Stock Price Synchronicity and Analyst Coverage in Emerging Markets*
Kalok Chan
and
Allaudeen Hameed
January 2005
* Chan is from the Department of Finance, Hong Kong University of Science and Technology, Clear Water Bay,
Hong Kong, Tel: 852-2358-7680, Fax: 852-2358-1749, kachan@ust.hk. Hameed is from the Department of Finance
and Accounting, National University of Singapore, Singapore 117592, Tel: 65-6874-3034, Fax: 65-6779-2083,
Allaudeen@nus.edu.sg. Hameed is grateful to Academic Research Grant, National University of Singapore for
financial support. We thank Bernard Yeung, Campbell Harvey, and participants at HKUST, McMaster University,
UNC at Chapel Hill, and the 2002 Korean Finance Association Meetings for useful comments. We are also grateful
to an anomyous referee, whose suggestions have greatly improved the paper. The authors gratefully acknowledge
the contribution of Thomson Financial for providing earnings per share forecast data, available through the
Institutional Brokers Estimate System. This data has been provided as part of a broad academic program to
encourage earnings expectations research. Any errors are our own.
Stock Price Synchronicity and Analyst Coverage in Emerging Markets
Abstract
This paper examines the relationship between the stock price synchronicity and analyst activity in
emerging markets. Contrary to the conventional wisdom that security analysts specialize in the
production of firm-specific information, we find that securities which are covered by more analysts
incorporate greater (lesser) market-wide (firm-specific) information. Using the R-square statistics of the
market model as a measure of the synchronicity of stock price movements, we find that more analyst
coverage leads to an increase in stock price synchronicity. Furthermore, after controlling for the
influence of firm size on the lead-lag relation, we find that the returns on a high analyst-following
portfolio lead returns on a low analyst-following portfolio more than vice versa. We also find that the
aggregate changes in the earnings forecast of the high analyst-following portfolio affect the aggregate
returns of the portfolio itself as well as those of the low analyst-following portfolio, whereas the
aggregate changes in the earnings forecasts of the low analyst-following portfolio have no predictive
ability. Finally, when the forecast dispersion is high, the effect of analyst coverage on stock price
synchronicity is reduced.
2
This paper investigates the informational role of security analysts in a number of emerging
markets. We collect information from I/B/E/S International on analyst activity and examine whether
analysts help to generate market-wide information or firm-specific information.
Our paper is motivated by the finding in Morck, Yeung and Yu (2000) that stock prices move
together more in emerging markets than in developed markets, which suggests that less firm-specific
information is produced in emerging markets. Their interpretation is that in emerging markets, weak
property rights discourage informed trading and therefore prevent firm-specific information from being
incorporated into stock prices. In addition, the recent financial crisis in Asia and other emerging markets
shows that the dissemination of firm-specific information to the public investors is inadequate. This lack
of firm-specific information in emerging markets can be attributed to a number of factors. First, there are
few regulations and little enforcement of information disclosure in the emerging markets. Second, there
is a low degree of voluntary disclosure and corporate transparency. Third, many companies in emerging
markets are group-affiliated or family-owned, and it is difficult to collect reliable information on such
companies.
This raises an important issue about the role of security analysts in the information production
process in emerging markets. The incentive of security analysts to collect private information about
individual companies has been discussed in many theoretical papers (Admati and Pfleiderer (1986),
Diamond and Verrecchia (1981), Grossman and Stiglitz (1976)). A number of papers present empirical
evidence on the role of analysts in the U.S. market (Brown (1988), O’ Brien (1988) and O’ Brien and
Bhushan (1990), Brown (1988)). Nevertheless, most of these studies examine analyst activity in the U.S.
and seldom investigate their behavior in emerging markets, where the incentives to collect information
might be different from those in the developed markets. An exception is Chang, Khanna, and Palepu
(2001), who examine analyst activity around the world, including in a number of emerging markets.
They show that country-specific variables influence the extent of analyst activity and the accuracy of
analysts’ forecasts. They also show that in emerging markets, the earnings of business groups are more
difficult to forecast than the earnings of non-business groups.
1
An important but unexplored issue is the nature of the information that is produced by security
analysts. In particular, although the availability of firm-specific information is shown to affect external
financing and efficiency of capital markets (Durnev, Morck and Yeung (2000)), the role of security
analysts in producing firm-specific information in emerging markets is unclear. As “information
intermediaries” who issue earnings forecasts of individual companies, security analysts specialize in the
production of firm-specific information. However, Piotroski and Roulstone (2003) find that in the U.S.,
although the presence of insiders and large institutional investors has the net effect of increasing the
amount of firm-specific information that is incorporated into stock prices, security analysts decrease that
amount. In other words, in developed markets, security analysts do not have an advantage over insiders
and institutional investors in accessing firm-specific information.
On a theoretical basis, it is unclear whether the presence of security analysts increases marketwide or firm-specific information in emerging markets. Given the lack of publicly available companyspecific news due to less stringent requirements for information disclosure in these markets, the benefits
to be gained from collecting firm-specific information might be high so that there are more incentives for
analysts to collect such information. Therefore, the poor protection of investors’ property rights in
emerging markets may lead to greater investor demand for analysts who produce firm specific
information. This is supported by the empirical evidence in Lang, Lins and Miller (2004) who find that
the additional monitoring provided by analyst coverage increases firm value, especially in countries with
low levels of shareholder rights protection. On the other hand, Morck, Yueng and Yu (2000) argue that
weak property rights discourage informed risk arbitrage based on firm-specific information.
Consequently, the payoff to analysts who produce firm specific information may be too low because it is
not possible to arbitrage on them. Therefore, due to the difficulty associated with collecting firm-specific
information in emerging markets, the information that a security analyst collects might have more
macroeconomic content than firm-specific details.
Using stock return synchronicity as a proxy for the amount of firm-specific information that is
impounded into stock prices, we examine its relationship with the level of analyst activity. If analysts
2
mainly generate firm-specific information, we should observe a negative association between the
synchronicity of stock price movements and the number of security analysts. If analysts generate marketwide information, then we should observe a positive association. Using R2 statistics from the market
model as a measure of synchronicity of stock price movement, we find that greater analyst coverage
increases stock price synchronicity. Therefore, our results for emerging markets are similar to those of
Piotroski and Roulstone (2003) for the U.S. market.
An alternative interpretation of our results is that more analyst coverage lessens the amount of
firm-specific noise. If firm-specific stock price movements reflect noise, then the presence of more
security analysts decreases the level of noise, and consequently increases stock return synchronicity. We
therefore examine the information that is contained in stock prices and earnings forecasts of firms
followed by more or fewer analysts. First, we examine the lead-lag relationship among stock returns of
low analyst-following and high analyst-following firms. Controlling for the firm size effect, we find that
the returns of high analyst-following firms lead returns of low analyst-following portfolio, which supports
the conjecture that firms which are followed by more analysts are faster in incorporating market-wide
information into their stock prices than are firms followed by fewer analysts. We also compare the
information contained in the aggregate change in earnings forecasts across firms in low analyst-following
and high analyst-following portfolios. The aggregate change in earnings forecasts in the two portfolios is
a direct measure of news about the systematic information. We find that the aggregate change in
earnings forecasts in a high analyst-following stock portfolio affects aggregate returns of the portfolio
itself as well as the aggregate returns of the low analyst-following stock portfolio. In contrast, the
aggregate change in earnings forecasts in the low analyst-following stock portfolio does not provide
information about the returns on either of the two portfolios. Overall, our evidence is consistent with the
explanation that the information which is produced by security analysts has more market-wide content.
This paper is organized as follows. Section 1 reviews previous work on stock return
synchronicity and analyst activity. Section 2 discusses the construction of the variables. Section 3
presents the empirical methodologies and analysis, which is followed by concluding remarks in Section 4.
3
1. Previous Research Work
1.1. Stock Price Synchronicity
A common measure that is used to analyze stock price synchronicity is the R2 statistic from the
market model. A high R2 from the market model indicates a high degree of stock price synchronicity.
According to Roll (1988), individual stocks in the U.S. exhibit low R2 statistics, which suggests that much
firm-specific information is incorporated into stock prices. However, Roll also finds that firm-specific
stock price movements are generally not associated with identifiable news releases, which suggests either
that the financial press misses a great deal of relevant information that is generated privately or that price
fluctuations are purely due to noise trading.
The synchronicity of stock price movement is studied in several papers. Morck, Yeung and Yu
(2000) find less synchronicity in economies where the government better protects private property rights.
Their interpretation of this finding is that strong property rights promote informed arbitrage, which leads
to the inclusion of more firm-specific information and less co-movement in stock returns across firms.
Wurgler (2000) shows that the efficiency of capital allocation across countries is negatively correlated
with synchronicity in domestically traded stock returns. Durnev, Morck and Yeung (2000) show that
firms that exhibit less synchronicity tend to use more external financing and allocate capital more
efficiently. Their interpretation is that for a firm with higher firm-specific price variation, informed
arbitrageurs will focus more on the company so that stock prices will track fundamentals closely. This
will reduce information asymmetry problems that impede external financing and distort capital spending
decisions.
A related issue in the study of stock return synchronicity is whether firm-specific stock price
variations reflect noise, thus pushing stock prices to deviate from their fundamental values (DeLong et al
(1990), Shleifer and Vishny (1997)). Because noise trading is firm specific, it tends to decrease the
synchronicity of stock price movements. Durnev, Morck, Yeung, and Zarowin (2001) examine the
relationship between firm-specific stock price variation and accounting measures of stock price
4
informativeness. They define firm-specific price variation as the portion of a firm’s stock return variation
that is unexplained by market and industry returns. They also define price informativeness as the amount
of information that stock prices contain about future earnings, which is estimated from a regression of
current stock returns against future earnings. Their measures of informativeness are (i) the aggregated
coefficients on future earnings and (ii) the marginal variation of current stock returns explained by future
earnings. They document empirical evidence that firm-specific stock price variability is positively
correlated with both measures of stock price informativeness. Hence, they support the argument that
stock price synchronicity is related to the flow of firm-specific information.
1.2 Analyst Activity
Because security analysts collect and disseminate information about firms, their activities are
closely related to theoretical literature on information acquisition (Grossman and Stiglitz (1980),
Diamond and Verrecchia (1981), Verrecchia (1982), Admati (1985), Admati and Pfleiderer (1986), and
Bhushan (1989)). The typical investor holds a small stake in a given firm and therefore has little
incentive and limited resources to produce independent information about the firm’s outlook.
Consequently, there is a demand for security analysts who produce information for typical investors.
Many previous papers examine how various firm characteristics can influence either the
aggregate demand or the supply of analyst services. On the supply side, Bhushan (1989) argues that
larger companies tend to attract a larger number of analysts, presumably because there are significant
fixed costs in following a company, and the payoff from following a company is related to its size.
Furthermore, analysts have an incentive to follow firms with high trading volumes (Alford and Berger
(1999)) as there will be more brokerage commissions. The correlation between firm returns and market
returns is also likely to affect the supply of analyst services (Bhushan (1989)). For a given level of
information costs relating to macro variables, the marginal information acquisition cost for a firm will be
low if the correlation between the firm and the market returns is high. Therefore, a higher correlation
between firm returns and market returns leads to a lower information acquisition cost so that there will be
5
an increase in the supply of analyst services.
On the demand side, analyst activity is related to the corporate ownership structure. There is
likely to be a greater demand for analyst services in firms in which the ownership structure is widely
dispersed. When there is an increase in the concentration of ownership, the acquisition of analyst services
is not cost effective for small investors. As Ball, Kothari and Robin (1998) posit, when ownership is
concentrated, information is likely to be communicated through private channels, thus decreasing the role
of financial analysts. However, concentrated ownership by institutional investors such as pension funds
and money managers may increase the demand for analyst services, because institutional investors who
perform fiduciary roles use analyst reports as evidence of their due diligence (O’ Brien and Bhushan
(1990). Moyer, Chatfield and Sisneros (1989) provide evidence that the number of analysts who follow a
given firm is inversely related to the portion of the firm that is held by insiders, and positively related to
measures of institutional shareholdings. Another variable that will affect the aggregate demand for
analyst services is a firm’s return variability. Assuming that the public information flow is constant, there
is more private information when the return variability is higher. Consequently, the number of analysts
who are following a given firm is positively related to its return variability (Bhushan (1989)).
1.3 Earnings forecasts and stock prices
One of the determinants of stock price variation is the revaluation of a security based on the
expectations of future earnings. Some earlier papers show that stock prices are related to current
earnings and future earnings (Ball and Brown (1968) and Beaver, Clarke and Wright (1979)). Besides
examining how stock prices include information about future earnings, they also look at the accuracy of
analyst forecasts. For example, Brown and Rozeff (1978) and Brown et al. (1987a) show that analyst
forecasts are superior to the time-series model in forecasting earnings.
As analysts often specialize by industry and their knowledge about a particular industry can be
applied to all companies within that industry (O’ Brien (1990)), company specific events can have
implications on the earnings of other companies in the same industry. Chandra, Procassini and Waymire
6
(1999) investigate the relation between industry-wide information disclosures by the trade association in
the semiconductor industry and both share prices and analyst forecasts. They investigate the industry
reports that contain industry data on new orders and shipments, and document significant stock price
movements on the release dates of industry reports by the trade association each month.
2. Construction of Variables
We investigate the relationship between stock return synchronicity and analyst coverage. Stock
return synchronicity and analyst coverage are affected by several firm-specific and economy-wide factors,
and it is important that that we control for these factors. In this section, we discuss the variables used in
our analyses.
Two sources of data were used in the construction of the variables. The first was Standard &
Poor’s (formerly the International Finance Corporation) Emerging Markets Database (EMDB), which
covers more than 2,000 stocks from 45 emerging markets. This database compiles the monthly and
weekly closing stock prices, dividends, shares outstanding, market capitalization, trading volume, and
other financial statement information such as earnings and book value. The second data source was
I/B/E/S International, which provides data on analyst activity and their earning forecasts for a large
number of companies around the world. We report the number of analysts who are following each
company, which will be used as our measure of the extent of analyst coverage.
2.1. Stock return synchronicity (SYNCH)
Our measure of stock price synchronicity follows Morck, Yeung and Yu (2000), and we estimate
the linear regression:
Ri ,t = β i0 + β i1 Rm,t + ε i ,t
(1)
where Ri ,t is return of stock i at week t and Rm ,t is market return at week t. In their analysis of U.S.
7
stocks, Roll (1988) and Piotroski and Roulstone (2003) include industry returns to explain stock returns in
the regression model. However, in emerging markets, it is problematic to include industry returns as an
additional factor because in some markets the economy is dominated by a few industries and it is difficult
to disentangle the industry effect from the market effect. Moreover, it is common for an industry in an
emerging economy to include only a few companies. Consequently, when industry returns are computed
using the few companies from an industry, they reflect company-specific news rather than industry news.
Following Morck, Yeung and Yu (2000), synchronicity can be defined as
SYNCHi ,t = log(
R2
1− R2
),
(2)
where R2 is the coefficient of determination from the estimation of equation (1) for firm i in year t.
SYNCHi,t is measured for each firm based on the weekly return observations of the year, provided that
there are a minimum of 40 weekly observations in the year. A high SYNCHi ,t indicates that the firm is
highly correlated with the market.
We hypothesize that stock return synchronicity is affected by the extent of analyst coverage,
trading volume, and firm capitalization, as discussed below.
2.1.1 Analyst coverage (ANALYST)
As security analysts frequently issue earnings forecasts for individual companies, it is reasonable
to expect that they acquire firm-specific information. However, given the difficulty associated with
collecting firm-specific information, the information that a security analyst collects might have a large
amount of macroeconomic content. Therefore, whether the presence of security analysts increases
market-wide information or firm-specific information in emerging markets becomes an empirical issue.
Likewise, the direction of the relationship between stock return synchronicity and analyst coverage is
uncertain. If security analysts tend to produce firm-specific information, then firms that are followed by
more analysts will exhibit lower stock return synchronicity. If security analysts tend to produce market-
8
wide information, then firms that are followed by more analysts will exhibit higher stock return
synchronicity.
2.1.2. Trading volume (VOLUME)
The level of trading volume of a stock affects stock return synchronicity because it influences the
speed of price adjustments. Actively traded stocks react to market information on a timely basis so that
their individual price movements are more synchronous with market movement. In contrast, infrequently
traded stocks experience a greater delay in their price reactions, which results in lower stock return
synchronicity.
2.1.3. Firm Size (SIZE)
As the stock market indices computed by EMDB are value-weighted, the market capitalization of
a company determines its component weight in the index. When the number of stocks within an index is
small, a few large companies dominate the market movement. Consequently, when R2 is estimated based
on the value-weighted index, we expect a positive relationship between stock return synchronicity and the
market capitalization of a company.
2.2. Analyst coverage (ANALYST)
The degree of analyst coverage, which is one of the variables that influence the stock return
synchronicity, is endogenous. We hypothesize that analyst coverage is affected by the following
variables: degree of investibility, firm size, stock return variability, trading volume, and stock return
synchronicity. The relationship between the degree of analyst coverage and these variables is discussed
below.
We measure the intensity of analyst activity as the average number of analysts who issued
earnings forecasts for a firm during a given calendar year. We gather data on the number of unique
analysts issuing forecasts through I/B/E/S International. Even for those firms that do not have earnings
9
forecasts being issued on them, they will also be included in our analysis. For the firms with no earnings
forecasts, this could mean that no analyst followed the firm or that the data for the firm were not captured
by I/B/E/S International. To avoid underestimating the number of analysts for those firms that are not
covered by I/B/E/S International, we also perform our analysis by excluding firms with zero analyst
coverage and find the results are generally robust.
2.2.1 Investibility Measure (INVEST)
We obtain an investibility measure (INVEST) for each company in various years from EMDB.
The investibility measure, which ranges from zero to one, reflects the extent to which foreign investors
can purchase stocks in emerging markets. Among the determinants of investibility are the foreign
ownership limits that are imposed by the government at the national level or the industry level. In
addition, the investibility measure reflects only the free float percentage available to public investors, as
the shares held by the government, strategic investors, or other corporations are excluded. If, for
example, a firm has a 30% foreign limit restriction but 40% of the firm’s shares are not publicly available,
the investibility measure is recorded as 60%. To a certain extent, the investibility measure is a reflection
of public ownership – the higher the percentage of public ownership, the higher the investibility measure.
As the demand for analyst services is expected to be higher for companies with widely dispersed
ownership (Bhushan (1989), Ball, Kothari, and Robin (1998)), a company’s investibility measure should
be positively related to the number of analysts who are following the company.
2.2.2 Firm size (SIZE)
Holding other things constant, the demand for analyst services is likely to be an increasing
function of firm size (SIZE). First, security analysts find that private information about a larger firm is
more valuable than the same information about a smaller firm, which creates greater incentives for
analysts to follow larger companies. Furthermore, the larger the firm size, the larger the number of
shareholders, so that there will be a greater demand for information that is produced by security analysts.
10
However, if a larger firm discloses more information publicly, this could be a substitute for the
information that an analyst could readily collect, therefore decreasing the demand for analyst services.
2.2.3 Stock return volatility (VOLATILITY)
Private information is more valuable for firms with higher return volatility (VOLATILITY). As
Bhushan (1989) explains, when return variability increases, so does the probability of there being large
deviations between the expected return conditional on both private and public information and the
expected return conditional just on public information. Therefore, the demand for analyst services is
likely to be an increasing function of a firm’s return variability.
2.2.4 Trading volume (VOLUME)
Alford and Berger (1999) suggest that trading volume (VOLUME) is a proxy for brokerage
commissions. Trading volume affects the incentives of security analysts to collect information. When
stocks are traded more frequently and thus generate more brokerage commissions, more security analysts
will be willing to supply more information. Furthermore, stocks with large trading volumes have more
diverse shareholders, and hence greater demand for security analysts.
2.2.5 Stock return synchronicity (SYNCH)
In Section 2.1.1 we discussed how stock return synchronicity (SYNCH) is affected by the extent
of analyst coverage. It is also possible that the causality runs in the reverse direction. Bhushan (1989)
argues that if the correlation between firm returns and market returns (R2) is high, then for a given level of
information costs relating to macro variables, the marginal information acquisition cost for a firm will be
low. Therefore, the information acquisition cost is likely to be an inverse function of the R2 , so that an
increase in stock return synchronicity leads to an increase in analyst services.
3. Empirical Analysis
11
3.1 Summary statistics
Table 1 presents a summary of the sample firms that are included in the study. The sample
includes firms from 25 countries with the sample years starting from 1993 and ending in 1999. Panel A
shows how the final sample is constructed. A total of 9392 firm-year observations are available from
I/B/E/S International, and a total of 12751 observations are available from the EMDB. Our final sample
includes 10820 observations, including 4638 observations of zero-analysts following. Panel B shows the
distributions of observations by year. There are fewer observations in the first two years (around 700
observations per year), but relatively more in later years (around 900 observations per year). Panel C
provides the frequency distribution of observations by industry. The manufacturing industry has the
largest number of observations (3135), followed by finance, insurance and real estate industry (1128), and
the other industries have fewer than 500 observations each.
Table 2 presents univariate statistics for the variables described above. For each country and
year, firms are partitioned into four groups based on the number of analysts following. The first group
contains firms with zero analysts following. The second to fourth groups contain firms with low,
medium, and high numbers of analysts following, within each country and year. The number of firm-year
observations is 4638, 1730, 2292, and 2160 in the four groups. On average, there are 3.9, 8.5, and 13.3
analysts in the second to fourth groups.
We estimate the R-square and stock return synchronicity measure for each company in each year
using weekly return observations that are denominated in either local currency or U.S. dollars. From the
market model we also calculate the standard deviation of residual returns. The EMDB computes two
market indices for each country: an investable index and a global index. The difference between the
global index and the investable index is that while the former is designed to represent the whole market of
the country, the latter is designed to represent the portion of that market which is available to foreign
investors. Therefore, with two possible currency denominations and two possible market indices, we
generate four different R-square and synchronicity measures for each company. However, as the results
are similar regardless of which currency or which market index is used, we present only the results based
12
on the R-square that is calculated using U.S. dollar-denominated returns and a global index for the
respective country. It is evident from Table 2 that the stock return synchronicity tends to increase for
those firms with more analysts following. The R-square statistics are 0.23, 0.23, 0.27, and 0.35, and the
synchronicity measures are -1.99, -1.85, -1.37, and -0.87 for groups of zero, low, medium, and high
numbers of analysts. Consequently, the residual standard deviation also declines with the number of
analysts following, being 0.0698, 0.0668, 0.0593, and 0.0532 in the four groups.
However, we also find that the number of analysts following is correlated with other variables.
The market capitalization (SIZE) is generally lower for firms with smaller numbers of analysts, with an
average of 337, 578, and 1412 million U.S. dollars from the groups of low to high numbers of analysts
following. The trading volume (VOLUME) is also higher for firms with more analysts following, with an
average monthly trading volume of 31 million, 40 million, and 87 million shares for the three groups. In
addition, the investibility measure (INVEST) is positively related to the number of analysts, whereas the
stock return variability (VOLATILITY) is negatively related to the number of analysts.
Overall, there is preliminary evidence that stock return synchronicity is related to analyst
coverage. However, these results only represent a univariate relationship, as we find that other variables
such as firm size and trading volume are also related to the extent of analyst coverage. In the next
section, we will investigate these relationships based on GMM estimation and simultaneous regressions.
3.2 The relationship between stock return synchronicity and analyst coverage
In this section, we will examine the relationship between stock return synchronicity and analyst
following. We first estimate the regression equations for explaining stock return synchronicity and the
number of analysts individually, and then estimate the two in a simultaneous equation framework.
3.2.1 GMM Estimation
13
We first estimate the equation that explains the stock return synchronicity for company i in year t:
SYNCH i ,t = α + β 1 * LOG (1 + ANALYSTi ,t ) + β 2 * LOG ( SIZE i ,t ) + β 3 * LOG (VOLUMEi ,t )
23
+
∑
λ k CDUM i ,k +
k =1
6
∑
l =1
δ l YRDUM i ,l +
8
∑φ
m INDDUM i , m
+ ε i ,t
(3)
m =1
The dependent variable SYNCHi.,t is the synchronicity measure of firm i, and is computed based on
equation (2). For the explanatory variables, ANALYSTi, t is the number of analysts covering company i at
year t, SIZEi, t is the natural log of market capitalization of firm i at year t, VOLUMEi, t is the natural log of
trading volume of company i at year t, and CDUM, YRDUM, and INDUM are dummy variables that
control for country, year, and industry fixed effects. The coefficient that is associated with
LOG(1+ANALYSTi, t) is predicted to have a positive sign if analysts tend to produce market-wide
information, and a negative sign if analysts tend to produce firm-specific information. The coefficient
that is associated with LOG(VOLUMEi,t,) is predicted to have a positive sign because more actively traded
companies are able to react to information more much rapidly and in a synchronous manner. The
coefficient that is associated with LOG(SIZEi,t,) is predicted to have a positive sign because larger firms
have higher weights in affecting market returns.
Panel A of Table 3 shows the GMM estimation results of Equation (3). All observations,
including those firms with zero analysts following, are included in the estimation.1 The t-statistics, shown
in brackets, are corrected for heteroskedasticity and autocorrelation based on the Newey-West adjustment
(1987).
There is strong evidence that stock return synchronicity (SYNCH) is significantly and positively
related to both the number of analysts following (ANALYST) and trading volume (VOLUME), with tstatistics of 4.82 and 28.58 respectively. In contrast, the coefficient of SIZE is insignificant (t-statistic of 0.48). One explanation for this is that the firm size and trading volume are highly correlated, and that
trading volume subsumes the firm size in explaining stock return synchronicity. Even if we exclude
1
We re-estimated all of the models excluding firms with zero analysts, and the results remain unchanged.
14
SIZE from the set of explanatory variables, the magnitudes and significance of coefficients on ANALYST
and VOLUME are very qualitatively similar.
The adjusted R-square measure is 0.34 regardless of
whether we include SIZE or not, which indicates that SIZE does not have an incremental explanatory
power for SYNCH.
When a firm has a higher proportion of market-wide information, it might have more marketwide volatility or less firm-specific volatility. To examine whether firms with wider analyst coverage
have smaller firm-specific volatility, we estimate another regression that is similar to Equation (3), but
replacing the stock return synchronicity (SYNCH) with the volatility of residual return (RESVAR) as the
dependent variable in the regression:
RESVARi ,t = α + β 1 * LOG( 1 + ANALYSTi ,t ) + β 2 * LOG( SIZE i ,t ) + β 3 * LOG( VOLUMEi ,t )
23
+
∑
k =1
λ k CDUM i ,k +
6
∑
δ l YRDUM i ,l +
l =1
8
∑φ
m INDDUM i ,m
(4)
+ ε i ,t
m =1
The results, which are also presented in Panel A of Table 3, show that the analyst coverage has a negative
impact on the residual variance, with a coefficient of -0.02 and a t-statistic of -4.79. This suggests that
firms with more analyst following tend to have less firm-specific volatility. There is also evidence that
the firm-specific volatility is negatively correlated with SIZE and positively correlated with VOLUME.
The adjusted R-square decreases from 0.44 to 0.33 when we exclude SIZE from the explanatory variables,
which indicates that SIZE is an important variable in explaining firm-specific volatility.
The third regression equation is for explaining the number of analysts who issued earnings
forecasts for company i in year t:
LOG( 1 + ANAYSTi ,t ) = α + β 1 * SYNCHi ,t + β 2 * LOG( SIZEi ,t ) + β 3 * LOG( VOLUMEi ,t )
+ β 4 * VOLATILITYi ,t + β 5 * INVESTi ,t +
23
∑
λ k CDUM k ,t +
k =1
8
∑
m =1
δ m INDDUMm ,t +
6
∑φ YRDUM
l
l ,t
+ ε i ,t
(5)
l =1
In addition to the variables that we defined in Equations (3) and (4), VOLATILITYi,t is the standard
deviation of (weekly) stock return of firm i at year t, and INVESTi,t is the investibility measure of firm i at
15
year t. The coefficient that is associated with SYNCHi,t is predicted to have a positive sign because the
marginal information acquisition for a firm will be smaller when it has a lot of market-wide information,
so more analysts will be covering the company. The coefficient that is associated with LOG (SIZEi,t ) is
predicted to have a positive sign because we expect a greater demand for security analysts for bigger
firms with more shareholders. The coefficient on LOG (VOLUMEi,t ) is predicted to have a positive sign
because more analysts can be supported by greater trading interests. The coefficient on VOLATILITYi,t is
expected to have a positive sign as a higher volatility means more information is produced, and this will
increase the demand for security analysts. The coefficient on INVESTi,t is expected to be positive, as a
higher degree of investibility is accompanied by a more dispersed ownership structure so that the demand
for analyst service is greater.
The empirical results in Table 3 show that the number of analysts following a stock (ANALYST)
increases with stock return synchronicity (SYNCH), with a coefficient of 0.02 and a t-statistic of 3.17. In
addition, the number of analysts following is also significantly and positively related to stock market
capitalization (SIZE), trading volume (VOLUME), and degree of investibility (INVEST). However, it is
interesting to note that higher stock volatility (VOLATILITY) reduces the extent of analyst coverage, even
after controlling for other factors. One possible explanation is that the volatile stocks in emerging
markets are especially risky, making them undesirable even in a well-diversified portfolio. Consequently,
there is also less demand for analyst coverage.
3.2.2 Two-stage Least Squares (2SLS) Estimation
As the stock return synchronicity and number of analysts affect each other simultaneously, GMM
parameter estimates of Equations (3) to (5) are likely to be biased and inconsistent. We therefore estimate
the coefficients based on two-stage least square (2SLS) estimation. Panel B of Table 3 presents the
results. Because the equation for explaining ANALYST is under-identified, the parameter estimates of
Equation (4) are not reported. We only report the parameter estimates of Equations (3) and (5), which
are just identified under 2SLS estimation. As the GMM estimation shows that SIZE is important in
16
explaining RESVAR but not SYNCH, we include SIZE as the explanatory variable only in Equation (4) but
not Equation (3) in 2SLS estimation. In general, the signs of the coefficient estimates under 2SLS are
similar to those under GMM estimation, although some of the coefficient estimates are even stronger. In
the equation that explains stock return synchronicity, the coefficient on ANALYST is 1.09 with a tstatistic of 11.53. In the equation that explains residual return volatility, the coefficient on ANALYST is 0.25, with a t-statistic of -10.37. Therefore, the result that the number of analyst is positively related to
stock return synchronicity and negatively related to residual return volatility is robust, thus confirming
that firms which have more analyst coverage have a higher proportion of market-wide information.
3.2.3 Alternative Interpretations
Another interpretation of the results is that they are related to the I/B/E/S data used in the sample.
If I/B/E/S primarily collects information on U.S. analysts who are following emerging markets, and given
that the cost of collecting firm-specific information for U.S. based analysts might be high because of
distance, language barriers, and accounting difference across countries, it will be of no surprise that U.S.
analysts produce predominately market-wide information in emerging markets.
The evidence on the performance of local versus foreign analysts is mixed. Orputt (2003) finds
that local analysts who cover European firms issue more accurate and timelier earning forecasts than nonlocal analysts. Malloy (2003) finds evidence which indicates that local U.S. equity analysts (located near
US cities) possess an information advantage over other analysts, particularly in firms in small and remote
areas. However, looking at analyst stock recommendations in Latin American emerging markets,
Bacmann and Bolliger (2001) report that foreign analysts outperform home country analysts. Chang
(2002) finds that expatriate analysts located outside of Taiwan (but whose firms have local operations)
outperform local analysts in their recommendations of Taiwanese stocks.
To investigate whether our results are driven by U.S. analysts, we obtained the names of the
brokerage firms that covered the stocks in our sample from the Broker Translation File provided by
I/B/E/S. The brokerage name was matched with entries from Nelson’s Directory of Investment Research
17
(1998), which lists the address of all brokerage firms. The official addresses listed in Nelson (1998) were
used to identify the location of the brokerage firm. The brokerage firm was classified as a U.S. firm if it
had a U.S. headquarters or if its primary office (address) was located there. For those brokerage firms that
are not covered by Nelson (1998), we conducted a web search to identify the broker’s headquarters.
Altogether, the U.S. analysts account for 16% of the total sample. We delete these U.S. analysts
from the sample and re-estimate Equations (3) to (5). The results, which are reported in Table 4, are
similar to those in Table 3. For example, in the 2SLS, in the equation that explains stock return
synchronicity, the coefficient on ANALYST is 1.15 with a t-statistic of 11.53. In the equation that explains
residual return volatility, the coefficient on ANALYST is -0.26 with a t-statistic of -10.29. Therefore, the
results that the number of analyst is positively related to stock return synchronicity and negatively related
to residual return volatility remain robust, thus confirming that our findings are not driven by U.S. analyst
coverage in emerging markets.
3.2.4 Robustness Check
A potential problem in our study is that in some emerging markets, a few large conglomerates can
account for a large part of the economy. In such cases, firm-specific information overlaps with marketwide information, and that the distinction between the two becomes unclear.
We check the robustness
of our results by examining two sub-samples. The first sub-sample excludes observations for a country in
a particular year if the top five companies account for more than 50% of total market capitalization of the
country in that year. The results are reported in Panel A of Table 5. The sub-sample is reduced to 8521
observations, which is 79% of the original sample. For the sake of brevity, only the GMM estimates of
Equations (3) and (4) are reported, as results based on 2SLS are similar. The overall results remain the
same, and we find that the number of analysts is positively related to stock return synchronicity and
negatively related to residual return volatility. The second sub-sample excludes diversified firms, which
are defined as firms having sales in more than two industries, based on two digit SIC codes. The relevant
data on the industries associated with each firm is obtained from Worldscope. We discard about 30% of
18
the firms based on this criterion. The results based on GMM estimation of Equations (3) and (4) are
reported in Panel B of Table 5. Again, there is a robust positive relationship between stock return
synchronicity and number of analysts. Therefore, the relationship between stock return synchronicity
and analyst activity is not due a few diversified companies dominating the economy
We also check the robustness of our results by estimating alternative specifications. First, we
construct an equally-weighted market index for each country and re-estimate the stock return
synchronicity using equally-weighed market returns. Second, we redefine analyst coverage by creating a
dummy variable that is equal to one if the number of analysts following a firm is higher the average of the
firm’s peer companies and zero if the number of analysts is lower. Third, we re-compute the R2 based on
bi-weekly returns and four-weekly returns and the leads and lags of market returns to adjust for
nonsynchronous trading. The results, which are not reported here, can be provided upon request.
Overall, the results that wider analyst coverage increases stock return synchronicity and decreases firmspecific information are robust.
3.3 Lead-lag relationship between returns
The evidence so far suggests that there is a positive relationship between stock return
synchronicity and the number of analysts following the firm. This is consistent with the hypothesis that
analysts tend to produce market-wide information, but an alternative explanation is that the firms which
are being followed by analysts tend to have less firm-specific noise. This is reflected as a higher
association between stock returns and market returns. Hence, an interesting alternative hypothesis is that
more analyst coverage reduces firm specific noise trading rather than increases market-wide information.
To ascertain whether firms that are followed by more analysts have more market-wide
information incorporated into their stock prices, we examine the lead-lag relationship among the returns
of three portfolios sorted by the number of analysts. The lead-lag relationship is studied extensively in
the literature. Lo and MacKinlay (1990) document a positive correlation in weekly returns on the stocks
of small firms and the lagged weekly returns of large firms. This is interpreted as evidence of differential
19
information production, and that firm size may proxy for the magnitude of information produced by
investors: the larger the firm, the greater the amount of information produced. Without market
imperfections of some sort, information transmission would be instantaneous. However, if investors or
market makers face sufficiently high costs that prevent them from instantaneously reacting to information,
then this information will be impounded in small-firm stock prices by small-firm investors only after
observing the past stock prices of large firms: that is, only after a lag. For example, Chan (1993) argues
that market makers do not observe the prices of the other stocks instantaneously, thus causing delays in
inferring market-wide information. Mech (1993) shows that the delay in the price adjustment is related to
high transaction costs. Badrinath, Kale and Noe (1995) argue that because of differential information
setup-costs and the legal restrictions that arise from the “prudent man” rule, informed investors invest
only a subset of traded firms. In this setting, uninformed investors can use the past prices of these stocks
to predict the prices of other stocks. Their empirical findings show that returns on stocks with high levels
of institutional ownership lead returns on stocks with lower levels of institutional ownership, and that the
lead-lag relation induced by the level of institutional ownership persists even when firm size is controlled.
Brennan, Jegadeesh and Swaminathan (1993) examine the effect of the number of investment analysts
who are following a firm on the speed of adjustment of the firm’s stock price to common information.
They find that returns on the portfolios of firms that are followed by many analysts tend to lead those of
firms that are followed by fewer analysts, even when the firms are approximately the same size. In
addition, partial adjustment of stock prices to common information can also generate positive portfolio
autocorrelations, as shown in Chordia and Swaminathan (2000).2
Given that firm size plays an important role in affecting the lead-lag relationship and that there is
a positive association between firm size and number of analysts following a firm, we need to control for
the influence of firm size when examining the impact of analyst coverage on the lead-lag relationship.
We divide our sample of firms in the following manner. For each year, firms in the same country are
2
Bouduokh, Richardson and Whitelaw (1994) provide a detailed discussion of various interpretations of
autocorrelation patterns in short-horizon portfolio returns.
20
first ranked according to the firm size at the end of the year. On the basis of this ranking, the firms are
divided into three portfolios, where the first portfolio consists of the smallest firms and the third portfolio
consists of the largest firms. Using this ranking, the firms within each of the three size-sorted portfolios
are further ranked on the basis of the number of analysts who are following the firms. Then, each firm is
placed into one of three analyst-following sub-portfolios, in which the first sub-portfolio has the smallest
number of analysts and the third sub-portfolio has the highest number. This method ensures that firms in
different analyst-following portfolios but in the same size portfolio vary only in terms of the number of
analysts but not in terms of the firm size.
We calculate the weekly returns for each sub-portfolio across different years for each country. To
examine the lead-lag relationship between low and high analyst-following portfolios of a particular firmsize portfolio within a country during a particular year, we estimate the following vector-autoregressive
(VAR) model:
K
R L , j ,t = a0 +
∑
K
a k R L , j ,t − k +
k =1
∑b R
K
R H , j ,t = a0 +
∑
k =1
k
H , j ,t − k
k =1
K
c k R L , j ,t − k +
∑d
k R H , j ,t − k
k =1
+ ε L ,t
(6)
+ ω L ,t
(7)
where RL , j ,t and RH , j ,t are weekly returns of the low analyst-following (L) portfolio and the high analystfollowing (H) portfolio in a particular firm size category j, where j = small, medium, and large, and K is
the number of lags that we examine. If coefficients bk (k = 1,2...K) are positive, then the past returns of
the high analyst-following portfolio have predictive ability for future returns of the low analyst-following
portfolio. Conversely, if coefficients c k (k = 1,2….K) are positive, then the past returns of the low
analyst-following portfolio have predictive ability for the future returns of the high analyst-following
portfolio. If analysts do produce market-wide information, then we expect that more systematic
information will be inferred from the firms with more analysts following. Therefore, we predict that the
21
high analyst-following portfolio will lead the low analyst-following portfolio more than vice versa, and
that the coefficients bk are positive and bigger than the coefficients ck .
We stack the portfolio returns of different countries together for the regression estimation.
Table 6 presents the results that are based on U.S. dollar returns, although the results that are based on
local currency returns are similar. The VAR is estimated for three lags (K=3). The choice of 3 lags is adhoc, but we also examine alternative specifications, and verify that the results are similar. The overall
results indicate that the lagged returns of the high analyst-following portfolio have predictive ability for
the returns of the low analyst-following portfolios in all three different size categories. In all three size
categories (small, medium, and large), the lagged returns of the high analyst-following portfolio have
significant predictive ability for the low-analyst portfolio in the first lag. In contrast, the predictive ability
of lagged returns of the low analyst-following portfolio for high analyst-following portfolio is much
weaker. Moreover, the lagged returns of the high analyst-following portfolio have predictive ability even
for its own future returns.3 This suggests that rather than responding to market-wide information
simultaneously, some firms in the high analyst-following portfolio are faster than the others. In contrast,
the lagged returns of the low analyst-following portfolio do not have predictive ability for its own future
returns. Overall, the asymmetric predictability between high and low analyst-following portfolios is
consistent with the hypothesis that firms with more analyst coverage increase the speed of their
adjustment to market-wide information.
3.4 Predictive ability of aggregate earnings forecasts
So far we have found a positive relationship between stock return synchronicity and the number
of analysts following a firm. Although we interpret our previous results as security analysts producing
3
As the lagged returns of the portfolio of high analyst following firms have predictive ability for its own future returns, the
residual returns in Equation (7) could be autocorrelated if we do not have enough lags of R H , j ,t on the right hand side. We
therefore include additional lags of R H , j ,t on the right hand side of the VAR model, and find that the results are robust.
22
market-wide information, an alternative interpretation is that a few leader analysts produce firm-specific
information which is subsequently mimicked by other analysts.
To see whether analysts follow each other in revising their earnings forecasts, we examine the
lead-lag relation in changes in earnings forecast. First, we extract all earnings forecast data in each month
for firms in our sample from I/B/E/S International. If there are multiple earnings forecasts issued for a
firm within a month, we use the mean forecast.. For each company and in each month, we compute the
monthly change in the one-year ahead forecasted earnings by taking the percentage change in the current
month’s forecast over the forecast in the previous month. To reduce the noise associated with individual
company level change in forecasts, we group the stocks into portfolios and compute the aggregate
(average) change in earnings forecasts for each portfolio. We sort the stocks within each country into
three portfolios based on the number of analysts following the firms, and then compute the average
percentage change in monthly earnings forecasts across all stocks in the low, medium, and high analystfollowing portfolios. The lead-lag relation in earnings forecast changes for the three portfolios are
estimated as follows:
∆FORECASTL,t = a0 + a1∆FORECASTL,t −1 + a 2 ∆FORECASTM ,t −1 + a3 ∆FORECASTH ,t −1 + ε L,t
∆FORECASTM ,t = b0 + b1∆FORECASTL,t −1 + b2 ∆FORECASTM ,t −1 + b3 ∆FORECASTH ,t −1 + ε M ,t
(8)
∆FORECASTH ,t = c0 + c1∆FORECASTL,t −1 + c 2 ∆FORECASTM ,t −1 + c3 ∆FORECASTH ,t −1 + ε H ,t
where ∆FORECASTL,t , ∆FORECASTM ,t and ∆FORECASTH ,t are the average percentage change in
earnings forecasts for the low, medium and high analyst-following portfolios at month t. We also
estimate Equation 8 in log specifications, where we use LOG(1+ ∆FORECASTi,t ) as the explanatory
variable. The results based on the linear and log specifications are presented in Panels A and B Table 7,
respectively. Based on either specification, we find strong evidence of analysts following each other in
revising the earnings forecasts. Of the nine lead-lag coefficients in Equation 8, six are statistically
significant in the linear specification and seven in the log specification. Furthermore, the forecast
revisions in the low analyst following portfolio trails the high analyst following portfolio more than the
23
vice versa. A test of the equality of the coefficients a3 and c1 is soundly rejected in both specifications.
Overall, the evidence indicates that analysts follow each other in revising their earnings forecasts.
However, it is insufficient to differentiate between analysts producing market-wide information and a few
analysts producing firm-specific information which is mimicked by other analysts. To distinguish
between the two possibilities, we further investigate the information content of the earnings forecasts
issued by analysts. If an earnings forecast contains market-wide information, it should lead to price
revisions not only for an individual company but also for other companies. However, if earning forecasts
reduce noise and do not contain market-wide information, then the forecast revisions will not lead to price
revisions of other companies. The following regression is estimated:
R j ,t = a0 + a1∆FORECASTL,t + a 2 ∆FORECASTM ,t + a3 ∆FORECASTH ,t + ε t ,
(9)
where Rj,t is the return of the low, medium, or high analyst-following portfolio j at month t, and
∆FORECASTL,t , ∆FORECASTM ,t and ∆FORECASTH ,t are the average percentage change in earnings
forecasts for the three portfolios at month t. If analysts produce market-wide information, then the
percentage change in earnings forecasts from the high analyst-following portfolio should have the best
predictive ability. We expect coefficient a1 to be not different from zero, and coefficient a3 to be
significantly positive. Again, we also estimate Equation 9 in log specifications, where we use
LOG(1+ ∆FORECASTi,t ) as the explanatory variable.
The results are presented in Table 8, in which Panel A and B contain the results based on the
linear and log specifications, respectively. Based on the linear specifications, we find that the coefficients
associated with ∆FORECASTH ,t are positive and significant in explaining the returns of the low and
medium analyst-following portfolios. The joint test of the null hypothesis that ∆FORECASTH ,t does not
affect returns on all three portfolios is rejected at conventional significance levels. In contrast, the
coefficients that are associated with ∆FORECASTL,t and ∆FORECASTM ,t are either negative or
insignificant in explaining returns of any of the portfolios. The joint tests reveal that revisions in
24
forecasted earnings for those firms with low and medium analyst coverage do not significantly influence
stock returns. The results based on log specifications are qualitatively similar, although we find that the
coefficients which are associated with LOG(1+ ∆FORECASTH ,t ) are significant in explaining the returns
of the medium and high analyst-following portfolios, but not low analyst- following portfolio. Since
changes in forecast of earnings for high analyst-following firms lead to revision of prices of other firms,
we infer that their earnings forecasts have macroeconomic content.
3.5 Effect of Forecast Dispersion on Stock Price Synchronicity
The last approach for us to evaluate the informational content in analyst forecasts is to examine
the effect of forecast dispersion on stock price synchronicity. Our logic is similar to that in Piotroski and
Roulstone (2003). So far, we have argued that the positive relationship between stock price synchronicity
and analyst coverage is mostly due to the market-wide information contained in the analysts forecasts. If
this is indeed the case, then higher forecast dispersion among analysts would indicate that there is less
agreement about the systematic component of their forecasts, which in turn will reduce stock price
synchronicity.
We modify regression (3) by including an interaction term as follows:
SYNCH i ,t = α + β1 * LOG (1 + ANALYSTi ,t ) + β 2 * ( LOG (1 + ANALYSTi ,t ) * DISPERSION i ,t ) +
β 3 * LOG ( SIZE i ,t ) + β 4 * LOG (VOLUMEi ,t ) +
k
k =1
8
6
23
∑ λ CDUM
i ,k
+
∑ δ YRDUM
l
l =1
i ,l
+
∑φ
m INDDUM i , m
+ ε i ,t
(10)
m =1
where DISPERSIONi,t is a forecast dispersion dummy variable, equal to 1 if the forecast dispersion is
higher than the average of other firms in the same country, year, and analyst coverage group, and 0 if
forecast dispersion is lower than the average of the peer group. Because no forecast dispersion is
available if no analyst follows a company, companies with zero analyst coverage are excluded from the
estimation.
The results are reported in Table 9. We have estimated equation (10) singularly using GMM and
25
jointly with equation (5) using 2SLS. Based on GMM estimation, regardless of whether we include SIZE
as the explanatory variable, the coefficient that is associated with the interaction term is significantly
negative. This indicates that when the forecast dispersion is high, the impact of the amount of analyst
coverage on stock price synchronicity is reduced.
We also replace stock price synchronicity (SYNCH) with the volatility of the residual return
(RESVAR):
RESVARi,t = α + β1 * LOG( ANALYSTi,t ) + β 2 * ( LOG(1 + ANALYSTi ,t ) * DISPERSIONi,t ) + β 3 * LOG( SIZEi,t )
+ β 4 * LOG(VOLUMEi,t ) +
23
∑ λ CDUM
k
k =1
6
i,k
+
8
∑δ YRDUM + ∑φ
l
i ,l
l =1
m INDDUMi ,m
+ ε i,t
(11)
m =1
The results, which are also presented in Table 9, indicate that the coefficient that is associated with the
interaction term is significantly positive. Therefore, when the forecast dispersion is higher, earnings
forecasts contain less market-wide information, and wider analyst coverage will reduce the decline in
firm-specific information.
4. Conclusion
This paper examines the relationship between stock return synchronicity and analyst activity in
emerging markets. Contrary to the conventional wisdom that security analysts specialize in the
production of firm-specific information, we find that security analysts predominantly produce marketwide information. First, using the R2 of a market model as a measure of the synchronicity of stock price
movement, we find that coverage by more analysts increases stock price synchronicity. Furthermore,
after controlling for the influence of firm size on lead-lag relations, we find that the returns of a high
analyst-following portfolio lead the returns of a low analyst-following portfolio more than vice versa. We
also find that the aggregate changes in earnings forecasts of a high analyst-following portfolio affect the
aggregate returns of all stocks, including those with low analyst following. In contrast, the aggregate
change in the earnings forecasts of a low analyst-following portfolio has little predictive content for the
26
returns of any portfolio. Finally, when the forecast dispersion is high, the effect of analyst coverage on
stock price synchronicity is reduced.
The results presented in our paper also have some implications for analyst activity in developed
markets. First, our work is related to Piotroski and Roulstone (2003) who find that although the presence
of insiders and large institutional owners in the U.S. have the net effect of increasing the amount of firmspecific information in stock prices, security analysts decrease the amount of firm-specific information.
Therefore, security analysts do not have any advantage over insiders and institutional investors in
producing firm-specific information. Our results based on emerging markets demonstrate that poor
information disclosure and lack of corporate transparency increases the cost of collecting firm-specific
information, so that security analysts generate their earnings forecasts based mostly on macroeconomic
information. Hence, one could also examine whether firms with poor corporate transparency in
developed markets have less firm-specific information discovered by security analysts.
Second, given the large market-wide information content in analyst forecasts, it might be
beneficial for analysts to learn from the forecasts of analysts covering different stocks. One could
therefore extend the study to developed markets and examine whether analysts will follow each other in
generating earnings forecasts for different firms. Furthermore, our results show that revision in earnings
forecast of one stock has predictive ability for returns of other stocks. A natural extension is to examine
whether earnings forecasts of stocks with large analysts coverage are more informative and have larger
price impact on other stocks in the same market (or industry) than stocks with small analyst coverage.
27
References
Admati, A., 1985, A noisy rational expectations equilibrium for multi-asset securities market,
Econometrica 53, 629-657.
Admati, A. and P. Pfleiderer, 1986, A monopolistic market for information, Journal of Economic Theory
39, 400-438.
Alford A., and P. Berger, 1999, A simultaneous equation analysis of forecast accuracy, analyst following,
and trading volume, Journal of Accounting, Auditing & Finance, 219-246.
Arbel, A., S. Carvell, and P. Strebel, 1983, Giraffes, institutions and neglected firms, Financial Analysts
Journal 39, 57-63.
Bacmann, J. and G. Bolliger, 2001, Who are the best? Local versus foreign analysts on the Latin
American stock markets, Working paper, University of Neuchatel.
Badrinath, S., J. Kale, and T. Noe, 1995, Of shepherds, sheep, and the cross-autocorrelations in equity
returns, Review of Financial Studies 8, 401-430.
Ball, R. and P. Brown, 1968, An empirical evaluation of accounting income numbers, Journal of
Accounting Research, Autumn, 159-178.
Ball, R., S. P. Kothari, and A. Robin, 1998, The effect of international institutional factors on properties
of accounting earnings, Working paper, University of Rochester.
Beaver, W., R. Clarke and W. Wright, 1979, The association between unsystematic security returns and
the magnitude of earnings forecast errors, Journal of Accounting Research, Autumn, 316-340.
Beaver, W., R. Lambert and D. Morse, 1980, The information content of security prices, Journal of
Accounting and Economics 4, 15-40.
Bhushan R., 1989, Firm characteristics and analyst following, Journal of Accounting and Economics 22,
255-274.
Boudoukh, J., M. Richardson, and R. F. Whitelaw, 1994, A tale of three schools: Insights on
autocorrelations of short-horizon stock returns, Review of Financial Studies 6, 799-824
Brennan, M., N Jegadeesh, and B. Swaminathan, 1993, Investment analysis and the adjustment of stock
prices to common information, Review of Financial Studies, 6, 799-824.
Brown, L.D., P.A. Griffin, R.L. Hagerman and M.E. Zmijewski, Security analyst superiority relative to
univariate time-series models in forecasting quarterly earnings, Journal of Accounting and Economics 9,
61-87.
Brown, L.D., and M.S. Rozeff, 1978, The superiority of analyst forecasts as measures of expectations:
Evidence from earnings, Journal of Finance, 1-16.
Chan, K., 1993, Imperfect information and cross-autocorrelation among stock returns, Journal of Finance
48, 1211-1230.
28
Chang, C., 2002, Information footholds: expatriate analysts in an emerging market, Working paper,
University of California, Berkeley.
Chordia, T., and B. Swaminathan, 2000, Trading volume and cross-autocorrelations in stock
returns, Journal of Finance 55, 913-935.
Collins, D., S.P. Kothari, and J. Rayburn, 1987, Firm size and the information content of prices with
respect to earnings, Journal of Accounting and Economics 9, 111-138.
Collins, D. and S. P. Kothari, 1989, An analysis of intertemporal and cross-sectional determinants of
earnings response coefficients, Journal of Accounting and Economics 11, 143-181.
DeLong, B., A. Shleifer, L. Summers, and R. Waldmann, 1990, Noise trader risk in financial markets,
Journal of Political Economy 98, 703-738.
Diamond, D. and R. Verrecchia, 1981, Information aggregation in a noisy rational expectations economy,
Journal of Financial Economics 9, 221-235.
Durnev, A., R. Morck, B. Yeung, and P. Zarowin, 2001, Does greater firm-specific return variation mean
more or less informed stock pricing? Working paper
Chang, J., T. Khanna, K. Palepu, 2000, Analyst activity around the world, Working paper, Harvard
Business School.
Freeman R., 1987, The association between accounting earnings and security returns for large and small
firms, Journal of Accounting and Economics 9, 195-228.
Grossman S. and J. Stiglitz, 1980, On the impossibility of informationally efficient markets, American
Economic Review 70, 393-408.
Hellwig, M., 1980, On the aggregation of information in competitive markets, Journal of Economic
Theory 22, 477-498.
Lang, M, K. Lins and D. Miller, 2004, Concentrated control, analyst following and valuation: do analysts
matter most when investors are protected least? Journal of Accounting Research, forthcoming.
Lo, A. and C. MacKinlay, 1990, When are contrarian profits due to stock market overreaction, Review of
Financial Studies 3, 175-206.
Morck, R., B. Yeung, and W. Yu, 2000, The information content of stock markets: why do emerging
markets have synchronous stock price movements? Journal of Financial Economics 59, 215-260.
Malloy, C, 2003, The geography of equity analysis, Working paper, London Business School.
Moyer, R. C., R. E. Chatfield, and P. H. Sisneros, 1989, Security analyst monitoring activity: Agency
costs and information demands, Journal of Financial and Quantitative Analysis 24, 503-512.
Nelson’s Directory of Investment Research, 1998, Nelson Publications.
Newey, W., and K. West, 1987, A simple positive definite, heteroscedasticity and autocorrelation
29
consistent covariance matrix, Econometrica, 55, 703-705.
Orputt, S., 2003, Local asymmetric information advantages: International evidence from analysts’
European firm earnings forecasts, Working paper, University of Chicago.
O’ Brien, P, 1988, Analysts’ forecasts as earnings expectations, Journal of Accounting and Economics 10,
53-83.
O’ Brien, P, and R. Bhushan, 1990, Analyst following and institutional ownership, Journal of Accounting
Research
O’ Brien, P., 1990, Forecast accuracy of individual analysts in nine industries, Journal of Accounting
Research, Autumn, 286-304.
Piotroski, J, and D. Roulstone, 2003, Analysts, institutional investors and insiders: What information do
they provide? Working paper, University of Chicago
Shleifer, A. and R. Vishny, 1997, The limits of arbitrage, Journal of Finance 52, 35-55.
Wurgler, J., 2000, Financial markets and the allocation of capital, Journal of Financial Economics 58,
187-214
30
Table 1
Number of firm-year observations in the sample
Panel A: By country
Country
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Argentina
Colombia
Hungary
Czech Republic
Chile
Greece
Israel
Pakistan
Peru
Philippines
Poland
Portugal
Sri Lanka
Turkey
Venezuela
Brazil
China
India
Indonesia
Mexico
South Africa
Taiwan
Thailand
Korea
Malaysia
Total
I/B/E/S Only
Sample
years
93-99
94-99
95-99
95-99
93-99
93-99
96-99
93-99
98-99
93-99
95-99
93-99
93-99
93-99
94-99
93-99
93-99
93-99
90-99
93-99
93-99
93-99
93-99
93-99
93-99
EMDB Only
Number of
firm-years
231
127
72
156
296
344
91
226
33
285
128
210
238
219
53
504
258
603
363
430
314
726
560
848
658
7973
Sample
years
93-99
93-99
93-99
94-99
93-99
93-99
97-99
93-99
93-99
93-99
93-99
93-98
93-99
93-99
93-99
93-99
93-99
93-99
93-99
93-99
93-99
93-99
93-99
93-99
93-99
31
Number of
firm-years
228
174
106
336
322
342
139
454
230
338
155
183
294
348
123
595
1103
883
352
527
451
640
493
1141
856
10813
I/B/E/S and EMDB
Matched Sample
Sample
Number of
years
firm-years
93-99
189
94-99
107
95-99
59
95-99
147
93-99
241
93-99
249
97-99
66
93-99
202
98-99
31
93-99
230
95-99
106
93-98
136
93-99
212
93-99
164
94-99
48
93-99
387
93-99
188
93-99
505
93-99
244
93-99
298
93-99
226
93-99
550
93-99
394
93-99
673
93-99
530
6182
Table 1 (continued)
Panel B: By Year (Based on Matched Sample)
1
2
3
4
5
6
7
Year
1993
1994
1995
1996
1997
1998
1999
Total
Number of firm-years
694
772
907
884
1029
981
915
6182
Panel C: By Industry (Based on Matched Sample)
Industry classification
1
2
3
4
5
6
7
8
9
Agriculture, forestry, and fishing
Mining
Construction
Manufacturing
Transportation, communication, electric, gas, and sanitary
services
Wholesale trade and retail trade
Finance, Insurance, and real estate
Services
Government and others
Total
32
Number of
firm-years
141
233
260
3135
491
317
1128
149
328
6182
Table 2
Summary Statistics
This table reports the summary statistics for the sample of portfolios with Zero, Low, Medium, and High analyst
following. The mean of a variable is calculated as the average across all firms and years, and its corresponding
standard deviation is indicated in the parenthesis. For each firm, we estimate a market model regression of weekly
U.S. dollar-denominated returns on the EMDB global market index returns of the respective country, obtaining the
R-square, standard deviation of residual returns (RESVAR), and synchronicity measure (SYNCH = log(
R( i )2
1− R( i )2
) ).
N is the number of firm-year observations in the group, ANALYST is the number of unique brokers/analysts
following the stock, SIZE is the market capitalization in millions of U.S. dollars, VOLUME is the trading volume in
millions of shares, INVEST is the investibility weight, and VOLATILITY is the standard deviation of weekly stock
returns.
N
ANALYST
R-SQUARE
RESVAR
SYNCH
SIZE
VOLUME
INVEST
VOLATILITY
Zero
Low
Medium
High
4638
0.2322
(0.2111)
0.0698
(0.0420)
-1.9993
(2.1175)
715.2157
(2134.1900)
47.3628
(212.6641)
0.4053
(0.4399)
1730
3.8867
(3.0137)
0.2302
(0.1876)
0.0668
(0.0446)
-1.8585
(1.9186)
337.1598
(743.9780)
31.1447
(91.2746)
0.4360
(0.4191)
2292
8.5301
(5.6567)
0.2769
(0.1900)
0.0593
(0.0304)
-1.3785
(1.5914)
558.6189
(1076.6300)
40.2651
(123.5912)
0.5448
(0.4117)
2160
13.3046
(8.0481)
0.3553
(0.2029)
0.0532
(0.0275)
-0.8730
(1.4301)
1412.2700
(2420.8100)
87.5798
(268.5701)
0.5578
(0.3862)
0.0808
(0.0456)
0.0770
(0.0479)
0.0709
(0.0353)
0.0675
(0.0326)
0
33
Table 3
Determinants of stock return synchronicity and analyst coverage for the full sample
SYNCH i ,t = α + β 1 * LOG (1 + ANALYSTi ,t ) + β 2 * LOG( SIZE i ,t ) + β 3 * LOG(VOLUMEi ,t ) +
23
∑
λ k CDUM i , k +
k =1
RESVARi , t = α + β1 * LOG(1 + ANALYSTi , t ) + β 2 * LOG( SIZEi , t ) + β 3 * LOG(VOLUMEi , t ) +
23
∑
6
∑
δ l YRDUM i ,l +
l =1
λk CDUM i , k +
k =1
6
∑
8
∑φ
m INDDUM i , m
+ ε i ,t (3)
m =1
δ l YRDUM i , l +
l =1
8
∑φ
m INDDUM i , m
+ ε i , t (4)
m =1
LOG (1 + ANALYSTi ,t ) = α + β 1 * SYNCH i ,t + β 2 * LOG ( SIZE i ,t ) + β 3 * LOG (VOLUMEi ,t ) + β 4 *VOLATILITYi ,t + β 5 * INVESTi ,t +
23
∑ λ CDUM
k
k ,t
k =1
8
+
∑δ
6
m INDDUM m,t
m =1
+
∑ φ YRDUM
l
l ,t
(5)
+ ε i ,t
l =1
where SYNCHi,t is the stock return synchronicity measure and is equal to log((R2/(1- R2)), where R2 is the R-square from the market model of regressing the stock
return of firm i against the stock market index of the country at year t, RESVARi,t is the volatility of the residual return estimated from the market model of
regressing stock return of firm i against the stock market index of the country at year t, LOG(1+ ANALYSTi, t) is the natural log of the number of analysts
covering company i at year t, LOG(VOLUMEi, t) is the natural log of trading volume of firm i at time t, LOG(SIZEi,t) is the natural log market capitalization of
firm i at year t, INVESTi,t is the investibility weight of firm i at year t, VOLATILITYi,t is the standard deviation of the stock return of firm i at year t, and CDUM,
YRDUM, and INDUM are dummy variables included to control for fixed effects representing the countries, years, and industries in the sample. Adj R-square is
the adjusted coefficient of determination. The sample period is from 1993 to 1999 for a total of 10,820 firm-year observations from 25 countries.
Panel A: GMM Estimation
SYNCH
INTERCEPT
-1.40
(-13.80)
SYNCH
LOG(ANALYST)
0.07
(4.82)
SYNCH
-1.42
(-16.79)
0.06
(4.85)
RESVAR
-2.20
(-85.82)
-0.02
(-4.79)
RESVAR
-2.86
(-124.73)
-0.04
(-10.81)
LOG(ANALYST)
-0.37
(-4.20)
0.02
(3.17)
LOG(SIZE)
-0.01
(-0.48)
-0.15
(-39.94)
0.12
(10.86)
34
LOG(VOLUME)
0.39
(28.58)
VOLATILITY
INVEST
Adj R-square
0.36
0.39
(38.28)
0.36
0.04
(10.70)
0.44
-0.06
(-20.14)
0.33
0.04
(3.86)
-0.15
(-5.59)
0.32
(10.11)
0.22
Table 3 (cont’d)
Determinants of stock return synchronicity and analyst coverage for the full sample
Panel B: 2SLS Estimation
INTERCEPT
SYNCH
LOG(1+ANALYST)
LOG(SIZE)
LOG(VOLUME)
VOLATILITY
INVEST
Adj R-square
Equations (3) & (5)
SYNCH
-2.54
(-17.27)
1.09
(11.53)
0.27
(16.84)
LOG(1+ANALYST)
0.27
UNIDENTIFIED
Equations (4) & (5)
RESVAR
-2.05
(-60.58)
-0.25
(-10.37)
-0.13
(-26.97)
LOG(1+ANALYST)
0.05
(13.41)
UNIDENTIFIED
35
0.35
Table 4
Determinants of stock return synchronicity and analyst coverage for a sub-sample of non-U.S. analyst coverage
SYNCH i ,t = α + β 1 * LOG(1 + ANALYSTi ,t ) + β 2 * LOG( SIZEi ,t ) + β 3 * LOG(VOLUMEi ,t ) +
23
∑
λ k CDUM i , k +
k =1
RESVARi ,t = α + β 1 * LOG(1 + ANALYSTi ,t ) + β 2 * LOG( SIZE i ,t ) + β 3 * LOG(VOLUMEi ,t ) +
6
∑
δ l YRDUM i ,l +
l =1
23
∑
λ k CDUM i ,k +
k =1
6
8
∑φ
m INDDUM i , m
+ ε i ,t (3)
m =1
∑
δ l YRDUM i ,l +
l =1
8
∑φ
m INDDUM i , m
+ ε i ,t (4)
m =1
LOG (1 + ANALYSTi ,t ) = α + β 1 * SYNCH i ,t + β 2 * LOG ( SIZE i ,t ) + β 3 * LOG (VOLUMEi ,t ) + β 4 *VOLATILITYi ,t + β 5 * INVESTi ,t +
23
∑ λ CDUM
k
k ,t
k =1
8
+
∑δ
6
m INDDUM m,t
m =1
+
∑ φ YRDUM
l
l ,t
(5)
+ ε i ,t
l =1
Where SYNCHi,t is the stock return synchronicity measure and is equal to log((R2/(1- R2)), where R2 is the R-square from the market model of regressing the
stock return of firm i against the stock market index of the country at year t, RESVARi,t is the volatility of the residual return estimated from the market model of
regressing the stock return of firm i against the stock market index of the country at year t, LOG(1+ ANALYSTi, t) is the natural log of number of analysts
covering company i at year t, LOG(VOLUMEi, t) is the natural log of trading volume of firm i at time t, LOG(SIZEi,t) is the natural log market capitalization of
firm i at year t, INVESTi,t is the investibility weight of firm i at year t, VOLATILITYi,t is the standard deviation of the stock return of firm i at year t, and CDUM,
YRDUM, and INDUM are dummy variables included to control for fixed effects representing the countries, years, and industries in the sample. Adj R-square is
the adjusted coefficient of determination. The sample period is from 1993 to 1999 for a total of 10,742 firm-year observations from 25 countries.
Panel A: GMM Estimation
SYNCH
INTERCEPT
-1.39
(-13.82)
SYNCH
LOG(1+ANALYST)
0.07
(4.60)
SYNCH
-1.42
(-16.79)
0.07
(4.63)
RESVAR
-2.20
(-85.80)
-0.02
(-4.74)
RESVAR
-2.86
(-124.58
-0.04
(-10.80)
LOG(1+ANALYST)
-0.33
(-3.94)
0.02
(2.88)
LOG(SIZE)
-0.01
(0.45)
-0.15
(-39.95)
0.11
(10.87)
36
LOG(VOLUME)
0.39
(10.69)
VOLATILITY
INVEST
Adj R-square
0.36
0.39
(38.31)
0.36
0.04
(10.69)
0.44
-0.06
(-20.17)
0.33
0.03
(3.67)
-0.14
(-5.62)
0.31
(10.43
0.22
Table 4 (cont’d)
Determinants of stock return synchronicity and analyst coverage for a sub-sample of non-U.S. analyst coverage
Panel B: 2SLS Estimation
INTERCEPT
SYNCH
LOG(1+ANALYST)
LOG(SIZE)
LOG(VOLUME)
VOLATILITY
INVEST
Adj R-square
Equations (3) & (5)
SYNCH
-2.50
(-17.34)
1.15
(11.53)
0.27
(17.05)
LOG(ANALYST)
0.27
UNIDENTIFIED
Equations (4) & (5)
RESVAR
-2.05
(-61.54)
-0.26
(-10.29)
-0.13
(-27.49)
LOG(ANALYST)
0.05
(13.50)
UNIDENTIFIED
37
0.35
Table 5 Determinants of stock return synchronicity and analyst coverage for sub-samples of well-diversified emerging market economies
SYNCH i ,t = α + β 1 * LOG(1 + ANALYSTi ,t ) + β 2 * LOG( SIZEi ,t ) + β 3 * LOG(VOLUMEi ,t ) +
23
∑
λ k CDUM i , k +
k =1
RESVARi ,t = α + β 1 * LOG(1 + ANALYSTi ,t ) + β 2 * LOG( SIZE i ,t ) + β 3 * LOG(VOLUMEi ,t ) +
6
∑
δ l YRDUM i ,l +
l =1
23
∑
λ k CDUM i ,k +
k =1
6
∑
8
∑φ
m INDDUM i , m
+ ε i ,t (3)
m =1
δ l YRDUM i ,l +
l =1
8
∑φ
m INDDUM i , m
+ ε i ,t (4)
m =1
where SYNCHi,t is the stock return synchronicity measure and is equal to log((R2/(1- R2)), where R2 is the R-square from the market model of regressing the stock
return of firm i against the stock market index of the country at year t, RESVARi,t is the volatility of residual return estimated from the market model of regressing
the stock return of firm i against the stock market index of the country at year t, LOG(1+ ANALYSTi, t) is the natural log of number of analysts covering company
i at year t, LOG(VOLUMEi, t) is the natural log of trading volume of firm i at time t, LOG(SIZEi,t) is the natural log market capitalization of firm i at year t,
INVESTi,t is the investibility weight of firm i at year t, VOLATILITYi,t is the standard deviation of the stock return of firm i at year t, and CDUM, YRDUM, and
INDUM are dummy variables included to control for fixed effects representing the countries, years, and industries in the sample. In Panel A, observations for a
country in a particular year is deleted if top five companies account for more than 50% of total market capitalization of the country in that year. In Panel B, we
delete firms that generates net revenue from more than two industries, based on their 2-digit SIC codes for their business segments. Adj R-square is the adjusted
coefficient of determination. The sample period is from 1993 to 1999 for a total of 8,521 and 7,280 firm-year observations from 25 countries in Panels A and B
respectively..
Panel A: Excluding country-years where the top five companies account for more than 50% of total market capitalization
SYNCH
INTERCEPT
-1.19
(-12.85)
SYNCH
LOG(1+ANALYST)
0.07
(5.04)
LOG(SIZE)
LOG(VOLUME)
0.33
(28.60)
-0.02
(-1.53)
0.35
(23.49)
0.36
-0.05
(-16.56)
0.33
0.04
(10.39)
0.45
SYNCH
-1.09
(-9.80)
0.08
(5.13)
RESVAR
-2.88
(-115.25)
-0.03
(-7.94)
RESVAR
-2.17
(-77.53)
-0.01
(-2.69)
-0.16
(-38.56)
38
VOLATILITY
INVEST
Adj R-square
0.36
Table 5 (cont’d)
Determinants of stock return synchronicity and analyst coverage for sub-samples of well-diversified emerging market economies
Panel B: Excluding firms that has businesses in more than 2 industries
SYNCH
INTERCEPT
-1.48
(-11.21)
SYNCH
LOG(1+ANALYST)
0.04
(2.37)
LOG(SIZE)
LOG(VOLUME)
0.42
(35.79)
-0.01
(-0.86)
0.43
(26.50)
0.34
-0.06
(-17.21)
0.32
0.03
(6.96)
0.42
SYNCH
-1.42
(-9.71)
0.04
(2.43)
RESVAR
-2.92
(-82.84)
-0.03
(-7.29)
RESVAR
-2.33
(-65.75)
-0.02
(-4.17)
-0.14
(-32.09)
39
VOLATILITY
INVEST
Adj R-square
0.34
Table 6
Vector Auto-Regressions for Weekly Size-Analyst Portfolios
The Vector Auto-Regressions (VAR) model is estimated for the weekly portfolio returns sorted on size and analyst following. S1 refers to the smallest portfolio
and S3 refers to the largest portfolio. The specification for the bi-variate VAR is as follows:
K
R L , j ,t = a0 +
∑
K
a k R L , j ,t − k +
k =1
∑b R
K
R H , j ,t = a0 +
∑
k
H , j ,t − k
k =1
+ ε L ,t
(6)
+ ω L ,t
(7)
K
c k R L , j ,t − k +
k =1
∑d
k R H , j ,t − k
k =1
where R L , j ,t and R H , j ,t are weekly returns of the low-analyst (L) portfolio and the high-analyst (H) portfolio of particular firm size j, where j = small, medium,
and large, and K is the number of lags that we examine. The weekly returns are denominated in U.S. dollars. Adj R-square is the adjusted coefficient of
variation. The Wald-test statistic tests the null-hypothesis:
K
K
k =1
k =1
∑ bk = ∑ ck . * denotes significance at the 10 percent level; ** denotes significance at the 5 percent
level; and *** denotes significance at the 1 percent level. The sample period is from 1993 to 1999 for a total of 6182 firm-year observations from 25 countries.
Independent variable
Size category
j = small
j = medium
j = large
Dependent
variable
RL,j,t
RH,j,t
RL,j,t
RH,j,t
RL,j,t
RH,j,t
Lag 1
0.009
0.048*
0.030
0.040
-0.039
-0.021
RL,j,t
Lag 2
0.070**
0.046
-0.004
-0.024
0.001
0.006
Lag 3
Lag 1
-0.022
-0.007
0.016
0.054*
-0.002
-0.030
0.088***
0.050
0.035
0.035
0.032
-0.039
40
RH,j,t
Lag 2
0.055**
0.096***
0.103***
0.150***
0.117***
0.097**
Lag 3
0.093***
0.095***
0.010
0.017
0.030
0.065**
Adjusted Rsquare
0.034
0.040
0.018
0.029
0.015
0.014
Wald test
7.46***
2.19
17.80***
Table 7
Vector Auto-Regressions of aggregate changes in earnings forecasts
The Vector Auto-Regressions (VAR) model is estimated for aggregate changes in earnings forecasts sorted into three portfolios based on number of analyst
following. ∆FORECASTL ,t ,and ∆FORECASTH ,t are average percentage changes in earnings forecasts for the low (L), medium (M), and high (H) analystfollowing portfolios at month t. The specification for the VAR(1) model is as follows:
∆FORECASTL ,t = a0 + a1∆FORECASTL ,t −1 + a2 ∆FORECASTM ,t −1 + a3∆FORECASTH ,t −1 + ε L ,t
∆FORECASTM ,t = b0 + b1∆FORECASTL ,t −1 + b2 ∆FORECASTM ,t −1 + b3∆FORECASTH ,t −1 + ε M ,t
(8)
∆FORECASTH ,t = c0 + c1∆FORECASTL ,t −1 + b2 ∆FORECASTM ,t −1 + b3∆FORECASTH ,t −1 + ε H ,t
for portfolios j=L, M and H. ∆FORECAST is based on percentage change in earnings forecast in Panel A and is measured by log transformation of the
percentage change in earnings forecast in Panel B. Adj R-square is the adjusted coefficient of variation. The Wald-test statistic tests the null-hypothesis:
a3 = c1 . * denotes significance at the 10 percent level; ** denotes significance at the 5 percent level; and *** denotes significance at the 1 percent level. The
sample period is from 1993 to 1999.
Panel A
∆FORECAST L, t −1
∆FORECAST M , t −1
∆FORECAST H , t −1
Adj R-square
Wald-test
∆FORECAST L, t
0.079
0.200***
0.355***
0.096
8.48***
∆FORECAST M , t
0.130***
0.193***
0.272***
0.172
∆FORECAST H , t
0.072**
0.086
0.343***
0.147
Panel B
log(1+ ∆FORECAST L, t −1 )
log(1+ ∆FORECAST M , t −1 )
log(1+ ∆FORECAST H , t −1 )
Adj R-square
Wald-test
log(1+ ∆FORECASTL ,t )
0.158***
0.286***
0.404**
0.159
5.50**
log(1+ ∆FORECASTM ,t )
0.135***
0.287***
0.339***
0.266
log(1+ ∆FORECASTM ,t )
0.045
0.101
0.453***
0.190
41
Table 8
Regressions of portfolio returns on aggregate changes in earnings forecasts
This table reports the coefficients of GLS regressions of portfolio returns (denominated in U.S. dollars) on aggregate changes in earnings forecast for three
analyst following sorted portfolios (low, medium, and high). The specification for the GLS regressions is as follows:
R j ,t = a 0 + a1, j ∆FORECAST L ,t + a 2, j ∆FORECASTM ,t + a 3, j ∆FORECAST H ,t + ε t
(9)
where R,jt is the return of low, medium, and high analyst-following portfolio j at month t, ∆FORECASTL,t , ∆FORECASTM ,t ,and ∆FORECASTH ,t are average
percentage changein earnings forecasts for the low (L), medium (M), and high (L) analyst-following portfolios at month t. Panel A is based on the specification
using ∆FORECAST , while Panel B is based on the specificiation using LOG(1+ ∆ FORECAST ). Adjusted R-square is the adjusted coefficient of variation.
K
K
k =1
k =1
The joint-test reports the Wald statistics for the test akL = akM = akH = 0 for k = 1,2 , and 3 . The Wald-test statistic tests the null-hypothesis: ∑ a1 k = ∑ a3 k .
* denotes significance at the 10 percent level; ** denotes significance at the 5 percent level; and *** denotes significance at the 1 percent level. The sample
period is from 1993 to 1999 for a total of 6182 firm-year observations from 25 countries.
∆FORECASTL ,t
RL,t
RM,t
RH,t
Joint-Test
RL,t
RM,t
R,Ht
Joint-Test
Panel A
∆FORECASTM ,t
∆FORECASTH ,t
-0.001
-0.001
-0.004
2.21
0.003
0.001
-0.001
1.77
0.019***
0.013**
0.009
8.62**
LOG(1+ ∆FORECASTL ,t )
Panel B
LOG(1+ ∆FORECASTM ,t )
LOG(1+ ∆FORECASTH ,t )
-0.003
-0.002
-0.004
2.35
-0.004
-0.006
-0.002
3.93
0.006
0.014***
0.012**
9.03**
42
Adj R-square
Wald-test
0.006
0.002
0.000
5.04**
Adj R-square
Wald-test
-0.000
0.003
0.003
5.36**
Table 9
Effect of forecast dispersion on the relationship between stock return synchronicity and analyst coverage
SYNCH i ,t = α + β 1 * LOG( ANALYSTi ,t ) + β 2 * LOG(( 1 + ANALYSTi ,t ) * DISPERSION i ,t ) + β 3 * LOG( SIZEi ,t )
23
6
8
k =1
l =1
m =1
(10)
+ β 4 * LOG( VOLUMEi ,t ) + ∑ λk CDUM i ,k + ∑ δ l YRDUM i ,l + ∑ φ m INDDUM i ,m + ε i ,t
RESVARi ,t = α + β 1 * LOG ( ANALYSTi ,t ) + β 2 * LOG ((1 + ANALYSTi ,t ) * DISPERSION i ,t ) + β 3 * LOG ( SIZE i ,t )
+ β 4 * LOG (VOLUME i ,t ) +
23
∑
λ k CDUM i ,k +
k =1
6
∑
l =1
δ l YRDUM i ,l +
8
∑φ
m INDDUM i , m
(11)
+ ε i ,t
m =1
LOG (1 + ANALYSTi ,t ) = α + β 1 * SYNCH i ,t + β 2 * LOG ( SIZE i ,t ) + β 3 * LOG (VOLUMEi ,t ) + β 4 *VOLATILITYi ,t + β 5 * INVESTi ,t +
23
∑ λ CDUM
k
k =1
8
+
∑δ
m =1
6
m INDDUM m,t
+
∑ φ YRDUM
l
l ,t
k ,t
(5)
+ ε i ,t
l =1
where SYNCHi,t is the stock return synchronicity measure and is equal to log((R2/(1- R2)), where R2 is the R-square from the market model of regressing the stock
return of firm i against the stock market index of the country at year t, RESVARi,t is the volatility of residual return estimated from the market model of regressing
the stock return of firm i against the stock market index of the country at year t, LOG(1+ ANALYSTi, t) is the natural log of number of analysts covering company
i at year t, LOG(VOLUMEi, t) is the natural log of trading volume of firm i at time t, LOG(SIZEi,t) is the natural log market capitalization of firm i at year t,
INVESTi,t is the investibility weight of firm i at year t, DISPERSIONi,t is a dummy variable that is equal to 1 if forecast dispersion is high, VOLATILITYi,t is the
standard deviation of the stock return of firm i at year t, and CDUM, YRDUM, and INDUM are dummy variables included to control for fixed effects representing
the countries, years, and industries in the sample. Equations (9) and (10) are estimated singularly using OLS or jointly with equation (5) using 2SLS. Only the
coefficients of non-dummy variables are reported. The t-statistics are reported in the parentheses. Adj R-square is the adjusted coefficient of determination. The
sample period is from 1993 to 1999 for a total of 6182 firm-year observations from 25 countries.
43
Table 9 (cont’d)
Panel A: GMM Estimation
INTERCEPT
LOG(ANALYST)
LOG(ANALYST)*
DISPERSION
LOG(VOLUME)
LOG(SIZE)
SYNCH
-2.00
(-13.63)
0.09
(2.05)
-0.06
(-3.72)
0.38
(18.63)
0.04
(2.03)
SYNCH
-1.83
(-14.18)
0.11
(2.74)
-0.07
(-3.81)
0.41
(25.44)
RESVAR
-2.24
(-56.02)
-0.03
(-3.10)
0.01
(2.83)
0.05
(10.45)
RESVAR
-2.83
(-71.54)
-0.12
(-10.23)
0.02
(3.72)
-0.03
(-7.21)
INTERCEPT
LOG(ANALYST)
LOG(ANALYST)*
DISPERSION
LOG(VOLUME)
-3.96
(-14.61)
1.13
(9.64)
-0.06
(-3.30)
0.29
(14.38)
Adj R-square
0.34
0.34
-0.15
(-27.20)
0.49
0.40
Panel B: 2SLS Estimation
LOG(SIZE)
Adj R-square
Equations (9) and (5)
SYNCH
LOG(ANALYST)
0.31
UNIDENTIFIED
Equations (10) and (5)
RESVAR
LOG(ANALYST)
-2.12
(-29.47)
-0.11
(-2.66)
0.01
(2.35)
0.05
(11.50)
UNIDENTIFIED
44
-0.14
(-21.66)
0.49