The comovement of investor attention
Michael S. Drake
Brigham Young University
mikedrake@byu.edu
Jared Jennings
Washington University in St. Louis
jaredjennings@wustl.edu
Darren T. Roulstone
The Ohio State University
roulstone.1@osu.edu
Jacob R. Thornock
University of Washington
thornocj@uw.edu
October 2014
Preliminary and Incomplete: Please do not cite or distribute
We thank Mary Barth (editor), an anonymous associate editor, and two anonymous referees for helpful comments
and suggestions. We thank Yung-Yu Chen for assistance in acquiring Google search data and Nick Guest for
research assistance. The financial support of the Fisher College of Business, Foster School of Business, Olin
Business School, and the Marriott School of Management is gratefully acknowledged.
Abstract
Prior literature has documented that investor attention is associated with the market’s pricing of
stocks. We measure comovement in investor attention: the extent to which investor attention
paid to a firm is explained by attention paid to the firm’s industry and to the market as a whole.
We then use attention comovement to help explain excess comovement in stock returns within in
an industry, i.e., the tendency of firm returns to move together more than can be explained by
comovement in firm fundamentals. We find that larger, more visible firms have higher attention
comovement and that, controlling for firm characteristics and prior determinants of return
comovement, attention comovement is positively associated with excess stock return and trading
volume comovement. Finally, we show that a prominent information release (a firm’s earnings
announcement) contributes to attention comovement. Our results have implications for
sentiment-based explanations of excess stock return comovement.
1
1. INTRODUCTION
Recent research demonstrates the importance of investor attention to the pricing of
stocks, providing consistent evidence that attention (inattention) to firm-specific information and
events is associated with positive (negative) capital market effects (e.g., Hirshleifer and Teoh
2003; Hirshleifer, Lim, and Teoh 2009; Drake, Gee, and Thornock 2014). Another stream of
research finds that stock returns comove in ways that fundamentals cannot fully explain,
prompting the investigation of behavioral explanations of excess return comovement (e.g.,
Kumar and Lee 2006; Barberis, Shleifer, and Wurgler 2005). This paper sits at the intersection
of these two literatures. Our research objectives are twofold. First, we investigate the extent to
which investor attention comoves (i.e., how much of a firm’s attention is explained by industryand market attention). We quantify attention comovement and investigate factors explaining the
level of attention comovement for a given firm. Second, we examine whether attention
comovement helps explain the excess comovement of stock returns documented in prior work.
The investor attention literature is based on the idea that attention is a limited resource.
That is, because investors cannot pay attention to all stocks or acquire and process all relevant
information, they must be selective about the particular stocks they choose to follow (Hirshleifer
and Teoh 2003). This literature has largely focused on firm-specific attention. 1 In this paper, we
extend the attention literature by estimating the extent to which firm-specific attention comoves
with industry/market attention. We conjecture that investor attention will comove as investors
collectively focus on similar firms and their correlated information flows. This notion follows
from a noise trader model, such as that in Barberis, Shleifer and Wurgler (2005), in which
1
For example, prior work employs firm-specific measures of attention, such as trading volume (Hou, Xiong, and
Peng 2009), past returns (Aboody, Lehavy, and Trueman 2010), 52-week highs (Li and Yu 2012), business press
articles (Da et al. 2011), or Google searches (Da et al. 2011; Drake, Roulstone and Thornock 2012) to identify firms
that are likely experiencing higher levels of investor attention.
2
different investor groups experience categorical or correlated sentiment, which in turn leads to
“sentiment-based” returns comovement. We extend the literatures on investor attention and
categorical sentiment by introducing the idea that investor attention will comove as investor
groups systematically seek out information for similar categorical stocks or as they experience
correlated shocks to the demand for information. To our knowledge, no prior study has explicitly
examined the concept of attention comovement, nor its capital market implications.
An intuitive category for categorical sentiment is industry. All else equal, we would
expect that firms in the same industry would have fundamentals and attention that move
together. To develop a measure of attention comovement, we follow a similar methodology to
the returns comovement literature to identify the amount of attention a firm receives that is
explained by industry and market-wide attention. We use four unique measures of the amount of
attention paid to a particular firm: (1) internet searches (i.e., Google search for a firm’s ticker),
(2) analyst forecast revisions, (3) business press articles, and (4) requests for a firm’s EDGAR
financial filings. We also compile a composite score of all four measures using factor analysis
and label the resulting factor “A-Score,” which is short for attention-score. We regress each of
our attention proxies on measures of industry and market attention, which are created by
aggregating firm-level measures within industries and across the market. We then use the R2
from these annual, firm-specific regressions as a measure of attention comovement—the greater
amount of firm-level attention that is explained by industry and market attention (i.e., higher R2),
the lower the amount of firm-specific attention paid to the stock. These measures capture the
comovement in firm attention with industry and market attention.
We conduct our analyses using a broad cross-section of 5,663 firm-year observations
from the 2007 to 2011 time period. We first provide evidence on the importance of
3
industry/market attention: we find that the average R2 from these regressions is 17.5%, ranging
from 9.9% for Google Internet searches to 25.2% for EDGAR financial filing searches. These
numbers reveal that a fair portion of the variation in firm-specific attention is explained by
variation in industry/market attention. In other words, some of the attention paid to a firm by
investors, analysts, and the business press is associated with attention paid to the firm’s industry
or the market in general. The implication of this result is that inferences in prior research
focusing on firm-specific attention are partially driven by industry/market attention.
Next, we examine factors that are related to attention comovement. We find that firms
with higher earnings comovement are more likely to have their attention comove with industry
and market attention, suggesting that comovement in firm fundamentals is positively related to
attention comovement. In addition, we find that, after controlling for the influence of
fundamentals comovement, attention comovement is higher for large firms and more visible
firms. This finding suggests that when investors are paying attention to a particular industry or
the market as a whole, they also concentrate on the larger, more visible firms within industries
and the market. For example, when investor attention moves to tech stocks, it implicitly falls on
the prominent players in the tech industry, such as Apple or Microsoft. On the other hand, we
find that institutional ownership is negatively related to attention comovement, suggesting that
attention comovement is negatively related to investor sophistication. Overall, the implication of
these findings is that attention comoves in excess of what is explained by comovement in firm
fundamentals and this excess comovement of attention is associated with firm characteristics.
The concept of categorical comovement has been proposed in prior research (Barberis et al 2005;
4
Morck et al 2000), but our operationalization and quantification of categorical attention is novel
to the literature. 2
Having identified the factors that are associated with cross-sectional variation in attention
comovement, we next investigate its capital market effects. In these analyses, we draw on the
returns comovement literature. The notion that the returns of related assets move in tandem is a
fundamental concept in asset pricing theory. Theoretically, in a well-functioning, frictionless
market, the comovement of asset prices should be tightly linked to comovement in asset
fundamentals. Early research documented that firm stock returns covary with both market and
industry returns (e.g., King 1966) and that firm-level earnings covary with industry and market
earnings (e.g., Ball and Brown 1967). The comovement in earnings is consistent with a
fundamentals-based explanation for the comovement in returns. However, recent research
documents that return comovement cannot be fully explained by comovement in fundamentals
(e.g., Barberis et al., 2005).
Subsequent comovement research focused on similarities in firm characteristics as an
explanation for return comovement. 3 These studies tend to identify categories within which
prices, which are the outcome of information acquisition and trading activities, comove. In other
words, prior research focuses on the outcome (excess return comovement) without necessarily
examining the information flows leading to that outcome. For example, Grullon et al (2014)
propose that firms sharing a lead underwriter comove because the lead underwriters share
information about these firms with investors; however, these information flows are not
2
Another way to distinguish our work is to note that comovement studies to date basically focus on finding
categories and showing that firms in the category comove with each other more than they comove with firms outside
the category. Our work is more like the traditional synchronicity literature (e.g., Morck et al. 2000), which
investigates determinants of the level of comovement within a particular category (in our case, industry).
3
These include similarities in size (Fama and French, 1993), the book-to-market ratios (Fama and French, 1993),
stock price (Green and Hwang 2009), industry membership (Kallberg and Pasquariello, 2007), stock index
membership (Barberis, Shleifer, and Wurgler, 2005), investment banking networks (Grullon, Underwood, and
Weston, 2014) and analyst following (Muslu et al 2014).
5
observable. In this study, we quantify the correlated actions of investors and information
intermediaries to gather and supply information about firms. Therefore, we can speak to the
information mechanisms that can influence excess returns comovement.
We find that comovement in investor attention is positively related to comovement in
returns and trading volume. 4 In other words, when the extent to which market participants pay
attention to a stock is strongly associated with their level of attention to the stock’s industry and
to the market in general, the stock’s returns and trading volume are more strongly linked to
industry and market-level returns and trading volume. These results hold after controlling for
comovement in firm fundamentals, as well as after controlling for firm characteristics that prior
research finds to be related to returns comovement. The implication of this finding is that the
correlated attention of investors and other market participants can affect comovement in asset
prices, an idea that is new to the literature.
An implication of these findings is that factors that draw attention to a particular firm
may also increase the amount of attention that investors pay to other firms that share a similar
categorical attribute, such as industry membership. For example, as earnings news draws
attention to a given firm, it may also draw attention to related firms. In our final set of tests, we
provide direct evidence on whether investors shift their attention based on significant firm events
within the industry. We find that earnings announcements also trigger significant increases in
investor attention for “peer” firms (other firms in the same industry). We also find that investor
attention is stronger for “peer” firms when the announcing firm’s attention comoves more with
other firms in the industry and market. This evidence is consistent with significant firm events
triggering investors to pay more attention to similar or related firms.
4
We extend prior work by examining trading volume as well as returns. As trading is necessary for price changes,
trading comovement should also be related to the attention paid to a firm; we document the existence this relation.
6
Our paper contributes to the literatures on investor attention, comovement in asset
prices/volume, and intra-industry information transfer. We introduce new measures of attention
and provide the first evidence on how firm-level attention varies with industry and market-level
attention. We document that attention has capital market effects: firms that are generally viewed
in the context of their industry and the market have stock returns and trading volume closely
linked to industry and market activity. Lastly, we provide some evidence that significant
corporate events, such as earnings announcements, trigger investor attention for other related
firms, which we define as firms that share the same industry.
2.
DATA, VARIABLES, AND SAMPLE
2.1.
Measures of Attention
Our analysis requires measures of investor attention to compute attention comovement.
To capture investor attention, we employ four state-of-the-art measures that have been used in
prior research: internet search volume, analyst forecast revisions, the count of business press
mentions for a particular firm, and the number of EDGAR downloads of a firm’s financial
statements. To calculate internet search volume, we obtain a proprietary database of investors’
weekly search activities made available by Google through their Google Trends application. 5
Google Trends tracks and reports the Search Volume Index (SVI), which is a relative measure of
user searches for a particular term on the Google search engine. 6 Following Da et al. (2011) and
Drake et al. (2012), we use ticker symbols as the search term for the firm which makes it more
5
We employ weekly rather than daily Google search data because it allows us to investigate a much broader sample
of firms. As discussed in Drake et al. (2012), daily Google search data is only available for larger, well-recognized
firms. Da et al. (2011) is another example of prior research using weekly Google search.
6
For example, the SVI for the term “super bowl” shows a strong increase in Google searches by users in January
and February of each year. http://www.google.com/trends/?q=super+bowl
7
likely that the user is searching for financial information, rather than for non-financial
information. 7 To control for the normal level of search volume in a firm, we remove from the
weekly SVI the median SVI over the past 8 weeks, following Da et al (2011). Thus, the variable
Googlei,w measures the abnormal level of internet search during the week for a particular firm.
Our second measure of attention, Analyst Fori,w, is equal to the count of earnings forecast
revisions made by sell-side analysts for a given firm-week. We obtain these data from the
I/B/E/S detail file. Following Frankel et al. (2006), we count the number of unique EPS (earnings
per share) revisions issued for each firm-week, including forecasts for all time horizons (i.e.,
earnings forecasts for one-year, two-year, and all other horizons are included in the Analyst
Fori,w variable). We count all revisions that are reported on the same date by the same analyst for
the same firm as a single analyst revision. We set all firm-weeks with no recorded analyst
revisions to zero as long as we are able to find some analyst coverage during that calendar year.
The third measure of attention is business press coverage (Mediai,w). We obtain business
press data from RavenPack, which provides data on business press coverage through the Dow
Jones (DJ) News Archives and Wall Street Journal (WSJ) (Drake, Guest and Twedt 2014). While
the data obtained from RavenPack is not comprehensive of all press outlets, the large circulation
of these two outlets makes them well suited to investigate market effects of the business press
(Tetlock 2007). The Mediai,w variable is equal to the number of articles issued by the business
press for each firm-week. We set all firm-weeks with no recorded news articles to zero as long as
we are able to find some media coverage during that calendar year.
Our fourth proxy for investor attention, Edgari,w, is the number of EDGAR downloads for
a given firm’s 10-Ks or 10-Qs during a given week. EDGAR is the SEC’s online repository for
7
In addition, we delete from our sample any ticker symbols with potential alternative meanings (e.g., ‘CAT,’
‘TOY,’ and ‘MAT’.
8
all SEC-mandated filings, among which are the 10-K (annual report) and 10-Q (quarterly report).
Drake et al. (2014) report that the 10-K and 10-Q reports are among the most highly downloaded
SEC filings. Therefore, we use the Edgari,w variable as a proxy for investor attention to financial
statements.
2.2.
Measures of Comovement
Our first objective is to examine the extent to which attention is an industry and market-
wide phenomenon. To achieve this objective, we employ a methodology similar to that used in
the returns comovement literature, which regresses firm returns on industry and market returns.
The explained variation (R2) from the model serves as the measure of comovement (e.g., Morck
et al. 2000). We do a very similar calculation with each of the attention measures previously
discussed. For each 52-week period prior to the fiscal year-end for a given firm, we estimate the
following regression using each attention measure for each firm-year:
FIRM ATTNi,w =
β0 + β1IND ATTNw + β2MARKET ATTN + εi,w
(1)
where i indexes firms and w indexes weeks. FIRM ATTN is one of the four measures of attention
described above: Googlei,w, Analyst Fori,w, Mediai,w, or Edgari,w. For each of these four
measures of attention, we also create an industry-level measure of attention, IND ATTNi,w, by
computing the value-weighted attention for the industry, defined by the two-digit SIC code, for a
given week. We also create a market-level measure of attention, MARKET ATTNi,w, for each of
the four attention variables by computing the value-weighted total attention for all sample firms
in a given week. We estimate equation (1) for each firm and fiscal period, requiring at least 45
weekly observations. To obtain an estimate of attention comovement, we take the log
transformation of the R2, similar to the approach in Morck et al (2000):
9
ATTN SYNCi,t =
log(1 + (R2/(1-R2)))
(2)
where R2 is the coefficient of determination from the firm-year estimation of equation (1). 8 We
compute a “SYNC” variable for each of the four attention variables – Sync Google, Sync Analyst,
Sync Media, and Sync Edgar – using this approach. By constructing the comovement measure in
this way, we can interpret increases in attention synchronicity variables as indicating that a given
firm’s attention is more closely tied to industry and market attention. In other words, higher
values of attention synchronicity serve as a proxy for higher attention comovement. 9
Once we calculate each of the attention synchronicity variables, we create a composite
attention synchronicity variable (Sync A-Scorei,t). We conduct a factor analysis with our four
attention synchronicity variables and retain the principal factor. 10 In our subsequent analyses, we
focus primarily on the results using the Sync A-Scorei,t variable because we believe that it best
captures the underlying construct of attention comovement.
We compute returns and volume comovement measures using a similar methodology.
We follow Piotroski and Roulstone (2004) and define the Sync Reti,t variable using the R2 from
the following regression calculated for each fiscal year end using the previous 52 weeks of data:
Reti,w =
ω + ω1Ind Reti,w + ω2Market Reti,w + ε
8
(3)
The log transformation allows a bounded variable—the R2 is bounded between 0 and 1—to take a continuous
form, and has been used in other studies that employ the R2 as a variable of interest (e.g., Piotroski and Roulstone
2004).
9
We refer to these variables as “synchronicity” variables because they are calculated in a similar manner to the
return synchronicity variables in Morck et al. (2000) and Piotroski and Roulstone (2004). We use the general term
“comovement” to refer to correlations in returns, trading volume, fundamentals, and our attention proxies across
firms in the same industry.
10
The results of principal factor analysis (untabulated) reveal that all four attention variables – Sync Googlei,t, Sync
Analysti,t, Sync Mediai,t, and Sync Edgari,t – converge to a single significant underlying factor with an Eigenvalue
greater than 1.0.
10
The Reti,w variable is the weekly return for firm i in week w. The Ind Reti,w variable is the valueweighed industry return (two-digit SIC code) in week w. The Market Reti,w variable is the valueweighted market return for week w. Similar to the attention synchronicity variables, we estimate
equation (3) for each firm and fiscal period, requiring at least 45 weekly observations. To obtain
an estimate of return synchronicity, we take the log transformation of the R2, similar to the
approach in Morck et al (2000):
Sync Reti,t =
log(R2/(1-R2))
(4)
We calculate stock turnover synchronicity, Sync Turni,w, in the same fashion. We define stock
turnover as the percentage of shares traded during the week.
In the prior literature, SYNC RET is often interpreted as a measure of how much firmspecific information is incorporated into a firm’s stock price: firms with high (low) SYNC RET
have lower (greater) amounts of firm-specific information in price relative to industry and
market related information. 11
2.3.
Data and Sample
We obtain data from seven different databases including Compustat (financial data),
CRSP (market data), IBES (analyst data), Thomson Reuters (institutional ownership data),
RavenPack (business press data), Google Trends (Google search data), and EDGAR (SEC filing
data). The intersection of these datasets results in 5,663 firm-year observations for the 2007 to
11
Other scholars interpret low SYNC RET as indicating a poor information environment, e.g., Teoh, Yong, and
Zhang (2009). Our paper, by linking SYNC RET with attention comovement, suggests that information does play a
role in determining SYNC RET.
11
2011 time period. 12 For selected analyses conducted at the firm-week level, these databases
yield a sample of 218,331 firm-week observations.
3.
RESULTS
3.1.
Attention Comovement
3.1.1. Descriptive Statistics and Correlations
Table 1, Panel A reports the descriptive statistics for all variables measured at the firmyear level and used to investigate the comovement of investor attention. Across the four
measures, EDGAR downloads of financial statements appear to exhibit the highest average
comovement (mean raw R2 = 25.2% and mean Sync Edgari,t = 0.347), followed by the
comovement of analyst forecast revisions (mean raw R2= 20.1% and Sync Analysti,t = 0.265),
business press mentions (mean raw R2 = 14.7% and Sync Mediai,t = 0.211), and internet searches
(mean raw R2 S= 9.9% and Sync Googlei,t = 0.138). We note that the average raw R2across the
four measures is 17.5%. We also note that the median firm in our sample is relatively large with
a market capitalization of 1.3 billion; one fourth of our sample firms are members of the S&P
500. We further note that the majority of firms are profitable (the 25th percentile for ROA is
above zero). The median firm has a book-to-market ratio of 0.545 (Bk/Mkti,t), sales growth of
5.6 percent (Sales Growthi,t), has an institutional ownership percentage of approximately 79
percent (Insti,t), and is followed by 7 sell-side analysts (# Analystsi,t). Table 2, Panel A presents
pairwise correlations and shows that the attention comovement measures are highly correlated
with each other, with correlation coefficients ranging from 33% to 62%. Each of the four
individual attention comovement measures is strongly positively associated with the composite
12
Our sample is constrained to the 2007 to 2011 time period due to the availability of SEC EDGAR and Google
search data.
12
attention comovement score (Sync A-Scorei,t), indicating a common attention factor among the
variables.
3.1.2. Determinants of Attention Comovement
We now turn to examining the determinants of attention comovement. The objective of
these analyses is to better understand why investors give individual attention to some firms while
viewing other firms primarily as members of an industry or market. We do this by regressing
attention comovement on comovement in firm fundamentals (proxied by earnings) and other
firm characteristics. By so doing, we are able to identify which firm characteristics are associated
with a firm’s attention being strongly (or weakly) associated with industry and market-wide
attention. The regression is specified as follows:
ATTN SYNCi,t =
δYEAR + δ1ROAi,t + δ2 Mkt Vali,t + δ3 Bk/Mkti,t + δ4 Sales Growthi,t
(5)
+ δ5 Insti,t + δ6 #Analystsi,t + δ7 Std ROAi,t + δ8 Abs(Reti,t)
+ δ9 Stk Turni,t + δ10 Earn Synci,t + δ11 SP 500i,t + δ12 Pricei,t, + ε,
where ATTN SYNC is one of five variables – Sync A-Scorei,t, Sync Googlei,t, Sync Analysti,t, Sync
Mediai,t, and Sync Edgari,t – as defined above and all independent variables are defined in
Appendix A. The unit of observation here is the firm-year, capturing the general association
between firm characteristics and attention comovement. We include year fixed effects to
account for macroeconomic differences and assess statistical significance using standard errors
that are clustered by firm.
We emphasize that attention comovement is a new construct and thus, our analysis in this
section is exploratory in nature. That being said, we do offer the following predictions with
respect to selected firm characteristics. First, we expect firms whose fundamentals track closely
with the industry and market to have attention that tracks closely with the industry and the
13
market. Thus, we predict a positive association between Sync Earni,t (which measures the
comovement of a firm’s return-on-assets with the firm’s industry-level return-on-assets) and our
measures of attention comovement. Similarly, firms whose fundamentals are more volatile or
idiosyncratic will receive attention that is more volatile and idiosyncratic, suggesting lower
attention comovement with the industry and market. Thus, we expect a negative coefficient on
Std ROAi,t.
Second, investors prefer to invest in securities with which they are familiar (Merton,
1987; Lehavy and Sloan 2008). This intuition, together with the fact that attention is a limited
resource (Hirshleifer and Teoh, 2003), suggests that investors are more likely to pay attention to
larger, “household name” stocks. Consistent with this intuition, prior research finds that larger,
more visible stocks are positively associated with Google searches (Drake at al., 2012), analyst
following (Bhushan, 1989), business press coverage (Bushee et al. 2010), and EDGAR requests
(Drake et al., 2014). Thus, we predict that large, visible stocks will also receive even greater
levels of investor attention when the industry/market receives attention. For example, investors
who allocate attention to “tech” stocks in general, will also likely allocate attention to “Apple,”
because Apple is such a visible, important player in the industry and in the market as a whole.
This is an example of what we mean by attention comovement. On the flip side, attention paid to
a small cap stock, that is not a visible stock in the industry, is likely related to something unique
about the particular firm and not to the industry in general. Thus, in model (3) we expect a
positive coefficient on variables that capture elements of firm size and visibility such as Mkt
Vali,t, Sales Growthi,t, Abs (Reti,t), ROAi,t, # Analysts, SP 500i,t, and Stk Turni,t.
Third, following a similar line of intuition, individual investors who have fewer resources
are more likely to follow large, visible stocks than institutional investors. That is, individual
14
investors are more likely to pay attention to the visible firms in a given sector than the smaller,
lesser known companies in the sector. Thus, we expect a negative coefficient on Insti,t as firms
with lower institutional ownership and greater individual ownership will have higher attention
comovement with the industry and market.
Fourth, we include variables that capture possible comovement with firms outside a given
firm’s industry. For example, Barberis et al. (2005) show that firms in the S&P 500 index have
returns that comove with other index members regardless of their industry affiliation. Thus,
comovement with industry (as measured by our attention synchronicity variables) may be lower
for members of the S&P 500. On the other hand, membership in the S&P 500 is another
indicator of size and visibility; thus, we have no clear prediction for the sign of the coefficient on
this variable). We also include Pricei,t in the regression as Green and Hwang (2009) find that
firms’ stock returns comove with those of other firms with similar prices.
In summary, we include multiple variables to capture firm size and visibility, investor
sophistication, and possible categories of comovement outside of industry membership. Given
our focus on investor attention, we expect firm size and visibility to be the primary drivers of
attention comovement in excess of that explained by comovement in fundamentals.
Table 3 presents the results of estimating equation (5). In column (1), the dependent
variable is the composite attention comovement score (Sync A-Scorei,t). In each of the subsequent
columns (columns 2 through 5), the dependent variable is one of the individual attention
comovement measures (i.e., Sync Googlei,t, Sync Analysti,t, Sync Mediai,t, and Sync Edgari,t). We
focus our discussion primarily on the Sync A-Scorei,t results in column (1). We find that the
explanatory power of the model is 38.4 percent, thus, approximately one-third of the variation in
15
attention comovement is explained by variation in firm-characteristics. 13 We also find that
earnings comovement (Sync Earni,t) is strongly related to attention comovement, suggesting that
firms with higher comovement in their fundamentals also have higher comovement in investor
attention. Remaining explanatory variables can be interpreted as explaining attention
comovement in excess of that associated with how well a firm’s fundamentals are related to
those of the firm’s industry and market.
We find that composite attention comovement is positively associated with the firm’s
market value (Mkt Vali,t) and book-to-market ratio (Bk/Mkti,t), indicating that large, value firms
receive attention that is more likely to comove with industry and market attention. We find that
attention comovement is positively related to membership in the S&P 500 (SP 500i,t), consistent
with these firms receiving proportionally more industry and market attention due to their
visibility within the industry and market (and despite the fact that members of this index tend to
comove with each other regardless of industry membership). Finally, we find that attention
comovement is negatively related to institutional holdings (Insti,t), suggesting that less
sophisticated investors (e.g., individual investors) are more likely to pay attention to an
individual firm when there is higher attention being paid to the firm’s industry and to the market
as a whole.
The results for the components of the Sync A-Scorei,t variable are reported in columns (2)
through (5) and are generally consistent with the coefficients reported in column (1). While only
the coefficients on firm size (Mkt Vali,t), institutional ownership (Insti,t), and earnings
synchronicity (Sync Earni,t) are significant with the same sign in all five regressions, in no case
do we find that a significant coefficient flips signs. In summary, we show that a firm’s attention
13
When we include the lag of Sync A-Scorei,t as an additional independent variable we find that all results are
qualitatively similar. The inclusion of Sync A-Scorei,t-1 as an additional independent variable increases the R2 to over
80%
16
comoves more with industry and market attention when the firm is more visible, has less
sophisticated ownership, and is viewed by investors to be a value stock.
3.1.3. Attention Comovement and Market Comovement
We next examine the association between attention comovement and stock market
comovement. Stock return comovement should be, according to asset pricing theory, driven by
the underlying fundamentals of the firm. However, prior research has shown that returns tend to
comove in excess of what should be explained by fundamental comovement (Barberis et al.
2005). Thus, other behavioral factors that appear to be associated with market comovement are
incremental to an explanation based on fundamentals.
As discussed earlier, we follow Piotroski and Roulstone (2004) and define the Sync Reti,t
and Sync Turni,t variables using the R2 from annual, firm-specific regressions of stock returns and
stock turnover on industry and market-level stock returns and stock turnover.
We examine whether variation in attention comovement incrementally explains variation
in market comovement using the following regressions:
Sync Reti,t =
§YEAR +§1 ATTN SYNCi,t +§2 ROAi,t +§3 Mkt Vali,t +§4 Bk/Mkti,t
(6)
+§5 Sales Growthi,t +§6 Insti,t +§7 #Analystsi,t +§8 Std ROAi,t
+§9 Abs(Reti,t) +§10 Stk Turni,t +§11 Earn Synci,t +§12 SP 500i,t
+§13 Pricei,t +ε
Sync Turni,t =
λ§YEAR +λ1 ATTN SYNCi,t +λ2 ROAi,t +λ3 Mkt Vali,t +λ4 Bk/Mkti,t
+λ5 Sales Growthi,t +λ6 Insti,t +λ7 #Analystsi,t +λ8 Std ROAi,t
(7)
+λ9 Abs(Reti,t) +λ10 Stk Turni,t +λ11 Earn Synci,t +λ12 SP 500i,t
+λ13 Pricei,t +ε
where all variables are as defined above and in Appendix A. The models presented in equation
(6) and (7) include year fixed effects and statistical significance is assessed using standard errors
17
clustered by firm. In these models, our primary variable of interest is ATTN SYNCi,t, which
captures the association between our market-based comovement measures (Sync Reti,t and Sync
Turni,t variables) and attention comovement, after controlling for the comovement in firm
fundamentals (proxied by earnings comovement) and other firm characteristics associated with
stock return and trading volume synchronicity.
Panel A of Table 4 presents the results from estimating equation (6), with Sync Reti,t as
the dependent variable. In Column (1), we find that the coefficient on Sync A-Scorei,t is equal to
0.1460 and is significant at the 1% level, providing evidence that attention comovement is
positively associated with return comovement. In Columns (2) through (5), we include each
individual attention synchronicity variable separately. Without exception, we find a positive and
significant coefficient on each of the attention synchronicity variables. The coefficients range
from a low of 0.1429 on the Sync Googlei,t variable and a high of 0.3796 on the Sync Analysti,t
variable. The results provide consistent evidence of a positive relation between returns
comovement and measures of attention comovement, even after controlling for comovement in
fundamentals.
We note that the sign of the coefficients on the control variables are consistent with
expectations and/or the prior literature (e.g., Piotroski and Roulstone, 2004). For example, we
find a positive coefficient on firm size (Mkt Vali,t), institutional ownership (Insti,t) and earnings
synchronicity (Sync Earni,t) as well as a negative coefficient on earnings volatility (Std ROAi,t),
consistent with Piotroski and Roulstone (2004). We also find that growth firms (Sales Growthi,t),
glamour stocks (Bk/Mkti,t), firms with higher stock prices at the fiscal year end (Pricei,t), firms
that are part of the S&P 500 (SP500i,t), and firms with higher returns during the fiscal year
(Abs(Reti,t)) are less likely to have synchronous returns.
18
Panel B of Table 4 reports the results from estimating equation (7) with Sync Turni,t as
the dependent variable. Similar to the results in Panel A, we find a positive and significant
coefficient on Sync A-Scorei,t. In addition, we find a positive and significant coefficient for each
of the other attention synchronicity variables presented in columns (2) through (5). The
coefficients in columns (2) through (5) range between 0.1428 (Sync Googlei,t) and 0.6165 (Sync
Analysti,t). These results suggest that stock turnover synchronicity can be partially explained by
the synchronicity of investor attention, after controlling for comovement in fundamentals and for
other important firm characteristics. We note that the control variables are relatively consistent in
explaining both returns and stock turnover synchronicity.
Overall, the results in Table 4 indicate that the comovement of investor attention is
associated with the comovement of the firm’s returns and trading volume. Put another way,
firms that receive attention even when investors are not searching for information about industry
peers and when analysts and the media are not forecasting or writing about industry peers, have
returns and trading volume that are less related to the returns and trading volume of their industry
or the market. From an economic standpoint, the magnitude of the relation between attention
comovement and return/trading comovement is plausible given that a one standard deviation
increase in Sync A-Scorei,t is associated with an increase in Sync Reti,t (Sync Turni,t) of 0.1346
(0.1992) standard deviations.
3.2.
Attention comovement around important corporate events
A stream of studies examines the intuition that information announced by one firm can be
informative to investors in related firms, a phenomenon often called information transfer (e.g.,
Foster 1981; Han, Wild and Ramesh 1989; Ramnath 2002; Thomas and Zhang 2008). These
studies show that the earnings announcement for a particular firm can lead to a significant
19
market reaction for related firms. The theory behind intra-industry information transfer is that
economic events that drive one firm’s news are often related to economic events at peer firms
(e.g., firms in the same industry). An untested underlying assumption in this literature is that
investor attention comoves, as the earnings news for a given firm draws investor attention to
related firms.
We build on the information transfer literature and use the setting of intra-industry
earnings news as an alternate way to show the phenomenon of attention comovement.
Specifically, in our final set of tests, we extend prior research by investigating whether earnings
announcements also trigger investor attention in related (“peer”) firms. We also examine
whether the relation between investor attention and a peer firm’s earnings announcement is
explained, in part, by the extent to which the firm’s general attention comoves with industry and
market attention.
3.2.1. Descriptive Statistics
In the analyses above, we employed firm-year level data; for these analyses, we change
the unit of observation to the firm-week, allowing us to more precisely focus on investor
attention around a corporate announcement. Further, in order to more precisely identify “peer”
firms, we identify related firms as those operating in the same four-digit SIC code as a given
firm. 14 Table 1, Panel B reports the descriptive statistics for all variables measured at the firmweek level. Here we find that average weekly abnormal Google search (Googlei,w) is 0.004, with
a standard deviation of 0.157. The average number of analyst forecast revisions for a firm-week
14
We use four-digit SIC code to identify firms that are most likely to have similar operations. In addition, we use
four-digit SIC codes to limit the number of peer firm-weeks for a single earnings announcement. Using four-digit
SIC codes, we identify approximately 30% of all firm-week observations to be weeks in which peer firms announce
earnings. Making the industry definition broader (e.g., two-digit SIC code) will increase this percentage
considerably and reduce the power of our tests because a much higher percentage of firm-weeks become weeks in
which peer firms announce earnings.
20
(Analyst Fori,w) is 1.106, with a median of 0 revisions. The average number of media articles
(Mediai,w) is 17.82, with a median of 6 articles per firm-week. The average number of EDGAR
downloads (Edgari,w), is 90 per firm-week, with a median of 59 downloads. Similar to our
calculation of Sync A-Scorei,t, we calculate a composite attention variable, A-Scorei,w, by
identifying the principal factor using the four previously defined attention variables. 15 We
present the Pearson correlations for our variables in Panel B of Table 2. For the most part, each
of the attention variables are positively related to the other attention variables. For example, the
correlations between Mediai,w, Analyst Fori,w, and Edgari,w range from 27% to 58%. The
Googlei,w variable is positively related to the other measures, but to a lesser degree. The AScorei,w variable is positively related to each of the individual attention variables, except for the
Googlei,w variable.
3.2.2. Peer firm investor attention around earnings announcements
We first investigate whether the earnings release of an “announcing firm” triggers
attention in a peer firm by regressing separately each of the five weekly attention variables – AScorei,w, Googlei,w, Analyst Fori,w, Mediai,w, and Edgari,w – on an indicator variable for the week
of the firm’s earnings release (EA) and an indicator for the week of a peer firm’s earnings
announcement (Peer EA):
ATTNi,w= σFIRM + σ1 EAi,w + σ2 Peer EAi,w + σ3 Mkt Vali,q + σ4 Bk/Mkti,q + σ5 Sales
Growthi,q + σ6 ROAi,q + σ7 Insti,q + σ8 Uncertaintyi,q + σ9 Compi,q,
+ σ10 MOMi,q + ε.
Where ATTNi,w is one of the five attention measures and all other variables are as previously
defined in the text and Appendix A. In addition to the earnings announcement indicator
15
The results of principal factor analysis (untabulated) reveal that all four attention variables – Googlei,w, Analyst
Fori,w, Mediai,w, and Edgari,w – converge to a single significant underlying factor with an Eigenvalue greater than
1.0.
21
(8)
variables, we include controls for firm size (Mkt Vali,q ), book-to-market, sales growth, return-onassets, institutional ownership, analyst uncertainty (Uncertaintyi,q), the competitiveness of the
firm’s industry (Compi,q ), and recent stock return performance (MOMi,q). Since the level of
attention a firm receives is likely to be highly correlated through time, we also include firm fixed
effects in the model. We assess statistical significance using standard errors clustered by week.
An intuitive way to view equation (8) is that we are regressing the weekly level of
attention paid to a firm (say, Coca Cola), on an indicator variable for the week in which Coke
announces earnings (EAi,w) and an indicator variable for the week in which a peer firm (say,
PepsiCo) announces earnings (Peer EAi,w), along with control variables. A significantly positive
coefficient on Peer EAi,w indicates that attention comoves across related firms around important
corporate announcements. In equation (8), we include the EAi,w variable as attention for a firm
should increase during the week that firm releases its own earnings. We also include control
variables for firm characteristics that could be associated with investor attention. Lastly, we
delete all firm-week observations in which the firm and another peer firm in the industry both
announce earnings. When these two events happen simultaneously, it becomes more difficult to
understand whether the increase in attention is related to the firm’s earnings announcement or the
peer firm’s earnings announcement.
Table 5 presents the results of estimating equation (8). Again, in column (1) we present
the results using A-Scorei,w as the dependent variable. In each of the subsequent columns (2)
through (5), the dependent variable is one of the individual attention variables. Across all five
regressions we find that the coefficient on EAi,w is positive and significant, confirming that
investor attention significantly increases during the week of a firm’s earnings announcement.
Further, the coefficient on PEER EAi,w is positive and statistically significant for all measures of
22
attention. This suggests that investors’ attention for a given firm increases when a peer firm
releases earnings to the public. That is, these results provide additional evidence for the
comovement of attention across firms within an industry and provide a source for some of that
comovement: firm information releases spark attention both for the disclosing firm and for peer
firms in the same industry.
The coefficient on EAi,w, captures the increase in investor attention for a particular firm
when that firm announces earnings and provides a benchmark against which we can assess the
economic significance of the increase in investor attention observed when a peer firm announces
earnings. We take the ratio of the coefficient on the PEER EAi,w variable to the coefficient on the
EAi,w variable in the same model specification. Higher values of this ratio are consistent with
greater levels of investor attention comovement. This ratio is equal to 6.9% when the AScorei,w is the dependent variable and ranges between 2.0 percent for analyst forecasts (column
(3)) and 43.2 percent for Google searches (column (2)). We interpret these results as suggesting
an economically meaningful relation between peer firm announcements and attention given to a
firm.
In our final set of tests, we estimate equation (8) separately for two subsamples: firms
with low attention comovement (below the Sync A-Scorei,t sample median) and firms with high
attention comovement (above the Sync A-Scorei,t sample median). The objective of this test is to
investigate whether the positive relation between investor attention and a peer firm’s earnings
announcement is conditional on the extent to which the firm’s attention generally comoves with
that of firms in the industry.
In Table 6 we present results with A-Scorei,w as the dependent variable. For firms with
low and high levels of attention comovement, we find a positive and significant coefficient on
23
Peer EAi,w, consistent with the results presented in Table 5. However, we find that the
coefficient on Peer EAi,w for the low attention comovement subsample is 0.0834 and that the
coefficient on Peer EAi,w for the high attention comovement subsample is 0.1503. 16 This
represents an 80% higher level of information transfer for the high attention comovement
subsample. A Chow F-test confirms that these coefficients are statistically different from each
other at the 1% level. This finding suggests that firms with investor attention that is generally
more strongly associated with the amount of attention paid to their industry, experience greater
increases in investor attention when a peer firm in that industry discloses news to the public. In
other words, firms that receive more (less) firm-specific attention are less (more) likely to
receive attention when firms in their industry release news. Our results support the idea that
significant firm events can trigger investors to pay attention to other firms that are in the industry
and that this relation is stronger for firms with greater industry and market attention
comovement.
4.
CONCLUSION
This paper investigates the extent to which the amount of attention a firm receives from
investors and other market participants comoves with the amount of attention paid to their
industry and the market as whole. We also examine the capital market impact of this attention
comovement. We find that approximately one-fifth of the variation in firm-specific attention is
explained by industry- and market-attention. Further, we identify specific firm characteristics
that are related to the level of this attention co-movement. We find that attention comovement is
positively related to earnings comovement (a proxy for the comovement in firm fundamentals)
16
In untabulated analyses, we confirm that the results are inferentially similar for each of the individual attention
variables when examining the differences in the coefficients on the Peer EAi,w variable.
24
and firm visibility. We also show that the comovement in attention is positively associated with
the comovement in stock returns and trading volume, suggesting that comovement in stock
returns and trading outcomes is partially driven by the actions of investors who view individual
firms in the context of categories such as industry. This finding is consistent with “sentiment”
explanations for return comovement like those discussed in Barberis et al. (2005). Finally, we
document that an important information event (earnings announcements) can increase attention
for related firms, once again suggesting an industry component to attention. We document that
this effect is more pronounced for firms that are more likely to receive attention at the same time
as other firms in the industry and market.
Our results suggest that information flows (proxied by our attention measures) help
explain comovement in capital market outcomes (i.e., returns and trading volume). This finding
is consistent with the arguments in Barberis et al. (2005) that comovement in returns is driven by
investors categorizing firms according to similar characteristics, trading subsets of stocks rather
than individual stocks, and information diffusion across stocks occurring at different rates for
stocks in different categories. We show that when information flows for a stock are focused on
the stock and are not explained by industry and market information flows, returns and trading
volume for the stock are more idiosyncratic. While we cannot establish a causal relation
between attention comovement and return/trading volume comovement, we believe it unlikely
that return/trading volume comovement drives attention comovement. Thus our results are
suggestive of a causal relation from attention to returns and trading volume. Future research can
continue to explore the characteristics and events associated with these forms of comovement as
well as the implications of their relation.
25
References
Aboody, D., R. Lehavy, and B. Trueman, 2010, “Limited Attention and the Earnings
Announcement Returns of Past Stock Market Winners.” Review of Accounting Studies,
15: 317-344
Ball, R. and P. Brown, 1967, “Some Preliminary Findings on the Association between the
Earnings of a Firm, its Industry, and the Economy.” Journal of Accounting Research, 5:
55-77
Barberis, N., A. Shleifer, and J. Wurgler, 2005, “Comovement.” Journal of Financial
Economics, 75: 283-317
Bushee, B., J. Core, W. Guay, and S. Hamm, 2010, “The Role of the Business Press as an
Information Intermediary.” Journal of Accounting Research, 48(1): 1-19.
Barron, O., D. Byard, and O. Kim, 2002, “Changes in Analysts’ Information around Earnings
Announcements.” The Accounting Review, 77: 821-846.
Barron, O., O. Kim, S. Lim, and D. Stevens, 1998, “Using Analysts’ Forecasts to Measure
Properties of Analysts’ Information Environment.” The Accounting Review, 73: 421-433.
Bhushan, R., 1989, “Firm Characteristics and Analyst Following.” Journal of Accounting and
Economics, 11(2-3): 255-274.
Da, Z., J. Engelberg, and P. Gao, 2011, “In Search of Attention.” Journal of Finance, 66(5):
1,461-1,499.
Drake, M., N. Guest, and B. Twedt, 2014, “The Media and Mispricing: The Role of the Business
Press in the Pricing of Accounting Information.” The Accounting Review, 89(5): 16731701.
Drake, M., K. Gee, and J. Thornock, 2014, “March Market Madness: The Impact of Valueirrelevant Events on the Market Pricing of Earnings News,” forthcoming, Contemporary
Accounting Research.
Drake, M., D. Roulstone, and J. Thornock, 2012, “Investor Information Demand: Evidence from
Google Searches around Earnings Announcements.” Journal of Accounting Research,
50(4): 1,001-1,040.
Fama, E. and K. French, 1993, “Common Risk Factors in the Returns on Stocks and Bonds.”
Journal of Financial Economics, 33(1): 3-56.
26
Foster, G., 1981, “Intra-Industry Information Transfers Associated with Earnings Releases.”
Journal of Accounting and Economics, 3(3): 201-232.
Frankel, R., S.P. Kothari, and J. Weber, 2006, “Determinants of the Informativeness of Analyst
Research.” Journal of Accounting and Economics, 41(1-2): 29-54.
Green, T. and B. Hwang, 2009, “Price-Based Return Comovement” Journal of Financial
Economics, 93(1):37-50.
Grullon, G., S. Underwood, and J. Weston, 2014, “Comovement and Investment Banking
Networks.” Journal of Financial Economics, 113(1): 73-89.
Han, J., J. Wild, and K. Ramesh, 1989, “Managers’ Earnings Forecasts and Intra-Industry
Information Transfers.” Journal of Accounting and Economics, 11(1): 3-33.
Hirshleifer, D. and S. Teoh, 2003, “Limited Attention, Information Disclosure, and Financial
Reporting.” Journal of Accounting and Economics, 36:337-386.
Hirshleifer, D., S. Lim, and S. Teoh, 2009, “Driven to Distraction: Extraneous Events and
Underreaction to Earnings News.” Journal of Finance, 64(5): 2,289-2,325.
Hou, K., L. Peng, and W. Xiong, 2009, “A Tale of Two Anomalies: The Implications of Investor
Attention for Price and Earnings Momentum.” Working paper, The Ohio State
University.
Kallberg, J. and P. Pasquariello, 2007, Time-Series and Cross-Sectional Excess Comovement in
Stock Indexes.” Journal of Empirical Finance, 15(3): 481-502.
King, B., 1966, “Market and Industry Factors in Stock Price Behavior.” Journal of Business, 39:
139-190.
Kumar, A. and C. Lee, 2006, “Retail Investor Sentiment and Return Comovements.” Journal of
Finance, 61(5): 2,451-2,486.
Lehavy, R. and R. Sloan, 2008, “Investor Recognition and Stock Returns.” Review of Accounting
Studies, 13(2-3): 327-361.
Li, J. and J. Yu, 2012, “Investor Attention, Psychological Anchors, and Stock Return
Predictability.” Journal of Financial Economics, 101: 401-419.
Merton, R., 1987, “A Simple Model of Capital Market Equilibrium with Incomplete
Information.” Journal of Finance, 42: 483-510.
27
Morck, R., B. Yeung, and W. Yu, 2000, “The Information Content of Stock Markets: Why Do
Emerging Markets have Synchronous Price Movements?” Journal of Financial
Economics, 58(1-2): 215-260.
Muslu, V., M. Regbello, and Y. Xu, 2014, “Sell-Side Analyst Coverage and Stock
Comovement.” Forthcoming, Management Science.
Piotroski, J. and D. Roulstone, 2004, “The Influence of Analysts, Institutional Investors, and
Insiders on the Incorporation of Market, Industry, and Firm-Specific Information into
Stock Prices.” The Accounting Review, 79(4): 1,119-1,151.
Ramnath, S., 2002, “Investor and Analyst Reactions to Earnings Announcements of Related
Firms: An Empirical Analysis.” Journal of Accounting Research, 40(5): 1,351-1,376.
Teoh, S., Y. Yang, and Y. Zhang, 2009, “R2 and Market Efficiency” Working Paper, University
of California-Irvine.
Tetlock, P., 2007, “Giving Content to Investor Sentiment: The Role of the Media in the Stock
Market.” Journal of Finance, 62: 1,139-1,168.
Thomas, J. and F. Zhang, 2008, “Overreaction to Intra-Industry Information Transfers?” Journal
of Accounting Research, 46(4): 909-940.
28
Appendix A
Variable
# Analystsi,t
Abs(Reti,t)
Analyst Fori,w
A-Scorei,w
Bk/Mkti,q
Bk/Mkti,t
Compi,q
EAi,w
Edgari,w
Googlei,w
Insti,q
Insti,t
Mediai,w
Mkt Vali,q
MOMi,q
MVEi,t
Peer EAi,w
Pricei,t
ROAi,q
ROAi,t
Sales Growthi,q
Sales Growthi,t
Std ROAi,t
Stk Turni,t
Sync Analysti,t
Sync A-Scorei,t
Sync Earni,t
Sync Edgari,t
Sync Googlei,t
Definition
equal to the number of analysts following firm i in year t.
equal to the buy and hold abnormal monthly return for firm i in fiscal year t.
equal to the number of analyst forecasts issued in week w for firm i.
a factor analysis using the Googlei,w, Analyst Fori,w, Mediai,w, and Edgari,w variables to
identify the common factor and calculate the A-Scorei,t variable for firm i during week
w.
equal to the book value of equity divided by the market value of equity for firm i in
quarter q.
equal to the book value of equity divided by the market value of equity for firm i in
year t.
equal to one minus the Herfindahl Hirschman Index, which is calculated at the annual
level. We first divide the firm’s annual sales by the total industry (4-digit SIC code)
sales then square the ratio. We then sum the squared ratio by industry.
equal to one during the week firm i announces earnings and zero otherwise.
equal to the number of searches for firm i's 10-Ks or 10-Qs through the Edgar search
platform in week w.
equal to the number of google searches for firm i in week w less the median number of
google searches in the previous eight weeks for firm i.
equal to the percentage of firm i's shares held by institutional investors in quarter q
and equal to zero if missing.
equal to the percentage of firm i's shares owned by institutions in year t.
equal to the number of news articles that mentioned firm i in week w.
equal to the market value for firm i calculated at the end of the fiscal quarter q.
equal to the buy and hold daily return during quarter q-1 for firm i.
equal to the market value of equity for firm i at the end of fiscal year t.
equal to one for firm i in week w if a peer firm (i.e., firm in the same four-digit SIC
code) announces earnings in week w and zero otherwise.
is equal to the stock price for firm i at the end of fiscal year t.
equal to net income before extraordinary items in quarter q scaled by total assets in
quarter q-4 for firm i.
equal to net income for firm i in year t scaled by total assets in year t-1.
equal to the sales in quarter q divided by sales in quarter q-4 for firm i.
equal to sales in year t divided by sales in year t-1 for firm i.
equal to the standard deviation of return on assets between year t-4 and t for firm i.
equal to the stock turnover for firm i in year t.
calculated similarly to the Sync Googlei,t variable using the number of analyst
forecasts issued during week w for firm i.
a factor analysis using the Sync Googlei,t, Sync Analysti,t, Sync Mediai,t, and Sync
Edgari,t variables to identify the common factor and calculate the Sync A-Scorei,t
variable for firm i during year t.
equal to the logarithmic transformation, defined as log(R2 / (1 - R2)), of the R2 from a
regression of the firm's return on assets on a value-weighted industry index of ROA
using quarters t-11 through t.
calculated similarly to the Sync Googlei,t variable using the number of searches for 10Ks or 10-Qs during week w for firm i.
equal to the logarithmic transformation of the R2, defined as log(R2 / (1 - R2)), using
the following model using data over the 52 week period prior to the fiscal year end for
firm i in year t: Googlei,w = α + α Ind Googlei,w + Mkt Googlei,w + ε, where Googlei,w
is equal to the Google search volume for firm i in week w, Ind Googlei,w is equal to the
value weighted Google search volume for industry j (defined as two digit SIC code) in
week w, and the Mkt Googlei,w is equal to the value-weighted google search volume
29
Sync Mediai,t
Sync Reti,t
Sync Turni,t
Uncertaintyi,q
for all firms with Google search volume in week w.
calculated similarly to the Sync Googlei,t variable using the number of media articles
written during week w for firm i.
calculated similarly to the attention synchronicity variables using returns for firm i
during week w.
calculated similarly to the attention synchronicity variables using stock turnover for
firm i during week w.
equal to the average squared analyst forecast error for firm i during quarter q, as
presented in Barron, Kim, Lim and Stevens (1998) and Barron, Byard and Kim
(2002). Specifically, we compute the variable with the following equation: (1 – (1/N))
* D + SE, where N is the number of analysts following the firm, D is the dispersion of
analyst forecasts, and SE is the squared error in the mean forecast.
30
Table 1
Descriptive statistics.
Panel A - Attention synchronicity sample descriptive statistics
Variable
N
Mean
Std Dev
25th Pctl
50th Pctl
75th Pctl
Sync A-Scorei,t
5,663
-0.062
0.903
-0.578
-0.334
0.092
Sync Googlei,t
5,663
0.138
0.342
0.021
0.053
0.121
Sync Analysti,t
5,663
0.265
0.316
0.043
0.147
0.373
Sync Mediai,t
5,663
0.211
0.415
0.032
0.086
0.206
Sync Edgari,t
5,663
0.347
0.392
0.098
0.233
0.454
i,t
5,663
0.099
0.149
0.020
0.052
0.114
Analyst R2i,t
5,663
0.201
0.192
0.042
0.137
0.311
i,t
5,663
0.147
0.184
0.031
0.082
0.186
Edgar R2i,t
5,663
0.252
0.199
0.093
0.208
0.365
Sync Reti,t
5,663
-0.347
1.134
-0.952
-0.211
0.424
Sync Turni,t
5,663
-1.180
1.421
-2.006
-1.002
-0.168
ROAi,t
5,663
0.033
0.120
0.006
0.042
0.088
Mkt Vali,t
5,663
5,773
14,272
384
1,289
3,982
Bk/Mkti,t
5,663
0.662
0.538
0.320
0.545
0.848
Sales Growthi,t
5,663
1.074
0.247
0.957
1.056
1.154
Insti,t
5,663
0.729
0.227
0.600
0.787
0.905
# Analystsi,t
5,663
8.820
6.959
3.000
7.000
13.000
Std ROAi,t
5,663
0.062
0.087
0.015
0.032
0.072
Abs(Reti,t)
5,663
0.362
0.367
0.120
0.259
0.485
Stk Turni,t
5,663
0.029
0.021
0.015
0.023
0.036
Sync Earni,t
5,663
0.256
0.377
0.020
0.102
0.321
SP 500i,t
5,663
0.250
0.433
0.000
0.000
1.000
Pricei,t
5,663
28.016
22.743
10.790
23.110
38.870
2
Google R
2
Media R
31
Panel B - Weekly attention descriptive statistics
Variable
N
Mean
Std Dev
25th Pctl
50th Pctl
75th Pctl
A-Scorei,w
218,331
-0.002
0.988
-0.737
-0.134
0.616
Googlei,w
218,331
0.004
0.157
-0.016
0.000
0.015
Analyst Fori,w
218,331
1.106
2.048
0.000
0.000
1.000
Mediai,w
218,331
17.815
34.574
2.000
6.000
18.000
Edgari,w
218,331
89.963
100.466
32.000
59.000
107.000
EAi,w
218,331
0.020
0.141
0.000
0.000
0.000
Peer EAi,w
218,331
0.350
0.477
0.000
0.000
1.000
Mkt Vali,q
218,331
7,262
16,910
685
1,900
5,344
Bk/Mkti,q
218,331
0.599
0.461
0.300
0.497
0.776
Sales Growthi,q
218,331
1.087
0.280
0.956
1.060
1.173
ROAi,q
218,331
0.012
0.031
0.003
0.012
0.025
Insti,q
218,331
0.779
0.189
0.680
0.820
0.926
Uncertaintyi,q
218,331
0.132
0.544
0.002
0.010
0.043
Compi,q
218,331
0.767
0.197
0.695
0.821
0.906
MOMi,q
218,331
0.024
0.231
-0.109
0.019
0.139
Panel A includes the descriptive statistics for the attention synchronicity sample. Panel B includes
descriptive statistics for the weekly attention sample. All variables are defined in Appendix A. All
variables are winsorized at the 1st and 99th percentile.
32
33
34
Table 3
Determinants of attention synchronicity
ROAi,t
Ln(Mkt Vali,t)
Bk/Mkti,t
Sales Growthi,t
Insti,t
Ln(# Analystsi,t)
Std ROAi,t
Abs(Reti,t)
Stk Turni,t
Sync Earni,t
SP 500i,t
Pricei,t
Intercept
# Observations
R2
Sync A-Scorei,t
Sync Googlei,t
0.0866
0.6818
0.2417***
7.4988
0.1282***
3.6697
-0.0549
-1.6126
-0.6363***
-6.2271
0.1044***
3.2935
-0.0623
-0.4624
0.0223
0.8278
0.2716
0.2731
0.4174***
5.0424
0.2609***
3.8079
-0.0011
-0.7920
-1.7078***
-10.7835
0.0084
0.1797
0.0575***
3.9172
0.0159
0.9982
0.0082
0.5924
-0.1390***
-2.6245
0.0029
0.2119
0.0267
0.4604
0.0107
0.9545
0.2009
0.4022
0.0841**
2.1902
0.0251
0.9113
-0.0008
-1.0688
-0.2210***
-3.1144
Sync
Analysti,t
0.0534
1.2446
0.0488***
5.4825
0.0386***
3.6086
-0.0466***
-3.6019
-0.1940***
-6.1230
0.0850***
7.2955
-0.1049**
-2.0457
-0.0096
-0.9771
-0.2097
-0.6475
0.1403***
5.4797
0.0741***
3.2271
0.0003
0.7950
-0.1257***
-2.7272
5,663
0.3840
5,663
0.0900
5,663
0.2990
Sync Mediai,t
Sync Edgari,t
0.0251
0.3810
0.1075***
6.0387
0.0486**
2.3729
-0.0051
-0.3108
-0.2613***
-5.4913
-0.0155
-0.9110
-0.0057
-0.0880
0.0034
0.2804
-0.0418
-0.0921
0.1916***
4.0767
0.0878***
2.6442
-0.0006
-1.0071
-0.4080***
-4.7211
-0.0044
-0.0828
0.1033***
8.4188
0.0467***
3.9175
-0.0020
-0.1145
-0.2437***
-6.0112
0.0336***
2.7798
0.1184*
1.8060
0.0387***
3.1789
0.2889
0.7134
0.1953***
5.5717
0.1145***
4.3885
-0.0007
-1.3233
-0.3957***
-6.5432
5,663
0.2500
5,663
0.3720
All variables are defined in Appendix A. T-statistics are presented below the coefficient estimates.
Standard errors are clustered by firm. Year indicator variables are included in the regression but
the coefficient outputs are suppressed for brevity. *, **, *** indicate statistical significance at
10%, 5%, and 1%.
35
Table 4
Relation between the synchronicity of attention and the synchronicity of market variables
Panel A - Synchronicity of returns on the synchronicity of attention
Sync A-Scorei,t
Sync Reti,t
0.1460***
6.5328
Sync Googlei,t
Sync Reti,t
Sync Reti,t
Sync Reti,t
0.1429***
3.2437
Sync Analysti,t
0.3796***
6.8216
Sync Mediai,t
0.2049***
4.3952
Sync Edgari,t
ROAi,t
Ln(Mkt Vali,t)
Bk/Mkti,t
Sales Growthi,t
Insti,t
Ln(# Analystsi,t)
Std ROAi,t
Abs(Reti,t)
Stk Turni,t
Sync Earni,t
SP 500i,t
Pricei,t
Intercept
# Observations
R2
Sync Reti,t
0.0452
0.3059
0.3775***
17.2605
0.1896***
5.5211
-0.1949***
-3.2006
0.5037***
6.0842
0.0189
0.5993
-0.6911***
-3.4743
-0.2040***
-5.5694
0.9124
0.9114
0.2525***
6.2939
-0.4444***
-7.5517
-0.0022**
-2.3140
-3.7930***
-24.6686
0.0567
0.3805
0.4046***
19.1334
0.2061***
6.0502
-0.2041***
-3.3407
0.4306***
5.2521
0.0337
1.0765
-0.7040***
-3.5313
-0.2023***
-5.4439
0.9234
0.9216
0.3015***
7.1910
-0.4099***
-6.8157
-0.0022**
-2.3297
-4.0108***
-27.1770
0.0376
0.2540
0.3943***
18.6407
0.1937***
5.7602
-0.1852***
-3.0485
0.4845***
5.8924
0.0019
0.0602
-0.6604***
-3.2961
-0.1971***
-5.3854
1.0317
1.0426
0.2602***
6.5587
-0.4345***
-7.4386
-0.0024***
-2.6096
-3.9947***
-27.1395
0.0527
0.3552
0.3908***
18.0111
0.1984***
5.7500
-0.2019***
-3.3029
0.4643***
5.5817
0.0373
1.1832
-0.6991***
-3.4958
-0.2014***
-5.4400
0.9607
0.9530
0.2742***
6.7669
-0.4243***
-7.1279
-0.0022**
-2.3233
-3.9588***
-26.4659
0.2159***
4.5043
0.0588
0.3971
0.3905***
17.9893
0.1983***
5.8101
-0.2025***
-3.3152
0.4634***
5.6316
0.0269
0.8544
-0.7258***
-3.6392
-0.2091***
-5.6755
0.8897
0.8861
0.2713***
6.6032
-0.4310***
-7.2146
-0.0022**
-2.2908
-3.9570***
-26.5124
5,663
0.4250
5,663
0.4180
5,663
0.4240
5,663
0.4200
5,663
0.4200
36
Panel B - Synchronicity of stock turnover on the synchronicity of attention
Sync A-Scorei,t
Sync Turni,t
0.2205***
9.2606
Sync Googlei,t
Sync Turni,t
Sync Turni,t
Sync Turni,t
0.1428***
2.7807
Sync Analysti,t
0.6165***
10.9741
Sync Mediai,t
0.2977***
6.5921
Sync Edgari,t
ROAi,t
Ln(Mkt Vali,t)
Bk/Mkti,t
Sales Growthi,t
Insti,t
Ln(# Analystsi,t)
Std ROAi,t
Abs(Reti,t)
Stk Turni,t
Sync Earni,t
SP 500i,t
Pricei,t
Intercept
# Observations
R2
Sync Turni,t
0.2771*
1.7029
0.3912***
16.4274
0.0559
1.4256
-0.2606***
-3.5398
0.4176***
4.3000
-0.0504
-1.2318
-1.1736***
-5.4375
-0.2249***
-4.6430
1.3334
1.1520
0.2062***
4.6020
-0.1241**
-1.9659
0.0007
0.6897
-3.9276***
-23.6071
0.2950*
1.7846
0.4363***
18.6552
0.0819**
2.0996
-0.2739***
-3.6893
0.2971***
3.0068
-0.0278
-0.6692
-1.1911***
-5.4760
-0.2215***
-4.4898
1.3646
1.1527
0.2863***
6.3327
-0.0702
-1.0746
0.0006
0.5496
-4.2727***
-26.2198
0.2633
1.6215
0.4144***
17.9495
0.0604
1.5758
-0.2440***
-3.2929
0.3969***
4.1183
-0.0797*
-1.9282
-1.1226***
-5.2001
-0.2140***
-4.4122
1.5226
1.3130
0.2118***
4.8875
-0.1123*
-1.7988
0.0003
0.2556
-4.2267***
-26.3604
0.2887*
1.7590
0.4125***
17.3170
0.0697*
1.7802
-0.2712***
-3.6575
0.3551***
3.5972
-0.0227
-0.5538
-1.1856***
-5.4563
-0.2209***
-4.5092
1.4058
1.1933
0.2413***
5.3545
-0.0927
-1.4384
0.0006
0.6433
-4.1827***
-25.5612
0.3163***
6.3580
0.2976*
1.8162
0.4119***
17.2094
0.0694*
1.7779
-0.2721***
-3.6796
0.3544***
3.6042
-0.0379
-0.9259
-1.2248***
-5.5943
-0.2322***
-4.7359
1.3019
1.1185
0.2365***
5.2555
-0.1028
-1.5956
0.0007
0.6667
-4.1791***
-25.5668
5,663
0.3880
5,663
0.3770
5,663
0.3890
5,663
0.3810
5,663
0.3800
All variables are defined in Appendix A. T-statistics are presented below the coefficient estimates.
Standard errors are clustered by firm. Year indicator variables are included in the regression but the
coefficient outputs are suppressed for brevity. *, **, *** indicate statistical significance at 10%, 5%,
and 1%.
37
Table 5
Attention comovement around corporate events
EAi,w
Peer EAi,w
Ln(Mkt Vali,q)
Bk/Mkti,q
Sales Growthi,q
ROAi,q
Insti,q
Uncertaintyi,q
Compi,q
MOMi,q
Intercept
# Observations
R2
A-Scorei,w
1.6849***
85.1948
0.1171***
5.5146
0.1690***
7.4287
0.3690***
16.1936
-0.0226*
-1.7693
-0.4146***
-3.3353
-0.0659
-1.5229
0.0136***
3.4049
-0.2128***
-6.2837
0.0722**
2.4098
-1.3466***
-8.0048
Ln(Googlei,w)
0.0125***
3.7710
0.0054***
3.0120
0.0025
1.2335
-0.0008
-0.3808
0.0010
0.5915
0.0127
0.7779
-0.0088*
-1.8643
-0.0013*
-1.7463
0.0103
1.3433
-0.0038
-1.1602
-0.0329**
-2.0137
Ln(Analyst Fori,w)
1.6529***
149.1949
0.0323***
3.1441
0.0445***
3.7464
0.0971***
7.5402
-0.0075
-0.9127
-0.2454***
-3.2852
0.1648***
7.2279
0.0088***
2.7836
0.0249
0.9869
-0.0195
-0.9959
-0.0927
-1.0070
Ln(Mediai,w)
1.5664***
85.3333
0.1387***
8.2155
0.2853***
12.4873
0.2281***
12.0653
0.0625***
3.6777
-0.6473***
-4.8491
0.0811**
2.0096
0.0247***
4.8050
-0.0401
-0.9792
-0.0381
-1.1546
-0.3902**
-2.2942
Ln(Edgari,w)
0.3591***
12.7822
0.0995***
3.2232
0.0817***
2.7795
0.4803***
15.7751
-0.0842***
-4.6991
-0.0619
-0.4391
-0.4019***
-6.5471
-0.0011
-0.2814
-0.4451***
-12.0318
0.2014***
4.7314
3.8920***
18.0400
218,331
0.690
218,331
0.068
218,331
0.358
218,331
0.630
218,331
0.613
All variables are defined in Appendix A. T-statistics are presented below the coefficient estimates. Standard errors are clustered by
week. Firm indicator variables are included in the regression but the coefficient outputs are suppressed for brevity. *, **, ***
indicate statistical significance at 10%, 5%, and 1%.
38
Table 6
Attention comovement around corporate events partitioned on attention synchronicity.
A-Scorei,w
EAi,w
Peer EAi,w
Ln(Mkt Vali,q)
Bk/Mkti,q
Sales Growthi,q
ROAi,q
Insti,q
Uncertaintyi,q
Compi,q
MOMi,q
Intercept
# Observations
R2
Low Synci,t
1.7534***
95.2568
0.0834***
4.6049
0.1388***
7.2844
0.2731***
14.4710
-0.0231**
-2.0740
-0.3910***
-3.8201
-0.0558
-1.4478
0.0073
1.5707
-0.3977***
-8.3793
0.0590**
2.5068
-0.3075*
-1.6989
High Synci,t
1.5816***
60.2263
0.1503***
6.7030
0.1738***
5.9617
0.4512***
16.0645
-0.0228
-1.3825
-0.3434
-1.6287
-0.1525**
-2.3782
0.0173***
3.5440
-0.0084
-0.2021
0.0802**
2.0289
-1.7742***
-6.7978
109,143
0.5680
109,188
0.7190
Difference
P-Value
-0.172***
<0.001
0.0669***
<0.001
All variables are defined in Appendix A. T-statistics are presented below the coefficient estimates.
Standard errors are clustered by week. Firm indicator variables are included in the regression but the
coefficient outputs are suppressed for brevity. *, **, *** indicate statistical significance at 10%, 5%,
and 1%.
39