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:rui.guo@ruc.edu.cn. Jia: Hanqing Advanced Institute of Economics and Finance, Renmin University ofChina. Email: dun.jia@ruc.edu.cn. Sun: Hanqing Advanced Institute of Economics and Finance, RenminUniversity of China. Email: xi.sun@ruc.edu.cn. 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