What are They Meeting For?

A Tale of Two FOMC Announcements

Oliver Boguth, Vincent Gregoire, and Charles Martineau∗

January 14, 2016

ABSTRACT

In an effort to increase transparency, the Chair of the Board of Governors holds a pressconference following some, but not all, Federal Open Market Committee (fomc) an-nouncements since 2011. Press conferences are scheduled long in advance and thereforeindependent of economic conditions. Using evidence from equity and derivative marketsas well as media coverage and Google searches leading up to the announcements, we showthat this has led markets to lower their expectations of major monetary policy announce-ments on days without press conference. In turn, these expectations constrain the actionsof a committee that wants to avoid surprising markets.

JEL Classification: G10, G14, G18, E58.

Keywords: Federal Reserve, Press Conferences, Transparency, Uncertainty, Attention.

∗Boguth: W. P. Carey School of Business, Arizona State University, PO Box 873906, Tempe, AZ 85287-3906.Gregoire: Department of Finance, University of Melbourne, Level 12, 198 Berkeley Street, Carlton VIC, 3068,Australia. Martineau: Sauder School of Business, University of British Columbia, 2053 Main Mall, VancouverBC, Canada, V6T 1Z2. We thank Frank Hatheway of nasdaq omx for providing access to the TotalView-itch data and Compute Canada (WestGrid) for high-performance computing support. Martineau gratefullyacknowledges financial support from the Montreal Exchange and the nasdaq omx Educational Foundation.

“[T]he importance of the timing of a first decision to raise rates is something thatshould not be overblown, whether its September or December or March.”

– Janet L. Yellen, Chairwoman of the Board of Governors (June 17th, 2015)

The Federal Open Market Committee (fomc), the monetary policy-making body of

the U.S. Federal Reserve System (Fed), meets regularly to discuss the state of the

economy and monetary policy. Following the meetings, it communicates their decisions

to the public. Because asset prices are intimately linked to macroeconomic conditions,

market participants closely follow the news and quickly update prices to reflect the new

information.1 As a result, these announcements often move asset prices intensely.2

However, the mechanism through which the fomc presents information has changed

over time. While it was left to market participants to infer decisions from the Fed’s open

market operations until 1994, policy decisions are now announced in a press statement.

Since May 2011, in an effort to “provide additional transparency and accountability”

(Bernanke, 2011), the Chair of the Board of Governors additionally holds a press con-

ference (pc) following half of the announcements. Importantly, the decision to hold a pc

does not depend on macroeconomic conditions, as the schedule for both announcements

and pcs for a year is typically released in June of the previous year.3

We ask if the decision to hold press conferences following some meetings has led to

two classes of fomc announcements. It is conceivable that the Fed defers important

decisions for meetings when it has the opportunity to provide explanations and con-

text in a pc, and support their decisions with the release of the economic projection

materials.4

1A large literature documents returns surrounding and in response to macroeconomic news announcementsfor various asset classes. See, for example, Jones, Lamont, and Lumsdaine (1998), Fleming and Remolona(1999), Andersen, Bollerslev, Diebold, and Vega (2003), and Savor and Wilson (2013).

2Bernanke and Kuttner (2005) and Ozdagli and Weber (2015) estimate that a surprise decrease in thefederal funds rate of 25 basis points increase stock prices by one percent, whereas the analysis in Bjornlandand Leitemo (2009) suggests an even bigger impact. For bond markets, Cook and Hahn (1989) show thatchanges in the federal funds target rate are associated with changes in interest rates in the same direction.

3The only exception occurred when press conferences were first introduced in 2011. The schedule for 2011was announced on March 24, 2011, which is 5, 13, and 32 weeks before the respective press conferences.

4These materials contain the economic projections of Federal Reserve Board Members and Federal ReserveBank Presidents about growth, unemployment, and inflation, as well as their projections of future policydecisions. The materials are released only on days with press conferences, either 15 minutes prior to the pc

1

The official position of the Fed is that all meetings and announcements are equally

important, irrespective of press conferences. When asked if it is good “that the market

expects big news to come when you have a press conference and no news to come when

you don’t have one,” Chairwoman Yellen replied that she “would really like to strongly

discourage the expectation that policy moves can only occur when there’s a scheduled

press conference” (Yellen, 2014).5

In the opening quote of this paper, however, in June 2015 Chairwoman Yellen sug-

gests a first interest rate raise in “September [2015] or December [2015] or March [2016]”

(Yellen, 2015a). These three meetings have in common that they all are followed by a

press conference. The fomc also meets in July 2015, October 2015, and January 2016,

each without press conference following the announcement of their decisions.

If this latter quote is representative of a new approach of the Fed to time monetary

policy decision to coincide with press conferences, it would seriously question that pcs

increased transparency. Since pcs provide little incremental information relative to the

prior press statement, it is hard to imagine that the resulting gain in transparency

outweighs the loss caused by delaying decisions during fomc meetings without a press

conference.

Using evidence spanning three asset classes, we document striking differences in

markets’ expectations and reactions to fomc announcements with and without press

conference. We first show that average returns of the S&P 500 in the 30 minutes

immediately following the fomc announcement are large and positive on days with press

conference, averaging 29 basis points. This estimate is highly statistically significant

and robust to possible outliers and bootstrapped small-sample statistics. In contrast,

announcement returns are on average negative on days without pc. The difference in

announcement returns between pc and non-pc days is highly significant at 55 basis

points. This finding is robust to controlling for changes to the unemployment rate and

inflation, the two variables the fomc is mandated to manage, as well as past market

(2011-2012) or at the beginning of the pc (2013-2015). In both cases, at least 30 minutes pass between releasesof fomc announcements and economic projection materials.

5A similar exchange again occurred nine months later (Yellen, 2015b).

2

returns.

We argue that this ex-post reaction to fomc announcements can be used to proxy

for the ex-ante market expectation of Fed’s actions. The reasoning relies on the obser-

vation that throughout our sample similar information was revealed on both types of

announcements. In particular, the federal funds target rate, one of the main drivers

of equity prices in fomc announcements, remained unchanged at 0 to 25 basis points.

Since 2011, the fomc has therefore repeatedly surprised markets positively, with the

magnitude of the surprise directly proportional to ex-ante expectations of target rate

increases. The large market returns following announcements with press conference

then correspond to large ex-ante market expectations of rate increases.6

Two aspects about our analysis are important to emphasize. First, these findings

are about the market reaction to the fomc announcement. They are not returns

in anticipation of the announcement, as in Lucca and Moench (2015), nor do they

necessarily present profitable trading opportunities. Second, we analyze announcement

returns conditional on press conferences taking place, but the returns do not include

any information provided during the actual press conferences. Our study is silent

about the incremental information revealed in press conferences relative to the fomc

announcement.

We then confirm these findings in option markets, where we use the option-implied

volatility of the S&P 500, as measured by the vix index, to proxy for uncertainty associ-

ated with monetary policy. The vix drops sharply by more than 4% at announcements

if a pc is scheduled. In contrast, the vix remains virtually unchanged on days without

pcs. These findings suggest that uncertainty about monetary policy is high leading

up to announcements with press conference, and the Fed provides valuable information

to reduce this uncertainty. On days without pc, however, markets do not expect any

monetary policy changes, uncertainty is therefore low even before the announcement,

6For example, Kuttner (2001), Bernanke and Kuttner (2005), and Ozdagli and Weber (2015) show howimportant target rate announcements are for asset prices. Gurkaynak, Sack, and Swanson (2005) confirm thatrate announcements are important, but argue that the future path of policy is also important.

3

and the Fed does not surprise markets.7

Both changes in stock prices and volatility in response to fomc announcements

are only indirect measures of true expectations of changes to monetary policy. Using

a direct measure for expected interest rate changes, obtained from overnight index

swaps, we confirm our findings. On days with press conferences the market-implied

probability of rate changes is on average 2.4 percentage points, or about 25 percent,

higher than on announcement days without pc. In other words, markets expect more

relevant information on days with press conferences.

Of course, even though our evidence is derived from three different financial mar-

kets, we expect those markets to be at least partially integrated, and therefore provide

correlated evidence. We overcome this potential criticism and provide an independent

test by examining different measures of attention around fomc announcements.

We show that media attention significantly increases before announcements with

press conference. The effect is large and holds both for a low-frequency measure based

on articles in the print editions of major newspapers as well as for intraday newswire

articles. Just allowing conditional means to vary between pc and non-pc days explains

up to 34% of the variation in our media attention measures. A similar picture emerges

for a broader attention measure based on Google search volume in the week prior to

fomc announcements, where the number of searches for “fomc” is significantly higher

before announcements with pc than without.

Taken together, our evidence strongly supports the hypothesis that market expecta-

tions of relevant changes to monetary policy are lower and the press releases convey less

price-relevant information on fomc announcement days without press conferences. In

other words, in the markets’ view, the introduction of pcs separated fomc announce-

ments into important and lesser ones.

Interesting dynamics arise when markets disagree with the Fed about the impor-

tance of non-pc announcements. Clearly, when markets expect no changes to monetary

7An alternative interpretation for days without pc is that markets expect monetary policy decisions, butthe actual announcement does not contain any relevant information. Our findings for returns of the S&P 500are more consistent with the former interpretation.

4

policy, any action from the Fed would be interpreted as a surprise. However, the Fed

is frequently thought of as averse to surprising markets.8 This aversion imposes a con-

straint on the Fed’s actions, and market expectations can therefore become self-fulfilling.

This tension also increases the Fed’s incentives for the kind of informal communication

studied in Cieslak, Morse, and Vissing-Jorgensen (2015).

Press conferences were introduced with the intention to increase transparency. While

this goal is plausibly achieved for the main fomc announcements that are followed

by a pc, our evidence suggests that the reduced information revealed at second-rate

announcements decreases transparency at these intermediate times.

The implications of this new fomc announcement regime are difficult to gauge.

While transparency is frequently viewed as positive (Bernanke, 2013), it is less clear

whether increased transparency really results in lower price volatility or in prices that

better reflect fundamental values. See, for example, Amato, Morris, and Shin (2002),

Goldstein and Sapra (2013), Goldstein and Yan (2015), and Banerjee, Davis, and

Gondhi (2015).

Our findings suggest that with the advent of press conferences, there are now effec-

tively fewer meaningful fomc announcements. By itself, this does not need to be bad

for investors or overall welfare. For example, Abel, Eberly, and Panageas (2013) show

that investors that face information and transactions costs can be better off monitoring

their investments less frequently.

Taken to the extreme, our evidence raises the question why the fomc meets and

makes policy announcements on days without scheduled press conferences. If the ob-

jective of the fomc is to increase transparency while simultaneously limiting market

surprises and maintaining flexibility of action, it should consider following the practice

of holding press conferences after every meeting, as adopted by the European Central

Bank, the Bank of Japan, Sweden’s Riksbank and Norway’s Norges Bank.

8For example, in a recent paper Stein and Sunderam (2015), Jeremy Stein, a former member of the Boardof Governors, assumes that the central bank is averse to bond-market volatility. See also Cieslak, Morse,and Vissing-Jorgensen (2015) for a detailed discussion. In the press, a survey by the Wall Street Journal“underscores just how much work it would take for the Fed to create expectations of a rate increase at ameeting without a news conference” (Zumbrun, 2015).

5

I. The Federal Open Market Committee

The fomc is the monetary policy-making body of the U.S. Federal Reserve System.

It oversees the nation’s open market operations, i.e., purchases and sales of United

States Treasury and Federal Agency Securities, which affect the cost and availability of

money and credit in the U.S. economy, under the statutory dual mandate of maximum

employment and stable prices.9 By law, the fomc must meet at least four times a year.

Since 1981, however, eight regularly scheduled meetings have been held each year at

intervals of five to eight weeks.10 Prior to 1994, changes to the federal funds rate were

not announced and market participants had to infer it by observing the size and type

of open market operations. Starting in 1994, the fomc publicly announces their policy

decisions, and the announcement dates and times are publicly known in June of the

previous year.

Starting in 2011, the Chairman of the Board of Governors holds a press conference

following half of the fomc announcements. Importantly, just like the announcements

themselves, press conferences are also scheduled at least six months in advance, and the

decision to hold a press conference therefore does not depend on economic or market

conditions. Press conferences last on average 55 minutes and consist of an opening

statement by the Chairman of the Board of Governors followed by a question and

answer session with financial journalists.11

Table I provides an overview of the fomc announcements, their times, and the

starting time of the associated press conference. In total, our sample is comprised of 37

announcements, 19 with and 18 without press conference. After some initial irregulari-

9The fomc is composed of the seven members of the Board of Governors and five of the twelve ReserveBank presidents. While the president of the Federal Reserve Bank of New York serves on a continuous basis,the presidents of the other Reserve Banks serve one-year terms on a rotating basis.

10Members may also be called on to participate in special meetings if circumstances require consultation orconsideration of an action between these regular meetings. For the purpose of this paper, we are not interestedin these special meetings.

11In 2011 and 2012, fomc announcements with pc were made at 12:30 pm, followed by the release of theeconomic projection materials at 2:00 pm and the pc beginning at 2:15 pm. Announcements without pc weremade at 2:15 pm. Since 2013, fomc announcement always occur at 2:00 pm. If there is a press conference, theeconomic projection materials are released at the beginning of the pc at 2:30 pm.

6

ties, since June 2012 press conferences have followed every other fomc announcement.

A. Significant FOMC Announcements

Throughout our sample, the federal funds target range remained constant at 0 to 0.25

percent. Nevertheless, our sample contains some changes in monetary policy by means

of quantitative easing (qe) to help reviving the U.S. economy following the financial

crisis. We now list some of the key fomc announcements since 2011.

June 22, 2011 (pc): the Fed announces the end of qe2 (Fawley and Neely, 2013).

September 21, 2011 (no pc): the Fed announces Operation Twist, which consisted of

purchasing $400 billion of Treasuries with long maturities and selling an equal

amount with shorter-term maturities.

June 20, 2012 (pc): the Fed announces that it will continue Operation Twist.

September 13, 2012 (pc): the Fed announces qe3.

December 12, 2012 (pc): the Fed announces the expansion of qe3.

June 19, 2013 (pc): During the pc, Chairman Bernanke suggests a gradual “tapering

down” of the bond purchases could begin in the months to come.

September 18, 2013 (pc): the Fed decided to hold off on “tapering”.

October 29, 2014 (no pc): the Fed announces the halt of bond purchases.

June 17th, 2015 (pc): Chairman Yellen specifies three possible dates for a first raise

in interest rates (Yellen, 2015a).

All but two of these key fomc announcements occurred on days with press confer-

ence, suggesting that the Fed might time important announcements to coincide with

pcs. In the following section, we describe the data and methodology used to formally

test this conjecture.

7

II. The Role of Press Conferences for Market Expectations

In this section, we present the main empirical findings of the paper. We use measures

from equity, options, and interest rate swap markets to show that there have been

two classes of fomc announcements since 2011. Only events with a press conference

are associated with large expectations of important monetary policy decisions and a

significant resolution of uncertainty at the time of the announcement.

A. Stock Market Announcement Returns

We begin our analysis by showing that stock market reactions to fomc announcements

differ across days with and without press conference. If markets are efficient, these

announcement returns measure the unexpected component of the announcement. We

argue that, specific to our sample, these surprises can also be used to proxy for the

expected part of the announcement.

Our identification relies on the observation that there is little variation in the total

information content, expected and unexpected, of announcements in our sample. In

particular, the federal funds target rate, the single most closely watched number associ-

ated with fomc announcements, has remained at its lower bound of 0 to 0.25 percent.

Any decisions regarding this rate can therefore be thought of as binary: rates can either

remain unchanged or increase.12

Since unexpected rate increase typically lead to a drop in equity prices (Kuttner,

2001, Bernanke and Kuttner, 2005), in this scenario prices will always rise when the

Fed announces that rates remain low. The magnitude of the rise, however, depends on

the market’s ex-ante expectations that rates would increase. For example, if markets

are certain that rates will not change, an announcement of no increase should not affect

prices. If on the other hand markets have a large expectation of a rate increase, any

announcement of constant rates should be considered a large positive surprise, and

12In practice, unconventional monetary policies, such as large-scale asset purchases, can be used to effectivelyovercome the zero lower bound (Swanson, 2015). On the other end, target rates could increase by more than25 basis points.

8

stock prices should therefore increase.

We retrieve intraday data for the S&P 500 tracking exchange traded fund, spy,

from the nasdaq TotalView-itch database. TotalView-itch tracks all orders that are

submitted, canceled, or executed on nasdaq, which allows us to construct the entire

limit order book.13 We define the spy price as the midpoint of the best outstanding

bid and ask quotes, and convert this time-series of prices into one-second midquote

returns.14 We further restrict our sample to regular trading hours (i.e., 9:30 am to 4:00

pm est).

Figure 1 plots the cumulative return of spy around the fomc announcement, start-

ing 24 hours (6.5 trading hours) before and ending 1.75 hours after the announcement.

Returns are normalized to zero at the announcement. The ending point of 1.75 hours is

chosen to avoid potential effects from overnight returns. As shown in Table I, between

August 2011 and January 2013, announcements without press conference were made at

2:15 pm, or 1.75 hours before market close.

Panel A groups all fomc announcements from April 2011 to October 2015. Inter-

estingly, the return in the 24 hours leading up to the announcement is very flat, and

the pre-fomc announcement drift of Lucca and Moench (2015) is absent in this more

recent period. Consistent with the conjecture that fomc announcements throughout

our sample contained good news for equity markets, there is a small return of about 10

basis points immediately following the announcement, or up to 20 basis points in the

hour after the announcement. The 95% confidence interval, plotted in gray, suggests

that this effect is not statistically significant.

A striking pattern emerges in Panel B, where we separate fomc announcements

into ones with and without press conference. When there is a pc (blue solid line),

prices increase by economically large and statistically significant 40 basis points after

the announcement. In contrast, fomc announcements without pc (red dashed line)

13Importantly, this data is not limited to nasdaq listed securities. According to Fidessa Fragmentation Index(2015), between 20 and 25 percent of the total market turnover of spy during our sample occurs on nasdaq.The average bid-ask spread of spy in our data is 1.1 cents, or less than 0.01 percent. Clearly, the quoting andtrading activity we observe is sufficient to obtain meaningful prices.

14We ignore order sizes of the best quotes, and, more generally, the depth of the limit order book.

9

are accompanied by a drop in prices of about 10 basis points during a volatile period

following the announcement.

We formally test the main insights from Figure 1 in Table II. The table provides

estimates of moments and associated statistical tests of announcement returns, which

we define as the cumulative return of spy in the 30 minutes event window starting at

the announcement. The choice of 30 minutes follows Ozdagli and Weber (2015), and

further ensures that the announcement returns are not affected by information released

during the press conference. The full sample results in Panel A show a positive aver-

age announcement return of 3 basis points. On days with press conference, this figure

rises to 29 basis points, while it is -26 on days without pc. Based on the asymptotic

distribution, mean returns for all announcements and for non-pc announcements are

insignificant. Announcements returns with press conference are both significantly pos-

itive and significantly larger than those without. On days with pc, they range from

-0.40 to 1.16 percent, with only 4 out of 19 observations (21%) negative.

Our evidence is based on a rather small sample containing only 19 (18) observations

for pc (non-pc) events. When working with small samples, natural concerns about

the sample distribution of the test statistic and the effect of possible outliers arise.

We address these concern in two ways. First, in addition to asymptotic standard

errors, we also provide bootstrapped standard errors and p-values for the test that

mean announcement returns are less than or equal to zero. All bootstrapped results

are based on 1,000,000 samples. The bootstrapped standard errors closely resemble the

asymptotic ones, and the p-values confirm the findings. In particular, p-values near

zero suggest that announcement returns on pc days are both significantly positive and

significantly greater than on non-pc days.

Second, to investigate the potential impact of outliers, Panel B repeats the analysis

on a trimmed sample that excludes both the largest and smallest announcement return

observations. Point estimates for the means are, with one exception, little affected.

On non-pc days, average returns rise from -0.26 to -0.16, the minimum increases from

10

-2.45% to -0.75%, and the standard deviation declines from 0.64 to 0.32. This implies

that the sample was affected by one very large negative observation. Crucially, even

in the trimmed sample, the statistical inference remains unchanged. Announcement

returns on days with press conference are significantly positive, and larger than those

on days without press conference.

We test whether the announcement return differences between pc and non-pc days

can be explained by different economic environments in Table III. The first two speci-

fications regress announcement returns on two indicator variables, one for pc and one

for non-pc days, or a constant and one indicator variable. These two tests confirm the

results from Table II under the additional assumptions ordinary least square regressions

require about the error distribution. Just allowing for differences in averages between

pc and non-pc days explains 19% of the variation in announcement returns.

In the third specification we add monthly changes in seasonally adjusted consumer

price index (∆CPI) and unemployment (∆UNEMP ) to control for the economic envi-

ronment. These variables are the most natural candidates to influence expected mone-

tary policy, as they correspond to the fomc’s target measures under its dual mandate.

Data are obtained from the U.S. Bureau of Labor Statistics, and we always use the most

recently announced data. In the fourth specification, we further control for the cumula-

tive return of the S&P 500 over the 21 trading days ending three days before the event,

(RS&P). The specific window is chosen to avoid overlap with both the previous and the

current fomc meetings. The daily S&P 500 returns are from Thomson Reuters Tick

History as supplied by the Securities Industry Research Centre of Asia-Pacific (sirca).

Of the control variables, changes in the unemployment rate are significantly neg-

atively and the prior 21-day S&P 500 returns significantly positively related to an-

nouncement returns. The signs are consistent with our interpretation of the dependent

variable. Following improvements in the state of the economy, such as a decrease in

the unemployment rate or rising stock prices, markets expect a tightening in monetary

policy. Announcements to keep policy unchanged therefore result in large positive sur-

11

prises. Importantly, none of the control variables has any impact on the coefficient on

the press conference indicator. The marginal impact of press conferences on announce-

ment returns is very stable across specifications, ranging from 55 to 57 basis points,

and highly statistically significant.

To investigate if the announcement returns are persistent, Figure 2 shows spy re-

turns over a longer window of two trading days before and after the event. Two notewor-

thy patterns emerge. First, the market increases by about 35 basis points over a short

window on the morning of the day before fomc announcement days with pc. Despite

the small sample, standard error bounds indicate that this movement is statistically

significant. Since it occurs prior to the actual news release, this drift appears similar

to the one identified in Lucca and Moench (2015). The timing, however, is different.

The rise is prices ends about 24 hours prior to the announcement, the time when the

pre-fomc announcement drift in Lucca and Moench (2015) would start. We speculate

that some investors, now aware of the pre-fomc announcement drift, attempt to front-

run it. Second, the differences in announcement effects we document are persistent for

at least 48 hours. The point estimates indicate that prices revert slowly following the

negative reaction to non-pc announcements, but standard errors are large.

B. Revelation of Uncertainty at FOMC Announcements

The use of ex-post stock market announcement returns to proxy for ex-ante expec-

tations relies on the key assumption that the total information content of all fomc

announcements in our sample in comparable. We now provide more direct evidence by

showing that fomc announcements reduce uncertainty about monetary policy on days

with press conference, but not on other days.

We follow Beber and Brandt (2009) and Amengual and Xiu (2015) and use the

option implied volatility index, vix, as proxy for uncertainty associated with mone-

tary policy. With the arrival of new information, we generally expect uncertainty to

decrease.15 But volatility would change little if announcements merely confirm what

15Beber and Brandt (2009) show a general link between resolution of macroeconomic uncertainty and changes

12

markets already expected, or if announcements provide little price-relevant informa-

tion. If on the other hand uncertainty in the market was large, and the announcement

resolves this uncertainty, we expect large declines in the vix.

Figure 3 shows cumulative changes in the vix around the fomc announcement,

starting 24 hours prior and ending 1.75 hours after. The intraday vix data is provided

by Thomson Reuters Tick History. Across all fomc announcements (Panel A), the

vix exhibits the expected pattern. There is little time-series variation prior to the

announcement, but the vix drops sharply by about 2% when the new information

arrives. The release of the Fed’s monetary policy decisions clearly reduces uncertainty

in the stock market.

A striking contrast emerges in Panel B, which splits up fomc announcements into

ones that are followed by a press conference (blue solid line) and ones that are not

(red dashed line). While announcements with press conference see an average drop

of over 4% in the volatility index, uncertainty remains virtually unaffected by fomc

announcements without pc.

Table IV formally tests this finding. We first regress the level of the vix five minutes

before the fomc announcement on a pc indicator and a non-pc indicator, and then

on an intercept and a pc indicator (specifications 1 and 2).16 There is no significant

difference in the pre-annoucement vix levels on pc and non-pc days on average, or

after controlling for macroeconomic variables and market conditions.

The last four columns show how the vix changes at the announcement, looking at

the vix level 30 minutes following the announcement compared to 5 minutes before.

Regression (5) shows that the vix decreases by a statistically and economically highly

significant 4.4% on days with pc, and remains unchanged on days without pc. Including

control variables affect neither the economic magnitude nor the statistical inferences of

in the vix index. Amengual and Xiu (2015) are specifically interested in large downward jumps in the vix, andargue that in addition to resolving uncertainty, the Fed usually intervenes in hard times, effectively providinga put option to markets. For our purposes, the distinction between both interpretations is secondary.

16We use the vix level five minutes before the announcement to ensure that our findings are not affected byeither information leakage prior to the announcement, or possible data errors with regard to the exact fomcannouncement time. The results are robust to using the exact announcement time.

13

the impact of press conferences.

The large decrease in option-implied volatility suggests that a lot of uncertainty in

equity markets is resolved at the time of the announcement on days with pc. In contrast,

when there is no pc, uncertainty does not change around fomc announcements. In

turn, this implies that fomc announcements communicate price-relevant information

only on pc days, and markets correctly expect no relevant monetary policy changes on

days without pc.

The argument that important monetary policy decisions should reduce uncertainty

in the market is general and, in contrast to the evidence using announcement returns,

does not rely on the specific sample. This allows us to investigate if the segregation

of fomc announcements is a new effect caused by press conferences, or if historically

quarter-end announcements have always implicitly carried a higher weight.

Figure 4 shows changes in the vix around fomc announcements from January 2006

to March 2011, separately for the first (solid blue line) and second (dashed red line)

announcements in each calendar quarter. In short, there is no difference. Therefore,

there is no evidence to suggest that the timing of press conferences simply reflects a

previously existing pattern. Instead, the separation into important and less important

fomc announcements seems to be caused by the advent of press conferences.

C. Ex-Ante Implied Probabilities of Target Rate Changes

Both announcement returns and changes in volatility are ex-post measures that might

be affected by the content of the announcement. We now validate these results using

a pure ex-ante measure from derivative markets that directly captures the expected

gravity of fomc announcements, that is, the expected volatility of monetary policy.

We measure the ex-ante expectations of rate changes using the Overnight Index Swap

(ois) Implied Probability provided by Bloomberg. These probabilities are estimated by

stripping the ois term structure into forward rates, which in turn are used to compute

the expected Federal Funds Effective Rate (FFER) before and after the next fomc

14

meeting.17

We compute the probability of a rate change as:

Prob(Increase) = max([FFER+ − FFER−] /0.25, 0

)Prob(Decrease) = max

([FFER− − FFER+

]/0.25, 0

)Prob(Change) = Prob(Increase) + Prob(Decrease)

where FFER− and FFER+ are the expected FFER before and after the next meeting

respectively. This definition assumes that, at any given point in time, only one of

Prob(Increase) or Prob(Decrease) can be non-zero, and that rate changes occur only

in steps of 25 basis points.18

For each fomc meeting, we obtain the ois implied probability computed on the

previous day. To test whether press conferences affect the probability of rate changes,

we regress changes in the ois implied probability onto changes in an indicator variable

for press conferences and control variables. Performing our test on changes rather than

levels avoids concerns that variables might be non-stationary in-sample.

Our findings are summarized in Table V. In the first three columns, the dependent

variable is the difference in market-implied probabilities of changes in interest rates

from the prior fomc announcement to the current one. The first specification only

contains an intercept and changes in the pc indicator variable, ∆PC, which can take

one of three values: one if the announcement has a pc while the previous did not, minus

one for the opposite case, and zero if both the current and prior announcement were

followed by or not followed by pcs.

It shows that, on average, the probability of rate changes is 2.4 percentage points

17Alternatively, the probabilities could be extracted from federal funds futures prices around the announce-ment, as suggested by Kuttner (2001). These contracts trade with monthly maturities, aligned by calendarmonth, and pay off the average daily federal funds futures rate over the time of the contract. However, toobtain surprises and therefore expectations, Kuttner’s approach requires the use of federal fund futures pricesfrom the end of the announcement day. This is not suitable for our purposes, as those end-of-day prices containinformation revealed during the press conference.

18Even though the federal funds target rate is at its zero lower bound, we observe positive probabilities ofrate decreases following fomc announcements. This is because rates are targeted to stay within an interval, inour case 0 to 0.25 percent, rather than a specific number, and the ois pricing is based on specific market ratesrather than target rates.

15

higher on days with press conferences than on those without. The estimate is statisti-

cally significant and economically large. With an average probability of rate changes of

10.3% in our sample, this estimate suggests that press conferences are associated with

a 25% increased probability of Fed action (11.5% on pc days, 9.1% on non-pc days).

When controlling for changes in inflation and unemployment rates in specifications

(2) and (3) the coefficient on the pc indicator is unaffected, and press conferences

remain associated with a higher probability of rate changes. As before, inflation is not

significantly related to the probability of changing interest rates, while unemployment

negatively predicts the probability of rate changes. The coefficient on the S&P 500

return is significantly negative, suggesting that changes in market prices reflect the

altered probabilities of interest hikes.

Since the target rate has been at its zero lower bound throughout our sample, we

also perform the tests on the narrower probability of target rate increases. The results,

shown in columns (4)-(6) of Table V, confirm the previous findings. On days with press

conferences, the probability of rate increases is between 2.1 and 2.3 percentage points

higher than on days without press conferences. Relative to the unconditional average

of a rate increase of 3.1% in our sample, this corresponds to a 50 percent increase in

probability on press conference days relative to non-pc days.19

Overall, the prices of the ois, combined with the reactions of equity and option

markets to fomc announcements, paint a clear picture that markets expect big changes

in monetary policy only following fomc meetings with press conferences, and view the

remaining announcements as less important.

III. When do People Pay Attention to FOMC Announcements?

The previous section suggests that financial markets expect more important monetary

policy decisions prior to fomc announcements with press conferences, and the informa-

19The unconditional mean of the probability of target rate increases of 3.1% might appear low consideringthat target rates are at their zero lower bound. However, in the first half of our sample, markets simply didnot expect increases in rates. This view is also supported by the observed unconventional measures to furtherloosen monetary policy. Conditional on expected target rate increases, the mean rises to 7.6%.

16

tion revealed during the announcement complies with the market’s expectations. Even

though we present evidence from three different markets, we expect them to be at least

partially integrated. As an additional independent test, we now show that different

measures of attention, in particular media coverage spanning multiple frequencies and

Google search volume, are also higher on announcement days with press conferences.

This confirms that the perceived uncertainty associated with the information is higher

on announcement days with press conference than without.

A. Printed News

We begin our analysis with a measure of media attention that is based on low-frequency

printed news in the Wall Street Journal (wsj) and the New York Times (nyt) around

fomc announcements. To measure daily news intensity, we follow Fisher, Martineau,

and Sheng (2015) and divide the number of articles related to the fomc or mone-

tary policy by the total number of articles published in the morning edition of each

newspaper.20 We then average daily intensity over four-day windows ending on the an-

nouncement day and difference the observations to look at meeting-to-meeting changes

in news intensity.21 Fisher, Martineau, and Sheng (2015) provide a detailed overview

over the construction of macroeconomic media attention indices and their statistical

properties.

Our findings are summarized in Table VI. We present three groups of results where

the dependent variables are the changes in news intensity for wsj, nyt, and a com-

bined index that is computed as the average of the two standardized series. The first

specification within each group only contains an intercept and a pc indicator variable.

It shows that, on average, the media intensity in the wsj is 70 basis points higher

on days with pc than on days without. The estimate is economically large since the

variation associated with press conferences represents around 25% of the unconditional

mean (2.7%). The indicator variable alone explains 12.4 percent of the total variation

20In particular, we search for the following key words: ((federal reserve or federal open market committeeor fomc) and (interest rate or monetary or inflation or economy or economic or unemployment)).

21We can include the announcement day since all the news considered are from the morning editions.

17

in media coverage. Moreover, the estimate is highly statistically significant, with a

p-value of 0.01.

A similar picture emerges for our media attention measure for the nyt. While

the coefficient is much smaller (20 basis points), it is important to keep in mind that

the nyt is a general news media outlet that is less focused on economic and financial

coverage than the wsj. Accordingly, the average proportion of relevant articles is lower

at 1.0%, and the estimated coefficient is therefore again sizable relative to the mean.

Overall, control variables have little impact on our estimates, and the combined sample

confirms the findings.

As expected, media attention is greater prior to fomc announcements with press

conference. This suggests that journalists view fomc announcements with press con-

ference as more important than those without, and reinforces our previous findings.

B. Intraday Newswire

We next move to a high-frequency measure of media attention that is based on intra-

day newswires from RavenPack’s global macroeconomic news database in the hours

before and the minutes after fomc announcements. We collect from this database a

comprehensive sample of news stories from the Dow Jones News Wire about two en-

tities, the Federal Reserve and the Federal Open Market Committee.22 We eliminate

intraday news that are not classified as full-article.23 The articles are timestamped,

which allows us to count the number of articles released during a specific time period.

RavenPack further assigns to each news a relevance score between 0 and 100 to capture

the predominance of the entity mentioned in the news. We select news articles with

a minimum relevance score of 90, and make sure not to double count news that are

relevant for both entities.

Our findings are summarized in Table VII. We present two groups of results with

different dependent variables. The first counts newswire articles in the 24 hours leading

22This also includes the wsj, wsj online edition, Barron’s, and MarketWatch.23The news fall into three categories: news-flash, tabular, and full-article. News-flash is a single sentence

released after an important announcement, while tabular news only contain a table with statistics or figures.

18

up to each fomc announcement, and the second in the 30 minutes after.

We expect the number of articles in the 24 hours prior to the announcement to

be correlated with the data from the printed news sample. As before, our coefficient

estimate on ∆PC is positive, highly significant, and unaffected by controls. We find

that on days with pc the raw article count increases by more than 70, which again

is very large compared to the unconditional mean of 156. These results confirm those

obtained from the printed news.

We next turn our attention to articles published after the announcement. Our

sample contains only full-length articles published within 30 minutes of the fomc an-

nouncements. These would likely be articles written during the fomc embargo period

in which accredited news agencies get access to pre-release material in order to publish

timely reports once the press release is made public. We therefore expect our variable

to proxy for both the ex-ante interest, represented for example by the number of re-

porters present, and the unexpected component of the press release, i.e., news-worthy

information. As before, we limit ourselves to a 30-minute window following the an-

nouncement in order to avoid any overlap with an actual press conference, which would

mechanically increase media articles. The coefficient estimate on ∆PC confirms the

previous findings: it is again significantly positive in all specifications. Article count

increase by about 12 on days with pc relative to days without pc, compared to an

unconditional mean of 47. This is further evidence in support of more attention and

importance attached to fomc announcements followed by pc.

C. Google Search Volume

We conclude our analysis with the search volume measure from Google Trends data.

Google Trends measures the popularity of a subject based on key words searches coming

from the population in Googles search engine. This data is more appropriately used to

identify awareness of the general population of specific keywords or topics rather than

their popularity. Data obtained from Google Trends have previously been used in asset

pricing, for example to study the effects of investor attention (Da, Engelberg, and Gao,

19

2011) and to obtain broad sentiment measures (Da, Engelberg, and Gao, 2015).

In particular, the weekly Google Trend search volume index (svi) is calculated by

dividing the number of searches for a specific keyword (“fomc”), by the total number

of searches in a geographic area (“global”), and rescaling the resulting series so that

the maximum is 100.

While our previous measures looked at the attention from media only, the search

volume index proxies for the overall level of interest in the general population. For

example, Da, Engelberg, and Gao (2011) argue that their svi likely measures the at-

tention from retail investors. We use the svi in the last full week prior to each fomc

meeting, and again analyze meeting-to-meeting changes in the svi.24

The findings mirror the previous ones. Search volume for “fomc” is higher by about

three points prior to announcements with pc than without, and the findings are robust

to controls.

IV. Conclusion

In an effort to increase transparency, the Chair of the Board of Governors now holds a

press conference following half of the scheduled fomc announcements. We show that

this information practice has unintended effects that curtail the range of actions the

Fed can take and counteract the declared transparency goal.

Building on evidence spanning three asset markets and multiple attention measures,

we show that the practice to hold press conferences on some, but not all, days has led

to the perception of two classes of fomc announcement. Markets expect important

monetary policy decisions only on announcement days with press conference, and in

turn, market reactions suggest that the Fed reveals significantly less price-relevant in-

formation on days without press conferences.

These expectations imply that any major monetary policy decision on a day with-

24Concerns might arise if some fomc announcements are later in the week than others. In our sample, thevast majority of announcements fall on a Wednesday, only three on a Tuesday and two on a Thursday. Sincethe two Thursday announcements are followed by a pc while the three Tuesday announcements are not, weconjecture that any bias would lower the significance of our results.

20

out press conference would be a surprise to markets. For a committee that is generally

believed to be averse to surprising markets, the current information practice thus makes

big announcements on days without press conference very costly, constraining the pos-

sible monetary policy decisions. Naturally, these constraints diminish information flow

and reduce transparency.

Taken to the extreme, our evidence raises the question why the fomc meets and

makes policy announcements when there is no press conference. Resolving the con-

straints on actions and the associated reduced transparency requires that markets per-

ceive all fomc announcement equal. This could be achieved by only meeting four times

a year, or by removing press conferences completely. However, to maintain their goal of

increased transparency, the Fed should instead consider holding press conferences after

every meeting, as many other central banks do.

21

References

Abel, Andrew B., Janice C. Eberly, and Stavros Panageas, 2013, Optimal inattention to the stockmarket with information costs and transactions costs, Econometrica 81, 1455–1481.

Amato, Jeffery D., Stephen Morris, and Hyun Song Shin, 2002, Communication and monetary policy,Oxford Review of Economic Policy 18, 495–503.

Amengual, Dante, and Dacheng Xiu, 2015, Resolution of policy uncertainty and sudden declines involatility, Working Paper, University of Chicago, Chicago, IL.

Andersen, Torben G., Tim Bollerslev, Francis X. Diebold, and Clara Vega, 2003, Micro effects ofmacro announcements: Real-time price discovery in foreign exchange, American Economic Review93, 38–62.

Banerjee, Snehal, Jesse Davis, and Naveen Gondhi, 2015, When transparency improves, must pricesreflect fundamentals better?, Working Paper, Northwestern University, Evanston, IL.

Beber, Alessandro, and Michael W. Brandt, 2009, Resolving macroeconomic uncertainty in stock andbond markets, Review of Finance 13, 1–45.

Bernanke, Ben S., 2011, Transcript of Chairman Bernanke’s Press Conference, April 27, 2011,http://www.federalreserve.gov/mediacenter/files/FOMCpresconf20110427.pdf, accessed Nov. 18,2015.

, 2013, Communication and monetary policy, Speech at the National Economists Club AnnualDinner, Herbert Stein Memorial Lecture Washington, DC, November 19, 2013.

, and Kenneth N. Kuttner, 2005, What explains the stock market’s reaction to Federal Reservepolicy?, Journal of Finance 60, 1221–1257.

Bjornland, Hilde C., and Kai Leitemo, 2009, Identifying the interdependence between US monetarypolicy and the stock market, Journal of Monetary Economics 56, 275–282.

Cieslak, Anna, Adair Morse, and Annette Vissing-Jorgensen, 2015, Stock returns over the FOMCcycle, Working Paper, University of California at Berkeley, Berkeley, CA.

Cook, Timothy, and Thomas Hahn, 1989, The effect of changes in the federal-funds rate target onmarket interest-rates in the 1970s, Journal of Monetary Economics 24, 331–351.

Da, Zhi, Joseph Engelberg, and Pengjie Gao, 2011, In search of attention, Journal of Finance 66,1461–1499.

, 2015, The sum of all FEARS: Investor sentiment and asset prices, Review of Financial Studies28, 1–32.

Fawley, Brett, and Christopher J. Neely, 2013, Four stories of quantitative easing, Federal ReserveBank of St. Louis Review pp. 51–88.

Fidessa Fragmentation Index, 2015, http://fragmentation.fidessa.com, accessed Nov. 18, 2015.

Fisher, Adlai J., Charles Martineau, and Jinfei Sheng, 2015, Media attention, macroeconomic fun-damentals, and stock market activity, Working Paper, University of British Columbia, Vancouver,Canada.

Fleming, Michael J., and Eli M. Remolona, 1999, Price formation and liquidity in the US treasurymarket: The response to public information, Journal of Finance 54, 1901–1915.

22

Goldstein, Itay, and Haresh Sapra, 2013, Should banks’ stress test results be disclosed? An analysis ofthe costs and benefits, Foundations and Trends in Finance 8, 1–54.

Goldstein, Itay, and Liyan Yan, 2015, Information diversity and complementarities in trading andinformation acquisition, Journal of Finance 70, 1723–1765.

Gurkaynak, Refet S., Brian Sack, and Eric T. Swanson, 2005, Do actions speak louder than words?The response of asset prices to monetary policy actions and statements, International Journal ofCentral Banking 1, 55–93.

Jones, Charles M., Owen Lamont, and Robin L. Lumsdaine, 1998, Macroeconomic news and bondmarket volatility, Journal of Financial Economics 47, 315–337.

Kuttner, Kenneth N., 2001, Monetary policy surprises and interest rates: Evidence from the fed fundsfutures market, Journal of Monetary Economics 47, 523–544.

Lucca, David O., and Emanuel Moench, 2015, The pre-FOMC announcement drift, Journal of Finance70, 329–371.

Ozdagli, Ali, and Michael Weber, 2015, Monetary policy through production networks: Evidence fromthe stock market, Working Paper, University of Chicago, Chicago, IL.

Savor, Pavel, and Mungo Wilson, 2013, How much do investors care about macroeconomic risk?Evidence from scheduled economic announcements, Journal of Financial and Quantitative Analysis48, 343–375.

Stein, Jeremy C., and Adi Sunderam, 2015, Gradualism in monetary policy: A time-consistencyproblem?, Working Paper, Harvard University, Cambridge, MA.

Swanson, Eric T., 2015, Measuring the effects of unconventional monetary policy on asset prices,Working Paper, University of California at Irvine, Irvine, CA.

Yellen, Janet L., 2014, Transcript of Chair Yellen’s Press Conference, December 17, 2014,http://www.federalreserve.gov/mediacenter/files/fomcpresconf20141217.pdf, accessed Nov. 18,2015.

, 2015a, Transcript of Chair Yellen’s Press Conference, June 17, 2015,http://www.federalreserve.gov/mediacenter/files/FOMCpresconf20150617.pdf, accessed Nov.18, 2015.

, 2015b, Transcript of Chair Yellen’s Press Conference, September 17, 2015,http://www.federalreserve.gov/mediacenter/files/fomcpresconf20150917.pdf, accessed Nov. 18,2015.

Zumbrun, Josh, 2015, WSJ survey: Most economists don’t see all FOMC meetings as ’live’, TheWall Street Journal, http://www.wsj.com/articles/wsj-survey-most-economists-dont-see-all-fomc-meetings-as-live-1444312805, accessed Nov. 18, 2015.

23

Figure 1. Cumulative SPY Returns Surrounding FOMC Announcements

This figure shows cumulative returns on spy in event time (trading hours) around fomc an-nouncements, with calendar time hours noted in parenthesis. The window starts 6.5 tradinghours (24 calendar hours) before and ends 1.75 trading hours after the event. Returns arenormalized to zero at the announcement. Panel A groups all announcements. Panel B sep-arates announcements into those with press conference (blue solid line) and those without(red dashed line). The dashed vertical lines indicate the time of the fomc announcement.95% confidence intervals are also shown for each series. The sample period is April 2011 toOctober 2015.

-6.5hrs(-24hrs)

-3.25hrs 0Announcement

1.75hrs(1.75hrs)

−0.003

−0.002

−0.001

0.000

0.001

0.002

0.003

0.004

0.005

Ret

urn

onS

PY

Panel A: All Announcements

-6.5hrs(-24hrs)

-3.25hrs 0Announcement

1.75hrs(1.75hrs)

−0.008

−0.006

−0.004

−0.002

0.000

0.002

0.004

0.006

0.008

Ret

urn

onS

PY

Panel B: Announcements with and without press conference

With Press Conference Without Press Conference

Figure 2. Cumulative SPY Returns Surrounding FOMC Announcements (Long Window)

This figure shows cumulative returns on spy in event time (trading hours) around fomcannouncements with and without press conference, with calendar time hours noted in paren-thesis. The window starts 13 trading hours (48 calendar hours) before and ends 13 tradinghours (48 calendar hours) after the event. Returns are normalized to zero at the announce-ment. The dashed vertical lines indicate the time of the fomc announcement. 95% confidenceintervals are also shown for each series. The sample period is April 2011 to October 2015.

-13hrs(-48hrs)

-6.5hrs(-24hrs)

0Announcement

6.5hrs(24hrs)

13hrs(48hrs)

−0.010

−0.005

0.000

0.005

0.010

0.015

Ret

urn

onS

PY

With Press Conference Without Press Conference

Figure 3. Changes in the vix Surrounding FOMC Announcements

This figure shows cumulative changes in the vix index in event time (trading hours) aroundfomc announcements, with calendar time hours noted in parenthesis. The window starts 6.5trading hours (24 calendar hours) before and ends 1.75 trading hours after the event. vixchanges are normalized to zero at the announcement. Panel A groups all announcements.Panel B separates announcements into those with press conference (blue solid line) and thosewithout (red dashed line). The dashed vertical lines indicate the time of the fomc announce-ment. 95% confidence intervals are also shown for each series. The sample period is April2011 to October 2015.

-6.5hrs(-24hrs)

-3.25hrs 0Announcement

1.75hrs(1.75hrs)

−0.05

−0.04

−0.03

−0.02

−0.01

0.00

0.01

0.02

0.03

Ret

urn

onV

IX

Panel A: All announcements

-6.5hrs(-24hrs)

-3.25hrs 0Announcement

1.75hrs(1.75hrs)

−0.08

−0.06

−0.04

−0.02

0.00

0.02

0.04

0.06

Ret

urn

onV

IX

Panel B: Announcements with and without press conference

With Press Conference Without Press Conference

Figure 4. Changes in the vix Surrounding FOMC Announcements (2006-2011)

This figure shows cumulative changes in the vix index in event time (trading hours) aroundfomc announcements between January 2006 to March 2011, with calendar time hours notedin parenthesis. The sample contains 42 observations. The window starts 6.5 trading hours(24 calendar hours) before and ends 1.75 trading hours after the event. vix changes arenormalized to zero at the announcement. Events are separated into the first (red dashed line)and second (blue solid line) announcements in each calendar quarter. The dashed verticallines indicate the time of the fomc announcement. 95% confidence intervals are also shownfor each series.

-6.5hrs(-24hrs)

-3.25hrs 0Announcement

1.75hrs(1.75hrs)

−0.06

−0.04

−0.02

0.00

0.02

0.04

0.06

Ret

urn

onV

IX

1st Announcement in Quarter

2nd Announcement in Quarter

Table I

Calendar of FOMC Announcements

This table shows the scheduled time of the FOMC announcements and the press conferencesbetween April 2011 and October 2015.

Date Time pc Date Time pc

04/27/2011 12:30 14:15 09/18/2013 14:00 14:3006/22/2011 12:30 14:15 10/30/2013 14:0008/09/2011 14:15 12/18/2013 14:00 14:3009/21/2011 14:15 01/29/2014 14:0011/02/2011 12:30 14:15 03/19/2014 14:00 14:3012/13/2011 14:15 04/30/2014 14:0001/25/2012 12:30 14:15 06/18/2014 14:00 14:3003/13/2012 14:15 07/30/2014 14:0004/25/2012 12:30 14:15 09/17/2014 14:00 14:3006/20/2012 12:30 14:15 10/29/2014 14:0008/01/2012 14:15 12/17/2014 14:00 14:3009/13/2012 12:30 14:15 01/28/2015 14:0010/24/2012 14:15 03/18/2015 14:00 14:3012/12/2012 12:30 14:15 04/29/2015 14:0001/30/2013 14:15 06/17/2015 14:00 14:3003/20/2013 14:00 14:30 07/29/2015 14:0005/01/2013 14:00 09/17/2015 14:00 14:3006/19/2013 14:00 14:30 10/28/2015 14:0007/31/2013 14:00

Table II

The FOMC Announcement Return

This table reports average fomc announcement returns with associated standard errors andp-values for the test that announcement returns are less than or equal to zero, as well asits standard deviations and selected percentiles of the distribution. We report values forthe whole sample, as well as samples split into days with and without press conference.The difference between the two is also reported. Announcement returns are defined as thecumulative 30-minute log return of spy (in %) starting at the fomc announcement. PanelB repeats the analysis on trimmed samples that omit the smallest and largest observation.Standard deviations and asymptotic standard errors are based on usual asymptotic statisticalinference, and bootstrapped standard errors and p-values are based on 1,000,000 bootstrapsamples. N is the number of observations in each subset of events. The sample period isApril 2011 to October 2015.

All pc No pc Difference

Panel A: Full SampleMean 0.025 0.292 -0.257 0.550Std. Error (asympt.) (0.10) (0.10) (0.15) (0.18)Std. Error (bootstr.) [0.10] [0.09] [0.15] [0.17]p-value (bootstr.) 0.42 0.00 0.95 0.00Std. Deviation 0.600 0.425 0.637Minimum -2.450 -0.396 -2.45025th Percentile -0.171 0.064 -0.385Median 0.086 0.259 -0.09175th Percentile 0.269 0.481 0.135Maximum 1.156 1.156 0.353Proportion <0 0.405 0.211 0.611N 37 19 18

Panel B: Trimmed SampleMean 0.063 0.282 -0.158 0.440Std. Error (asympt.) (0.07) (0.09) (0.08) (0.12)Std. Error (bootstr.) [0.07] [0.08] [0.08] [0.11]p-value (bootstr.) 0.17 0.00 0.97 0.00Std. Deviation 0.402 0.356 0.324Minimum -0.751 -0.326 -0.75125th Percentile -0.145 0.069 -0.384Median 0.086 0.259 -0.09175th Percentile 0.266 0.473 0.124Maximum 1.034 1.034 0.201Proportion <0 0.400 0.176 0.625N 35 17 16

Table III

FOMC Announcement Return Regressions

This table reports ols regression estimates for fomc announcement returns with associatedstandard errors and p-values for the test that coefficients are less than or equal to zero.Announcement returns are defined as the cumulative 30-minute log return of spy (in %)starting at the fomc announcement. PC (non-PC) is an indicator variable equal to one(zero) if there is a press conference after the announcement, and zero (one) otherwise. ∆CPIand ∆UNEMP are the latest monthly changes in CPI and unemployment rates publiclyannounced prior to the fomc announcement. RS&P is the S&P 500 total return in the 21-day interval ending 3 days before the announcement (in %). Asymptotic heteroscedasticityrobust and bootstrapped standard errors are presented in parenthesis and square brackets,respectively, and bootstrapped p-values in italics. The bootstrapped distribution is based on1,000,000 samples. Adjusted R2 and the number of observations N are also reported. Thesample period is April 2011 to October 2015.

Announcement Returns(1) (2) (3) (4)

Intercept -0.257 -0.308 -0.421(0.15) (0.17) (0.20)[0.12] [0.15] [0.14]0.99 0.99 1.00

PC 0.292 0.550 0.565 0.568(0.09) (0.17) (0.17) (0.16)[0.12] [0.17] [0.17] [0.16]0.01 0.00 0.00 0.00

non-PC -0.257(0.15)[0.12]0.99

∆CPI -0.123 -0.165(0.32) (0.31)[0.35] [0.31]0.66 0.71

∆UNEMP -0.697 -1.419(0.46) (0.68)[0.71] [0.70]0.85 0.98

RS&P 0.059(0.04)[0.02]0.00

Adjusted R2 0.193 0.193 0.169 0.290N 37 37 37 37

Table IV

Changes in VIX at FOMC Announcements

This table reports ols regression estimates of the vix 5 minutes before each fomc announce-ment and changes in vix around announcements with associated standard errors and p-valuesfor the test that coefficients are less than or equal to zero. Changes in vix are defined asreturns (in %) from 5 minutes before the fomc announcement until 30 minutes after. PC(non-PC) is an indicator variable equal to one (zero) if there is a press conference after the an-nouncement, and zero (one) otherwise. ∆CPI and ∆UNEMP are the latest monthly changesin CPI and unemployment rates publicly announced prior to the fomc announcement. RS&P

is the S&P 500 total return in the 21-day interval ending 3 days before the announcement(in %). Asymptotic heteroscedasticity robust and bootstrapped standard errors are presentedin parenthesis and square brackets, respectively, and bootstrapped p-values in italics. Thebootstrapped distribution is based on 1,000,000 samples. Adjusted R2 and the number ofobservations N are also reported. The sample period is April 2011 to October 2015.

VIX −5 minutes ∆VIX(1) (2) (3) (4) (5) (6) (7) (8)

Intercept 18.308 18.590 19.678 -0.148 0.354 0.784(1.83) (2.09) (2.40) (0.82) (1.06) (1.24)[1.49] [1.78] [1.76] [0.93] [1.03] [1.05]0.00 0.00 0.00 0.56 0.37 0.23

PC 17.271 -1.036 -1.055 -1.078 -4.386 -4.238 -4.481 -4.490(1.05) (2.11) (2.11) (1.97) (0.99) (1.29) (1.22) (1.19)[1.45] [2.08] [2.09] [1.97] [0.90] [1.30] [1.20] [1.17]0.00 0.69 0.70 0.71 1.00 1.00 1.00 1.00

non-PC 18.308 -0.148(1.83) (0.82)[1.49] [0.93]0.00 0.56

∆CPI -1.195 -0.790 3.670 3.830(2.76) (2.31) (2.08) (2.11)[4.23] [4.00] [2.44] [2.38]0.60 0.56 0.07 0.06

∆UNEMP 1.449 8.381 10.060 12.802(6.25) (7.68) (3.84) (4.77)[8.71] [8.89] [5.01] [5.27]0.44 0.16 0.02 0.01

RS&P -0.567 -0.224(0.56) (0.27)[0.27] [0.16]0.98 0.92

Adjusted R2 -0.022 -0.022 -0.081 0.003 0.202 0.202 0.278 0.293N 37 37 37 37 37 37 37 37

Table V

Probability of Interest Rate Changes before FOMC Announcements

This table reports ols regression estimates for meeting-to-meeting changes in Overnight In-dex Swap (ois) Implied Probability of interest rate changes with associated standard errorsand p-values for the test that coefficients are less than or equal to zero. We use the proba-bility of a change (∆Prob(Change)) or increase (∆Prob(Increase)) in federal funds rates, asmeasured one day prior to each fomc meeting. ∆PC can take one of three values: one ifthe announcement has a press conference while the previous one did not, minus one for theopposite case, and zero if both the current and prior announcement were followed by or notfollowed by press conferences. ∆CPI and ∆UNEMP are the latest monthly changes in CPIand unemployment rates publicly announced prior to the fomc announcement. RS&P is theS&P 500 total return in the 21-day interval ending 3 days before the announcement. Asymp-totic heteroscedasticity robust and bootstrapped standard errors are presented in parenthesisand square brackets, respectively, and bootstrapped p-values in italics. The bootstrappeddistribution is based on 1,000,000 samples. Adjusted R2 and the number of observations Nare also reported. The sample period is April 2011 to October 2015.

∆Prob(Change) ∆Prob(Increase)(1) (2) (3) (4) (5) (6)

Intercept 0.001 -0.023 -0.006 0.002 -0.016 -0.002(0.01) (0.02) (0.02) (0.01) (0.02) (0.01)[0.01] [0.02] [0.02] [0.01] [0.01] [0.01]0.48 0.93 0.66 0.42 0.89 0.58

∆PC 0.024 0.026 0.027 0.021 0.022 0.023(0.01) (0.01) (0.01) (0.01) (0.01) (0.01)[0.01] [0.01] [0.01] [0.01] [0.01] [0.01]0.04 0.02 0.01 0.03 0.02 0.01

∆CPI 0.045 0.050 0.048 0.052(0.05) (0.04) (0.05) (0.04)[0.05] [0.04] [0.04] [0.04]0.18 0.13 0.12 0.09

∆UNEMP -0.217 -0.106 -0.148 -0.061(0.09) (0.08) (0.08) (0.05)[0.10] [0.10] [0.09] [0.08]0.98 0.87 0.96 0.77

RS&P -0.927 -0.732(0.30) (0.33)[0.29] [0.25]1.00 1.00

Adjusted R2 0.052 0.127 0.298 0.058 0.105 0.250N 36 36 36 36 36 36

Table VI

Printed Media Attention Before FOMC Announcements

This table reports ols regression estimates for meeting-to-meeting changes in printed mediaattention with associated standard errors and p-values for the test that coefficients are lessthan or equal to zero. Daily news intensity is measured as the number of articles relatedto monetary policy divided by the total number of articles. Printed media attention is thefour-day average of the daily news intensity measure ending on the announcement date. Wecompute printed media attention seperately for the Wall Street Journal (WSJ) and the NewYork Times (NYT), and also compute a combined index. ∆PC can take one of three values:one if the announcement has a press conference while the previous one did not, minus one forthe opposite case, and zero if both the current and prior announcement were followed by or notfollowed by press conferences. ∆CPI and ∆UNEMP are the latest monthly changes in CPIand unemployment rates publicly announced prior to the fomc announcement. RS&P is theS&P 500 total return in the 21-day interval ending 3 days before the announcement. Asymp-totic heteroscedasticity robust and bootstrapped standard errors are presented in parenthesisand square brackets, respectively, and bootstrapped p-values in italics. The bootstrappeddistribution is based on 1,000,000 samples. Adjusted R2 and the number of observations Nare also reported. The sample period is April 2011 to October 2015.

∆WSJ ∆NYT ∆Combined(1) (2) (3) (4) (5) (6) (7) (8) (9)

Intercept -0.000 -0.000 0.004 0.000 -0.000 0.001 0.004 -0.011 0.180(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.12) (0.16) (0.14)[0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.12] [0.16] [0.14]0.51 0.52 0.08 0.43 0.63 0.30 0.49 0.53 0.10

∆PC 0.007 0.007 0.007 0.002 0.002 0.002 0.248 0.255 0.269(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.13) (0.13) (0.11)[0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.13] [0.12] [0.11]0.01 0.00 0.00 0.03 0.02 0.01 0.02 0.02 0.01

∆CPI -0.009 -0.008 -0.001 -0.001 -0.364 -0.308(0.01) (0.01) (0.00) (0.00) (0.36) (0.31)[0.01] [0.01] [0.00] [0.00] [0.49] [0.42]0.82 0.83 0.65 0.63 0.77 0.77

∆UNEMP -0.014 0.013 -0.009 -0.002 -0.676 0.547(0.02) (0.02) (0.01) (0.01) (0.85) (0.78)[0.02] [0.02] [0.01] [0.01] [0.98] [0.90]0.75 0.24 0.87 0.61 0.75 0.27

RS&P -0.224 -0.057 -10.220(0.06) (0.02) (2.02)[0.06] [0.02] [2.74]1.00 0.99 1.00

Adjusted R2 0.124 0.102 0.358 0.068 0.048 0.148 0.072 0.041 0.286N 36 36 36 36 36 36 36 36 36

Table VII

Intraday Media Attention Around FOMC Announcements

This table reports ols regression estimates for meeting-to-meeting changes in the numberof intraday newswire articles around fomc announcements with associated standard errorsand p-values for the test that coefficients are less than or equal to zero. We consider articlespublished in all Dow Jones online platforms separately for the 24 hours before and the 30 min-utes after each announcement. ∆PC can take one of three values: one if the announcementhas a press conference while the previous one did not, minus one for the opposite case, andzero if both the current and prior announcement were followed by or not followed by pressconferences. ∆CPI and ∆UNEMP are the latest monthly changes in CPI and unemploy-ment rates publicly announced prior to the fomc announcement. RS&P is the S&P 500 totalreturn in the 21-day interval ending 3 days before the announcement (in %). Asymptoticheteroscedasticity robust and bootstrapped standard errors are presented in parenthesis andsquare brackets, respectively, and bootstrapped p-values in italics. The bootstrapped distri-bution is based on 1,000,000 samples. Adjusted R2 and the number of observations N arealso reported. The sample period is April 2011 to October 2015.

∆News [−24h : 0] ∆News [0 : +30m](1) (2) (3) (4) (5) (6)

Intercept 2.901 7.229 24.303 0.558 -1.450 1.528(15.36) (18.50) (19.08) (3.00) (3.54) (3.97)[15.25] [18.86] [18.63] [3.01] [3.50] [3.49]

0.42 0.35 0.10 0.43 0.67 0.33∆PC 71.452 72.587 73.870 12.078 12.609 12.833

(15.17) (13.61) (12.07) (3.17) (2.82) (2.60)[15.93] [14.97] [13.80] [3.14] [2.77] [2.58]

0.00 0.00 0.00 0.00 0.00 0.00∆CPI -120.144 -115.168 -22.966 -22.098

(52.95) (48.73) (10.98) (10.16)[59.30] [54.63] [10.99] [10.24]

0.98 0.98 0.98 0.99∆UNEMP -112.882 -3.825 -54.603 -35.579

(92.70) (104.73) (18.92) (22.44)[119.10] [117.83] [22.05] [22.02]

0.83 0.51 0.99 0.95RS&P -9.109 -1.589

(3.58) (0.72)[3.60] [0.67]0.99 0.99

Adjusted R2 0.340 0.383 0.460 0.270 0.399 0.463N 36 36 36 36 36 36

Table VIII

Google Search Volume Before FOMC Announcements

This table reports ols regression estimates for meeting-to-meeting changes in the SearchVolume Index (SVI) prior to each fomc announcement with associated standard errors andp-values for the test that coefficients are less than or equal to zero. The weekly SVI measureis from obtained Google Trends for the search term “fomc”. ∆PC can take one of threevalues: one if the announcement has a press conference while the previous one did not,minus one for the opposite case, and zero if both the current and prior announcement werefollowed by or not followed by press conferences. ∆CPI and ∆UNEMP are the latestmonthly changes in CPI and unemployment rates publicly announced prior to the fomcannouncement. RS&P is the S&P 500 total return in the 21-day interval ending 3 days beforethe announcement. Asymptotic heteroscedasticity robust and bootstrapped standard errorsare presented in parenthesis and square brackets, respectively, and bootstrapped p-values initalics. The bootstrapped distribution is based on 1,000,000 samples. Adjusted R2 and thenumber of observations N are also reported. The sample period is April 2011 to October2015.

∆SVI(1) (2) (3)

Intercept -0.247 -1.774 -1.450(1.27) (1.22) (1.35)[1.27] [1.61] [1.72]0.57 0.87 0.80

∆PC 3.114 3.267 3.291(1.29) (1.20) (1.18)[1.32] [1.28] [1.28]0.01 0.01 0.01

∆CPI 1.201 1.296(5.42) (5.50)[5.07] [5.06]0.41 0.41

∆UNEMP -16.153 -14.088(8.99) (9.31)

[10.17] [10.88]0.95 0.90

RS&P -17.242(33.80)[33.25]

0.70

Adjusted R2 0.108 0.116 0.094N 36 36 36