,0
1,0
2,0
3,0
4,0
Average Sentiments Unemployment Rate
250
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
significant UMP steps, the NEER correlates with the level of uncertainty in the Fed’s
communications.
9. Policy Implications
The Fed implemented more unconventional monetary policies during the COVID-19
pandemic than during the dot-com and GFC crises. Moreover, it did so in a concise
time window due to the abrupt upward slope of adverse shocks to the economy that
the COVID-19 restrictions generated. The Fed’s experience in crisis-specific
communication and UMP tools acquired during the dot-com and GFC also
contributed to understand better and address the COVID-19 crisis. To be successful,
the Fed’s UMP steps needed to be supported by clear and transparent
communications and engagement with both the financial markets and the public.
We show that both the supervised and unsupervised learning methods we
employ demonstrated that the Fed’s communications during the COVID-19 crisis
sharply differ from those of previous crises. Comparing the terms, sentiments, and
topics conveyed by the Fed’s communications with COVID-19 and financial data
confirms that the Fed adopted a specific communication strategy during the COVID-
19 crisis that also differs from the one adopted during the GFC and dot-com crises.
We conclude that the Fed is getting better at using its communications to manage
crises.
Our analysis determines that this communication policy consists of conveying
optimism to the public during the worst periods of the pandemic while discussing
(and implementing) earlier than in the previous crisis (Figure 14) unconventional
monetary policies by justifying their importance in mitigating risks and uncertainties
(Figures 16 and 17).
Another critical finding corroborates the Fed’s forward-looking ability and its
appropriate use of communication to convey a determined sentiment and justify UMP
before each wave of the virus or each worsening of the financial conditions due to the
virus’s spillovers.
While we are still far from recovery, our results show that communications
regarding the adopted policies and emergency programs allowed them to be
perceived as a useful tool supporting economic recovery.
The Fed may have conducted a specific communication strategy for the COVID-
19 crisis. This potential strategy conveyed less uncertainty and more optimism to the
public while promoting UMP measures for managing the crisis situation. We interpret
this behavior as conveying optimism without affecting transparency. The Fed’s timely
communications, together with its actions, succeeded in stabilizing financial markets.
10. Conclusion
This paper provides a comprehensive analysis of central bank communication during
the past two decades, emphasizing the COVID-19 pandemic.
We show that both the supervised and unsupervised learning methods we
employ determined that the Fed’s communications during the COVID-19 crisis
sharply differ from those of previous crises. Comparing the terms, sentiments, and
topics conveyed by the Fed’s communications with COVID-19 and financial data
251
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
confirms that the Fed adopted a specific communication strategy during the COVID-
19 crisis that also differs from the one adopted during the GFC and dot-com crises.
During the COVID-19 pandemic, the Fed’s communications emphasized topics
of health, social welfare, and UMP interventions, which appear to be related to the
conveyed sentiments. The Fed’s communications regarding COVID-19 and UMP
typically touch on the topics of financial volatility, uncertainty, and stability.
The content, sentiment, and timing of the Fed’s communications changed in the
COVID-19 crisis compared to previous crises. In particular, the sentiments of the Fed’s
communications significantly changed during the COVID-19 crisis compared to the
GFC. Following the GFC, communicating about UMP became a “new normal” in the
Fed’s minutes and chairman speeches. Interestingly, we have shown that a negative
financial stability sentiment usually precedes conventional monetary policy
accommodation, except under the ZLB.
COVID-19 caused structural changes in the Fed’s communication content. The
Fed may have implemented a specific communication policy for the COVID-19 crisis
that contrasted with its communication policy during the dot-com and GFC crises.
11. References
Benchimol, J., Kazinnik, S., and Saadon, Y., 2020a. Communication and transparency
through central bank texts. Paper presented at the 132nd Annual Meeting of the
American Economic Association, January 3–5, 2020, San Diego, CA, United States.
Benchimol, J., Kazinnik, S., and Saadon, Y., 2020b. Text mining methodologies with R:
An application to central bank texts. Occasional Paper #2020.01, Bank of Israel.
Bernanke, B.S., 2020. The new tools of monetary policy. American Economic Review
110(4), 943–983.
Bianchi, F., Faccini, R., Melosi, L., 2020. Monetary and fiscal policies in times of large
debt: Unity is strength. CEPR Discussion Paper #14720.
Blei, D. M., Ng, A. Y., and Jordan, M. I., 2003. Latent Dirichlet allocation. Journal of
Machine Learning Research 3, 993–1022.
Craig. B. R, Phelan, T., Siedlarek, J.-P., 2021. Modeling behavioral responses to
COVID-19. Economic Commentary, Federal Reserve Bank of Cleveland 2021(05),
1–6.
Correa, R., Garud, K., Londono, J. M., and Mislang, N., 2021. Sentiment in central
banks’ financial stability reports. Review of Finance 25(1), 85–120.
Daly, M. C., 2020. Is the Federal Reserve contributing to economic inequality? FRBSF
Economic Letter 2020(32), 1–7.
Erasmus, R., and Hollander, H., 2020. A forward guidance indicator for the South
African Reserve Bank: Implementing a text analysis algorithm. Working Paper
WP04/2020, Stellenbosch University, Department of Economics.
Chetty, R., Friedman, J. N., Hendren, N., and Stepner, M., 2020. The economic impacts
of COVID-19: Evidence from a new public database built using private sector data.
NBER Working Paper #27431.
Giacomini, R., and White, H., 2006. Tests of conditional predictive ability. Econometrica
74(6), 1545–1578.
252
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Guerrieri, V., Lorenzoni, G., Straub, L., Werning, I., 2020. Macroeconomic implications
of COVID-19: Can negative supply shocks cause demand shortages? NBER
Working Paper #26918.
Haldane, A., McMahon, M., 2018. Central bank communications and the general
public. AEA Papers and Proceedings 108, 578–583.
Hansen, S., McMahon, M., Tong, M., 2019. The long-run information effect of central
bank communication. Journal of Monetary Economics 108(C), 185–202.
Henry, E., 2008. Are investors influenced by how earnings press releases are written?
Journal of Business Communication 45(4), 363–407.
Loughran, T., and McDonald, B., 2011. When is a liability not a liability? Textual
analysis, dictionaries, and 10-Ks. Journal of Finance 66(1), 35–65.
Loughran, T., and McDonald, B., 2016. Textual analysis in accounting and finance: A
survey. Journal of Accounting Research 54(4), 1187–1230.
Nyman, R., Kapadia, S., Tuckett, D., Gregory, D., Ormerod, P., and Smith, R., 2018.
News and narratives in financial systems: Exploiting big data for systemic risk
assessment. Bank of England Working Paper #704, Bank of England.
12. Appendix
12.1 Unconventional Monetary Policy Dictionary
Our targeted lexicon (presented in Table 2) was constructed by collecting words
related to unconventional monetary policies from Fed communications using topic
modeling and Bag-of-Words (BoW); see Benchimol et al. (2020b).
Table 2. Unconventional Monetary Policy Lexicon
asset purchases
depreciation pressure
market disrupt
risk premium
helicopter
direct lending
market functioning
securities purchases
QE
ELB
monetary base
stagflation
securities purchases
foreign exchange reserve
monetary stimulus
support
balance sheet
forward guidance
money supply
support liquidity
business support
funding
negative policy
supporting corporat
credit facilit
insolvency
negative rate
swap line
credit impair
intervention
NIRP
unconventional
deferral
lending facilit
quantitative easing
ZLB
deflation
lower bound
relaxing regulatory
Source: Words and root words were extracted mainly from Fed communications.
253
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
12.2 COVID-19 Dictionary
Table 3 was constructed essentially from terms related to COVID-19 that appeared in
both media (e.g., Google Trends search queries) and recent Fed communications
(BoW) using the same methodology as in Table 2.
Table 3. COVID-19 Lexicon
acute
elderly
infect
pandemic
severe acute
cases
emergency
infection
pneumonia
sickness
confin
epidem
infection rate
quarantine
spreading
contagio
epidemic
lockdown
relief
syndrom
corona
hcov
mask
reproduction rate
testing
coronavirus
health
medical
respirator
vaccin
covid
hospital
morbid
respiratory
virus
death
hubei
morbidity rate
sars
wave
disabilit
human
mortal
sars cov
wuhan
disease
illness
ncov
sarscov
disorder
inception rate
outbreak
sars-cov
Source: Words and root words were extracted mainly from media and Fed communications.
12.3 Full Sample Figures
Figure A1. Sentiment Scores in FFR Announcements Full Sample
Notes: Solid black lines represent sentiment score values. Source: Benchimol et al. (2020a).
254
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Figure A2. Sentiment Scores for the Fed’s FOMC Minutes - Full Sample
Notes: Solid black lines represent sentiment score values. Source: Benchimol et al. (2020a).
Figure A3. Sentiment Scores for Fed Chairman Speeches Full Sample
Notes: Solid black lines represent sentiment score values. Source: Benchimol et al. (2020a).
255
Covid Economics 79, 27 May 2021: 218-256
COVID ECONOMICS
VETTED AND REA L-TIME PAPERS
Figure A4. Sentiment Scoring of Main Fed CommunicationsFull Sample
Notes: Solid black lines represent sentiment score values. Source: Benchimol et al. (2020a).
Figure A5. Topic Analysis of Main Fed Communications
Notes: For clarity and robustness, we restrict attention to the six most frequently discussed topics.
256
Covid Economics 79, 27 May 2021: 218-256
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
We review several existing text analysis methodologies and explain their formal application processes using the open-source software R and relevant packages. Several text mining applications to analyze central bank texts are presented.
Article
Full-text available
This paper applies algorithmic analysis to financial market text-based data to assess how narratives and sentiment might drive financial system developments. We find changes in emotional content in narratives are highly correlated across data sources and show the formation (and subsequent collapse) of exuberance prior to the global financial crisis. Our metrics also have predictive power for other commonly used indicators of sentiment and appear to influence economic variables. A novel machine learning application also points towards increasing consensus around the strongly positive narrative prior to the crisis. Together, our metrics might help to warn about impending financial system distress.
Article
The expansion of central bank communications and the increased use thereof as a policy tool to manage expectations have led to an area of research, semantic modelling, that analyses the words and phrases used by central banks. We use text-mining and text-analysis techniques on South African Reserve Bank monetary policy committee statements to construct an index measuring the stance of monetary policy: a forward guidance indicator (FGI). We show that, after controlling for market expectations, FGIs provide significant explanatory power for future changes in the repurchase interest rate (the primary monetary policy instrument). Their out-of-sample predictive power is, however, weak. Furthermore, we show that FGIs are primarily driven by inflation expectations, which highlights the strong link between the SARB’s communication strategy and its inflation targeting mandate. In fact, we observe a systematic anti-inflation bias in the communicated stance of monetary policy—both absolutely and asymmetrically. Overall, Monetary Policy Committee (MPC) statements reflect relevant information on the inflationary stance and policy decisions of the South African Reserve Bank (SARB), but, since forecasts are conditional on current information, they provide unreliable forward guidance. Given this finding, MPC statements should emphasize the conditional nature of the SARB’s stance, and what that implies for the future path of the policy rate.
Article
To overcome the limits on traditional monetary policy imposed by the effective lower bound on short-term interest rates, in recent years the Federal Reserve and other advanced-economy central banks have deployed new policy tools. This lecture reviews what we know about the new monetary tools, focusing on quantitative easing (QE) and forward guidance, the principal new tools used by the Fed. I argue that the new tools have proven effective at easing financial conditions when policy rates are constrained by the lower bound, even when financial markets are functioning normally, and that they can be made even more effective in the future. Accordingly, the new tools should become part of the standard central bank toolkit. Simulations of the Fed’s FRB/US model suggest that, if the nominal neutral interest rate is in the range of 2–3 percent, consistent with most estimates for the United States, then a combination of QE and forward guidance can provide the equivalent of roughly 3 percentage points of policy space, largely offsetting the effects of the lower bound. If the neutral rate is much lower, however, then overcoming the effects of the lower bound may require additional measures, such as a moderate increase in the inflation target or greater reliance on fiscal policy for economic stabilization. (JEL D78, E31, E43, E52, E58, E62)
Article
Why do long-run interest rates respond to central bank communication? Whereas existing explanations imply a common set of signals drives short and long-run yields, we show that news on economic uncertainty can have increasingly large effects along the yield curve. To evaluate this channel, we use the publication of the Bank of England’s Inflation Report, from which we measure a set of high-dimensional signals. The signals that drive long-run interest rates do not affect short-run rates and operate primarily through the term premium. This suggests communication plays an important role in shaping perceptions of long-run uncertainty.
Article
Despite the recent revolution in central bank communication, it is unclear the general public have benefitted from these enhanced central bank communications. We first analyze the growth of communication and some reasons that the public may have missed the revolution. We then discuss the desirability of greater communication with the public. The case for feasibility of such communication examines observational and experimental evidence following the recently adopted use of a layered communication at the Bank of England. We find that, while more research is needed, there are compelling reasons for central banks to continue trying to target communication at the public.
Article
Relative to quantitative methods traditionally used in accounting and finance, textual analysis is substantially less` precise. Thus, understanding the art is of equal importance to understanding the science. In this survey, we describe the nuances of the method and, as users of textual analysis, some of the tripwires in implementation. We also review the contemporary textual analysis literature and highlight areas of future research. © 2016 University of Chicago on behalf of the Accounting Research Center