monetary announcement premium in chinaecon.hkbu.edu.hk/eng/doc/monetary announcement... · evidence...
TRANSCRIPT
Monetary Announcement Premium in China
Rui Guo, Dun Jia, Xi Sun∗
This Draft: November 4, 2018
Abstract
This paper documents a pre-announcement premium of Chinese equity market an-ticipating its central bank PBOC’s monthly data release announcement of monetaryaggregates. This premium more than doubles the size of total equity premium inChina. We then present a model characterizing investors’ information acquisition deci-sion. When investors with limited attention find optimal to learn about monetary dataprior to announcement, attentive learning drives down the forecast uncertainty andboosts up equity prices. We stress the unique importance to study China. Exploitingthe randomness in PBOC’s announcement timing, this paper provides the identifica-tion of the causal link rationalized in Ai and Bansal (2018) by which the reduction ofuncertainty accounts for the U.S. FOMC announcement premium. Chinese evidencesuggests that delayed release of monetary data triggers pre-announcement attentivelearning which decreases forecast uncertainty and generates pre-drift of equity returns.We find the intensity of learning as proxied by traffics to PBOC’s website has increasedbefore announcement. Cross-sectionally, stock prices of small firms and large growthfirms are particularly sensitive to incoming monetary announcement.
JEL codes: E44, E52, G12, G14
Key Words: Equity Premium, Monetary Policy, Announcement, Macro-finance
∗Guo: Hanqing Advanced Institute of Economics and Finance, Renmin University of China. Email:[email protected]. Jia: Hanqing Advanced Institute of Economics and Finance, Renmin University ofChina. Email: [email protected]. Sun: Hanqing Advanced Institute of Economics and Finance, RenminUniversity of China. Email: [email protected]. We benefit from discussions with Hengjie Ai, Tao Zha,Jun Qian ”QJ”, Christopher Polk, Xiaoji Lin, and Hongda Zhong. We thank comments from participantsat various conferences and seminars. All errors are ours. This paper was previously circulated under thetitle of “Monetary vs. Non-monetary Macro News: Announcement Premium in China”.
1
1 Introduction
In this paper, we study how stock market reacts to the anticipated central bank’s an-
nouncement regarding its monetary policy stance. From asset pricing perspective, this paper
joins the endeavor to examine the important question on how stock price is incorporating
information about economic fundamentals.1 To the macroeconomic interests, by focusing
on the pre-announcement period, we better understand the “ex-ante” impact of monetary
policy on equity markets through an information channel.
Our exploration of the research question is framed within Chinese context. Echoing the
evidence of pre-FOMC drift of U.S. stock returns in Lucca and Moench (2015), we document a
sizable pre-announcement premium of Chinese equity market in anticipation of the monetary
aggregates data as published every month by People’s Bank of China (PBOC), the central
bank of China.2 This premium more than doubles the magnitude of total equity premium
in China. Evidence suggests that at least qualitatively, the pre-announcement premium of
the two markets share similar characteristic features.
This paper demonstrates that by studying China, it helps better understand the common
mechanism behind equity market’s reactions to announcements regarding monetary policy
practice. We present a model that enriches upon the key ingredient of recursive utility with
preference for early resolution of uncertainty as proposed by Ai and Bansal (2018). They
rationalize that ex-ante reduction of investors’ uncertainty about U.S. monetary policy gener-
ates the pre-FOMC announcement premium. In this paper, by modeling the specifics about
the announcement environment in China and further building in the information decision
prior to announcement faced by investors with limited attention in spirit of Sims (2003), our
model delivers the identification of uncertainty reduction as the driver of pre-announcement
premium, which serves as the common ground to account for the pre-announcement premium
for both China and U.S. markets.
PBOC’s announcements routine is marked by two distinctive features. First, monetary
aggregates data are published in a “quasi-scheduled” fashion with randomness in announce-
ment timing. That is, entering a month, the market expects an announcement to be made
about monetary data but the exact date and time of the announcement for that month is
1For example, early works that explored the efficiency of financial markets date back to Fama (1965,1970), Grossman and Shiller (1981), Shiller (1981), and Cochrane (1991).
2Lucca and Moench (2015) showed the pre-FOMC drift of U.S. stock market returns based on a sampleending in early 2011. However, Gilbert et al. (2018) recently note the pre-FOMC equity premium disappearsstarting from 2011 due to some unidentified reasons. Importantly, the announced measure of Chinese mon-etary policy stance, i.e. the monetary aggregates data are quantities. These data differ from the monetarypolicy instruments operated by the U.S. Federal Reserve Board, which are largely interest rate based. Thispaper put the PBOC’s announcements of monetary aggregates data and the U.S. FOMC announcementsthat publish the adjustment of short-term rate targets as comparable monetary announcements.
2
largely unpredictable. Second, conditional upon announcement, monetary statistics pub-
lished are “backward-looking”, which is about realizations of previous month. These two
environment parameters contrast with the U.S. counterparts such that FOMC announce-
ments are pre-scheduled, and the published interest rate adjustment action is the real-time
information released within the announcement window.3 Correspondingly, investors in our
model gets larger utility loss as forecast uncertainty about monetary data grows bigger if
announcement is kept delayed and the announced central bank’s signal is not real-time infor-
mative. Therefore, the duration of time lapsed waiting for incoming announcement indirectly
measures the size of accumulated uncertainty. With investors constrained by limited atten-
tion, they find it optimal to allocate more attention to ongoing monetary conditions prior to
announcement when uncertainty is accumulated too much. As a result, decreased forecast
uncertainty conditional on attentive learning boosts up equity prices ex-ante.
With Chinese data, we thus provide the empirical test of the model-implied causal iden-
tification by exploiting the exogenous variations in PBOC’s announcement timing across
events. Evidence suggests that pre-announcement premium in China is mainly driven by the
delayed releases of monetary data. More postponed arrivals of announcements, by triggering
larger uncertainty reduction, generate greater size of equity premium before announcement.
In addition, it appears the intensity of attentive learning to monetary announcement in
China as proxied by traffics to PBOC’s website during announcement window has increased
before announcement. Hence, we conclude that the pre-announcement uncertainty reduction
is causal for explaining ex-ante reactions of stock markets.
In addition, examining the unique performance of Chinese stocks itself for this fast grow-
ing country during windows of major macro announcements is intriguing. First, at the
cross-sectional dimension, China sees its great heterogeneities in stock returns in response
to PBOC’s monetary announcements. We show it is the portfolios of small and medium-cap
stocks and those of large but growth firms that exhibit price reactions to monthly announce-
ments about monetary aggregates data. Also, returns of these portfolios are particularly
sensitive to the timeliness of announcement arrival. However, little heterogeneity is detected
for the U.S. market in response to FOMC announcements (Lucca and Moench, 2015). As
suggested by our model, these findings imply that smaller firms and growth firms in China
may be affected more by the risk of market liquidity and credit condition brought by changes
in monetary policy stance.4 Second, we note that unlike many other markets, China’s stock
3Though, during unusual times, emergent FOMC meetings can take place and extra statements andannouncements were sporadically issued. For example, irregular issuance of FOMC statements in years of2007-2008 during the financial crisis.
4For example, they are more likely to be financially constrained and subject to credit misallocationproblem. See Song et al. (2011).
3
market does not respond at all to the U.S. FOMC statement releases.5 This can be due to
the micro-structure of Chinese stock market that various market frictions are still outstand-
ing, e.g. miscellaneous trading restrictions, periodic regulatory interventions, and sizeable
fraction of noisy traders etc., which provides the disconnection of Chinese market from the
co-movements of international financial markets. (Carpenter et al., 2017).
Important to note that this paper distinguishes itself from the large literature that aims
to identify the impacts of monetary policy shocks on equity market and other dimensions
of the economy.6 Rather, we disentangled the effects of anticipation about monetary policy
stance on stock market returns ex-ante regardless of the realized nature of policy shocks.
Our results show that it’s not the content of to-be-announced monetary data but the antic-
ipation of announcement per se that drives up the equity returns. With this respect, this
paper contributes to the literature by exploring the information channel through which the
performance of stock market can be affected by monetary policy ahead of time.
Related Literature. This paper is related to four strands of literature. First, this
paper aligns itself with the stream of works that explores the asset pricing implications of
macro announcements. Savor and Wilson (2013) find that the U.S. equity market exhibits
larger excess returns and Sharpe ratios on days of data releases for inflation, unemployment
and various interest rates. By studying the U.S. stock markets, Lucca and Moench (2015)
detect the pre-announcement premium in response to FOMC statements but find little ev-
idence for the intra-day equity premium using high frequency data. They show that the
U.S. stock market kicks off its pre-drift precisely one day before the FOMC statement day.
Ai and Bansal (2018) provides a theory that under certain regularlity conditions, a range
of non-expected utility functions with probability distortions can deliver positive premium
in anticipation of macro risk. Bollerslev et al. (2016) give an elegant identification that
uncovers the relationship between trading volume and return volatility in the window of
incoming macro announcements. Balduzzi and Moneta (2017), Altavilla et al. (2017), and
Philippe et al. (2017) locate the announcement premium in the treasury markets, bond fu-
ture markets, and foreign exchange markets respectively. Our paper is the first one that
provides empirical evidence on Chinese equity market’s reactions to monetary and non-
monetary macro announcements, and finds that pre-announcement premium is associated
with monetary announcements and equity market only.
5German DAX, British FTSE 100, French CAC40, Spanish IBEX, Swiss SMI, Canadian TSX index, andJapanese NIKKEI 225 all have been documented in Lucca and Moench (2015) to exhibit the pre-FOMCannouncement premium.
6For example, to name just a few, Bernanke and Kuttner (2005) considers shocks to the U.S. monetarypolicy on asset prices while Romer and Romer (2004) identifies the non-neutrality of monetary policy forthe macroeconomy.
4
Second, at firm level, a rich literature dated back to Beaver (1968) has documented higher
excess returns on the announcement day of a firm’s corporate earnings. In addition, both the
pre-announcement and post-announcement drifts of equity returns are identified around the
day of corporate earning announcements (Barber et al., 2013; Bernard and Thomas, 1989;
Frazzini, 2006). This paper applies a similar event study on the aggregate stock market to
examine its reactions to announcements at the macro level.
Third, this paper joins the works that examines the Chinese stock market efficiency in
terms of reflecting information about economic fundamentals. Franklin et al. (2017) finds
that firms listed on Chinese stock market perform worse than the listed Chinese firms on other
markets and unlisted Chinese firms. Carpenter et al. (2017) argue that China’s equity market
is increasingly efficient due to rounds of financial reform and development, which implies
the Chinese stock market well reflects firms’ future profit potential. Plus, Chinese market,
because of its disconnection to the rest of the world financial markets, provides the hedging
opportunities for international investors. This paper provides additional evidence regarding
China’s equity market performance relative to the U.S. market within the observation window
of its central bank’s announcement.
Fourth, this paper is also closely related to the literature that explores implications for
asset pricing and macroeconomic policy based on frictions of imperfect information and
uncertainty. Following Sims (2003), Peng and Xiong (2006) and Kacperczyk et al. (2016),
investors with limited attention can endogenously choose whether or not and how much
attention should be paid to learn about a variable of interest due to the fact that information
processing is costly. In line with Coibion and Gorodnichenko (2015), we also highlight the
importance that information updating among rational agents through attentive learning
has non-negligible impacts. Also, in our paper, uncertainty reduction a few days prior to
announcement is the key to generate pre-announcement premium. We provide additional
asset pricing evidence suggesting that uncertainty variations are important forces that could
shift the equilibrium in line with Bloom (2009, 2014). However, to examine higher frequency
uncertainty changes within the announcement window, our measure of uncertainty is proxied
by stock market uncertainty aggregated from higher frequency return blocks for no option-
based implied volatility index nor text-based uncertainty proxies as in Baker et al. (2016) is
readily available up to daily frequency.
The rest of the paper is structured as follows. We discuss the selection of announcement
events and data sources in Section 2. Section 3 presents the main empirical findings re-
garding the equity premium associated with monetary announcements. Section 4 presents a
model that defines the identification of uncertainty reduction as driver of pre-announcement
premium. Section 5 provides the empirical test of the causal link by studying the sensitiv-
5
ity of China’s stock market reaction to the timeliness of PBOC’s data releases of monetary
aggregates. Section 6 looks into cross-sectional heterogeneities in stock returns. Section 7
presents additional empirical results. Section 8 concludes.
2 Data
In this section, we summarize the data used for identifying the potential responses of
China’s equity market to a wide range of macroeconomic announcements published by PBOC
and other statistical agencies. Importantly, we focus on the announcements of data releases,
which specifically deliver the latest statistics about different facades of Chinese economy to
the public, i.e. the “macro news”.7
2.1 News Categories
We select a range of macroeconomic variables that have data regularly published by
different agencies through public announcements. Broadly, we can categorize the selected
macro variables into four groups about monetary-related statistics, trade competitiveness,
real-sector productivity and activitiy, and aggregate price indices.8 We also examine whether
or not Chinese equity market responds to the U.S. FOMC statement releases. Thus we group
the FOMC statements into the fifth category of news that are originated from outside the
country.
Finally, we come up with 10 macroeconomic announcements that span all the selected
macro variables of the five categories. In the following, we discuss each group of macro news
in details. In general, most of these announcements are made public in a monthly frequency
with a few exceptions noted below. Typically, the announcement made in month t publishes
data about the realized value of a given macro variable for month t− 1.9
1. Monetary-related Statistics Announcements. We are interested in the macro
news for releasing China’s monetary aggregates data, which are indicative of the stance
7Macro news that are ruled out here, for example, are general discussions or comments on Chineseeconomy and financial markets made by public figures such as officials or scholars, who may be affiliatedwith the government, the Communist Party of China (CPC), or a research institution etc.
8We notice that the very closely related macro data are routinely released at the same time throughthe same piece of announcement. Data releases in China can take forms like conference press releases, newsarticles published on the website of the data agency, or public statements etc.
9Exceptions are that for certain variables for some unusual time, the month t data may be published inthe end of month t. For example, the releases of Manufacturing Purchasing Managers Index (PMI) numberoccasionally appear to be the exceptional cases.
6
of Chinese monetary policy and overall credit condition.10 Data on the monetary
aggregates including levels and growths of M0, M1, M2 are all published by PBOC in
one announcement on its website along with other monetary and financial statistics
including the balance of aggregate loans and deposits, monthly averaged interest rate
and total size of interbank loan, and balance of foreign reserves.11 To avoid the abuse
of terminology, we simply label the announcements that publish the most updated
monetary aggregate data and other credit statistics as M2 announcements. 12
2. Trade Data Announcements. Published by the General Administration of Customs
of the People’s Republic of China (GACC), values of China’s imports and exports along
with the GACC’s discussions of the trade competitiveness of China are all available
every month. We call these news releases the TRD announcements.
3. Real-sector Productivity and Activity Announcements. We consider four ma-
jor macro variables related to the real side of the economy and their associated data
announcements: fixed assets investment excluding rural households (FAI), value added
of the industrial enterprises above the designated size (VAI), profits of the indus-
trial enterprises above the designated size (INP), and the manufacturing purchasing
managers index (PMI). All these statistics are published by the National Bureau of
Statistics of China (NBS) every month.13
10The growth rate of M2 is one of most critical policy instruments. In mid of interest rate liberalizationprocess in 2010s, various interest rates including the overnight repo rates, short-term government yields,and the SHIBOR interbank loan rates have developed their importance when gauging the Chinese monetarypolicy stances. See text in Chen et al. (2016) and Liu et al. (2017). However, examining announcementsof interest rate adjustments does not square well with the purpose of our study because announcements ofinterest rate adjustment are often times released in an unexpected way by the PBOC. Therefore, we are notable to distinguish the anticipation effects from the effects originated from the unexpected shocks.
11On the same day since November 2012, these statistics are published along with the balance of TotalSocial Financing (TSF) though TSF data is announced in a separate news release on the website. TSF datawere online either few seconds or a few hours before or after the monetary aggregates data releases.
12We also note the quarterly publication of China’s Monetary Policy Report (MPR) of PBOC. Technically,MPR does not fit well into the category of announcement that are about data releases, which specificallycommunicates with the public about the a particular data statistics. Rather, MPR is a comprehensivecollection of PBOC’s assessments about the soundness of credit market, the macroeconomic and financialstability, and the associated necessity for further adjustment of monetary policy stance. Therefore, MPRis not directly comparable to other major central banks’ policy statements that specifically highlight thea policy instrument target or decision of a monetary policy committee on policy moves, i.e., the FOMCstatement by U.S. Federal Reserve Board of Governors (FRB) or European Central Bank (ECB)’s MonetaryPolicy Accounts. For completeness, we examined the stock market reactions to these Monetary Policy Reportannouncements but found little pre-announcement responses.
13Since 2011, FAI and VAI numbers were published at the same time. These two statistics then wereannounced together with some other measures of the real economy including the total retail sales of consumergoods, the national real estate development and sales statistics, and the private fixed asset investment.Note that the GDP growth rate, a quarterly series, is published along with the data release of all these
7
4. Aggregate Price Indices Announcements. The NBS announcements of three
aggregate price indexes are included: the consumer price index (CPI), the producer
price index (PPI), and the sales price index of residential real estate in 70 large and
medium-sized cities (RST).
5. FOMC Announcements. FOMC meetings that discuss the relevance of U.S. mon-
etary policy changes are regularly held eight times a year and the associated FOMC
statements are issued accordingly after the meeting. An FOMC statement is often
times available to the U.S. public around 2:15 PM in the U.S. Eastern time. Ac-
counting for the China-U.S. time difference, details about the FOMC news are fully
accessible by the Chinese market around 2:15 to 3:15 AM of the next day depending
on whether the U.S. Daylight Saving Time applies.
2.2 Data Sources
We move on to discuss how we identify the relevant announcement events for our empirical
study. Our sample is restricted to a period of January, 2011 to June, 2017. We made this
choice primarily for three reasons. First and foremost, by focusing on these years, we abstract
from a period of global crisis of financial market turmoil and economic downturn since 2007-
2009. A number of countries underwent credit and liquidity distress which were coupled with
fiscal and monetary stimulus, all of which could be of the first order importance to drive the
stock market valuation worldwide. China enjoyed the benefits from its integration to the
global financial system while at the same time bore the cost of excessive turbulences as well.
In addition, China provided a massive stimulus package of 4 trillion RMB (roughly US $ 586
billion) to its economy and provided sufficient liquidity support to its financial markets for
years of 2008 to 2010. Hence our sample selection helps us better isolate the effects of macro
news during a quieter period from the effects of big macro shocks on China’s stock market
of turbulent years.
The second merit of focusing on recent years is that for the purpose of studying how
efficient its stock market incorporates information, China’s equity market could have in-
creasingly developed its maturity after rounds of financial reforms upon entering 2011. Last
but not least, in the post-2010 period, most macro data are firstly communicated to the
public through internet. The internet news vendor then enables us with good precision to
tell on what day and at what time the first piece of information about an updated data is
transmitted to the markets. We relegate Section 7 to discuss the robustness of our baseline
results using different sample periods.
aforementioned statistics every three months.
8
It is crucial to extract a list of dates and release times for the macro announcements we
consider with great precision. We thus employed two separate methods to cross-check the
relevant news events. The timing information for all the selected data release announcements
are first downloaded from the Bloomberg Economic Calendar (BEC). For the concern that
BEC might imperfectly collect information of news published in China, we then apply a
self-coded web-crawl algorithm that automatically collects the date and time of each piece
of macro announcements that firstly appeared on its publisher’s official website.14 We find
that apart from trivial differences regarding the timing of the NBS’s managed announce-
ments, there is no discrepancy between our web-crawled dates and times and those readily
available on BEC. We summarize the differences between crawled dates and BEC dates for
the announcements managed by NBS in Table A2 in the Appendix. We proceed with our
empirical study using the timing information from the BEC database as our benchmark
dates and time for the selected announcement events.
To measure the stock market reactions to news, we obtain the daily open and close price
series for the Wind A Share Index, which is constructed by incorporating A shares of all firms
listed on Shanghai and Shenzhen Stock Market Exchanges. Thus this index is considered the
most comprehensive measure of stock performances for the Chinese market. For robustness
checks, we also examined the Shanghai Stock Market Exchange Composite Index (SSE) Index
and the Shenzhen Stock Exchange Component Index (SZSE) index. These index series are
downloaded from Wind Data Feed Services. Then we construct various measures of daily
returns based on the price index data. Further, we search for robustness of our results
by looking into the intra-day data sourced from the RESSET High Frequency Database
separately for Shanghai and Shenzhen exchange.
To calculate the excess returns, we use the daily 10-year treasury bond yield series as the
risk free rate, also from Wind Data Feed Services. The one-year bank time deposit rate and
the one-month repo rate are examined as alternative risk-free rates. These risk-free rates are
obtained from CSMAR Economic and Financial Database. To examine other asset markets’
reactions, prices of CSI300 A share future, gold future, and RMB exchange rates against
major foreign currencies are obtained from RESSET database.
2.3 Timing of News
Our sample ranges from January, 2011 to June, 2017 for 1577 trading days. We finalize
with 693 macro announcement events. Table A3 summarizes all the announcements that we
14We didn’t perform the crawling exercise for the FOMC events and we adjusted the FOMC time forDaylight Saving Time on our own.
9
consider in our empirical study with their publishers, related statistics that are published
at the same time, and the starting month with data release of regular frequency. Since the
timing of macro news is critical for us to identify the announcement-related premium, we give
an extensive summary of the day and time details associated with date releases. Additional
tables of data summaries are left in the Appendix for reference.
We define the day of a data announcement as the first trading day that Chinese financial
markets have access to the macro news. Table 1 shows the day distribution of selected
macro announcements. PBOC does not pre-communicate with the market the whole set of
M2 announcement dates for the year ahead of time. However, market should expect each
month a new announcement will be made regarding the most recent monetary aggregates
data. We call this publication routine “quasi-scheduled”.15 We see in the table that 75 %
of the monetary aggregates data published by PBOC were announced between the 8th and
the 14th day of a month, sooner or later, though extremely rare to see the number delayed
beyond the mid of the month. Graphically, we do the box-plot of the distribution of day of
month regarding the M2 announcements in Figure 1. The vertical height of the box denotes
the percent of M2 announcements that fall into a two-day bin. The solid line approximates
a continuous probability density function of the discrete distribution. We confirm from the
graph that around 50 % of M2 announcements fall on days between the 11th and 14th day
of a given month along with a peak day for monetary aggregates data release on the 12th.
In sum, we see that though PBOC is more likely to release data around the second week of
the month, ex-ante there is no such a distinctive day of month on which the market should
confidently assign a significant probability as the announcement day.
However, TRD announcements made by GACC are pre-scheduled by which the market is
notified of announcement date of the year ahead of time. According to the data summary, we
see the TRD announcements made by GACC always deliver trade statistics before the 15th
day of the month. Regarding the announcements related to the real-sector productivity and
activity, these statistics managed by the NBS are announced following a pre-fixed schedule
since 2007. In the table, 75 % of FAI and VAI data were published in the first half of
a month. Three quarters of INP announcement is scheduled to be available on the 27th
day of the month near the end of a month. The PMI announcements are found to be
routinely published on the first day of a month but occasionally, we see PMI data in the end
of a month. For those rare cases, month t PMI data can be published prompy in the end
of month t. We note that these real sector-related statistics about month January will be
always postponed to be announced in March instead of February. This could be related to
15By contrast, the FRB routinely releases the eight FOMC meeting dates of the year no later than January.Therefore these FOMC announcements are considered pre-scheduled.
10
the fact that Chinese Spring Festival holiday season often falls into February. Also, these
postponed announcements made in March only state the aggregated but not the individual
numbers for the months of January and February. The CPI and PPI are mostly published
before the 11th day of the month. The RST data is scheduled to be published mostly on
the 18th day of a month. With time-difference conversion adjusted, the FOMC statement
release dates were found to be evenly distributed over a given month.
Table 2 gives the summary for the day of week distribution across different announce-
ments. We can see that though M2 are often times issued on weekdays, a portion of 33% of
them fall on Fridays respectively. On the other hand, announcements like PMI, TRD, and
RST are evenly distributed on each day of a week.
Table 3 shows the distribution of point of time of data releases for all selected announce-
ments. In general, all these macro data can be published on either weekdays or weekends
and within, before or after trading hours. M2 news are mostly published after trading hours
if on weekdays. Also, a considerable portion of these announcements fall between weeks that
are after trading hours on Friday till a few minutes before the trading sessions of the next
trading Monday. The trade data, real-sector statistics, and the price indices announcements
are regularly made available within trading hours and sometimes the news can be published
during weekends. Exceptions are with the PMI data, which mostly are announced around
9:00 AM on weekdays.
Importantly, across the table results, by comparing announcements made by central bank
of China and those by FRB, we see drastic differences in the day and the point of time in
a day for releasing the news. PBOC’s M2 and announcements can be made public on any
day within a week, whereas the FOMC statement releases predominantly fall on Thursdays
of early AM in local Beijing time (Wednesdays P.M. in the U.S. Eastern Time). Moreover, a
large fraction of the monetary-related announcements in China are made out of the trading
sessions including periods of post-trading hours and between-weeks of weekends. However,
the FOMC statements are issued within trading hours and on weekdays.
11
Table 1: Day of Month Distribution of Announcements
M2 TRD FAI VAI INP PMI CPI PPI RST FOMC
Min 8 8 9 9 3 1 8 8 17 125.Perctl 11 8 11 11 27 1 9 9 18 14Median 12 10 13 13 27 1 9 9 18 1975.Perctl 14 10 15 16 27 1 11 11 18 28Max 18 15 21 21 29 31 20 20 26 31Mode 11 8 13 13 27 1 9 9 18 28
No. Events 78 78 71 65 38 79 78 78 76 52
Notes: Sample: January, 2011 to June, 2017. This table shows the day of month distribution ofannouncements by their percentile cut-off days of a month. The number i in a cell denotes the i-thday of a month. Min: the earliest day of month for a data release in the sample; Max: the latest dayof data release; Percentiles: percentiles of the day of month distribution; Median: 50 % percentilecut-off. Mode: the day of month with most announcements. Numbers reported are rounded up ifthey contain decimal points.
Table 2: Day of Week Distribution of Announcements
M2 TRD FAI VAI INP PMI CPI PPI RST FOMC
Mon .10 .17 .13 .14 .08 .11 .10 .10 .18Tue .18 .13 .18 .18 .16 .15 .18 .18 .13Wed .13 .13 .20 .20 .08 .15 .13 .13 .14 .08Thu .18 .18 .08 .08 .18 .13 .17 .17 .12 .88Fri .33 .14 .25 .26 .21 .16 .26 .26 .21 .04Sat .03 .13 .10 .08 .11 .13 .09 .09 .12Sun .05 .13 .06 .06 .18 .16 .08 .08 .09
No. Events 78 78 71 65 38 79 78 78 76 52
Notes: Sample: January, 2011 to June, 2017. This table shows the percentage of announcements (indecimals) made in each day of week for a given data publication.
12
Figure 1: Distribution of Day of Month for M2 Announcements
Notes: Sample: January, 2011 to June, 2017. This figure plots the distribution of day of all M2 announcements ina given month. Each bin spans over two consecutive days. The vertical height of the box denotes the percent (%) of M2announcements that fall into a two-day bin. The solid line approximates a continuous probability density function of thediscrete distribution.
13
Tab
le3:
Tim
eD
istr
ibu
tion
of
Macro
econ
om
icA
nn
ou
ncem
ents
M2
TRD
FAI
VAI
INP
PM
ICPI
PPI
RST
FOM
C
Sam
ple
2011M
1-2
017M
6
Weekday
before
tradin
ghours
No.
An
ns.
358
54
Avg.
An
n.
Tim
e8:4
09:0
02:0
3
Weekday
within
tradin
ghours
No.
An
ns.
19
60
62
58
27
67
67
63
Avg.
An
n.
Tim
e10:2
810:4
011:1
411:0
99:3
99:3
69:3
69:3
2
Mon-T
hurafter
tradin
ghours
No.
An
ns.
31
11
1A
vg.
An
n.
Tim
e15:5
715:3
415:4
015:4
0
Betw
een
weeks
No.
An
ns.
28
20
11
911
23
14
14
16
Avg.
An
n.
Tim
e15:4
110:2
813:2
713:2
69:4
39:0
09:3
09:3
09:3
3
Tota
l81
81
74
68
38
81
81
81
79
54
Note
s:T
his
tab
led
epic
tsth
eti
me
dis
trib
uti
on
of
macr
oec
on
om
ican
nou
nce
men
ts.
Sh
an
gh
ai
an
dS
hen
zhen
Sto
ckE
xch
an
ges
are
op
enfo
rtr
ad
ing
from
Mon
day
toF
rid
ay,
wit
hca
llau
ctio
nfr
om
9:1
5to
9:2
5,
conti
nu
ou
sau
ctio
nin
9:3
0-
11:3
0an
d13:0
0-
15:0
0.
Inte
nt
ord
ers
for
blo
cktr
ad
esare
acc
epte
db
etw
een
9:3
0an
d11:3
0an
dagain
bet
wee
n13:0
0an
d15:3
0,
wh
ile
exec
uti
on
ord
ers
an
dfi
xed
-pri
ceord
ers
for
blo
cktr
ad
esare
acc
epte
dfr
om
15:0
0to
15:3
0.
Sp
ecia
lb
lock
trad
ese
ssio
ns
are
hel
don
an
ad
hoc
basi
sb
etw
een
15:0
0an
d17:0
0.
Base
don
this
,w
eca
tegori
zeth
ean
nou
nce
men
tsin
tofo
ur
gro
up
sby
the
tim
eof
thei
rre
lease
:(1
)an
nou
nce
men
tsre
lease
db
efore
trad
ing
hou
rson
wee
kd
ays;
(2)
an
nou
nce
men
tsre
lease
dw
ith
intr
ad
ing
hou
rs(w
eals
oin
clu
de
an
nou
nce
men
tsre
lease
db
etw
een
morn
ing
sess
ion
an
daft
ern
oon
sess
ion
);(3
)an
nou
nce
men
tsre
lease
daft
ertr
ad
ing
hou
rsfr
om
Mon
day
toT
hu
rsd
ay;
(4)
an
nou
nce
men
tsre
lease
db
etw
een
mark
etcl
osu
reon
Fri
day
an
dm
ark
etop
enn
ess
on
nex
tM
on
day.
Th
ista
ble
rep
ort
sth
enu
mb
erof
an
nou
nce
men
tsof
each
gro
up
san
dth
eaver
aged
rele
ase
tim
ew
ith
inth
egro
up
.
14
3 Monetary News: Pre-announcement Premium
In this section, we present the main empirical findings of this paper. First, we document a
pre-announcement drift of China’s stock market returns in response to M2 announcements.
However, the stock market appears not responsive to a range of non-monetary news. Second,
we show that such monetary pre-announcement premium is persistent, sizable, and is not
driven by the news content, that is, whether monetary data is signaling a credit tightening
or easing.
3.1 Equity Market Responses: Monetary Announcements
We examine the performance of China’s equity market in the announcement window of
an M2 announcement by estimating a baseline empirical model specified as below:
Exrett = β0 +T∑
i=−T
βiItM2−i + βxXt + υt (1)
A period t corresponds to a day. Exrett denotes the log excess return constructed from the
Wind A Share Index. The baseline Exrett is measured by the close-to-close daily returns
based on daily close prices. We will show that using open-to-close returns, alternative risk-
free rates, and expanding the announcement window would not affect our main results. Our
explanatory variables ItM2−i are dummy variables that equal to one if day t is the i-th trading
day before (or, after if i is negative) an M2 announcement. i = 0 denotes the exact day on
which an announcement is available to the public, tM2.
For the complications that an M2 news can be announced off the trading hours whether
it’s on weekdays or between-weeks, we thus align the return data of the first trading day
that the equity market has access to the news with the dummy variable ItM2= 1 when
i = 0. In other words, the M2 data may be announced either after the trading hours of day
tM2− 1 or before the trading hours of day tM2, or within the trading sessions of day tM2. In
addition, we set the length of announcement window to be 2T + 1 days. Ceteris paribus, the
coefficients βi is interpreted as the mean excess return difference on the i-th day before or
after the announcement day relative to the average daily excess return outside all windows
of M2 announcements. We check for the robustness of results by controlling for additional
covariates in vector Xt. In specific, the year, month and weekday dummies are included in
Xt to capture the calendar effect.
Table 4 reports the coefficient estimates of Equation (1) based on our baseline sample
of January, 2011 to June, 2017. With T = 3, results in (1) suggest that the coefficients βi
15
for i ∈ [−T, T ] are insignificant except for the ones associated with the dummy variables
ItM2−1. Precisely, it implies that the excess return on the previous trading day before an M2
announcement is 43 basis points (bps) higher than the mean daily return of all days outside
the seven trading day announcement windows. It is important to note that we don’t find
a significant premium on the M2 announcement day. These results suggest the presence
of a pre-announcement equity premium before the M2 news is publically announced. This
finding is largely consistent with Lucca and Moench (2015), which documents a sizeable
excess return accumulation of the U.S. equity market since the day right before the day of
FOMC statement release.
In addition, Column (2) shows that the additional equity premium generated by the pre-
drift of stock market on day tM2 − 1 is robust with a realized magnitude of 44 bps when we
measure the daily excess returns using open-to-close returns. Columns (3) and (4) replace the
benchmark risk-free rate of daily 10-year government bond yield by the one-year bank time
deposit rate and the 3-month moving averages of one-month repo rate respectively. These
estimation results yield a consistent number of 43 bps as the pre-announcement premium rel-
ative to no news scenario. Column (5) presents an similar estimate of the pre-announcement
premium when we extend the length of the M2 announcement window to 11 trading days
of T = 5.
In Table A6 of Appendix, we also show that constructing China’s equity market returns
using alternative stock market index such as the Shanghai Stock Exchange Composite (SSE)
Index or Shenzhen Stock Exchange Component Index (SZSE) Index for estimation does not
alter our main results of a sizable pre-announcement premium for M2 news.
3.2 Equity Market Responses: Non-monetary Announcements
Then we examine if the pre-announcement premium associated with news releases of
monetary aggregates data can be found for other non-monetary macro news as well. Table
5 reports the results based on our baseline dummy regression of Equation (1) by looking
into windows of a range of other announcement events. The table results exhibit that no
statistically and economically significant pre-announcement premium is associated with non-
monetary announcements. Since most of the dummy coefficient estimates are statistically
insignificant, we only discuss a few noted findings of interest in the following.
For the lead term of post-announcement day tAnns+3 regarding to the INP news about
industrial production data, the partial effect is significantly different from zero. Also, in
response to the TRD announcements releasing China’s trade statistics, the Chinese stock
market reacts with a large 40 bps in excess returns relative to no news daily returns. We
16
Table 4: Wind A Share Index Returns in Windows of M2 Announcements
(1) (2) (3) (4) (5)VARIABLES Baseline Open-to-Close 1Y Bank Rate 1M Repo Rate 11 Day Window
ItM2−5 0.18(0.20)
ItM2−4 -0.16(0.22)
ItM2−3 0.33 0.36* 0.33 0.33 0.34(0.20) (0.21) (0.20) (0.20) (0.21)
ItM2−2 0.23 0.21 0.23 0.23 0.24(0.18) (0.16) (0.18) (0.18) (0.18)
ItM2−1 0.43** 0.44*** 0.43** 0.43** 0.44**(0.17) (0.17) (0.17) (0.17) (0.18)
ItM2 0.16 0.08 0.16 0.16 0.17(0.18) (0.17) (0.18) (0.18) (0.19)
ItM2+1 -0.16 -0.08 -0.16 -0.16 -0.15(0.18) (0.17) (0.18) (0.18) (0.18)
ItM2+2 0.04 0.04 0.04 0.04 0.05(0.21) (0.19) (0.21) (0.21) (0.21)
ItM2+3 0.01 -0.04 0.01 0.01 0.02(0.19) (0.17) (0.19) (0.19) (0.19)
ItM2+4 0.12(0.19)
ItM2+5 -0.01(0.22)
Year FE YES YES YES YES YESMonth FE YES YES YES YES YESWeekday FE YES YES YES YES YESConstant -0.28 0.00 -0.28 -0.28 -0.28
(0.22) (0.20) (0.22) (0.22) (0.22)
Observations 1,577 1,577 1,577 1,577 1,577R-squared (%) 1.82 1.89 1.82 1.82 1.95
Notes: Sample: January, 2011 to June, 2017. This table reports dummy variable regression results ofEquation (1) for different specifications. The dependent variable is the log excess return constructedfrom the Wind A Share Index. Announcement dummy ItM2−i equals to one if the i-th trading day isbefore (or, after if i is negative) an M2 announcement. We align the return data of the first tradingday that the equity market has access to the news with the dummy variable ItM2 = 1 when i = 0.Columns (3) and (4) replace the benchmark risk-free rate of daily 10-year government bond yield bythe one-year bank time deposit rate and the 3-month moving averages of one-month repo rate re-spectively. ***Significant at 1%, **significant at 5%, *significant at 10%. Robust standard errors areshown in parentheses.
see that as the nature of macro risk has been released to the public, the market is found to
be reacting to these updated statistics conditional upon the arrival of news. Therefore, we
should not label these estimated partial effects as pre-announcement premium, which is our
main focus of this paper. In addition, we see coefficients of lag terms tAnns−3, the boundary
of the pre-announcement window, related to data releases of value added of industrial enter-
prises and residential real estate are only marginally significant at 10 % significance level. It
is weak to draw conclusion that China’s stock market also exhibits premium prior to VAI
17
and RST announcements.
Table 5: Wind A Share Index Returns in Windows of Other Macro Announcements
(1) (3) (4) (5) (6) (7) (8) (9) (10)Announcement M2 TRD VAI FAI INP PMI CPI PPI RST
ItAnns−3 0.33 0.16 0.40* 0.25 -0.21 -0.26 0.17 0.17 -0.35*(0.20) (0.20) (0.22) (0.22) (0.22) (0.23) (0.14) (0.14) (0.19)
ItAnns−2 0.23 0.26 0.31 0.24 -0.11 -0.20 -0.10 -0.10 0.04(0.18) (0.20) (0.22) (0.21) (0.20) (0.22) (0.21) (0.21) (0.21)
ItAnns−1 0.43** 0.17 0.03 0.03 -0.19 0.25 -0.02 -0.02 0.03(0.17) (0.19) (0.19) (0.18) (0.20) (0.17) (0.21) (0.21) (0.21)
ItAnns 0.16 0.40** 0.05 0.10 0.01 0.25 0.05 0.05 -0.28(0.18) (0.19) (0.23) (0.22) (0.23) (0.22) (0.23) (0.23) (0.21)
ItAnns+1 -0.16 -0.03 0.10 0.04 0.10 0.25 0.17 0.17 0.06(0.18) (0.18) (0.19) (0.18) (0.19) (0.21) (0.18) (0.18) (0.21)
ItAnns+2 0.04 0.06 -0.09 -0.05 -0.10 0.28 -0.01 -0.01 0.04(0.21) (0.19) (0.23) (0.22) (0.19) (0.18) (0.19) (0.19) (0.18)
ItAnns+3 0.01 0.17 0.09 0.10 0.56*** 0.09 0.04 0.04 -0.10(0.19) (0.14) (0.17) (0.17) (0.16) (0.18) (0.18) (0.18) (0.20)
Year FE YES YES YES YES YES YES YES YES YESMonth FE YES YES YES YES YES YES YES YES YESWeekday FE YES YES YES YES YES YES YES YES YESConstant -0.28 -0.29 -0.29 -0.28 -0.25 -0.28 -0.27 -0.27 -0.23
(0.22) (0.22) (0.22) (0.21) (0.21) (0.21) (0.21) (0.21) (0.21)
Observations 1,577 1,577 1,577 1,577 1,577 1,577 1,577 1,577 1,577R-squared (%) 1.82 1.65 1.59 1.43 1.59 1.85 1.36 1.36 1.58
Notes: Sample: January, 2011 to June, 2017. This table reports dummy variable regression results ofEquation (1) for different specifications with the exception that announcement windows are associatedwith non-monetary news. The dependent variable is the log close-to-close excess return constructed fromWind A Share Index. Announcement dummy ItAnns−1 equals to one if the i-th trading day before (orafter if i is negative) a particular type of announcement. We align the return data of the first trading daythat the equity market has access to the news with the dummy variable ItAnns = 1 when i = 0. ***Sig-nificant at 1%, **significant at 5%, *significant at 10%. Robust standard errors are shown in parentheses.
3.3 Duration of Monetary Pre-announcement Premium
Having established that the pre-announcement premium in China is associated with
monetary news only, we then look into details about this monetary premium. We check
the possibility that the pre-drift of China’s equity market may have kicked off several days
prior to an M2 announcement day. Our empirical strategy helps pin down the duration of
pre-announcement premium in days. In specific, we test the null hypothesis that the daily
excess returns do not react to the M2 announcement any earlier than the day right before
the announcement day. Respectively, we use generalized dummy variables to denote those
trading days that fall into a j-trading day window with j = 2, 3, 5 before M2 announcement
day tM2. Then we estimate the following model:
Exrett = β0 + βjItM2−1,j + βxXt + υt (2)
18
βj can be interpreted as the average daily return of those days that fall into the j-day window
before an M2 announcement day, relative to daily returns that are outside of these windows.
Estimation results are summarized in Table 6.
In Column (1), we firstly drop all the dummy variables other than ItM2−1, and use the
estimation results as reference point for cross-specification comparison. The back of envelope
calculation says if the window premium is solely driven by the tM2 − 1 premium of 40 bps,
the daily announcement premium in window j should be approximately 40/j bps. Columns
(2)(3) and (4) respectively shows the coefficient estimates of βj when j = 2, 3, 5, which are
all statistically positive and much larger than 40/j bps. Therefore, we conclude that the pre-
announcement premium to M2 news does not only realize on the exact day of one day before
the announcement. Importantly, by Column (3), the magnitude of 3-day window premium
estimate is the largest among the j-trading day windows prior to the announcement day.
Then we consider some restrictions to our M2 news sample. It’s possible that announce-
ments made in different point of time of a year may generate different size of “news” impacts
on equity returns. We have this hypothesis as findings at the firm-level suggest that the
stocks react stronger in response to interim (e.g. quarterly) corporate earnings announce-
ments (Barber et al., 2013). By the same token, we explore in the following the news
effects of February M2 announcements only, along with those announcements that release
the quarterly monetary aggregates data.
There are important reasons to study the M2 announcements made in February. Febru-
ary announcement of M2 data contains monetary aggregate measures for the month of Jan-
uary, the first data point of a new year. Hence, the February M2 news may somewhat predict
the PBOC’s policy moves for the entire year at least partially. To capture the news effects of
quarterly announcements of monetary aggregate data, we then restrict our M2 news sam-
ple to announcements made in four months: February, April, July and October. All these
announcements summarize monetary aggregate data for the previous quarter. We thus con-
jecture that investors are drawing excessive attention to these important interim monetary
announcements relative to others, which leads to a larger pre-announcement equity premium
for these months.
In Column (5) of Table 6, we present the estimation results when we exclusively focus on
February stock market returns and the February M2 announcements. Results find that the
three-day window pre-drift of equity returns is more than twice larger than our estimate of
three-day window daily premium when all M2 announcements are considered as shown in
Column (3). Column (6) then shows that the three-day window pre-announcement premium
related to quarterly M2 announcements are of the same magnitude at 63 bps in excess
returns. Column (6) implies that the pre-announcement premium is not driven by the
19
effects of February M2 announcements alone. More importantly, both columns of estimates
lend credence to our conjecture that a stronger pre-announcement premium is associated
with the those interim M2 news.
Table 6: Wind A Share Index Returns in Windows Prior to M2 Announcements
(1) (2) (3) (4) (5) (6)VARIABLES All Anns All Anns All Anns All Anns Feb Qtr. Anns
ItM2−1,1 0.40**(0.17)
ItM2−1,2 0.31**(0.13)
ItM2−1,3 0.33*** 0.69** 0.65***(0.11) (0.29) (0.23)
ItM2−1,5 0.19**(0.10)
Constant -0.26 -0.27 -0.28 -0.29 0.23 -0.02(0.21) (0.21) (0.21) (0.21) (0.47) (0.28)
Year FE YES YES YES YES YES YESMonth FE YES YES YES YES YESWeekday FE YES YES YES YES YES YESObservations 1,577 1,577 1,577 1,577 116 504R-squared (%) 1.50 1.53 1.70 1.48 5.80 4.06
Notes: Sample: January, 2011 to June, 2017. This table reports the dummy variableregression results of Equation (2). “All Anns sample” columns summarize the results con-sidering all M2 announcements in our sample; “Feb” present results estimated from asample restricted to M2 news issued on Februaries only; “Qtr.Anns” present results esti-mated from a sample that covers M2 announcements of February, April, July, and Octo-ber. The dependent variable is the log close-to-close excess return constructed from theWind A Share Index. We align the return data of the first trading day that the equitymarket has access to the news to the day tM2. Announcement dummy ItM2−1,j equals toone for the trading days in a j-trading-day window before an M2 Announcement. ***Sig-nificant at 1%, **significant at 5%, *significant at 10%. Robust standard errors are shownin parentheses.
3.4 Evidence from High Frequency Data
We then look into the higher frequency data to confirm the pre-drift of China’s stock
market returns in response to M2 announcements. In particular, we check if the equity
market index kicks off drifting a few days prior to reaching its peak. We plot two graphs of
cumulative returns constructed from Shenzhen (SZSE) and Shanghai (SSE) index in Figure
2. These plots clearly yield findings that are consistent with our regression analysis in
previous sections.
Across the subplots, the solid lines denote the average cumulative intra-day returns based
on SZSE and SSE Composite Index over all seven-day windows centering at the M2 an-
nouncement days for a period of January, 2011 to December, 2016.16 The one standard
deviation confidence band are drawn along with the cumulative returns. Timing of an-
16RESSET High Frequency Database stops updating data beyond the end of 2016.
20
nouncements are aligned to daily returns such that the grey bar marks the first trading day
on which the market has access to the most updated monetary aggregates data. The dashed
line captures the average cumulative returns over all “no-news” seven-day intervals with
none of any announcement day falling in the windows. Strikingly, we see that the equity
return regardless of stock exchanges starts accumulating roughly three days prior to the M2
announcement until it reaches the peak on the announcement day, i.e. the precise evidence
of the pre-announcement drift.
Regarding the magnitude of pre-announcement premium in response to M2 news, it
shows that the cumulative returns reach to a peak of roughly 90 bps. This averages out
to roughly 30 bps per day of equity premium within a three-day window relative to a no
announcement premium of slightly above zero, which squares well with our estimates in
Table 6 though the latter is obtained based on a comprehensive A-share market index.
Figure 2: Cumulative Chinese Stock Market Returns Around M2 Announcements
(a) Shenzhen Stock Exchange (b) Shanghai Stock Exchange
Notes: Sample: January, 2011 to December, 2016. This figure shows the average cumulative log return over five-minutes blocks on the Shenzhen Stock Exchange Component (SZSE) Index and Shanghai Stock Exchange Composite (SSE)Index of a seven-day announcement window. The solid line is the average cumulative return across all seven-day windowscentering on the first trading day when the market has access to the M2 announcements as shaded by a vertical grey bar. Thedashed line denotes the average cumulative returns of seven-day windows with none of any announcement day included. Theshadow areas mark +/− 1 standard deviation around average returns.
3.5 Pre-announcement Premium and Total Equity Premium
We then ask the question up to what percent this pre-announcement premium regarding
news about monetary aggregates data can account for the total premium of Chinese equity
market. We measure the size of daily pre-announcement premium using the estimated pre-
announcement premium per day for a three-day window prior to an M2 announcement.
Table 7 summarizes the results.
21
Per Panel (a) of the table, the average daily close-to-close excess return of Wind A Share
Index is about a scant number of 2 bps, with an annualized return of approximately 5 %. Our
daily pre-announcement premium associated with the three-day M2 announcement window,
once annualized with a factor of 36 (12 times a year), scales China’s equity premium by a
multiple of 2.17. The Sharpe Ratio for a trading strategy of buy-and-hold the Wind A Share
Index for three days prior to the M2 announcements for twelve times a year yields a large
number of 1.15, which is more than six times of the average Sharpe Ratio of 0.18 based on the
buy-and-hold the market index throughout the year. Therefore, the benchmark monetary
pre-announcement premium is sizable in both absolute and risk-adjusted terms.
On the right hand side of the Table, Panel (b) collects the calculated Chinese equity
premium and the pre-announcement premium using open-to-close returns. In specific, the
scale multiple of the pre-announcement premium drops to fractions of 45 %. The relative
ratio of Sharpe Ratio is shrunk to 1.22. Intuitively, though the pre-announcement premium
gets larger, it is the greater increases in total equity premium that drive down the relative
magnitude of pre-announcement. In sum, these results emphasize the critical importance of
studying monetary announcements as holding stocks a few days before the updated monetary
aggregates data are announced brings about a significantly large equity premium.
Table 7: China’s Equity Premium and Pre-announcement Premium of M2 News
(a) Close-to-close Returns (%) (b) Open-to-close Returns (%)
No.Obs Daily average Annualized S.R. Daily average Annualized S.R.
All trd day 1577 0.02 4.89 0.18 0.13 32.45 1.32M2 Anns. 234 0.30 10.63 1.15 0.41 14.71 1.61
Scale/Ratio 2.17 6.39 .45 1.22
Notes: This table presents excess log returns of Wind A Share index earned in three-day Pre-M2trading windows comparing to the average level with different measurements denoted at the top ofeach panel. “Annualized” stands for cumulative annual excess return, assuming there are 250 trad-ing days in a calendar year. “M2 Anns.” presents respective returns earned in the three-day pre-M2trading window. “All trd day” presents respective returns earned in all trading days of the samplerange. “S.R.” denotes the annualized Sharpe ratio on pre-M2 window returns. Since there are 12three-day window per year, we calculate the annualized announcement Sharpe ratio as the per daySharpe ratio times
√36. “Scale/Ratio” shows the scale or the ratio of returns earned in the three-day
pre-M2 trading window to those earned in all trading days. Panel (A) summarizes results based onclose-to-close returns; Panel (b) summarizes results based on open-to-close returns. Returns shownare all in percentage.
3.6 Announcement Premium and Content of Monetary News
If the equity market well anticipates the direction of monetary aggregate changes prior
to the actual announcement, we should be able to see that lax or tightening of monetary
policy moves could have affected the documented pre-announcement equity drift. In this
section, we present evidence to establish that the sizable equity premium associated with
22
M2 announcements is driven by the presence of an incoming news rather than the content
of news.
We use three different proxies to characterize the content of an M2 announcement.
First, using year-over-year (YOY) M2 growth rate gM2,t, we take the difference of actual
YOY M2 growth rate and that of the previously realized number ∆gM2,t = gM2,t − gM2,t−1
as the baseline measure of announcement content. It proxies for how much more lax the
monetary policy stance or the credit condition is relative to that of previous month ex-
post. In addition, we construct the “unexpected” innovations to the stance of monetary
aggregates εM2,t = gM2,t − gM2,t, where gM2,t denotes the market expected M2 growth rate,
as measured by the Bloomberg Survey number. In addition, we directly take the difference
of Bloomberg Survey data on the M2 growth rate E(∆gM2,t) = gM2,t − gM2,t−1 as the third
measure of news content. Though the surveyed forecasts do not directly correspond to the
announcement contents ex-pose, this measure is at least suggestive of the ex ante anticipated
news content in the market.
A positive (negative) of ∆gM2,t, εM2,t, or E(∆gM2,t) is considered extra ease (tightening) of
monetary policy or credit condition. In Figure 3, conditional on whether the baseline measure
of news content ∆gM2,t is positive (dashed line) or negative (solid line), we plot the average
cumulative stock market returns constructed from SZSE and SSE market index around the
M2 announcements after the risk-free rate is subtracted. We label those announcements with
released data suggesting ∆gM2,t > 0 simply as “good news” whereas ∆gM2,t ≤ 0 as “bad
news”, though no welfare criterion is imposed here to specifically differentiate good from
bad. shows that prior to M2 announcements, regardless of stock exchanges, market price
index exhibits cumulative drifting no matter if the market will receive a lax or tightened M2
number ex-pose. Though it shows the realized mean pre-announcement cumulative returns
associated with ∆gM2,t > 0 are relatively higher on day tM2 − 1, mean cumulative return of
either scenario falls into the one standard deviation confidence band of the other. Hence, in
terms of the size of pre-announcement premium, no statistically significant difference can be
discerned here with or without loosened M2 numbers. Interestingly, after the announcements
are made on day tM2, the post-announcement market index moves in two opposite directions.
Cumulative returns conditional on goods news keep going up whereas equity price fluctuates
and drops conditional upon a bad news of tightened M2 growth rate.
We then estimate the following specification to test the null that the pre-announcement
premium is not affected by the content of M2 announcements.
Exrett = β0 + β1ItM2−1,j + β2 · ItM2−1,j · ContenttM2+ βxXt + υt (3)
23
Figure 3: Cumulative Returns Around M2 Announcements by ∆gM2,t
(a) Shenzhen Stock Exchange (b) Shanghai Stock Exchange
Notes: Sample: January, 2011 to December, 2016. This figure shows the average cumulative log return over five-minutes blocks on the Shenzhen Stock Exchange Component (SZSE) Index and Shanghai Stock Exchange Composite (SSE)Index of a seven-day announcement window with restricted samples of announcements with ∆gM2,t > 0 (good news, dashedline) and ∆gM2,t ≤ 0 (bad news, solid line). Average cumulative returns across all seven-day windows are centered on thefirst trading day when the market has access to the M2 announcements as shaded by a vertical grey bar. The dotted lineat the bottom denotes the average cumulative returns of seven-day windows with none of any announcement day included.The shadow areas mark +/ − 1 standard deviation confidence band around average returns conditional on good or bad newscontained in M2 announcements.
ItM2−1,j are dummy variables to denote those days that fall into a j-trading day window before
the M2 announcement day tM2. We check j = 1, 3 by focusing on the return reactions
during the one-day or three-day window prior to the announcement. ContenttM2on the
announcement day tM2 is measured by monthly numbers of ∆gM2,t, εM2,t, or E(∆gM2,t). The
coefficient associated with the interaction term β2 thus gives the estimate of additional gain
or loss, if any, due to the news content.
We summarize the results in Table 8, which shows that across specifications of Columns
(2) to (4), the one day ahead pre-announcement premium is not affected by the content of
news regardless of how we measure the “attractiveness” of news content. More importantly,
we are ruling out the possibility that the news content, if by any chance, leaked into or well
anticipated by the market before the announcement, does not shift the pre-announcement
equity returns by any margin. Therefore, we maintain that the equity premium associated
with incoming M2 announcements are market reactions due to expectation changes, rather
than responses to the directional information content contained in monetary announcements.
In addition, as we have discussed in 2, additional statistics other than M2 including M1,
total outstanding loan balance (Loan), deposit balance (Deposit) are also contained in the
statement of M2 announcement. On the same day, statistics giving the balance of total social
financing (TSF) are also published within hours of M2 announcement via a separate state-
ment. We further explore the possibility that investors’ earned pre-announcement premium
24
may be driven by the announcement content that is associated with statistics other than M2.
Columns (5) to (8) presents coefficient estimate about the interaction term of dummy ItM2−1
and measure of news content regarding the related data numbers, i.e. monthly difference of
YOY growth rates: ∆gM1,t, ∆gLoan,t, ∆gDeposit,t, and ∆gTSF,t. Given that TSF data were
published quarterly only until the end of 2016, we have fewer day return observations for
the regression. Results show that neither of these four other news content measure would
affect the size of pre-announcement premium.
Table 8: Announcement Premium: Content of Announcements
(1) (2) (3) (4) (5) (6) (7) (8)VARIABLES cls2cls cls2cls cls2cls cls2cls cls2cls cls2cls cls2cls cls2cls
ItM2−1 0.43** 0.48*** 0.46*** 0.45*** 0.43** 0.48*** 0.45** 0.68*(0.17) (0.18) (0.18) (0.17) (0.17) (0.18) (0.18) (0.36)
ItM2−1 ·∆gM2,t 0.38(0.25)
ItM2−1 · εgM2,t 0.43(0.27)
ItM2−1 · E[∆gM2,t] 0.41(0.52)
ItM2−1 ·∆gM1,t 0.03(0.06)
ItM2−1 ·∆gLoan,t 0.59(0.66)
ItM2−1 ·∆gDeposit,t 0.11(0.18)
ItM2−1 ·∆gTSF,t 0.10(0.10)
Year FE Yes Yes Yes Yes Yes Yes Yes YesMonth FE Yes Yes Yes Yes Yes Yes Yes YesWeekday Yes Yes Yes Yes Yes Yes Yes Yes
Observations 1,577 1,577 1,577 1,577 1,577 1,577 1,577 520R-squared 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.06
Notes: This table reports the dummy variable regression results of Equations (1) and (2). The depen-dent variable is the log close-to-close excess return constructed from the Wind A Share Index. We alignthe return data of the first trading day that the equity market has access to the news to the day tM2.”Weekday FE”: the weekday fixed effects controls. See text for the definitions of variables. Announce-ment dummy ItM2−1,j equals to one for the trading days in a j-trading-day window before a M2 An-nouncement. ***Significant at 1%, **significant at 5%, *significant at 10%. Robust standard errors areshown in parentheses.
4 Model
In this section, we present a model that generates a positive equity premium while the
market is anticipating an incoming announcement about the monetary aggregates data. The
key mechanism of the model is that investors’ uncertainty about the monetary policy prac-
25
tice is reduced over days prior to the announcement, i.e. the exact channel of uncertainty
reduction first highlighted in Ai and Bansal (2018). However, our model generalizes in the
sense that the forecast uncertainty can be affected by the attentive learning via information
acquisition about monetary policy, which is an endogenous decision made by investors who
are constrained by limited attention a la Sims (2003). In addition, when building in the
particular environment settings associated with Chinese markets, which says the monetary
data can be announced on any day of a monthly cycle (timing randomness) and are pub-
lished with day lags (backward-looking announcement), the model framework well explains
the characteristic pre-announcement premium in China. Our model predicts that size of
forecast uncertainty depends on the timeliness of monetary announcement arrival, which
gives the identification for the causal link between uncertainty reduction and magnitude of
pre-announcement premium. This model also has rich implications on the cross-sectional
heterogeneities in equity return reactions to monetary announcements. While our model is
quite consistent with Chinese data, we show the model naturally rationalizes the U.S. case
when the announcements are pre-scheduled and are about the real-time monetary policy
actions.
In specific, we set up a consumption-based asset pricing model by working with a general
form of recursive utility (Kreps and Porteus, 1978; Epstein and Zin, 1989). In line with Ai
and Bansal (2018), the implied preference for early resolution of uncertainty is critical to
associate the variation in uncertainty with jumps in equity returns within the announcement
window. We show that the expected excess stock return decreases and the current equity
price increases in the reduction of investors’ forecast uncertainty about money growth. While
the actual money growth data is announced by the central bank in a monthly cycle in form
of a public signal, the announcement is backward-looking such that the announced data is
regarding the money growth realized in the past. Subject to the information cost, investors
may choose whether or not to pay additional attention by privately learning about the money
growth before the announcement arrival. As a result, conditional on attentive learning that
reduces the forecast uncertainty over time, our model yields a positive pre-announcement
equity premium.
4.1 Equity Premium and Uncertainty
Our dynamic model is discrete-time and each period t ≥ 0 corresponds to a day.17 A
representative household maximizes its life-time utility Vt(ct, zt) evaluated on day t defined
over real consumption ct and the certainty equivalent of the day t expected continuation
17We don’t differentiate between trading and non-trading days in the model.
26
value zt
Vt(ct, zt) = maxct,xt,bt∞t=0
[(1− β)c1−ξt + β(EtV 1−α
t+1 )1−ξ1−α ]
11−ξ . (4)
where zt = (EtV 1−αt+1 )
11−α . ξ and α respectively indexes the inverse of household’s elasticity
of inter-temporal substitution and the coefficient of relative risk aversion. We work with
parameter values of α > 1 > ξ such that household prefers early resolution of uncertainty in
line with Ai and Bansal (2018).
Household chooses consumption ct, equity holding xt, and the position of a risk-free bond
bt that pays one unit of consumption good in period t+1. β ∈ (0, 1) is the subjective discount
factor. Utility maximization is subject to a daily budget constraint
ct + qtxt +bt
Rft+1
= (qt + yt)xt−1 + bt−1 (5)
where qt is the per share equity price. yt is the dividend payout per share of equity. The
gross rate of risk-free bond return is given by Rft+1 known as of day t. By definition, the
rate of return from the risky equity investment is Rt = qt+ytqt−1
. A portfolio investment with
share of holdings φt invested in equity and 1 − φt in the riskless bond on day t gives an
aggregate market rate of return for tomorrow RW,t+1 = Rft+1 + φt(Rt+1 − Rf
t+1). With the
beginning-of-day total wealth Wt = (qt + yt)xt−1 + bt−1, budget constraint is equivalently
given by Wt+1 = (Wt − ct)RW,t+1.
In addition, we impose the quantity theorem of money holds in equilibrium such that
the total income in nominal terms is spent using the demanded real balance of monetary
aggregate Mt with yt = ψMt. Constant ψ > 0 measures the velocity of per unit money
transactions. While household’s dividend income is consumed every period yt = ct, it follows
that consumption ct is proportional to the real balance of money given by
ct = ψMt (6)
Defining mt the log growth rate of real money balance such that mt = log(Mt/Mt−1). Equa-
tion (6) gives ct+1
ct= emt+1 . Also, in equilibrium, the aggregate holding of risk-free bond bt
has to be zero with φt = 1 and thus RW,t+1 = Rt+1.
Further, solving for the first order condition with a imposed constant equity price-
dividend ratio χ = qtyt
for simplicity, we have the key asset pricing equation
1 = Et[βθe−ξθmt+1Rθt+1] (7)
27
where related terms can be factored into a stochastic discount factor Ωt|t+1 = βθe−ξθmt+1Rθ−1t+1
such that 1 = Et[Ωt|t+1Rt+1] and θ = 1−α1−ξ . The equation says that the equity return is shifted
by investors’ forecasts on day t about future money growth rates mt+1.
We show in the appendix that the expected excess equity return in log EXt+1|t is deter-
mined by the risk-aversion α weighted investors’ forecast variance V ar(mt+1|It) about future
money growth conditional on the information set It on day t.
EXt+1|t = log(EtRt+1)− log(Rft+1) = αV ar(mt+1|It) (8)
Equation (8) suggests that heightened forecast uncertainty measured by larger forecast vari-
ance requires increased expected excess return in compensation. We further prove that
greater forecast variance raises the expected excess return by shrinking the current equity
price qt.
qt = ct[e− log β−(1−ξ)mt+1|t+
(1−ξ)(α−1)2
V ar(mt+1|It) − 1]−1 (9)
Equation (9) shows that qt is a function of investors’ expected mean mt+1|t = E(mt+1|It)and the forecast variance V ar(mt+1|It). ct is the day t consumption level which is on the
optimal consumption growth path. Current stock price jumps if investors anticipates a higher
money growth rate mt+1|t thus consumption growth in equilibrium, given the elasticity of
intertemporal substitution 1ξ> 1. In addition, greater forecast uncertainty about future
money growth adds a risk premium which decreases the current stock price. As investors
prefers early resolution of uncertainty α > 1 > ξ, larger uncertainty requires a current price
discount to earn a higher expected excess return. In addition, the current day t excess return
EXt = log(Rt)− log(Rft ) where Rt = qt+ct
qt−1increases in the current equity price qt with Rf
t ,
qt−1 pre-determined and ct optimized.
Proposition 1 Given recursive utility with preference for early resolution of uncertainty,
greater forecast uncertainty raises the expected excess return while decreases the current equity
prices along with the realized excess equity return.
We then proceed to construct the environment of announcement arrivals and the infor-
mation problem faced by investors. As a result, the forecast uncertainty V ar(mt+1|It) will be
endogenously determined and affected by the information decision, by which excess returns
and equity prices are affected.
28
4.2 Environment of Monetary Announcements
We construct the environment of monetary announcement arrivals by featuring the
two characteristics of Chinese market, i.e. randomness of announcement timing and the
backward-looking nature of data releases through announcements.
Log money growth rate mt has been shown to determine the consumption growth in
equilibrium which affects equity price over time. We further specify the supply of monetary
aggregates that makes the growth rate of money mt evolve over time via a stationary AR(1)
process as follows
mt = ρmt−1 + (1− ρ)µ+ et (10)
where ρ ∈ (0, 1) governs the persistence of money growth rate and et ∼ N(0, σ2e) captures
the innovations to the money supply process. µ ≥ 0 denotes the mean money growth rate.
In equilibrium, the supply of money aggregates meets the demand of money by absorbing
the aggregate consumption.
We assume there is a central bank who closely monitors, manages, and observes mt on a
daily basis. However, the exact data point of mt on day t is available to the market investors
only periodically through monthly announcements made by the central bank. We give the
definition of “timing randomness” associated with quasi-scheduled announcements in the
following:
Definition 1 (Quasi-scheduled Announcements with Timing Randomness)
Suppose integer number i = 1, 2, ... indexes the i-th month and ti denotes the end day of
month i such that ti = i ·N where a month is consisted of N days. A monthly announcement
is considered quasi-scheduled with timing-randomness if (1) the exact announcement day
tAi of any month i such that tAi = ti−1 + T with T ∈ 1, ..., N each month is drawn from
a given distribution with Cumulative Density Function F (T ) and (2) T is unknown to
investors entering month i.
By definition, we see PBOC of China announces the monetary aggregates every month
but the exact date of announcements is unknown to the market until the announcement
day. The market knows there will be a data release announcement but the day might fall
on any day of the month.18 We thus call the monetary announcement routine in China
quasi-scheduled with timing randomness.
We further introduce the notion of “backward-looking announcements”. Though there
18Though in the data summary we see the date mostly falls within a range of day 11 to 14 of a month,PBOC never pronounces that there will be a pre-scheduled fixed date for making such announcement.
29
is one monetary announcement made every month, the announcement however publishes an
outdated monetary aggregate number of mt that is not associated with the announcement
day. We model the announcements as public signals made on day tAi of month i about data
mti−1, the money growth realized on the end day of previous month i− 1. For example, on
May 12th 2017, PBOC of China announced the statistics about monthly growth of monetary
aggregates of April relative to March.19 Specifically, we define the public signals in the
following:
stAi = mti−1+ ηtAi (11)
where ηtAi ∼ N(0, σ2η) captures the measurement error shocks to the signal about mti−1
.20
Note that this signal structure differs from those commonly seen real-time signal by which the
announcement releases data about the realization on the announcement day t. For instance,
the FOMC statement in the U.S. publishes the most recent federal funds rate target and
discount rate discussed from the just convened FOMC meeting, which reflects the monetary
policy moves around the announcement time. In later sections, we modify the setting to
allow for pre-scheduled announcements and real-time data releases, we show the reaction of
U.S. equity market to FOMC announcements can be rationalized in this unified framework
as well.
In sum, the market is subject to some information frictions. That is, the real money
growth mt evolves every day according to Equation (10) regardless of whether or not the
announcement arrives. Nonetheless, even if the investors incorporate the new information
conditional upon the announcement arrival on day tAi of month i, they only have access to
outdated realization of mti−1. For illustration purposes, we draw a time line in the following
to recapitulate the environment of our model.
19Daily money growth rate in our model can be easily converted to month-over-month growth rate to beperfectly aligned to the reality of announcement environment.
20In reality, the statistics of GDP deflater corresponding to pt in our model, and nominal balance ofmonetary aggregates Mt are routinely released through separate announcements and are computed by dif-ferent statistical agencies across countries. Our modelling choice of signal structure can be thus regardedas a composite signal that aggregates various sources of statistics relevant for investors to compute the realmoney balance growth subject to a given measurement error term. Though PBOC publishes the level andgrowth of nominal balance of money aggregates, it may be reasonable model the signal about real balancegrowth only. Besides the great simplicity, it is safe to assume the the real balance changes month to monthare mostly driven by nominal changes of M0/M1/M2 for aggregate price may move rigidly.
30
Figure 4: Time Line of Monetary Announcements
Calender Date
ti−1
last day of month i− 1ti
last day of month iti+1
last day of month i+ 1
signal about mti−1signal about mti
tai tai+1
According to Figure 4, note that the number of days that measures the interval length
of between two consecutive announcements tAi+1 − tAi is at least one day and no longer than
2 · N days of two months. Before lay out our theoretical results, we introduce some extra
notations. Conditional on having received an announcement, e.g. one made on day tAi , we
denote the forecast of mt conditional on the new information with mt|tAi = E(mt|ItAi ) and
conditional forecast variance V ar(mt|ItAi ).
Without loss of generalities, we assume time evolves as t enters month i+ 1. Prior to the
next announcement to be made on tAi+1 = ti+T , investors’ forecast of mti+x for x ≤ T ∈ [1, N ]
of day ti + x in month i+ 1 conditional on the past announcement made on day tAi is given
by
mti+x|tAi = ρN+xmti−1|tAi + (1− ρN+x)µ (12)
Equation (12) results from the AR(1) structure of mt. It says the conditional forecast has
rolled the forecast of mti−1|tAi conditional on the information from the past announcement
for N days throughout the month of i plus x days in month i+ 1 up until day ti + x. We see
for ρ ∈ (0, 1), as x goes larger, i.e. the further of the day being forecast from the end day
of previous month on which we had information about, the weight associated with outdated
conditional forecast ρN+x shrinks. On the other hand, investors forecasts get closer to the
unconditional mean µ, the a priori mean or the prior belief about money growth.
Regarding the conditional forecast variance, we have the following equation hold:
V ar(mti+x|ItAi ) = ρ2(N+x)V ar(mti−1|ItAi ) + V ar(
N+x−1∑j=0
ρjeN+x−j)
= ρ2(N+x)V ar(mti−1|ItAi ) + (1− ρ2(N+x))
σ2e
1− ρ2(13)
Similarly, we see as x increases, the forecast variance as weighted averages of the outdated
31
conditional forecast variance and the unconditional variance of mt tilts towards to latter.
In general, these equations imply that as time moves forward, the contribution of outdated
information contained in the past announcement to formation of forecasts is reduced over
time. With the following lemma, we then summarize the key proposition that associates size
of forecast uncertainty and the time lapsed since the past announcement till the arrival of
next announcement.
Lemma 1 Assuming central bank’s signal is informative, (1) forecast variance regarding
money growth rate of the last day of previous month conditional on the announcement is
smaller than the prior uncertainty of mt given its AR(1) structure, i.e. V ar(mti−1|ItAi ) <
σ2e
1−ρ2 . (2) Using constant γ to denote the size of reduction of prior uncertainty in percent
such that V ar(mti−1|ItAi ) = (1− γ) · σ2
e
1−ρ2 , γ ∈ (0, 1) is independent of future date t > ti−1.
This lemma directly results from the Bayesian updating formula such that a priori uncer-
tainty is reduced when beliefs are updated with a signal with finite variance. Rearranging
Equation (13), it thus gives
V ar(mti+x|ItAi ) =σ2e
1− ρ2(1− γρ2(N+x)) (14)
Equation (14) says as x goes larger, prior to next announcement, the informativeness of past
announcement that helps reduce forecast uncertainty shrinks over time. It is easy to show
that∂V ar(mti+x|ItA
i)
∂x> 0. This suggests that investors’ forecasts about real money balance
growth become increasingly uncertain over time. It further implies that if the announcement
is delayed as T goes larger, the uncertainty regarding mti+T keeps climbing even higher.
Proposition 2 Forecast uncertainty increases over time until announcement as∂V ar(mti+x|ItA
i)
∂x> 0 for x ≤ T ∈ [1, N ]. More delayed is the arrival of incoming an-
nouncement, the larger is the pre-announcement forecast uncertainty for∂V ar(mti+T |ItA
i)
∂T> 0.
We further discuss two scenarios under which the forecast uncertainty about mti+x
evolves. First, we present a baseline case if investors are generically waiting without pre-
learning about money growth rate until the arrival of next monetary announcement, i.e.
scenario of inattention. Then we construct the model block in which investors may or may
not pay attention to the money growth prior to the arrival of next announcement, i.e. sce-
nario of rational inattention. We show attentive learning of pre-announcement period drives
down the forecast uncertainty and generates the pre-drift of equity prices.
32
4.3 Inattention to Money Growth Rate
In this section, we examine a scenario in which investors are generically not paying any
attention to learn about money growth rate during the period of between announcements.
In specific, only the arrival of the monetary announcements would affect investors’ evolving
conditional expectation and forecast variance. Upon the arrival of announcement tAi+1 =
ti + T , investors’ forecast of mti+T conditional on the new information delivered is given by
mti+T |tAi+1= ρT mti|tAi+1
+ (1− ρT )µ (15)
Note the new announcement would directly revise the back-cast of mti by updating mti|tAiwith mti|tAi+1
. Equation (15) suggests that the new forecast of mti+T rolls the forecast mti|tAi+1
for T days into month of i+ 1. Similarly, the updated conditional variance is given by
V ar(mti+T |ItAi+1) = ρ2TV ar(mti |ItAi+1
) + (1− ρ2T )σ2e
1− ρ2(16)
Similar to Equations (12) and (13), as T rises with more delayed arrival of announcement,
conditional forecast and forecast variance turn toward to the a priori moments of µ and σ2e
1−ρ2 .
Now we look into how mti|tAi is updated by mti|tAi+1given the new announcement. Applying
the Bayes’s rule, the updated forecast about mti on the last day of month i linearly combines
the forecast that is carried over and the announcement signal as weighted by their relative
informativeness. We thus have the following
mti|tAi+1= (1− κ)mti|tAi + κstAi+1
(17)
where κ =1/σ2
η
1/σ2η+1/V ar(mti |ItA
i)
captures the Kalman gain, which measures how much the
updated forecast should be weighted towards the new announcement signal. It shows that
more precise signal of smaller σ2η or the less precise carried forecast of larger V ar(mti|tAi )
raises the Kalman gain. The updated conditional forecast variance about mti similarly
satisfies the following
1
V ar(mti |ItAi+1)
=1
V ar(mti |ItAi )+
1
σ2η
(18)
Equation (18) implies that more precise of the signal with smaller ση, the revised forecast
uncertainty V ar(mti |ItAi+1) is further reduced.
By Equations (13) and (16), we can re-express the conditional forecast of mti+T based
on past announcement using V ar(mti+T |ItAi ) = ρ2TV ar(mti |ItAi ) + (1−ρ2T ) σ2e
1−ρ2 . As a result,
33
the forecast uncertainty changes for day tAi+1 = ti + T in month i + 1 due to arrival of new
announcement ∆ti+TV ar is given by
∆ti+TV ar = V ar(mti+T |ItAi+1)− V ar(mti+T |ItAi )
= ρ2T [V ar(mti |ItAi+1)− V ar(mti |ItAi )]
= −ρ2T
σ2η
V ar(mti |ItAi+1) · V ar(mti |ItAi ) < 0 (19)
The inequality directly comes from Equation (18) for the fact that an announcement signal
is informative such that certain precision is taken in to reduce the carried uncertainty on the
announcement day such that ∆ti+TV ar < 0. Further, we have the following proposition
Proposition 3 Given backward-looking announcements with timing randomness, as the cen-
tral bank’s announcement is increasingly delayed, the reduction of investors’ forecast uncer-
tainty on the announcement day gets smaller for∂|∆ti+T
V ar|∂T
< 0. Less precise the central
bank’s signal is, a smaller reduction of uncertainty results such that∂|∆ti+T
V ar|∂σ2η
< 0.
Proposition 3 highlights the fact that the reduction of forecast uncertainty on the announce-
ment day depends on the initial size of decreased back-cast uncertainty regarding the money
growth rate realized in the past, which depreciates at a daily rate of ρ2 over T days until
announcement. More delayed the announcement is, the size of initial reduction is scaled
down further more.
4.4 Endogenous Learning with Limited Attention
In this section, we build in the endogenous information choice made by investors before
the arrival of incoming announcement. Information acquisition is a result of investors’ weigh-
ing the marginal gain of pre-announcement learning against the marginal cost. Our model
employs a framework of agents having limited attention, i.e. rational inattention. (Sims,
2003; Peng and Xiong, 2006; Mackowiak and Wiederholt, 2009) In the model, the learning
decision amounts to investors allocating the optimal amount of attention to the variable of
interest, the consumption growth rate and in equilibrium the log money growth rate, subject
to some information processing capacity constraint. In specific, conditional paying atten-
tion, investors learn about mt via some optimized private signals with the right amount of
residual uncertainty. Consequently, this pre-announcement attentive learning action reduces
the forecast uncertainty in absence of learning, and thus affects the equity prices.
We frame investors’ information problem in the following. Entering day t, investors de-
cide how much attention should be paid to learn about consumption growth rate and money
34
growth. By optimizing over the size of attention, attentive learning shrinks the forecast
variance carried into day t to a target size. Therefore, the optimized size of uncertainty re-
duction is equivalently proxied by the magnitude of optimal attention allocated by investors.
In line with the literature, we similarly assume that investors’ perfect learning is not attain-
able, which makes the size of uncertainty reduction as measured by the information entropy
changes each day capped from above by some finite information processing capacity κ > 0.
The actual usage of information capacity κt every day is thus bounded by [0, κ]. Note that
κt = 0 captures the case when investors decide to pay no attention at all to money growth
rate so that no optimized uncertainty reduction is obtained.
Importantly, we assume optimizing the attention incurs a flow cost of learning v > 0 per
using up one more bit of information capacity and a fixed cost ζ > 0 in numeraire of forecast
variance. The flow cost of paying attention can be interpreted as the opportunity cost when
investors paid attention to money growth while at the same time stay inattentive to other
variables of interest. The fixed cost can be rationalized as the cost that investors must pay
upfront for collecting private information in form of designing the data collection scheme,
setting up a good measurement, and preparing a research report etc. We further assume
that the fixed cost is only paid once in a monthly announcement cycle. That is, conditional
upon learning on day t with fixed cost paid, adjustment of daily attention in the following
days before announcement no longer incurs this cost. However, conditional on the arrival
of incoming announcement, the next learning cycle is triggered and the fixed cost scheme is
reset.21
Investors’ utility maximization problem can be approximately re-written as utility loss
minimization of sub-optimal consumption decisions relative to the case of optimal consump-
tion sequence. We show in the Appendix that with a second order approximation, in-
vestors’ objective function for maximizing recursive life-time utility amounts to the min-
imization of the following utility loss due to suboptimal consumption growth log ct for
min∆ log ctλt2E(∆ log ct−∆ log c∗t )
2. λt > 0 is a time-varying parameter that is independent on
consumption growth deviations and scales the utility loss due to suboptimal consumption.
Attention allocation problem builds on the information frictions that generates the deviation
of sub-optimal decision due to imperfect information from optimal decision of perfect infor-
mation. While we have show in Section 4.1 the optimal consumption decisions conditional
on information set It, in the following, we further optimize over the information structure
by pinning down the optimal attention allocated to minimize the utility deviation.
In equilibrium, log money growth rate amounts to consumption growth mt = ∆ log ct.
21For example, each monthly announcement cycle, as it approaches the data releases dates, investorswould start collecting relevant data for devoting attention.
35
Using mt to denote the growth of monetary supply that satisfies the sub-optimal consumption
growth in equilibrium and m∗t for the optimal money growth rate given perfect information,
we thus have the following minimization problem
minκt
λt2E(mt −m∗t )2 + ζ · ικt + vκt
s.t. H(mt)−H(mt) = κt (20)
0 ≤ κt ≤ κ (21)
Choice variable κt denotes the size of attention allocated to learn about money growth rate.
Start paying attention with triggers the fixed cost ζ as indicator function ικt = 1. Otherwise
ικt = 0, if no attention is paid or the fixed cost has been paid since the past announcement
before arrival of the next. Once attention is allocated, attentive learning is modelled as
observing an optimized signal or setting up an measurement ft with ft = m∗t + ut. The
noise term ut is independent of the true state of money growth m∗t and i.i.d. distributed
with ut ∼ N(0, σ2m|f,t). σ
2m|f,t captures day t forecast variance of mt if attention is optimized.
Conditional on the optimized attention, our best estimate of the true state given by the
signal is E(m∗t |ft) = ft. Investors then align their action to the estimate such that mt =
E(m∗t |ft) = ft. H(x) = 12
log2 σ2x denotes the information entropy measure associated with
normally distributed random variable x. Therefore, H(mt) captures the forecast uncertainty
without attentive learning whereas H(mt) gives the information entropy after the attention
optimization.
Investors’ objective function for attention optimization can be further simplified to com-
pare the value of being inattentive to mt, πNt and the value of doing attentive learning πLt .
σ2m,t denotes the end of day t forecast variance if no attention is paid for day t. σ2
m|f,t therefore
measures the residual uncertainty about mt after attention is allocated. Hence, we rewrite
the information optimization problem in the following
maxπNt , πLt
= max−λt2σ2m,t,max
κt>0−λt
2σ2m|f,t − ζ − vκt (22)
s.t. σ2m|f,t = σ2
m,t2−2κt (23)
0 ≤ κt ≤ κ (24)
Note that attentive learning is irrelevant when σ2m,t <
vλt
where v = vlog(2)
. Equation (22) says
that conditional on learning, the marginal benefit from learning λtσ2m,t log(2)2−2κt decreases
in attention paid κt. If starting from κt = 0 with max marginal benefit still dominated by
36
the marginal cost of learning v, learning has no net value such that πLt < πNt if σ2m,t <
vλt
.
In the appendix, it can be shown that the relative gain from paying the right amount of
attention to money growth is given by
∆πt = πLt − πNt =λt2σ2m,t − (
v
2+v
2log2[
λtσ2m,t
v] + ζ) (25)
Denote σ2∗m,t >
vλt
to be the threshold such that investors are indifferent between atten-
tive learning and being inattentive with ∆πt = 0. The optimal attention decision can be
summarized by the following step function:
κ∗t =
κ if σ2
m,t > σ2∗m,t and κ < 1
2log2[
λtσ2m,t
v]
12
log2[λtσ2
m,t
v] if σ2
m,t > σ2∗m,t and κ ≥ 1
2log2[
λtσ2m,t
v]
0 if σ2m,t ≤ σ2∗
m,t
(26)
The solution function says that when the no-learning forecast uncertainty σ2m,t is not large
enough, investors find it unnecessary to be attentive but rather let go the accumulation of
forecast variance over time, thus κ∗t = 0. When the forecast uncertainty σ2m,t outsizes the
threshold σ2∗m,t, investors start paying attention to the money growth and the attentive learn-
ing from optimized attention allocation helps reduce the forecast uncertainty in absent of
learning σ2m,t to σ2
m|f,t by a factor 2−2κ∗t . However, as the σ2m,t is too large, the optimal atten-
tion associated with the information flow is capped by the maximal information processing
capacity of day t such that κ∗t = κ.
In Figure 5, we plot an illustrative example summarizing the decision rule for attentive
learning. The vertical distance captures the relative gain from attentive learning ∆πt. We
see that investors would not learn even if marginal benefits outweighs the marginal cost of
learning such that σ2m,t = v
λtdue to the non-zero fixed cost ζ > 0. However, across the range
of σ2m,t ∈ [ vt
λt, σ2
e
1−ρ2 ], the relative gain from learning increases in the magnitude of forecast
uncertainty in case of no learning until attentive learning brings more value to investors as
σ2m,t > σ2∗
m,t. Further, as σ2m,t grows over v
λt22κ, the optimal information flow has to be capped
by the processing capacity κ. Therefore, the marginal benefit of learning becomes larger as
κ∗t = κ, which leads to a steeper slope with a kink on the function ∆πt.
Lemma 2 For t ∈ [tAi , tAi+1],∀ month i with κ∗t > 0, given that ζ = 0 for t+x with t+x ≤ tAi+1,
it yields that σm,t2∗ = v
λt.
Lemma 2 says between announcements, since investors start paying attention after a fixed
cost is incurred, the fact that no more fixed cost is needed makes the threshold for atten-
tive learning shrink to σ2∗m,t = v
λt. According to Equation (25), this is a critical point for
37
Figure 5: Relative Gain from Attentive Learning with Rational Inattention
Notes: This figure plots an illustrative example of investors’ information acquisition decision as a function of σ2m,t,
the forecast variance if there is no attentive learning for day t. See text for definitions of variables. Y-axis denotes the relativegain from learning ∆πt = πLt − πNt . X-axis tick v
λt22κ marks the where information flow starts being capped by κ. X-axis tick
σ2e
1−ρ2 marks the unconditional variance of money growth rate mt.
∆πt = 0 by which the marginal benefit of learning offsets the marginal cost of learning v
as only the flow cost affects the attention allocation problem moving forward until the next
announcement arrival. Therefore, conditional on fixed cost is already paid in a monthly
announcement cycle, investors would do attentive learning prior to the next announcement
as long as σ2m,t >
vλt
. With the proof relegated in the Appendix, we then state the following
proposition:
Proposition 4 Regardless of whether ζ = 0,dσ2∗m,t
dλt< 0 and
dσ2∗m,t
dv> 0, that is, indifferent
investors will pay attention to learn about money growth and to reduce the growing fore-
cast uncertainty if ceteris paribus, (1) the marginal utility loss from suboptimal attention λt
increases or (2) flow cost of learning v decreases.
Proposition 4 results from the fact that the threshold in terms of the magnitude of forecast
uncertainty in absent of learning can be shifted by parameterization of the importance for
optimizing the utility loss λt and the information processing cost v. We relegate later sec-
tions to draw additional model implications for the cross-sectional dimension based on this
proposition.
Conditional on the attention allocation κ∗t per Equation (26), we summarize the difference
of updated forecast variance relative to that without learning based on the attentive learning,
38
i.e. magnitude of uncertainty reduction in the following.
σ2m,t − σ2
m|f,t =
σ2m,t(1− 2−2κ) if σ2
m,t > σ2∗m,t and κ < 1
2log2[
λtσ2m,t
v]
σ2m,t − v
λtif σ2
m,t > σ2∗m,t and κ ≥ 1
2log2[
λtσ2m,t
v]
0 if σ2m,t ≤ σ2∗
m,t
(27)
Equation (27) suggests that regardless of whether the information processing capacity is
binding, the optimized size of uncertainty reduction conditional on learning κ∗t > 0 increases
in the magnitude of uncertainty if no attentive learning is applied σ2m,t. Given Proposition
2, the following proposition results
Proposition 5 As V ar(mti+x|tAi ) accumulates over days of x prior to monetary announce-
ment x ≤ T , investors pay attention to the money growth rate data if the announcement
is more delayed such that carried forecast uncertainty without learning is greater than some
threshold on day ti + x, V ar(mti+x|tAi ) > σ2∗m,ti+x
. As a result, larger reduction of forecast
uncertainty is associated with announcements that are made public with more days of delay.
Hence, over time, the forecast variance regardless of whether attentive learning is at place
(no if σ2m|f,t = σ2
m,t), moves over time according to the following law of motion.
σ2m,t+1 = ρ2σ2
m|f,t + σ2e (28)
Given all our results, with certain parameterization, we give a graphical example of the
path of forecast variance evolution under two different scenarios (inattention vs. optimized
learning) in Figure 6. Starting from day ti the end of month i, forecast variance accumulates
over time entering month i+ 1. If the announcement is yet to arrive, absent of any form of
information, investors’ forecast uncertainty will keep climbing over time and gets converging
to the a priori uncertainty as date ti + x moves forward. This path of no announcement is
captured by the Magenta-colored dotted line.
In the scenario of generic inattention to money growth, conditional upon the arrival of
announcement on day tAi+1, we see the forecast variance on the announcement day gets a
sizable reduction, then it enters another round of uncertainty accumulation path as denoted
by the black solid line dotted with circles. Note that the revised and lower forecast variance
on day tAi+1 is because the back-casts between dates of ti and tAi+1 are updated. In addition,
Proposition 3 suggests that if we move the vertical red-dashed line of announcement day mark
to the right, a more delayed announcement gives a smaller reduction of uncertainty on the
announcement day, which is measured by the vertical distance between the Magenta-colored
dotted line and the black circled line.
39
Figure 6: Inattention vs. Attentive Learning: Uncertainty Reduction
Announcement Day
Notes: For given parameterization with fixed λt = λ, the vertical red dashed line marks the date of monetary an-nouncement. Magenta-colored dotted line denotes the path of uncertainty accumulation absent the arrival of announcementand any form of information. The black solid line dotted with circles captures the initial accumulation and reduction ofuncertainty conditional upon the information delivered through announcement only. The blue solid line dotted with plus signsnotes the path of uncertainty reduction with endogenous learning decision prior to monetary announcement. The horizontalgreen solid line marks the unconditional variance of real money growth in log mt.
Under the scenario of attentive learning based on attention optimization, we simulate a
path of forecast uncertainty starting from V ar(mti |IAi ) taking consideration of the learning
decision as characterized by Equation (26). As forecast uncertainty accumulates, investors
find optimal to learn prior to announcement as carried uncertainty grows over some threshold
on day tLi+1 by paying some fixed cost ζ > 0. The consequential reduction of uncertainty
via attentive learning is further capped by the information capacity κ. Since continuous
attention devotion is waived of any ongoing fixed cost, uncertainty will keep going down
though subject to daily constraint by converging to σ2m,t = v
λt. This declining process may
stop as long as convergence is complete even if the next announcement has not arrived.
However, once the announcement is realized on tAi+1, fixed cost for learning about next month
daily mt kicks in and investors stay rationally inattentive by letting uncertainty accumulate.
Important to note that our model generates the endogenous reduction of uncertainty due to
attentive learning which happens prior to the announcement, which consequently generates
the pre-announcement premium.
Ultimately, to draw implications on the equity prices, as implied by Equation (9), we
give out the theoretical account for the cause of pre-announcement premium in China.
Proposition 6 (Uncertainty Reduction and Pre-announcement Premium) Given
backward-looking monetary announcements with announcement timing randomness, exces-
sively accumulated market uncertainty triggers attentive learning prior to the announcement,
40
which drives down the forecast uncertainty about money growth rate and boosts up the equity
prices.
4.5 Discussion
In summary, we present a model that features both the within-month timing variation
of announcements and the backward-looking nature of monetary announcements given the
endogenous learning choice made by investors. As the expected equity excess return increases
and current stock price decreases in the forecast uncertainty, more attentive learning triggers
the reduction of forecast uncertainty before announcement, which leads to the pre-drift of
stock prices. In the next few sections, we proceed to present model-consistent evidence of
uncertainty reduction prior to monetary announcements in China.
Importantly, our model gives rich implications on the relationships between uncertainty
reduction and equity premium. First, more delayed announcement, by accumulating greater
carried uncertainty, induces more attentive learning that results in greater uncertainty re-
duction. This should lead to the correlation between delayed announcements and larger
associated pre-announcement premium. On the other hand, early arrivals of monetary an-
nouncements, by prediction, suggest that investors’ prior uncertainty about those monetary
data should not be accumulated big enough to trigger attentive learning let alone the con-
sequential uncertainty reduction. Therefore, we are motivated to look for evidence that the
pre-announcement premium might be largely driven by late announcements. This particular
model prediction then serves as our key identification for the causal link from uncertainty
reduction to the size of pre-announcement premium as proposed by Ai and Bansal (2018).
Specifically, by showing that PBOC’s announcement timing is hard to predict, announcement
timing randomness will generate variations of uncertainty reduction across announcement
events, by which we can check if size of premiums can be explained by the timeliness of
announcement arrival.
Second, we have shown in Proposition 4 that λt the measure of the importance of opti-
mizing utility loss. From a cross-sectional perspective, stock portfolios that are particularly
sensitive to monetary policy risk could mean larger λt’s for investors. As a result, for cer-
tain categories of stocks, they may exhibit excessively larger pre-announcement premium
due to larger gain from attentive learning. By exploiting the cross-sectional variations of
stock performance in the data, we present additional evidence to lend credence to the model
prediction on the cross-sectional dimension.
Third, our model suggests that the country-specific parameters regarding the announce-
ment environment for China and the U.S. may account for the quantitative but not the
41
qualitative differences of their pre-monetary announcement. U.S. exhibits relatively short
duration for few hours of pre-FOMC announcement premium according to Lucca and Moench
(2015) and conditional upon FOMC statement issuance, the stock market jumps in response
to uncertainty resolution (Ai and Bansal, 2018). Our theory about attentive learning squares
well with the U.S. evidence at least qualitatively. In particular, federal funds rate futures
and derivatives are actively traded by which forecast uncertainty may be hard to get accu-
mulated between FOMC announcements. Plus, the FOMC statement days are pre-scheduled
(without announcement timing randomness) and the statement precisely states the real-time
FRB’s decision on monetary policy (not backward-looking). Correspondingly, uncertainty
can be attenuated between announcements in this environment and the cost of pre-learn
(low ζ and v) and the benefit of waiting for more precise public communications (small ση)
can be huge. Therefore, U.S. investors may find it not optimal to learn about policy actions
too far ahead the scheduled date as the FOMC meetings are yet to hold. Hence, just a few
hours before the FOMC meetings, people are guessing what can be the policy adjustment
and fixed cost is paid for to acquire real-time information with increased attention to FRB’s
data. As a result, quick learning and somewhat reduction of uncertainty accounts for the
short-term pre-drift of U.S. stock market returns.
5 Identification: Timing-Dependent Premium
In this section, we further examine the model implications in the data. By exploiting the
random variation in the timing of PBOC’s monetary announcements, we present the identi-
fication of the causal link between uncertainty reduction and the pre-drift of equity returns
before monetary announcement. First, evidence shows that prior to M2 announcements,
investors’ forecast uncertainty as proxied by the stock return volatility declines. Second,
our empirical tests of the model propositions confirm that timeliness of monetary announce-
ment affects the magnitude of uncertainty reduction, which generates varied sizes of pre-
announcement premium across events. In the end, suggestive evidence shows that interests
and attention to PBOC’s actions, as measured by the intensity of PBOC’s web traffic, in-
creased a few days before the arrival of M2 announcement.
5.1 Correlations
According to model Proposition 6, the pre-announcement equity premium should be
coupled with with significant forecast uncertainty reduction. To identify forecast uncertainty
changes within just a few days of a monetary announcement window, uncertainty measures
42
based on low frequency forecast data is not directly testable. We therefore use the daily
stock return volatility aggregated over high frequency returns to approximate the investors’
forecast uncertainty in the model.22 We note that investors’ forecast uncertainty in our
model is about the growth rate of real money balance and in equilibrium, the consumption
growth rate, i.e. economic fundamentals. However, suggested by works that have shown
that stock market itself is an aggregated signal containing news about future movement of
economic fundamentals (Beaudry and Portier, 2006) and Chinese stock market well reflects
its profitability opportunities (Carpenter et al., 2017), using stock market volatility as our
empirical proxy for forecast uncertainty can be the best among few alternatives.
Figure 7 plots the mean daily return volatilities aggregated over five minutes trading
blocks about an average 11-day window of M2 announcement for the two Chinese stock
market exchange indexes, Shanghai (SSE) and Shenzhen (SZSE). The two volatility series
are normalized as divided by their corresponding unconditional means of daily volatility
excluding all days falling into the 11-day announcement windows. If this relative volatility
measure falls below one, it implies that the level of uncertainty on that day is smaller than
that of a non-announcement window daily average.
First, this figure shows that regardless of stock exchanges, across all M2 announcement
windows in our sample, there is a clear trend of uncertainty reduction. This is consistent
with our model implication that as uncertainty reduction is driven by attentive learning,
binding daily constraint of information flow can generate a gradual decline of uncertainty
over days prior to announcement. In terms of the correlation with equity returns, this
continuous reduction of uncertainty is consistent with pre-announcement drift of returns
for a duration of more than one day highlighted in Table 6. Second, drops of uncertainty
initialize from the peak level of return volatility which is about the fourth day prior to the
announcement. This piece of evidence echoes well with our model that uncertainty has to
climb over some threshold after which the reduction of uncertainty starts taking place. Third,
this relative uncertainty decays by hitting the bottom on the day before announcement. A
significantly lower degree of uncertainty relative to that of non-announcement days squares
well with our findings from Table 4 that only the coefficient estimate associated with tM2 − 1
in the regression analysis is significant. Lastly, it turns out the relative uncertainty stays
flat throughout the post-announcement period, which is aligned with the facts that no post-
monetary announcement drift of daily returns is found in China.
We further test the null hypothesis that the level of forecast uncertainty as proxied by
22This requirement rules out examining measures of uncertainty using forecast dispersion based onmonthly survey data such as Bloomberg surveyed forecasts about major economic and financial barome-ters for China. In addition, there is no option-based implied volatility index for Chinese stock exchanges ortext-based uncertainty proxies as in Baker et al. (2016) that is available up to daily frequency.
43
Figure 7: Stock Return Volatility in Windows of M2 Announcements
Notes: Sample: January, 2011 to December, 2016. This figure shows the average relative daily stock market returnvolatility to non-announcement window averages, which aggregates five-minutes return blocks per day on the Shenzhen StockExchange Component (SZSE) Index and Shanghai Stock Exchange Composite (SSE) Index around an M2 announcement.t = 0 marks the first trading day on which the market has access to the announcement as denoted by a vertical solid line.j captures the relative volatility |j| day after (before if j negative) the announcement. The solid line and the dashed linerespectively denote the relative volatility of the SSE and SZSE stock index.
the daily stock return volatilities prior to M2 announcement is no different from that of
other days outside the pre-announcement windows. Precisely, we estimate the following
model based on dummy indicator ItM2−1,j = 1 denoting a trading day that falls into a pre-
announcement window of j days from t− 1 to t− j for j = 1, 2, 3, 5:
Ret V olt = β0 + βjItM2−1,j + βxXt + υt (29)
Estimate of βj thus captures the size of daily stock return volatility Ret V olt before an
M2 announcement day, relative to a level of uncertainty that is outside these windows.
Table 9 summarizes the estimation results. It shows that the regression analysis yields the
similar picture as revealed in Figure 7: uncertainty is relatively smaller across a few days
before the arrival of an M2 announcement. While little difference can be discerned between
return volatilities of Shanghai and Shenzhen stock exchanges, we see across columns that
the shorter of the pre-announcement window length is associated with a even lower average
daily volatility. This finding implies that the forecast uncertainty keeps declining from a
44
higher level until reaching the bottom just one day prior to the announcement. Again, recall
the regression results in Table 4 about pre-drift of excess equity returns exhibit that the
pre-announcement premium is most significant on day tM2−1. Hence, a casual link between
the biggest drop in uncertainty and the largest jumps in excess returns can be noted here.
Table 9: Relatively Low Uncertainty Prior to M2 Announcements
(1) (2) (3) (4) (5) (6)VARIABLES SSE SSE SSE SZSE SZSE SZSE
ItM2−1 -0.13** -0.17***(0.05) (0.06)
ItM2−1,2 -0.10** -0.13**(0.05) (0.05)
ItM2−1,3 -0.07* -0.10**(0.04) (0.04)
Year FE Yes Yes Yes Yes Yes YesMonth FE Yes Yes Yes Yes Yes YesWeekday FE Yes Yes Yes Yes Yes YesConstant 1.24*** 1.21*** 1.21*** 1.63*** 1.59*** 1.60***
(0.09) (0.09) (0.09) (0.10) (0.10) (0.10)
Observations 1,458 1,458 1,458 1,458 1,458 1,458R-squared 0.32 0.32 0.32 0.27 0.27 0.27
Notes: Sample: January, 2011 to December, 2016. This table reports the dummyvariable regression results of Equation (29). The dependent variable is the intra-day return volatility constructed based on five minutes trading returns for SSE(Shanghai) or SZSE (Shenzhen) index. We align the return data of the first tradingday that the equity market has access to the news to the day tM2. For Columns(1) and (4), all other trading day dummies t − j beyond j = 1 are included ascontrol but not reported. Announcement dummy ItM2−1,j equals to one for thetrading days in a j-trading-day window before an M2 Announcement. “Year FE”,“Month FE”, and ”Weekday FE” represent the fixed effects controls for the year,month, and weekday respectively. ***Significant at 1%, **significant at 5%, *sig-nificant at 10%. Robust standard errors are shown in parentheses.
We move on to directly examine the association between the reduction of uncertainty prior
to announcement per our empirical measure based on stock return volatility, and the size of
pre-announcement equity premium. Per Proposition 6, we look for a positive correlation in
the data. For a given announcement q, we estimate partial effects of uncertainty reduction
on excess return by exploring cross announcement event variations.
Cumretj,q = β0 + β1∆Ret V olj,q + βxX + eq (30)
where ∆Ret V olj,q = Ret V oltM2−1,q − Ret V oltM2−j,q which measures the uncertainty
changes from the j-th day till the day right before the day of announcement q. Cumretj,q =∑ji=1ExrettM2−i,q captures the cumulative equity return obtained over the j day pre-
announcement window. Note that we have excess returns ExrettM2−i expressed in logs and
simply sum over the daily returns in the window. In the control variable list X, we examined
the robustness of our results by controlling the mean daily average stock market return for
the window. We examine the pre-announcement windows of three days and two days with
45
results summarized in Panel A and Panel B respectively in Table 10. Columns (2) and (5)
present estimation results by controlling for the year fixed effect relative to Columns (1) and
(4). These four columns show that larger uncertainty reduction (or increasingly accumulated
uncertainty) over time is positively (negatively) correlated with cumulative returns in the
j-day window. We didn’t control for additional fixed effects up to month and day given that
our total number of announcement observations is not great.23 According to Column (3) and
(6), having corrected for the level of stock return volatility, we see the estimated magnitude
of correlations based on SSE and SZSE index get closer to each other.
Table 10: Uncertainty Reduction and Size of Announcement Premium
Panel A: Cumret3,q
(1) (2) (3) (4) (5) (6)VARIABLES SSE SSE SSE SZSE SZSE SZSE
∆Ret V ol3,q -2.28** -2.07* -2.08** -1.54** -1.62** -2.07***(1.14) (1.04) (0.94) (0.72) (0.67) (0.77)
Mean Ret V ol3,q 1.07 1.40*(0.98) (0.79)
Year FE No Yes Yes Yes Yes Yes
Observations 72 72 72 72 72 72R-squared 0.16 0.24 0.27 0.06 0.22 0.26
Panel B: Cumret2,q
(1) (2) (3) (4) (5) (6)VARIABLES SSE SSE SSE SZSE SZSE SZSE
∆Ret V ol2,q -1.46* -1.75** -1.82** -1.40* -1.77** -1.78**(0.84) (0.80) (0.79) (0.83) (0.82) (0.80)
Mean Ret V ol2,q -0.57 -0.54(0.69) (0.48)
Year FE No Yes Yes Yes Yes Yes
Observations 72 72 72 72 72 72R-squared 0.09 0.19 0.21 0.08 0.21 0.22
Notes: Sample: January, 2011 to December, 2016. The dependent variable is thecumulative excess return obtained up to the day prior to the M2 announcement dayfor a window length of j days. Uncertainty reduction is measured by the differenceof intra-day return volatility constructed from high-frequency return data of SSE(Shanghai) and SZSE (Shenzhen) market index. ***Significant at 1%, **significantat 5%, *significant at 10%. Robust standard errors are shown in parentheses.
5.2 Causal Explorations based on Timing Randomness
In this section, we exploit the randomness in PBOC’s M2 announcement timing in order
to identify the role of uncertainty reduction as the trigger for pre-announcement drift of eq-
uity returns. Proposition 2 of our model highlights that size of investors’ forecast uncertainty
of day t can be simply measured by the duration of time lapsed from the past announce-
23We note that further controlling for month and weekday fixed effects attenuate the significance.
46
ment up to day t. Hence, by Proposition 5, delayed arrival of monetary announcement
with increasingly accumulated forecast uncertainty tends to trigger the attentive learning
among investors and generate larger pre-announcement premium. In following, we test if the
unexpected early or late arrivals of monetary announcement affect the size of uncertainty
reduction and in turn the magnitude of pre-announcement premium.
We first present the evidence that conditional on a range of real-time information avail-
able, the exact date of PBOC’s M2 announcement every month is largely unpredictable. This
randomness generates the exogenous variation in announcement timing across announcement
events. Second, we are able to show that more delayed announcement triggers larger un-
certainty reduction while exhibits greater pre-announcement premium. In sum, enabled by
the uniqueness of Chinese data, this paper directly tests the key mechanism of uncertainty
reduction as the main cause for pre-announcement premium as first laid out in Ai and Bansal
(2018).
To establish that the exact PBOC’s M2 announcement day in a month is not predictable,
we estimate the following empirical specification
DaytM2,q = β0 + β1 · Factorq + β2X + υq (31)
DaytM2,q gives the exact day of month for the day of announcement q, tM2,q ∈ [1, 31]. Smaller
of this measure gives that the announcement arrives earlier in a given month. We select a
range of variables of interest to be the predicting factor for the announcement day Factorq. In
specific, we use the level of M2 growth rate surveyed by the Bloomberg prior to announcement
with and without subtracting the realized M2 data for previous month, gM2,t, gM2,t−gM2,PM ;
the realized M2 growth rate for the previous month (PM), gM2,PM ; the day of month for the
previous announcement, DayPM ; and the announcement day of month corresponding to the
same month of previous year (PY), DayPY . We summarize the regression results in Table 11.
Across Columns (1) to (5), we see that none of the aforementioned predictors at least partially
predicts the timeliness of announcement arrival. Suspecting that multiple variables may be
jointly informative for predicting the forthcoming date, results from Column (6) suggest that
the null of no predictability cannot be rejected. In general, it is safe to conclude that based
on real-time information publicly available to the market, the precise announcement arrival
may be unexpectedly early or late. This piece of finding provides the necessary cross-event
exogenous variation by which we account for the pre-drift of equity returns with size of
pre-announcement uncertainty reduction.
47
Table 11: Predictability of M2 Announcement Timing
(1) (2) (3) (4) (5) (6)VARIABLES day day day day day day
gM2,t − gM2,PM -0.84 -0.62(0.61) (0.64)
gM2,PM 0.18 0.07(0.17) (0.19)
gM2,t 0.16(0.19)
DayPM -0.17 -0.15(0.14) (0.13)
DayPY 0.07 0.02(0.15) (0.15)
Year FE YES YES YES YES YES YESMonth FE YES YES YES YES YES YESWeekday FE YES YES YES YES YES YESObservations 78 78 78 78 78 78R-squared 0.00 0.02 0.03 0.02 0.11 0.16
Notes: Sample: January, 2011 to June, 2017. The dependent variableis the day of month associated with an announcement day. “Year FE”,“Month FE”, and ”Weekday FE” represent the fixed effects controls foryear, month, and weekday respectively. PM: previous month. PY: samemonth of previous year. gM2,PM : growth rate of M2 announced in theprevious month; gM2,t: bloomberg survey data of M2 growth rate. ***Sig-nificant at 1%, **significant at 5%, *significant at 10%. Robust standarderrors are shown in parentheses.
We continue to show that larger uncertainty reduction is associated with more delayed
monetary announcement in China, which is consistent with model Propositions 2 and 5.
Note that the magnitude of forecast uncertainty according to our model is endogenously
determined by attentive learning. Therefore, we are not able to uncover in the data the
equivalence of announcement timeliness and the size of uncertainty under the generic inat-
tention scenario. Rather, the equilibrium difference in size of uncertainty given attentive
learning is taken by equity investors relative to inattention can be identified if it’s indeed
dependent on the timeliness of announcement arrival. In specific, we first break our sample
of announcements into two groups in terms of their timeliness of arrival relative to some day
of month as day cutoffs Daycutoff , i.e. earlier than Daycutoff (DaytM2< Daycutoff ) and later
than Daycutoff group (DaytM2> Daycutoff ). Then we examine the performance of stock
market return volatility, our measure of forecast uncertainty Ret V olt in an announcement
window relative to days outside the announcement window, which is in line with our baseline
specification for identifying the pre-drift of excess equity returns per Equation (1).
Table 12 presents the magnitude of stock return volatility constructed from SSE and SZSE
market index during the announcement window conditional on whether the monetary data
is timely released in a given month. Across the two Chinese stock market exchanges, results
shown in both Panel A and Panel B suggest that most of the coefficient estimates for dummy
ItM2−1 are significantly below zero. While for the concern of table space, we suppress all
48
other dummy estimates which turn out to be statistically insignificant. Therefore, regardless
of announcement timing, investors’ forecast uncertainty is lower on the day before M2
announcements relative to an average day falling into a no-announcement window, which
is in line with results shown in Table 9. More importantly, Focusing on the group of early
released announcements, Columns (1) to (5) of both SSE and SZSE return volatilities all
suggest that as the monetary data is released earlier, the relative magnitude of uncertainty
prior to announcement decreases and eventually shrunk to a level of no significance if the
announcement day is before 10th of the month. According to Columns (6) to Columns
(10), evidence from the late announcements reveal that the relative uncertainty on average
is smaller than that associated with early delivered announcements. Though the coefficient
magnitude slightly increases as the announcement day moves even late in the month, the
size of lowered uncertainty prior to an announcement that delayed publishing monetary data
until 13th almost doubles the size of reduced uncertainty associated with an announcement
delivered before 10th of the month.
Given that uncertainty as proxied by stock return volatility decreases prior to announce-
ment, for robustness, we further test the null hypothesis that the timeliness of M2 announce-
ment of a month is not correlated with the size of pre-announcement uncertainty reduction.
Across announcement events, we estimate the following specification about the linear and
squared term of the day of month DaytM2,q associated with the announcement q date tM2,q
∆Ret V olj,q = β0 + β1DaytM2,q + β2 ·Day2tM2,q
+ βxX + eq (32)
where ∆Ret V olj,q = Ret V oltM2−1,q−Ret V oltM2−j,q measures the uncertainty changes from
j-th day till the day right before the day of announcement. We check j = 3, 5 looking for
evidence whether delayed announcement day with larger DaytM2,q is associated with greater
uncertainty reduction by focusing on β1. In addition to the level difference of volatilities, we
also consider the percent changes of uncertainty as the dependent variable.
Table 13 presents the estimation results for both Shanghai and Shenzhen stock exchanges
respectively in two panels. Panels 1 and 2 differ in the length of observation window of
j = 3, 5 days. Panels B are associated with results about the rate of uncertainty reduction.
Focusing on Columns (2) and (6), we find that one more day of monetary announcement
delay shrinks the stock market return volatility. In addition, as implied by the degree of
“convexity” effect associated with the squared term Day2tM2,q
, the magnitude of uncertainty
reduction per one day of delay gets smaller as the announcement day is further postponed.
However, this convex impact of announcement timeliness is comparatively trivial. For ex-
ample, DaytM2,q has to beat least the 25 (25-th day of a month) in order to completely
49
Table 12: Size of Uncertainty Reduction: Early vs. Late Announcements
Panel A: Shenzhen Stock Exchange Component Index
(1) (2) (3) (4) (5)VARIABLES Earlier than 10 Earlier than 11 Earlier than 12 Earlier than 13 Earlier than 14
ItM2−1 -0.09 -0.15* -0.15** -0.18** -0.19***(0.10) (0.09) (0.07) (0.07) (0.06) )
Observations 1,398 1,415 1,428 1,434 1,444R-squared 0.27 0.27 0.27 0.27 0.27
(6) (7) (8) (9) (10)VARIABLES Later than 10 Later than 11 Later than 12 Later than 13 Later than 14
ItM2−1 -0.21*** -0.22*** -0.23*** -0.27*** -0.23**(0.06) (0.07) (0.08) (0.10) (0.11)
Observations 1,453 1,446 1,429 1,416 1,410R-squared 0.27 0.27 0.27 0.27 0.27
Year/Month/Weekday FE Yes Yes Yes Yes Yes
Panel B: Shanghai Stock Exchange Composite Index
(1) (2) (3) (4) (5)VARIABLES Earlier than 10 Earlier than 11 Earlier than 12 Earlier than 13 Earlier than 14
ItM2−1 -0.04 -0.10 -0.12** -0.15** -0.13**(0.08) (0.07) (0.06) (0.06) (0.06)
Observations 1,398 1,415 1,428 1,434 1,444R-squared 0.32 0.32 0.32 0.32 0.32
(6) (7) (8) (9) (10)VARIABLES Later than 10 Later than 11 Later than 12 Later than 13 Later than 14
ItM2−1 -0.16*** -0.17*** -0.19** -0.19** -0.16(0.06) (0.06) (0.08) (0.10) (0.11)
Observations 1,453 1,446 1,429 1,416 1,410R-squared 0.32 0.32 0.32 0.32 0.32
Year/Month/Weekday FE Yes Yes Yes Yes Yes
Notes: Sample: January, 2011 to December, 2016. The dependent variable is the size of daily uncertainty, which is measuredby the day intra-day return volatility constructed from high-frequency return data of SSE (Shanghai) and SZSE (Shenzhen)market index. Announcement dummy ItM2−i equals to one if the i-th trading day is before (or, after if i is negative) an M2announcement. We align the volatility data to the first trading day that the equity market has access to the news with thedummy variable ItM2 = 1 when i = 0. Each column summarizes estimation results based on a restricted sample that includesonly trading days of non-announcement window days and the day windows of those announcements that fall into either earlyor late group. ***Significant at 1%, **significant at 5%, *significant at 10%. Robust standard errors are shown in parentheses.
nullify the linear effect of day delays on uncertainty reduction. Across panels, our evidence
is robust regardless of whether we are using level or rate of uncertainty changes as the de-
pendent variable. However, we see that the significance of estimated linear and convexity
effect survive alternative specifications only if the month and weekday fixed effects are not
included. Given our sample size, this can be explained by the potential multicolinearity. In
specific, it is likely that the arrival days of announcements are somewhat correlated with
some particular months of a year and certain weekdays. For example, it takes longer for
50
PBOC to publish its official statistics when it comes to months with national holidays. This
generates larger standard errors associated with our estimates. Nonetheless, according to
Columns (4) and (8), the sign and magnitude of point estimates with month and weekday
fixed effects controlled are within the range of estimates suggested by other columns.
Table 13: Return Volatility Reduction Across Announcements
SZSE index SSE index
(1) (2) (3) (4) (5) (6) (7) (8)
Panel 1A: ∆Ret V ol3,q
DaytM2,q -0.02 -0.53*** -0.51** -0.26 -0.03 -0.36** -0.28* -0.12(0.02) (0.20) (0.20) (0.35) (0.02) (0.17) (0.16) (0.31)
Day2tM2,q0.02** 0.02** 0.01 0.01* 0.01* 0.00
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Panel 1B: ∆Ret V ol3,q/Ret V oltM2−3,q
DaytM2,q -0.03 -0.51** -0.53** -0.30 -0.03 -0.32 -0.33* -0.18(0.02) (0.21) (0.20) (0.28) (0.02) (0.20) (0.19) (0.27)
Day2tM2,q0.02** 0.02*** 0.01 0.01 0.01* 0.01
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Panel 2A: ∆Ret V ol5,q
DaytM2,q -0.04 -0.57** -0.48* -0.36 -0.06 -0.60* -0.40 -0.46(0.04) (0.28) (0.25) (0.56) (0.04) (0.33) (0.26) (0.47)
Day2tM2,q0.02* 0.02* 0.01 0.02* 0.01 0.01
(0.01) (0.01) (0.02) (0.01) (0.01) (0.02)
Panel 2B: ∆Ret V ol5,q/Ret V oltM2−5,q
DaytM2,q -0.01 -0.40** -0.40** -0.32 -0.03 -0.46* -0.42* -0.56(0.02) (0.20) (0.20) (0.33) (0.03) (0.25) (0.25) (0.34)
Day2tM2,q0.02** 0.02* 0.01 0.02* 0.02 0.02
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Year FE Yes Yes Yes YesMonth FE Yes YesWeekday FE Yes YesObservations 72 72 72 72 72 72 72 72
Notes: Sample: January, 2011 to December, 2016. This table reports the dummy variable re-gression results of Equation (32). The dependent variable is the size of uncertainty reduction,which is measured by the j day difference in level and in growth rate regarding the intra-dayreturn volatility constructed from high-frequency return data of SSE (Shanghai) and SZSE(Shenzhen) market index. ***Significant at 1%, **significant at 5%, *significant at 10%. Ro-bust standard errors are shown in parentheses.
Therefore, collecting evidence drawn from sub-sample and cross-event regressions, we con-
clude that more delayed announcements are associated with larger reduction of uncertainty
prior to announcement. Given that announcement timing is largely random for market,
our finding leads to a causal interpretation that belated arrival of announcements triggers
greater reduction of forecast uncertainty. In the following, we further show that the size of
pre-announcement premium is time-dependent. As monetary announcement is increasingly
postponed, it generates reduction of forecast uncertainty while increases the excess equity
returns.
51
Conditional on sum-samples of monetary announcements as grouped by their timeliness
relative to some day cutoffs of a month, Table 14 reports the estimation results regarding
the stock performance within the announcement window of 2T + 1 days per specification of
Equation (1) with T = 3. We suppress point estimates regarding the day dummy variables
and leave with coefficient estimate of ItM2−1. In the upper panel, estimations are associated
with those announcements that arrived relatively early in the month. Results from Columns
(1) to (5) all suggest that if the monetary data is released relatively early, there is no excess
equity premium for the day prior to the announcement day relative to the non-announcement
days. In the lower panel, results associated with late announcement groups are presented.
However, Columns (6) to (10) highlight the fact that for those announcements announced late
in the month, size of pre-announcement premium is significantly greater than zero. Excess
equity returns are higher on the day before announcement relative to day returns outside the
announcement window. More importantly, across all the columns, we see the estimated one-
day ahead coefficient gets monotonically larger as data releases are increasingly postponed in
a month. Hence, we have shown that pre-announcement premium is largely time-dependent.
Compared with our baseline results shown in Table 4, we see that the excess return of 40
basis points on average realized prior to the day of announcement is completely driven with
those belated announcements in particular as the stock market reactions to early arrivals of
announcement is largely muted.
Next, for robustness, we specifically estimate the magnitude of pre-announcement pre-
mium per one day of announcement delay. The null is tested based on a specification that
identifies the time-dependent nature of pre-announcement premium. We run the following
regression by focusing on the coefficient estimate associated with the interaction term of our
measure of announcement timeliness Earlyt and the dummy for day tM2 − 1:
Exrett = β0 + β1ItM2−1 + β2 · Earlyt + β3 · ItM2−1 · Earlyt + βxXt + υt (33)
where the dummy variable ItM2−1 = 1 captures the day that is one day before the announce-
ment date.Earlyt = Daycutoff−Dayt measures the distance of day of month Dayt for a day t
relative to some day of month cutoff Daycutoff . If Earlyt > 0 (< 0), day t is relatively earlier
(later). When Earlyt is interacted with ItM2−1 = 1, a negative (positive) number identifies
an delayed (earlier) announcement with some number of days arrived late (early) for that
month. For the symmetry of this interaction term, the associated coefficient β3 not only
gives the estimate of discount when the announcement is one day earlier than median, but
also the additional premium for investors waiting one more day relative to a day cutoff. We
select the median day of month for all M2 announcements in our sample, i.e. Daycutoff = 12.
52
Table 14: Pre-announcement Premium: Early vs. Late Announcements
(1) (2) (3) (4) (5)VARIABLES Earlier than 10 Earlier than 11 Earlier than 12 Earlier than 13 Earlier than 14
ItM2−1 -0.11 0.09 0.23 0.22 0.22(0.38) (0.29) (0.21) (0.19) (0.18)
Win Ctrl YES YES YES YES YESYear FE YES YES YES YES YESMonth FE YES YES YES YES YESWeekday FE YES YES YES YES YES
Observations 1,512 1,529 1,544 1,550 1,563
(6) (7) (8) (9) (10)VARIABLES Later than 10 Later than 11 Later than 12 Later than 13 Later than 14
ItM2−1 0.46** 0.53*** 0.63*** 0.69*** 0.81**(0.18) (0.19) (0.20) (0.27) (0.32)
Win Ctrl YES YES YES YES YESYear FE YES YES YES YES YESMonth FE YES YES YES YES YESWeekday FE YES YES YES YES YES
Observations 1,571 1,564 1,547 1,532 1,526
Notes: Sample: January, 2011 to June, 2017. This table reports dummy variable regression results of Equa-tion (1) for different specifications. The dependent variable is the log excess return constructed from the WindA Share Index. Announcement dummy ItM2−i equals to one if the i-th trading day is before (or, after if iis negative) an M2 announcement. We align the return data of the first trading day that the equity markethas access to the news with the dummy variable ItM2 = 1 when i = 0. Each column summarizes estimationresults based on a restricted sample that includes only trading days of non-announcement three-day windowsand three-day windows of those announcements that fall into either early or late group. ***Significant at 1%,**significant at 5%, *significant at 10%. Robust standard errors are shown in parentheses.
Note that Daycutoff is a constant and the choice of it does not affect our point estimate per
se. Null hypothesis to be tested is that the equity market does not price in the duration of
waiting for the incoming monetary announcement.
Table 15 collects the estimation results for various specifications. Column (2) shows that
the size of jumps in excess returns on day tM2 − 1 decreases in the number of days the
announcement arrives earlier than the reference day, i.e. the 12-th of a month. With one
more day of announcement delay (ItM2−1·Earlyt = −1), the extra prior-day premium brought
in is 16 bps. Across results suggested by Columns (3) to (5), the coefficient estimate of β3 is
consistently robust and statistically positive. In terms of the magnitude of time-dependent
premium, one day earlier (delayed) of M2 announcement arrival, on average, is associated
with a additional cut (increase) of approximately 15 bps on tM2 − 1. For example, a back-
of-envelope calculation says there is no identifiable announcement premium associated with
monetary data releases that fall on the 9-th day of a month (that is, three days earlier than
our baseline cutoff 12-th), which is consistent with our findings in Table 14. To account for
the muted premium associated with early announcement arrivals, Proposition 2 highlights
the fact that the amount of forecast uncertainty is directly measurable by the time lapsed
53
into the next announcement. If the data release arrives too early in a month, investors are
yet to accumulate the amount uncertainty that can trigger attentive learning. Therefore,
equity prices are little affected.
Table 15: Early Termination of Waiting and Monetary Announcement Premium
(1) (2) (3) (4) (5)VARIABLES All Anns All Anns All Anns All Anns Excl. Feb
ItM2−1 0.40** 0.34** 0.37** 0.37** 0.34*(0.17) (0.16) (0.17) (0.17) (0.18)
ItM2−1 · Earlyt -0.16** -0.17** -0.16** -0.18**(0.08) (0.08) (0.08) (0.09)
Earlyt 0.00 0.00 0.01 0.01(0.02) (0.02) (0.03) (0.04)
Win dum Ctrls YES YES YESWeekday FE YES YES YES YES YESMonth FE YES YESYear FE YES YES
Observations 1,577 1,577 1,577 1,577 1,461R-squared (%) 0.70 0.89 1.61 2.00 1.89
Notes: Sample: January, 2011 to June, 2017. This table reports the dummyvariable regression results of Equation (33). “All Anns sample” columns summa-rize the results considering all M2 announcements in our sample; “Excl. Feb”present results estimated from a sample that excludes February M2 news. Col-umn (1) is a copy of results from Column (1) in Table 6 as our reference. Thedependent variable is the log close-to-close excess return constructed from WindA Share Index. ”Win dum Ctrls”: whether or not controls for other announce-ment day dummy variables ItM2−i up to a window length of 2T + 1 with T = 3.”Weekday FE”: the weekday fixed effects controls. See text for the definitions ofvariables. ***Significant at 1%, **significant at 5%, *significant at 10%. Robuststandard errors are shown in parentheses.
Important to note that we have shown the size of pre-announcement uncertainty reduction
and magnitude of equity premium both depend on the timeliness of monetary announcement
arrival. Given the exogenous variation in announcement timing, our empirical results thus
offer a causal interpretation to account for the pre-announcement premium, which is con-
sistent with our model implications. While exogenous timing should not directly link to
size of equity premium, prolonged waiting for announcement accumulates the right amount
of uncertainty that surpasses the attention threshold, which triggers the attentive learning.
Consequently, this lowered uncertainty leads to pre-announcement drift of equity returns.
For the completeness of having theory-consistent empirics, we further present suggestive
evidence on attentive learning before announcement given it is quite challenging to directly
measure the quantity of information flow in the data. As an approximation, we resort to an
indirect measure of investors’ attention allocated to PBOC’s announcement, i.e. web visit
traffic data to PBOC’s main website.24 We focus on traffic visits index constructed based
24We obtained web traffic data from from Alexa Internet, Inc., a company that monitors and estimateswebsite traffic data worldwide. We obtained three years of web traffic data about visits to PBOC’s mainsite: http://www.pbc.gov.cn.
54
on the web click visits initiated from mainland China. Reasonably, clicks to central bank’s
website should be considered as information acquisition initiative done by individuals who
have particular interests in knowing about PBOC in general and about its policy moves and
announcement, for example investors in the financial market among with other profession-
als.25 Focusing on a window of 2T + 1 days of T = 3 with the announcement day in the
center, we examine the mean web traffic intensity within the announcement window relative
to non-announcement daily average.
Figure 8 plots the relative web traffic changes in the window. It shows that visits to
PBOC’s website has increased before announcement until reaches its peak on the day of
announcement. The pre-announcement climbs in website visits can at least partially indicate
that attention among market professionals does pick up. In addition, the peak of clicks on tM2
reflects that professionals do look at PBOC’s websites looking for announced information,
which again suggests that our indirect measure based on website traffic data is informative
about learning intensity. Finally, we see that post-announcement traffic slides down and
returns to the non-announcement mean.
Figure 8: Relative PBOC Web Traffics in M2 Announcement Windows
Rel
ativ
e A
ttent
ion
Notes: Sample: October, 2016 to October, 2018. Three years of daily website visit traffic data are managed by andobtained from Alexa Internet, Inc. This figure plots the ratio of mean daily web visits of PBOC main webpage within andoutside the announcement windows of monetary aggregate data releases. tM2 marks the arrival day of M2 announcement asdenoted by the red dashed vertical line.
Hence, we show in this section that increased attention can be observed in the pre-
announcement window along with uncertainty reduction and pre-drift of equity prices. As
pre-announcement premium is mostly driven by more delayed monetary announcement, all
these evidence are strikingly consistent with our model predictions regarding the mechanism
25Professional financial market participants, scholars and government officials are most likely to makethese inquiry efforts.
55
of pre-announcement premium. Next we explore evidence in the cross-sectional dimension
to further lend credence to our model.
6 Cross-sectional Heterogeneities
In this section, we examine the potential heterogeneities in stock reactions to monetary
announcement. By sorting stocks into portfolios by the market value and book-to-market
ratio, we show that the magnitude of premium and the sensitivity to timeliness of an-
nouncement arrival vary across portfolios. Such heterogeneities in monetary announcement
premium distinguishes China’s equity market from the U.S. market. Importantly, we empha-
size that the cross-sectional heterogeneities in China are not driven by the firm ownership
differences, i.e. State-Owned-Enterprises (SOE) vs. Non-SOEs, a critical dimension that
often times noted in studies of Chinese economy and financial market (Fan et al., 2007; Song
et al., 2011).
We first sort the A-share stocks into five portfolios by the market value of a firm’s total
equity, i.e. Size Portfolios. Then we run the estimations according to specifications of
Equations (1) and (33) for each portfolio. Our null hypothesis is that the average size of pre-
announcement premium and the time-dependent effect should not differ across portfolios
of sizes. Panels A and B of Table 16 summarize the key estimation results. In Panel C
and D, we present additional evidence by break stocks into SOE and non-SOE groups with
each group sorted into five size portfolios. With estimates of coefficients associated with
other dummy indicators and control variables suppressed, we focus on the point estimates
of coefficients for ItM2−1 and its interaction term with Earlyt.
Panel A shows that portfolios of small and medium-size stocks exhibit positive and large
pre-announcement premium in response to monetary news. Jumps in excess returns can be as
high as about 70 bps per day across lowest four size portfolios. However, portfolio consisted of
big-cap stocks displays a coefficient on day tM2−1 that is not statistically different from zero.
Therefore, we conclude that big stocks have little reactions prior to monetary announcement
in China. On average, as shown in Table 4, for the whole market portfolio, jumps in excess
return of 40 bps on day tM2 − 1 is mainly driven by pre-announcement reactions of small
and medium capped stocks. Interestingly, Panel B also suggests that the portfolios of small
and medium-sized stocks are the ones that exhibit large sensitivity to the timely arrivals
of M2 announcements. In terms of the magnitude associated with time-dependent effect,
little difference can be discerned across the smallest four size portfolios. Nonetheless, large
stocks display insignificant sensitivities to the early or delayed announcement. Overall, not
only results of Panels A and B are consistent with each but conform well to our model-
56
based causal interpretation of China’s pre-announcement premium. Precisely, it is investors
holding portfolios of the small and medium capped stocks that exhibit increasing attention
to the belated monetary announcement that triggers uncertainty reduction and generates
the pre-announcement premium. By contrast, valuations of big stocks are found free of this
type of pre-announcement reactions.
Panel C and Panel D explore if the premium heterogeneities in size portfolios may be
explained by the critical differences of firm ownership. Results in Panel C confirm that small
and medium stocks exhibit the sizable pre-announcement premium regardless of whether
the firm is SOE or non-SOE. Looking into results suggested in Panel D, both quantitatively
and qualitatively, SOE and non-SOE differences do not alter our finding that small and
medium-cap stocks display sensitivities to announcement timing. However, it turns out that
the largest portfolio of non-SOE stocks rather than the SOE counterpart appears to have
moderate pre-announcement responses and delivers the timing-dependent premium. This
finding may be explained by the fact that the market value of non-SOE firm stocks even
grouped into its largest size portfolio is still relatively small compared to listed SOE firms
which are on average large caps. Hence, it is the size of equity value that drives this subtle
difference and generates heterogeneities in pre-announcement premium across portfolios.
Secondly, we regroup A-share stocks into five portfolios according to the firm’s book-to-
market ratio, i.e. BM Portfolios. Similarly as we do for size portfolios, we examine the size
of pre-announcement premium and portfolio’s sensitivity to announcement timing with and
without a further breakdown by firm’s ownership status. Regression results are summarized
in Table 17 of different panels. Results in Panels A and B suggest that only the portfolio
consisted of firms with smallest Book-to-Market ratio, the growth stocks exhibit the sizable
pre-announcement premium and the statistically significant sensitivity to timeliness of re-
leases of monetary aggregates data. Portfolios of mid and large BM ratio stocks, however,
are found with limited reactions to monetary announcement or marginally significant sen-
sitivity to announcement timing. According to Panels C and D, our results about growth
firms apply to the SOE firm sample. Pre-announcement premium can be obtained if invest-
ing in the portfolio of lowest BM ratio firms. On the contrary, We see that non-SOE firm
portfolios that cover medium and even largest BM ratio firms yield greater jumps in excess
return prior to announcement and the portfolio pre-announcement premium gets larger if
the announcement is increasingly delayed. Given that both dimensions of size and BM ratio
may shift the heterogeneities and results associated with different BM portfolios are sensitive
to firm ownership differences, we further do a twoway sort of stocks to clear up the findings.
The twoway sorting of stocks by size and BM ratios gives us nine portfolios. Relevant
regression coefficients regarding the day tM2 − 1 excess returns are similarly organized in
57
Table 16: Size Portfolio Regressions
Portfolio Small 2 3 4 Big
Avg. Size 2665.8 4134.8 6071.8 9977.6 57132.7
Panel A: Baseline
ItM2−1 0.78*** 0.72*** 0.73*** 0.67*** 0.22(0.21) (0.22) (0.22) (0.22) (0.14)
Panel B: Regression of Interaction Term
ItM2−1 0.70*** 0.65*** 0.66*** 0.60*** 0.19(0.20) (0.21) (0.21) (0.21) (0.14)
ItM2−1 · EarlytM2 -0.21** -0.20** -0.19* -0.19** -0.08(0.09) (0.09) (0.10) (0.09) (0.06)
Average Sizes: SOE vs. non-SOE
SOE sample: 2840.1 4056.5 5688.9 9669.3 68490.0non-SOE sample: 2587.5 4180.2 6313.2 10174.6 42278.3
Panel C: SOE vs. non-SOE: Baseline
SOE sample:ItM2−1 0.79*** 0.74*** 0.69*** 0.64*** 0.18
(0.21) (0.22) (0.22) (0.22) (0.14)
non-SOE sample:ItM2−1 0.77*** 0.71*** 0.75*** 0.69*** 0.30*
(0.22) (0.22) (0.22) (0.22) (0.15)
Panel D: SOE vs. non-SOE: Regression of Interaction Term
SOE sample:
ItM2−1 0.73*** 0.67*** 0.63*** 0.57*** 0.16(0.20) (0.21) (0.21) (0.21) (0.14)
ItM2−1 · EarlytM2 -0.17* -0.18** -0.16* -0.18* -0.06(0.09) (0.09) (0.09) (0.10) (0.06)
non-SOE sample:
ItM2−1 0.68*** 0.63*** 0.67*** 0.62*** 0.25*(0.21) (0.21) (0.21) (0.21) (0.15)
ItM2−1 · EarlytM2 -0.22** -0.21** -0.21** -0.21** -0.15**(0.09) (0.10) (0.10) (0.09) (0.07)
Win Ctrl YES YES YES YES YESYear/Month/Weekday FE YES YES YES YES YESObservations 1,577 1,577 1,577 1,577 1,577
Notes: Sample from 2011 to June 2017. The dependent variables are value weightedaverage excess return of each size portfolio. Size 0 is for the portfolio of stocks withthe smallest market value, Size 5 is for the largest. The breakpoint is derived fromthe market value quintiles of the whole market. The portfolios are monthly re-balanced. ***Significant at 1%, **significant at 5%, *significant at 10%. Robuststandard errors are shown in parentheses.
58
Table 17: Book-to-Market Ratio Portfolio Regressions
Portfolio Low 2 3 4 High
Avg. Adjusted BM -1.86 -0.96 -0.43 0.09 0.95
Panel A: Baseline
ItM2−1 0.66*** 0.32** 0.36*** 0.29* 0.29*(0.20) (0.15) (0.14) (0.16) (0.16)
Panel B: Regression of Interaction Term
ItM2−1 0.59*** 0.28* 0.33** 0.25 0.24(0.19) (0.14) (0.14) (0.16) (0.16)
ItM2−1 · EarlytM2 -0.18** -0.12* -0.08 -0.12* -0.11*(0.09) (0.06) (0.06) (0.07) (0.07)
Average Adj. BM Ratios: SOE vs. non-SOE
SOE sample: -1.9 -1.0 -0.4 0.1 0.9non-SOE sample: -1.8 -0.9 -0.4 0.1 1.0
Panel C: SOE vs. non-SOE: Baseline
SOE sample:ItM2−1 0.66*** 0.15 0.25* 0.23 0.32*
(0.21) (0.13) (0.13) (0.16) (0.17)
non-SOE sample:ItM2−1 0.67*** 0.62*** 0.61*** 0.44** 0.22
(0.20) (0.21) (0.20) (0.18) (0.15)
Panel D: SOE vs. non-SOE: Regression of Interaction Term
SOE sample:
ItM2−1 0.60*** 0.13 0.24* 0.20 0.28(0.21) (0.13) (0.13) (0.16) (0.17)
ItM2−1 · EarlytM2 -0.16* -0.08 -0.04 -0.08 -0.11(0.09) (0.06) (0.06) (0.07) (0.07)
non-SOE sample:
ItM2−1 0.60*** 0.55*** 0.54*** 0.37** 0.17(0.19) (0.20) (0.19) (0.18) (0.15)
ItM2−1 · EarlytM2 -0.19** -0.19** -0.17** -0.20** -0.13**(0.09) (0.09) (0.08) (0.08) (0.07)
Win Ctrl YES YES YES YES YESYear/Month/Weekday FE YES YES YES YES YESObservations 1,577 1,577 1,577 1,577 1,577
Notes: Sample from 2011 to June 2017. The dependent variables are valueweighted average excess return of each size portfolio. BM 1 is for the portfolio ofstocks with the smallest book-to-market ratio, BM 5 is for the largest. The break-point is derived from the market value quintiles of the whole market. The portfo-lios are yearly re-balanced. ***Significant at 1%, **significant at 5%, *significantat 10%. Robust standard errors are shown in parentheses.
59
Table 18. Key findings can be summarized in the following. First, regardless of firm own-
ership, market value of equity is the dominant dimension by which the pre-announcement
excess returns of a portfolio can be affected. Focusing on Panels A and B, we see that
within portfolios of small and medium cap stocks, differences in BM ratio would not create
extra heterogeneity in terms of the size and the sensitivity of pre-announcement premium
to announcement timing. Second, BM ratio is relevant for determining pre-announcement
premium only when portfolios consisted of large cap stocks. Panels A and B find that Port-
folios of largest cap stocks can generate additional equity premium prior to announcement
only if growth firms of smallest BM ratios are considered. Third, largest stocks of non-SOE
firms rather than SOE firms react to monetary announcement because non-SOE firms are
relatively smaller. As suggested by Columns of (7) to (9) of Panels C and D, given the same
corresponding BM ratios across subgroup portfolios of large cap stocks, portfolios of largest
stocks of SOE firms exhibit little pre-announcement premium and sensitivity to announce-
ment timing delayed announcements because their market value is much higher than that of
non-SOE firms of largest cap stocks.
In summary, we obtained results regarding the cross-sectional heterogeneities in stock
return reactions to M2 announcements at the portfolio level. Small and medium-cap stocks
in China are particular responsive to data releases of monetary aggregates data with respect
to size of pre-announcement and their sensitivities to timeliness of announcement arrival.
Portfolio consisted of large stocks generate similar patterns of pre-announcement premium
only if growth firms are included into the portfolio. Differences in firm ownership would not
alter these results. Recall that in our model, different asset allocations and consumption
paths up to day t are associated with different marginal gain λt of optimizing over attention
for minimizing the utility loss. In reality, investors with different asset exposures thus face
varied degrees of marginal benefit from attentive learning. Investors in portfolios of small
and medium cap stocks and large stocks of growth firms could be particularly sensitive to
the loss due to limited attention paid to learn about money growth rate. Therefore, large λt
necessitates timely attentive learning which reduces forecast uncertainty and delivers extra
returns to investors of these portfolios prior to announcement.
We offer a range of explanations for why these portfolios may be particularly sensitive
to not better knowing about monetary data and thus have higher λt . First, it is possible
that the easiness of credit in China as measured by money growth reflects the size of market
liquidity risk that affects trading of stocks of small and medium cap. Second, credit mis-
allocation is a key concern for smaller firms especially when non-SOE firms are relatively
smaller. However, larger firms and SOE firms in general have better access to formal cred-
60
Table 18: Size and BM Twoway Sorted Regressions
(1) (2) (3) (4) (5) (6) (7) (8) (9)Portfolio (S, L) (S, M) (S, H) (M, L) (M, M) (M, H) (B, L) (B, M) (B, H)
Avg. size 3057.1 3143.5 3242.9 6202.9 6154.8 6228.8 30893.6 40677.4 44285.2Avg. adusted BM -1.7 -0.6 0.5 -1.5 -0.4 0.6 -1.4 -0.3 0.8
Panel A: Baseline in Size & BM Portfolio
ItM2−1 0.77*** 0.76*** 0.76*** 0.78*** 0.73*** 0.64*** 0.35** 0.24* 0.23(0.21) (0.22) (0.22) (0.22) (0.22) (0.22) (0.15) (0.13) (0.16)
Panel B: Regression of Interaction Term
ItM2−1 0.70*** 0.68*** 0.69*** 0.70*** 0.66*** 0.58*** 0.30** 0.22 0.19(0.20) (0.21) (0.21) (0.22) (0.21) (0.21) (0.15) (0.14) (0.16)
ItM2−1 · EarlytM2 -0.19** -0.22** -0.20** -0.21** -0.20** -0.17* -0.14** -0.06 -0.11(0.09) (0.09) (0.09) (0.10) (0.09) (0.09) (0.07) (0.06) (0.07)
Size and BM Ratio: SOE vs. non-SOE
SOE sample:Avg. size 3157.2 3159.0 3385.9 5707.2 5944.7 5923.0 54027.7 54345.0 44296.9Avg. adusted BM -1.7 -0.5 0.6 -1.5 -0.4 0.7 -1.3 -0.2 0.8
non-SOE sample:Avg. size 3028.3 3137.0 3117.9 6398.1 6270.2 6548.5 19582.3 25359.2 44292.6Avg. adusted BM -1.7 -0.6 0.4 -1.5 -0.5 0.5 -1.5 -0.3 0.7
Panel C: SOE vs. non-SOE: Baseline
SOE sample:ItM2−1 0.81*** 0.78*** 0.78*** 0.74*** 0.70*** 0.62*** 0.20 0.17 0.27
(0.21) (0.22) (0.21) (0.22) (0.22) (0.21) (0.14) (0.13) (0.17)
non-SOE sample:ItM2−1 0.76*** 0.75*** 0.73*** 0.79*** 0.75*** 0.66*** 0.58*** 0.49*** 0.17
(0.22) (0.22) (0.22) (0.23) (0.22) (0.22) (0.19) (0.18) (0.15)
Panel D: SOE vs. non-SOE: Regression of Interaction Term
SOE sample:
ItM2−1 0.76*** 0.71*** 0.72*** 0.68*** 0.63*** 0.56*** 0.16 0.16 0.24(0.21) (0.21) (0.21) (0.22) (0.22) (0.21) (0.14) (0.13) (0.17)
ItM2−1 · EarlytM2 -0.14 -0.18** -0.18** -0.18* -0.19** -0.15* -0.09 -0.03 -0.10(0.09) (0.09) (0.09) (0.09) (0.09) (0.09) (0.06) (0.06) (0.07)
non-SOE sample:
ItM2−1 0.68*** 0.66*** 0.66*** 0.71*** 0.68*** 0.59*** 0.51*** 0.43** 0.12(0.21) (0.21) (0.21) (0.22) (0.21) (0.21) (0.18) (0.18) (0.15)
ItM2−1 · EarlytM2 -0.20** -0.24** -0.21** -0.22** -0.20** -0.19** -0.18** -0.17** -0.13**(0.09) (0.10) (0.10) (0.10) (0.10) (0.10) (0.08) (0.08) (0.06)
Win Ctrl YES YES YES YES YES YES YES YES YESYear/Month/Weekday FE YES YES YES YES YES YES YES YES YESObservations 1,577 1,577 1,577 1,577 1,577 1,577 1,577 1,577 1,577
Notes: Sample from 2011 to June 2017. The dependent variables are value weighted average excess return of each size-BMportfolio. (S, L) is for the portfolio of stocks with the smallest market value and lowest book-to-market value. The breakpointis derived from the market value quintiles of the whole market. The portfolios are yearly re-balanced. ***Significant at 1%,**significant at 5%, *significant at 10%. Robust standard errors are shown in parentheses.
its.26 Hence, increased money growth and credit expansion means that smaller and non-SOE
firms now face a larger pool of credit and loan. Also, it is very likely that small firms and
growth firms may be financially constrained. They are thus very responsive to the cost of
26See Song et al. (2011), Chen et al. (2016) for example.
61
borrowing changes that are indirectly linked to money growth. In terms of both credit avail-
ability and the cost of credit, portfolios of these stocks could have exposed themselves with
extra sensitivity to the numbers behind the monetary announcements.
7 Additional Results
In this section, we present additional results regarding China’s stock market reactions to
the U.S. FOMC announcements along with the responses of other Chinese asset markets to
domestic monetary announcements.
7.1 Return Responses: FOMC News
According to Column (11) of Table 5, we see that China’s equity market reactions to the
U.S. FOMC statement releases are completely muted. That is, for all lag and lead terms, no
excess return that is statistically different from zero can be realized. This finding contrasts
with the evidence documented in Lucca and Moench (2015) that the stock markets of a
number of advanced economies exhibit positive pre-drifts anticipating the incoming FOMC
announcements.
However, we note the sample difference of our paper (years of 2011-2017) against a pre-
2011 period in Lucca and Moench (2015). Therefore, our baseline sample features a period
when the U.S. Federal Funds rate, the key monetary policy instrument, is constantly fixed
near a zero lower bound for years since the end of 2008 until the end of 2015. Therefore,
it might be the reason that the U.S. FRB well managed the expectation of domestic and
international U.S. market investors by minimizing the limited risk of U.S. monetary policy
(Yellen, 2015). Hence, assuming this story is true, the implication is that for a sample with
more volatile interest rate changes like pre-2011, we should see China’s market reacts to
FOMC news. In Table 19, we show results based on estimations of Equation (1) using an
alternative sample of 2002-2010. However, according to Columns (3) and (6), the absence of
responsiveness of Chinese equity market to the FOMC announcements still holds. Hence, a
constant close-to-zero U.S. Federal Funds Rate indicating limited U.S. monetary policy risk
does not help explain the muted reaction of Chinese equity market to FMOC announcements.
We thus conclude that China’s equity market does not price in the risk of the incoming
U.S. monetary policy as delivered through FOMC announcements. It’s possible that despite
China has undergone a series of financial reforms, its integration to the global financial mar-
kets is yet to complete. More developed markets may consider changes in the U.S. interest
rates as an important risk that could potentially affect their domestic asset prices, capital
62
flows, exchange rates, international trade dynamics, and its real-sector growth through vari-
ous financial and trade linkages. China, due to the limited participation of Chinese investors
into foreign capital markets, actively managed its exchange rate and capital accounts, all
could be isolating China from the impacts of U.S. monetary policy risk.
Table 19: China: Equity Market Responses and FOMC Announcements
(1) (2) (3) (4) (5) (6)VARIABLES M2 FOMC FOMC M2 FOMC FOMC
2011-2017 2011-2017 2002-2010 2011-2017 2011-2017 2002-2010
ItAnns−1 0.44** 0.16 -0.24(0.17) (0.21) (0.25)
ItAnns−1,3 0.33*** -0.20 -0.06(0.11) (0.16) (0.15)
ItAnns−3 0.33 -0.42 -0.30(0.21) (0.32) (0.30)
ItAnns−2 0.23 -0.31 0.31(0.18) (0.29) (0.22)
ItAnns 0.16 -0.20 -0.16(0.18) (0.23) (0.24)
ItAnns+1 -0.16 0.06 -0.30(0.17) (0.25) (0.21)
ItAnns+2 0.04 -0.05 0.17(0.21) (0.26) (0.29)
ItAnns+3 0.02 0.20 -0.31(0.19) (0.19) (0.22)
Weekday FE YES YES YES YES YES YESConstant 0.07 0.13 -0.05 0.07 0.14 -0.07
(0.10) (0.11) (0.09) (0.10) (0.10) (0.09)
Observations 1,577 1,577 2,237 1,577 1,577 2,237R2(%) 1.03 0.84 0.93 0.91 0.55 0.52
Notes: Sample: January, 2011 to June, 2017. The dependent variable is the log close-to-closeexcess return constructed from Wind A Share Index. Announcement dummy ItAnns−1 equalsto one if the i-th trading day before (or after if i is negative) a particular type of announcement.We align the return data of the first trading day that the equity market has access to the newswith the dummy variable ItAnns = 1 when i = 0. ***Significant at 1%, **significant at 5%,*significant at 10%. Robust standard errors are shown in parentheses.
7.2 Return Responses: Other Markets
We further examine returns of other asset class in China in response to M2 announce-
ments. We examine the return responsiveness of the total return of 10 year government bond,
Chinese A share futures of 300 big stocks, gold future, along with exchange rates of Chinese
RMB against major currencies including US dollar, Euro and Japanese Yen. Apart from the
ex-post market reactions, we find no pre-announcement drift of returns across these asset
markets. Hence we see that the monetary announcement premium is uniquely associated
with equity market only in China.
63
Table 20: Other Markets’s Repsonse on M2 Announcements
(1) (2) (3) (4) (5) (6)VARIABLES R10Y,bond FurtureCSI300 FurtureGold EXUSD EXJPY EXEUR
ItM2−3 0.20 0.29 -0.01 -0.02 -0.12 0.02(0.29) (0.20) (0.11) (0.01) (0.10) (0.07)
ItM2−2 -0.20 0.20 0.06 -0.01 0.06 -0.05(0.39) (0.18) (0.10) (0.02) (0.07) (0.07)
ItM2−1 0.07 0.08 0.06 0.01 0.04 -0.01(0.32) (0.15) (0.10) (0.02) (0.07) (0.07)
ItM2 0.33 -0.05 -0.25** 0.02 -0.01 -0.09(0.32) (0.20) (0.12) (0.03) (0.07) (0.07)
ItM2+1 0.13 -0.22 0.03 0.02 0.04 0.06(0.32) (0.14) (0.10) (0.02) (0.06) (0.06)
ItM2+2 0.13 -0.05 -0.05 0.00 0.05 0.12**(0.33) (0.23) (0.11) (0.02) (0.06) (0.06)
ItM2+3 0.36 0.02 -0.17 -0.00 0.07 -0.06(0.31) (0.16) (0.14) (0.02) (0.07) (0.06)
Year FE YES YES YES YES YES YESMonth FE YES YES YES YES YES YESWeekday FE YES YES YES YES YES YESConstant 0.09 -0.30 0.18 -0.03 0.10 -0.01
(0.31) (0.20) (0.11) (0.02) (0.07) (0.07)
Observations 1,622 1,577 1,577 1,579 1,577 1,578R-squared (%) 2.67 1.50 2.95 2.75 2.58 1.62
Notes: Sample: January, 2011 to June, 2017. This table reports dummy variable regression resultsof Equation (1) for different specifications. The dependent variable is the log close-to-close excessreturn constructed from Wind A Share Index. ***Significant at 1%, **significant at 5%, *significantat 10%. Robust standard errors are shown in parentheses.
8 Conclusion
This paper documents a “pre-announcement drift” of Chinese equity market in response
to its central bank’s announcements of measures of monetary aggregates. For a period of
2011 to 2017, on average, Chinese A-share market climbs and realizes an excess return of 40
basis points per day in the three-day window prior to the day of announcement as followed
by a flattening-out after the announcement. This pre-announcement premium generates over
10 % annual excess return, which doubles the size of total equity premium in China.
We propose a theory for investors allocating limited attention to money growth rate and
in equilibrium the consumption dynamics. Pre-announcement premium is driven by the en-
dogenous information decision such that investors’ attentive learning keeps reducing their
forecast uncertainty prior to monetary announcement. By exploiting the quasi-scheduling na-
ture of Chinese monetary news, we show the premium is larger when the release of monetary
data is delayed. As the announcement timing provides exogenous variations, we provide
the exact test of Ai and Bansal (2018)’s theoretical account of the causal link of uncer-
tainty reduction and pre-announcement premium. At the cross-sectional, as differed from
64
the U.S. market, the pre-announcement premium and the associated sensitivity of returns
to the announcement date can be observed among small and medium-cap Chinese stocks.
Interestingly, the ownership differences such as SOE vs. non-SOE do not alter our main
cross-sectional results. It implies that smaller firms in China could be affected more by the
monetary risks.
We also find that the announcement premium in China is associated with the monetary
news only, which also exclusively applies to its equity market. Besides, China’s equity
market is largely immune from the risk of U.S. monetary policy changes when anticipating
FOMC announcements, whereas a range of advanced economies have exhibited the sensitive
responsiveness.
65
Appendix
A Other Summaries of News Timing
Table A4 reports the number of a particular announcement that falls on the same day
when other data releases are also announced. Out of the 78 M2 announcements, 13 pieces
shared the same day with FAI and VAI announcements, and 11 were co-released with CPI
and PPI. Since 2009, CPI and PPI data are released at the same time and PPI sometimes
preceded the CPI announcement by one day before 2009.
For comparisons, we also summarize the time distribution of an extended sample by in-
cluding macro announcements starting from January, 2000. Table A5 shows that by looking
at M2 announcements of a longer sample, we can infer that PBOC used to issue monetary
related data mostly during weekdays and even within trading hours before 2011. Further-
more, in early 2000s, PBOC tended to publish MPR reports before rather than after the
trading hours.
B Equity Market Responses to MPR Announcements
We examine if the stock market also reacts to other types of monetary-related news
published by PBOC, for example, the Monetary Policy Report (MPR) announcements.
Table 21 summarizes the key findings.
Columns (2) presents estimated coefficients using different sets of monetary announce-
ments looking for evidence of one-day ahead premium. We see that for the MPR announce-
ments, the pre-announcement premium on day tM2 − 1 is not distinguishable from zero.
Column (4) takes our estimation results of a three-day daily premium due to M2 news from
Table 6 as reference. Evidence again shows that the equity market does not respond with
positive premium to incoming MPR announcements. Results of Columns (3) and (6) show
the estimates of the pre-announcement premium by considering all dates of M2 and MPR
announcements. We see the estimated size of pre-announcement premium considering both
types of monetary announcements is of the similar magnitude of the premium associated
with M2 announcements only.
Overall, absence of pre-announcement premium for MPR announcements may be related
to the fact that the MPR report, though covering statistics about the conducts of Chinese
monetary policy, delivers much more complex information beyond the generic data release.
In addition, monetary aggregates data are published with higher frequency, which could be
more useful for investors to draw real-time asset pricing implications than an encyclopedia
66
style monetary policy report of quarterly issues.
Table 21: Wind A Share Index Returns in Windows Prior to M2 and MPR Announce-ments
(1) (2) (3) (4) (5) (6)VARIABLES M2 Ann. MPR Ann. M2 and MPR Ann. M2 Ann. MPR Ann. M2 and MPR Ann.
ItAnns−1 0.43** 0.30 0.40**(0.17) (0.30) (0.16)
ItAnns−1,3 0.33*** 0.27+ 0.33***(0.11) (0.16) (0.10)
Year / Month / Weekday FE YES YES YES YES YES YESConstant -0.28 -0.26 -0.29 -0.28 -0.25 -0.30
(0.10) (0.10) (0.10) (0.10) (0.09) (0.10)
Observations 1,577 1,577 1,577 1,577 1,577 1,577R2(%) 1.82 1.31 1.94 1.70 1.35 1.81
Notes: Sample: January, 2011 to June, 2017. This table reports dummy variable regression results of Equation (2). The dependent vari-able is the log close-to-close excess return constructed from Wind A Share Index. Announcement dummy ItM2−1,3 equals to one if dayt belongs to a three-trading-day window before an M2 or MPR Announcement. ***Significant at 1%, **significant at 5%, *significant at10%, + significant at 15%. Robust standard errors are shown in parentheses.
C Proofs
C.1 Proof of Equations (7)
For ease of notations, we first define quantities of mct and mVt+1 in the following:
mct =∂Vt∂ct
= (1− β)V ξt c−ξt
mVt+1 =∂Vt∂Vt+1
= βV ξt (EV 1−α
t+1 )α−ξ1−αV −αt+1
By the fact that V (ct, R(Vt+1)) is Homogeneous of degree one in ct and R(Vt+1), we can show
that the following equation hold.
Vt = mct · ct + E[mVt+1 · Vt+1]
Define Wt = Vtmct
, it follows that
Wt = ct + E[mVt+1 ·mct+1
mct· Wt+1]
Rearranging, we have
1 = E[mVt+1 ·mct+1
mct· Wt+1
Wt − ct]
67
In the following, we show the stochastic discount factor can be defined as Ωt|t+1 = mVt+1·mct+1
mct
and RW,t+1 = Wt+1
Wt−ctsuch that Wt = Wt at optimum. Maximizing equation (4) subject to
Equation (4), it yields that the first order condition regarding optimal wealth level of t + 1
is then for every state in the future
mVt+1 ·∂Vt+1
∂Wt+1
=mctRW,t+1
with marginal value of wealth at t given by ∂Vt∂Wt
= mct according to the Envelope Theorem.
It hence gives
1 = E[mVt+1 ·mct+1
mct·RW,t+1]
and Wt = Wt = Vtmct
at optimum. The stochastic discount factor is thus
Ωt|t+1 =βV ξ
t (EV 1−αt+1 )
α−ξ1−αV −αt+1(1− β)V ξ
t+1c−ξt+1
(1− β)V ξt c−ξt
=β[ct+1
ct]−ξ[
Vt+1
[EV 1−αt+1 ]
11−α
]ξ−α
The equilibrium investment return of the portfolio RW,t+1 can be expressed as
RW,t+1 =Wt+1
Wt − ct
=Vt+1/[(1− β)V ξ
t+1c−ξt+j]
Vt/[(1− β)V ξt c−ξt ]− ct
= β[ct+1
ct]−ξ[
Vt+1
[EV 1−αt+1 ]
11−α
]ξ−1−1
Now substitute out the ratio of future value relative to the certainty equivalence Vt+1
[EV 1−αt+1 ]
11−α
using RW,t+1, we have the following equations hold for all future states
Ωt|t+1 = β[ct+1
ct]−ξ[R−1
W,t+1/(β[ct+1
ct]−ξ)]
ξ−αξ−1
= βθ[ct+1
ct]−ξθRθ−1
W,t+1
where θ = 1−α1−ξ .
Hence, maximizing the utility per Equation (4) subject to the constraint Equation (5)
68
by exploiting the Envelope Theorem again. We can have
1 = E[mVt+1 ·mct+1
mct·Rt+1]
1 = E[mVt+1 ·mct+1
mct·Rf ]
C.2 Proof of Equation (8)
By Equation (7), we exploit the assumption that price-dividend ratio is constant χ.
1 = E[(β(1 + χ)
χ)θ · e(1−α)mt+1 ]
Given that the AR(1) process governing the real money growth has the lognormal structure,
we have the following
log[1 + χ
χ] = −[log β + (1− ξ)mt+1 +
(1− ξ)(1− α)
2V ar(mt+1)]
Note that the hat notations denote the conditional expectation and variance given the in-
formation set at time t about tomorrow’s money growth. Similarly, we can rewrite Equation
(??) such that
1 = E(βθe−ξθmt+1(1− χχ
emt+1)θ−1 ·Rf )
It yields the identity about the risk-free return given by
log(Rf ) = − log β + ξmt+1 +ξ − α− αξ
2V ar(mt+1)
Per the fact that Rt+1 = 1+χχ
ct+1
ct, the expected equity return follows
ERt+1 = E((1 + χ)
χemt+1)
It solves that
log(ERt+1) = − log β + ξmt+1 +ξ + α− αξ
2V ar(mt+1)
Hence, EXt+1 = log(ERt+1)− log(Rf ) follows.
69
C.3 Derivation of the Loss Function of Investors
Given that the objective function of investors’ value maximization problem Vt(ct, zt)
is homogeneous of degree one in arguments ct and zt. Conditional on realized previous
consumption of ct−1 > 0, it yields that
Vt(ct, zt) = ct−1Vt(ctct−1
,ztct−1
) = ct−1Vt(ect , ezt)
where ct = log[ ctct−1
] and zt = log[ ztct−1
]. Up to a second-order linear approximation of
ct, ztaround the couplets point (µ, 0) with µ governs the unconditional log growth rate
of real money balance, we have
Vt(ect , ezt) = Vt(e
µ, 1) + Vt,1eµ(ct − µ) + Vt,2zt +
Vt,11
2eµ(ct − µ)2 +
Vt,22
2z2t + Vt,12e
µ(ct − µ)zt
Note. Vt,1 and Vt,2 are partial derivatives of Vt with respect to the first and second arguments
evaluated at the centering couplets of (µ, 0). Vt,11, Vt,22, and Vt,12 are the evaluated second
order partials and cross-partials. Optimization over choice consumption ct then gives a first
order condition such that
Vt,1 + Vt,11(ct − µ) + Vt,12zt = 0
It implies the following identity holds at optimum:
ct = a+ bzt
where a = µ − Vt,1Vt,11
> 0, b = −Vt,12Vt,11
> 0. Now we linearize the value function Vt(ect , ezt)
around the optimum couplets (c∗t , z∗t ).
Vt(ect , ezt) = Vt(e
c∗t , ez∗t )− φt,c(ct − c∗t )2 − φt,z(zt − z∗t )2 + φt,cz(ct − c∗t )(zt − z∗t )
Note that first order conditions about ct and zt hold at optimum such that the first order
terms of the linearization cancel out to zeros. Other partials and cross-partials evaluated at
the optimum couplets are absorbed into terms φt,c = −ec∗t V∗t,11
2> 0, φt,z = −ez∗t V
∗t,22
2> 0, and
φt,cz = V ∗t,12ec∗t ez
∗t > 0. Substituting out zt as function of ct due to the first order condition,
70
we have the loss function L(ct, zt) = ct−1L(ct, zt) such that
L(ct, zt) = Vt(c∗t , z∗t )− Vt(ct, zt)
=λt2
(ct − c∗t )2
where λt = 2(φt,c + φt,zb2− φt,cz
b) ≥ 0 to ensure loss is non-negative. Imposing the equilibrium
condition of ct = ψMt
ptyields that log[ ct
ct−1] = log(Mt/pt) − log(Mt−1/pt−1) = mt, we end up
with the objective loss function of investors
L(ct, zt) =λt2
(mt −m∗t )2
C.4 Proof of Equation (26)
Conditional on paying attention to money growth rate such that κt > 0, we have the
value of attentive learning (L) given by the following maximization:
πLt = maxκt−λt
2σ2m,t2
−2κt − vκt − ζ
Marginal cost from from learning the private signal per unit of attention paid κt is simply
v. The marginal benefit from paying attention is given by λtσ2m,t ln(2)2−2κt , which decreases
in κt. As κt goes to zero, the marginal benefit converges to its max of λtσ2m,t log(2), it yields
to a corner solution κt = 0 if marginal cost dominates v > λtσ2m,t log(2). We work with
λtσ2m,t log(2) ≥ v to pin down an interior solution such that marginal benefit equalizes the
marginal cost: κ∗t = 12
log2[λtσ2
m,t log(2)
v] > 0. Hence, πLt is given by
πLt = − v
log(2)2− v
2log2[
λtσ2m,t log(2)
v]− ζ
While the no attentive learning value is given by πN = −λt2σ2m,t, we define the excess
value of learning ∆πt as
∆πt = πLt − πNt =λt2σ2m,t −
v
log(2)2− v
2log2[
λtσ2m,t log(2)
v]− ζ
Note ∂∆πt∂σ2m,t
= λt2− v
2v
λtσ2m,t[log(2)]2
λt log(2)v
= 12[λt − v
σ2m,t log(2)
] ≥ 0. Setting σ2m,t = v
λt log(2), we
have ∆πt = −ζ < 0. By Intermediate Value Theorem, we have σ2∗m,t such that πLt > πNt for
σ2m,t > σ2∗
m,t.
Considering binding constraint such that 12
log2[λtσ2
m,t log(2)
v] > κ, it yields that κ∗t = κ
71
C.5 Proof of Proposition 4
With excess value of learning ∆πt = 0 setting at σ2∗m,t, by the Implicit Function Theorem,
we have
dσ2∗m,t
dλt= − ∂∆πt/∂λt
∂∆πt/∂σ2∗m,t
< 0
It follows that the denominator is positive for σ2∗m,t >
vtλt
. The numerator is given by ∂∆πt∂λt
=σ2∗m,t
2− v
2v
λtσ2m,t[log(2)]2
σ2∗m,t log(2)
v= 1
2[σ2∗m,t − v
λt log(2)] > 0. Similarly, as
∂∆πt∂v
= − 1
log(2)2− 1
2log2[
λtσ2∗m,t log(2)
v] +
v
2
v
λtσ2m,t[log(2)]2
λ2tσ
2∗m,t log(2)
v2
= −1
2log2[
λtσ2∗m,t log(2)
v] < 0
we have
dσ2∗m,t
dv= − ∂∆πt/∂v
∂∆πt/∂σ2∗m,t
> 0
It shows that these two equations hold regardless of whether ζ = 0.
D Additional Tables
Table A1: Alternative Samples and Reactions to M2 Announcements
(1) (2) (3) (4) (5) (6)VARIABLES 2011-2017 2009-2017 2002-2010 2011-2017 2009-2017 2002-2010
ItM2−1 0.44** 0.31** -0.24(0.17) (0.15) (0.17)
ItM2−1,3 0.33*** 0.24** -0.13(0.11) (0.10) (0.11)
Win dum Ctrls YES YES YESWeekday FE YES YES YES YES YES YESConstant 0.07 0.02 -0.03 0.07 0.03 -0.04
(0.10) (0.09) (0.09) (0.10) (0.08) (0.08)
No. Obs 1,577 2,063 2,170 1,577 2,063 2,170R2(%) 1.03 0.64 0.68 0.91 0.48 0.47
Notes: This table reports the dummy variable regression results of Equations (1) and (2) for different sample periods. Thedependent variable is the log close-to-close excess return constructed from the Wind A Share Index. We align the returndata of the first trading day that the equity market has access to the news to the day tM2. ”Win dum Ctrls”: whether ornot controls for other announcement day dummy variables ItM2−i up to a window length of 2T + 1 with T = 3. ”WeekdayFE”: the weekday fixed effects controls. See text for the definitions of variables. Announcement dummy ItM2−1,j equalsto one for the trading days in a j-trading-day window before a M2 Announcement. ***Significant at 1%, **significant at5%, *significant at 10%. Robust standard errors are shown in parentheses.
72
Table A2: Cross-check of NBS Announcement Dates Using Two Data Sources
Year BEC Y / WC N -15 -11 -10 -9 -8 -7 -4 -3 -2 -1 0 1 2 BEC N / WC Y
2002 20 1 1 1 1 12 102003 25 3 16 1 92004 27 1 1 2 1 1 14 52005 27 5 242006 41 2 1 4 202007 25 9 28 12008 23 3 6 28 1 22009 37 6 19 22010 34 1 1 22 4 162011 15 1 1 1 1 58 1 22012 5 1 73 1 12013 4 76 12014 1 73 1 62015 1 1 64 4 122016 1 1 64 2 122017 1 32 1 6
Total 331 1 1 1 1 1 2 8 3 8 32 623 15 1 85
Notes: This table reports the differences of NBS announcement dates between two sources: Bloomberg Economic Cal-endar (BEC) and author-coded NBS website crawling algorithm (WC). ”BEC Y / WC N” denotes number of announce-ments that are included in Bloomberg but missed by web crawler, ”BEC N / WC Y” denotes exactly the opposite. Num-bers in the first row denote the number of days by which a Bloomberg announcement date leads the corresponding webcrawler date. We report the number of mismatched announcements in this table for a sample of Jan, 2002 to June, 2017.
Table A3: Snapshot of Selected Announcements
Ticker Publisher Key and Concurrent Released Statistics Starting Month of Regular Release
M2 PBC M0/M1/M2 Level and Growth Feb-2000Loan and Savings Balance: Level and GrowthInterbank Loan: Interest Rate and Balance
MPR PBC Monetary Policy Report May-2001TRD GACC Import/Export Growth Jan-2000FAI NBS Investment in Fixed Assets Jul-2002
Retail Sales of Consumer GoodsGDP Growth
VAI NBS Value Added of Industrial Enterprises Mar-2000INP NBS Profits of Industrial Enterprises Sep-2005PMI NBS Manufacturing/Non-manufacturing PMI Aug-2005CPI NBS CPI Jan-2000PPI NBS PPI Jul-2002RST NBS Price Indices of Residential Buildings Feb-2011FOMC U.S. FRB FOMC Statement Feb-1994
Notes: This table reports a summary of the selected announcements considered in this paper.
73
Table A4: Concurrent Macroeconomic Announcements
M2 MPR TRD FAI VAI INP PMI CPI PPI RST FOMC
M2 78 2 5 14 13 11 11 1MPR 26 3 1 2 2 1TRD 78 1 1 1FAI 71 65 20 20 3VAI 65 18 18 3INP 38PMI 79 2CPI 78 78PPI 78RST 76 4FOMC 52
Notes: Sample: January, 2011 to June, 2017. The crossing number of the table is the number of pair-wise concurrent announcements (on the same date). Note that the row or column sum does not have tobe equal to the total number of announcements for a given variable.
74
Tab
leA
5:
Tim
eD
istr
ibu
tion
of
Macro
econ
om
icA
nn
ou
ncem
ents
M2
MPR
TRD
FAI
VAI
INP
PM
ICPI
PPI
RST
FOM
C
Sam
ple
2000M
1-2
017M
6
weekday
before
tradin
ghours
No.
An
ns.
12
34
23
18
20
109
27
19
143
Avg.
An
n.
Tim
e8:4
38:3
58:4
78:4
79:0
08:3
18:4
52:2
5
weekday
within
tradin
ghours
No.
An
ns.
53
133
135
152
58
2166
144
63
Avg.
An
n.
Tim
e11:2
711:0
010:4
110:3
59:5
010:0
010:0
09:5
59:3
2
Mon-T
hurafter
tradin
ghours
No.
An
ns.
95
16
22
22
13
14
Avg.
An
n.
Tim
e16:0
517:5
816:3
815:2
015:2
020:0
017:4
015:0
020:5
3
betw
een
weeks
No.
An
ns.
51
14
33
13
15
12
34
16
17
16
Avg.
An
n.
Tim
e15:3
118:2
712:1
212:5
513:0
49:4
59:0
09:3
39:4
79:3
3
Tota
l211
64
211
168
189
70
146
212
181
79
147
Note
s:T
his
tab
led
epic
tsth
eti
me
dis
trib
uti
on
of
macr
oec
on
om
ican
nou
nce
men
ts.
Shan
gh
ai
an
dS
hen
zhen
Sto
ckE
xch
an
ges
are
op
enfo
rtr
ad
ing
from
Mon
day
toF
rid
ay,
wit
hca
llauct
ion
from
9:1
5to
9:2
5,
conti
nu
ou
sau
ctio
nin
9:3
0-
11:3
0an
d13:0
0-
15:0
0.
Inte
nt
ord
ers
for
blo
cktr
ad
esare
acc
epte
db
etw
een
9:3
0an
d11:3
0an
dagain
bet
wee
n13:0
0an
d15:3
0,
wh
ile
exec
uti
on
ord
ers
an
dfi
xed
-pri
ceord
ers
for
blo
cktr
ad
esare
acc
epte
dfr
om
15:0
0to
15:3
0.
Sp
ecia
lb
lock
trad
ese
ssio
ns
are
hel
don
an
ad
hoc
basi
sb
etw
een
15:0
0an
d17:0
0.
Base
don
this
,w
eca
tegori
zeth
ean
nou
nce
men
tsin
tofo
ur
gro
up
sby
the
tim
eof
thei
rre
lease
:(1
)an
nou
nce
men
tsre
lease
db
efore
trad
ing
hou
rsin
wee
kd
ays;
(2)
an
nou
nce
men
tsre
lease
dw
ith
intr
ad
ing
hou
rs(w
eals
oin
clu
de
an
nou
nce
men
tsre
lease
db
etw
een
morn
ing
sess
ion
an
daft
ern
oon
sess
ion
);(3
)an
nou
nce
men
tsre
lease
daft
ertr
ad
ing
hou
rsfr
om
Mon
day
toT
hu
rsd
ay;
(4)
an
nou
nce
men
tsre
lease
db
etw
een
mark
etcl
osu
reon
Fri
day
an
dm
ark
etop
enn
ess
on
nex
tM
on
day.
Th
ista
ble
rep
ort
sth
enu
mb
erof
an
nou
nce
men
tsof
each
gro
ups
an
dth
eaver
aged
rele
ase
tim
ew
ith
inth
egro
up
.
75
Table A6: Alternative Indices: Equity Returns in Windows of M2 Announcements
(1) (2) (3)VARIABLES Wind A Share Index SSE Composite Index SZSE Component Index
ItAnns−3 0.33 0.25 0.22(0.21) (0.16) (0.19)
ItAnns−2 0.23 0.14 0.20(0.18) (0.14) (0.16)
ItAnns−1 0.44** 0.26* 0.35**(0.17) (0.14) (0.17)
ItAnns 0.16 0.12 0.07(0.18) (0.15) (0.19)
ItAnns+1 -0.16 -0.14 -0.19(0.17) (0.14) (0.18)
ItAnns+2 0.04 -0.05 -0.08(0.21) (0.19) (0.21)
ItAnns+3 0.02 -0.01 -0.01(0.19) (0.15) (0.19)
Weekday FE YES YES YESConstant 0.07 0.04 0.09
(0.10) (0.08) (0.10)
Observations 1,577 1,577 1,577R-squared (%) 1.03 0.94 0.75
Notes: Sample: January, 2011 to June, 2017. This table reports dummy variable regressionresults of Equation (1) for different specifications. The dependent variable is the log close-to-close excess return constructed from different market indices. SSE Index: Shanghai StockExchange Composite (SSE) Index; SZSE Index: Shenzhen Stock Exchange Component Index(SZSE) Index. Announcement dummy ItM2−i equals to one if the i-th trading day before (or,after if i is negative) an M2 announcement. We align the return data of the first tradingday that the equity market has access to the news with the dummy variable ItM2 = 1 wheni = 0. ***Significant at 1%, **significant at 5%, *significant at 10%. Robust standard errorsare shown in parentheses.
76
References
Ai, H. and R. Bansal (2018): “Risk Preferences and The Macro Announcement Pre-
mium,” Econometrica, Forthcoming.
Altavilla, C., D. Giannone, and M. Modugno (2017): “Low frequency effects of
macroeconomic news on government bond yields,” Journal of Monetary Economics, 92,
31 – 46.
Baker, S. R., N. Bloom, and S. J. Davis (2016): “Measuring Economic Policy Uncer-
tainty,” The Quarterly Journal of Economics, 131, 1593–1636.
Balduzzi, P. and F. Moneta (2017): “Economic Risk Premia in the Fixed-Income
Markets: The Intraday Evidence,” Journal of Financial and Quantitative Analysis, 52,
1927–1950.
Barber, B. M., E. T. D. George, R. Lehavy, and B. Trueman (2013): “The
earnings announcement premium around the globe,” Journal of Financial Economics,
108, 118 – 138.
Beaudry, P. and F. Portier (2006): “Stock Prices, News, and Economic Fluctuations,”
American Economic Review, 96, 1293–1307.
Beaver, W. H. (1968): “The Information Content of Annual Earnings Announcements,”
Journal of Accounting Research, 6, 67–92.
Bernanke, B. S. and K. N. Kuttner (2005): “What Explains the Stock Market’s
Reaction to Federal Reserve Policy?” Journal of Finance, 60, 1221–1257.
Bernard, V. L. and J. K. Thomas (1989): “Post-Earnings-Announcement Drift: De-
layed Price Response or Risk Premium?” Journal of Accounting Research, 27, 1–36.
Bloom, N. (2009): “The Impact of Uncertainty Shocks,” Econometrica, 77, 623–685.
——— (2014): “Fluctuations in Uncertainty,” Journal of Economic Perspectives, 28, 153–76.
Bollerslev, T., J. Li, and Y. Xue (2016): “Volume, Volatility and Public News
Announcements,” Creates research papers, Department of Economics and Business Eco-
nomics, Aarhus University.
Bomfim, A. N. (2003): “Pre-announcement effects, news effects, and volatility: Monetary
policy and the stock market,” Journal of Banking and Finance, 27, 133–151.
77
Boyd, J. H., J. Hu, and R. Jagannathan (2005): “The Stock Market’s Reaction to
Unemployment News: Why Bad News Is Usually Good for Stocks,” Journal of Finance,
60, 649–672.
Carpenter, J. N., F. Lu, and R. F. Whitelaw (2017): “The Real Value of China’s
Stock Market,” Nber working papers, National Bureau of Economic Research, Inc.
Chen, K., P. Higgins, D. F. Waggoner, and T. Zha (2016): “Impacts of Monetary
Stimulus on Credit Allocation and Macroeconomy: Evidence from China,” NBER Working
Papers 22650, National Bureau of Economic Research, Inc.
Cochrane, J. H. (1991): “Volatility tests and efficient markets: A review essay,” Journal
of Monetary Economics, 27, 463 – 485.
Coibion, O. and Y. Gorodnichenko (2015): “Information Rigidity and the Expec-
tations Formation Process: A Simple Framework and New Facts,” American Economic
Review, 105, 2644–2678.
Dellavigna, S. and J. M. Pollet (2009): “Investor Inattention and Friday Earnings
Announcements,” Journal of Finance, 64, 709–749.
Epstein, L. G. and S. E. Zin (1989): “Substitution, Risk Aversion, and the Temporal
Behavior of Consumption and Asset Returns: A Theoretical Framework,” Econometrica,
57, 937–969.
Fama, E. F. (1965): “The Behavior of Stock-Market Prices,” The Journal of Business, 38,
34–105.
——— (1970): “Efficient Capital Markets: A Review of Theory and Empirical Work,” The
Journal of Finance, 25, 383–417.
Fan, J. P., T. Wong, and T. Zhang (2007): “Politically connected CEOs, corporate
governance, and Post-IPO performance of China’s newly partially privatized firms,” Jour-
nal of Financial Economics, 84, 330 – 357.
Franklin, A., J. Qian, C. Shan, and L. Zhu (2017): “Dissecting the Long-term Per-
formance of the Chinese Stock Market,” working paper, SSRN.
Frazzini, A. (2006): “The Disposition Effect and Underreaction to News,” Journal of
Finance, 61, 2017–2046.
78
Gilbert, T., A. Kurov, and M. H. Wolfe (2018): “The Disappearing Pre-FOMC
Announcement Drift,” Mimeo, Available at SSRN: https://ssrn.com/abstract=3134546.
Grossman, S. J. and R. J. Shiller (1981): “The Determinants of the Variability of
Stock Market Prices,” American Economic Review, 71, 222–227.
Kacperczyk, M., S. V. Nieuwerburgh, and L. Veldkamp (2016): “A Rational
Theory of Mutual Funds’ Attention Allocation,” Econometrica, 84, 571–626.
Kling, G. and L. Gao (2005): “Calendar effects in Chinese stock market,” Annals of
Economics and Finance, 6, 75–88.
Kreps, D. M. and E. L. Porteus (1978): “Temporal Resolution of Uncertainty and
Dynamic Choice Theory,” Econometrica, 46, 185–200.
Liu, Z., P. Wang, and Z. Xu (2017): “Interest-Rate Liberalization and Capital Misallo-
cations,” Working Paper Series 2017-15, Federal Reserve Bank of San Francisco.
Lu, L. and S. Liu (2008): “Are There Holiday Effect in China’ s Stock Market?” Journal
of Financial Research, 2, 014.
Lucca, D. O. and E. Moench (2015): “The Pre-FOMC Announcement Drift,” The
Journal of Finance, 70, 329–371.
Mackowiak, B. and M. Wiederholt (2009): “Optimal Sticky Prices under Rational
Inattention,” American Economic Review, 99, 769–803.
Mitchell, M. J. D. and M. L. L. Ong (2006): Seasonalities in China’s stock markets:
cultural or structural?, International Monetary Fund.
Peng, L. and W. Xiong (2006): “Investor attention, overconfidence and category learn-
ing,” Journal of Financial Economics, 80, 563–602.
Philippe, M., T. salehi Alireza, and V. Andrea (2017): “Exchange Rates and Mon-
etary Policy Uncertainty,” The Journal of Finance, 72, 1213–1252.
Romer, C. D. and D. H. Romer (2004): “A New Measure of Monetary Shocks: Deriva-
tion and Implications,” American Economic Review, 94, 1055–1084.
Savor, P. and M. Wilson (2013): “How Much Do Investors Care About Macroeconomic
Risk? Evidence from Scheduled Economic Announcements,” Journal of Financial and
Quantitative Analysis, 48, 343–375.
79
Shiller, R. J. (1981): “Do Stock Prices Move Too Much to be Justified by Subsequent
Changes in Dividends?” American Economic Review, 71, 421–436.
Sims, C. A. (2003): “Implications of rational inattention,” Journal of Monetary Economics,
50, 665 – 690, swiss National Bank/Study Center Gerzensee Conference on Monetary
Policy under Incomplete Information.
Song, Z., K. Storesletten, and F. Zilibotti (2011): “Growing Like China,” American
Economic Review, 101, 196–233.
Yellen, J. L. (2015): “Normalizing Monetary Policy: Prospects and Perspectives : a
speech at a research conference sponsored by the Federal Reserve Bank of San Francisco,”
Speech 840, Board of Governors of the Federal Reserve System (U.S.).
80