University of Southern California Dissertations and Theses

“What you see is all there is”: The effects of media co‐coverage on investors’ peer selection

(USC Thesis Other)

“What you see is all there is”: The effects of media co‐coverage on investors’ peer selection

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

“What You See is All There is”:
The Effects of Media Co-coverage on Investors’ Peer Selection

by

Jingjing Xia

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BUSINESS ADMINISTRA TION)

May 2018

Copyright 2018 Jingjing Xia

Acknowledgement
I am grateful to my co-chairs Mark Soliman and Sarah Bonner for their insights and
encouragement throughout the program. I am also greatly indebted to Maria Ogneva for her
valuable feedback on my works. This study has also benefited from the comments of Randolph
Beatty, Patricia Dechow, Mark DeFond, Shane Heitzman, Clive Lennox, Shelly Li, Tracie Majors,
Kenneth Merchant, Richard Sloan, Lorien Stice-Lawrence, K. R. Subramanyam, T. J. Wong, Mark
Young, and participants at the USC Accounting Research Forum. Finally, I thank all my fellow PhD
colleagues for their support and help over the years. The Dissertation Completion Fellowship from
USC Graduate School is gratefully acknowledged. All errors are my own.

ii

Table of Contents

Acknowledgement ii
List of Tables iv
Abstract v

1. INTRODUCTION ................................................................................................................................ 1
2. LITERATURE REVIEW AND HYPOTHESES DEVELOPMENT ................................................... 8
2.1. The Role of Media in Financial Markets ........................................................................................... 8
2.2. Media Co-coverage and Investors’ Peer Selection ............................................................................ 9
2.3. High Media Co-coverage Peers as Benchmarks .............................................................................. 12
3. SAMPLE AND RESEARCH DESIGN .............................................................................................. 14
3.1. Data .................................................................................................................................................. 14
3.2. Context-Driven Media Co-coverage ................................................................................................ 15
3.2.1. Purging the Confounding Effects of Common Economic Events ............................................. 15
3.2.2. Purging the Effects of Other Confounding Variables ............................................................... 16
3.3. Research Design .......................................................................................................................... 18
3.3.1. Media Co-coverage and Investors’ Peer Selection ................................................................... 18
3.3.2. High Media Co-coverage Peers as Benchmarks ....................................................................... 21
4. EMPIRICAL RESULTS ..................................................................................................................... 22
4.1. Descriptive Statistics ........................................................................................................................ 22
4.2. Media Co-coverage and Investors’ Peer Selection .......................................................................... 23
4.2.1. Reaction to Peers’ Earnings News and Media Co-coverage ..................................................... 23
4.2.2. Overweighting of Peers’ Information and Media Co-coverage ................................................ 24
4.3. High Media Co-coverage Peers as Benchmarks .............................................................................. 25
4.3.1. Beating the Highest Co-coverage Peer—A Future Performance Signal? ................................. 25
5. ROBUSTNESS CHECKS .................................................................................................................. 32
6. CONCLUSION ................................................................................................................................... 33
REFERENCES ........................................................................................................................................... 34
Appendix A. Variable Definitions .............................................................................................................. 37
Appendix B. Examples of News Articles with Unidentifiable Economic Events ...................................... 40

iii

List of Tables

Table 1. RavenPack Data-Filtering Steps ................................................................................................... 41
Table 2. Summary Statistics for Media Co-coverage Regression ............................................................... 42
Table 3. Descriptive Statistics ..................................................................................................................... 43
Table 4. Reaction to Early-Announcing Peers’ Earnings News and Media Co-coverage .......................... 44
Table 5. Overweighting High Co-coverage Peers’ Information ................................................................. 45
Table 6. High Media Co-coverage Peers as Benchmarks ........................................................................... 46
Table 7. Beating the Highest Co-coverage Peer and Future Returns .......................................................... 47
Table 8. Cross-Sectional Analysis .............................................................................................................. 49

iv
Abstract
Benchmarking with industry peers is ubiquitous in financial markets, yet relatively little is
known about investors’ peer selection process. This paper examines media co-coverage as a factor
in peer selection. An industry peer’s frequent co-appearance with a firm in media news articles can
increase the peer’s salience and highlight its association with the firm. In the presence of
information gathering and processing frictions, increased salience may cause investors to pay more
attention to, and even overweight signals from high media co-coverage peers. The empirical
evidence is consistent with this conjecture. First, investors attend more to information from high
co-coverage peers—a non-announcing firm’s stock price reacts more positively to the earnings
surprise of the early-announcing industry peer with higher media co-coverage. Second, investors
overweight signals from high co-coverage peers—earnings surprise of the highest co-coverage peer
negatively predicts the firm’s announcement return in the same quarter, which is consistent with a
correction of the initial overreaction to the highest co-coverage peer’s earnings news. In addition, I
find that media co-coverage distorts investors’ benchmarks—investors overreact to a firm’s
earnings announcement if its sales growth exceeds the early-announcing peer with the highest
media co-coverage. By documenting the economic consequences of investors’ (suboptimal) peer
selection, this paper contributes to prior literature that has mainly focused on developing
technologies that identify the most appropriate peers.
Keywords: media, benchmarking, industry peers, earnings announcements, market efficiency
v
1. INTRODUCTION
Benchmarking with industry peers is commonplace in financial statement analysis and
valuation.
1
Its popularity can be attributed to simplicity—it is what Bhojraj and Lee (2002) call a
“satisficing device” (Simon 1972) that produces reasonably relevant information without imposing
the high costs of a more complex analysis. Prior research has mainly focused on identifying the
“most appropriate” peers for benchmarking purpose.
2
However, relatively little is known about
which peers investors actually use. In this paper, I investigate one potential factor in investors’
peer selection—media co-coverage—and the associated economic consequences.
3
Specifically, I
examine (1) whether the frequency with which same-industry firms co-appear in media news
articles (i.e., media co-coverage) affects investors’ peer selection, and (2) whether investors use
peers with high media co-coverage efficiently in benchmarking.
The financial media plays an important role in capital markets (e.g., Fang and Peress 2009).
Prior studies suggest that media coverage can affect investors’ trading activities by directing their
attention to salient news covered by the press (e.g., Engelberg and Parsons 2011; Peress 2014),
and may have both positive (e.g., Fang and Peress 2009; Tetlock 2010; Peress 2008) and negative
effects (e.g., Tetlock 2007; Gurun and Butler 2012; Engelberg, Sasseville, and Williams 2012;
Hillert, Jacobs and Muller 2014) on the price discovery process.
The extant literature has mainly focused on individual firms’ media coverage. However,
the financial media also frequently co-covers multiple firms, especially industry peers, in the same
1
For example, peers are widely used by analysts to justify their target price of a firm (e.g., De Franco, Hope, and
Larocque 2015) and by investment banks in initial public offering (IPO) valuations (e.g., Paleari, Signori, and Vismara
2014), fairness opinions (e.g., DeAngelo 1990) and leveraged buyouts (e.g., Kaplan and Ruback 1995).
2
For example, Bhojraj and Lee (2002) identify peers for equity valuation based on a “warranted multiple.” Hoberg
and Phillips (2010, 2016) select peers based on the textual similarity in firms’ business descriptions in 10-K filings.
A recent study by Lee, Ma, and Wang (2015) identifies peers based on EDGAR co-search traffic. Other methods use
similarities in industry inputs and outputs (Fan and Lang 2000) and common analyst coverage (Kaustia and Rantala
2013).
3
In this paper, I use “peers” and “industry peers” interchangeably.
1

news article. Furthermore, although the media may co-cover same industry firms in the same
article because of their economic relatedness or some common economic events that involve the
co-mentioned firms (e.g., mergers and acquisitions), it may also co-cover industry peers for purely
contextual reasons. For example, Microsoft, a technology giant that develops a wide range of
products and services from computer software to video games, was co-covered with Yahoo!, a
web-service provider, in a 2008 Wall Street Journal article, not necessarily because they were the
most comparable peers to each other, but because their top executives happened to attend the same
conference (Crovitz 2008).
4
I term this type of co-coverage “context-driven” co-coverage.
5
More
formally, it is defined as media co-coverage that is not driven by the firms’ economic relatedness
or common economic events that affect the firms’ future cash flows and/or discount rates, but
occurs for contextual reasons to facilitate storytelling.
Although context-driven co-coverage may not always involve the most comparable peers,
it may still affect investors’ peer selection by increasing the salience of the co-covered peer. An
average firm may have many potential peers. Peers may also change as the firm adjusts its product
portfolio or enters into new sectors. However, bounded rationality due to limited resources (such
as time) (e.g., Simon 1972) may inhibit investors from identifying and tracking each and every
relevant peer. Therefore, investors may need to resort to a “satisficing” peer-selection strategy,
such as focusing only on the salient peers. Ample evidence suggests that investors attend more to
salient firms while overlooking less salient ones (e.g., Barber and Odean 2008; Fang and Peress
2009; Grullon, Kanatas, and Weston 2004; Lou 2014). Since media co-coverage can increase a
peer’s salience and highlight its association with the firm, investors are likely to pay more attention

4
Appendix B provides relevant quotes from the original news articles.
5
In the rest of the paper, I use “context-driven co-coverage,” “media co-coverage,” and “co-coverage”
interchangeably.
2

to peers with higher co-coverage. Furthermore, as humans are prone to overweighting salient
signals that they could easily retrieve at the time of decision-making—a phenomenon called “what
you see is all there is” (Kahneman 2011)—investors may over-extrapolate signals from high co-
coverage peers, leading to biased expectations about the firm and distortions in price discovery.
I investigate whether investors attend more to information from high co-coverage peers
through a firm’s stock price reaction to a peer’s earnings surprise. Prior research suggests that
early-announcing industry peers’ earnings news can be a relevant signal for investors to update
their expectations of the non-announcing firm (e.g., Foster 1981; Han and Wild 1990; Freeman
and Tse 1992) and that, on average, good news for the peer has positive implications for the firm
(e.g., Kim, Lacina and Park 2008, Wang 2014). I therefore hypothesize that a firm’s price reaction
to an early-announcing peer’s earnings surprise is more positive when the peer has higher media
co-coverage. In addition, I investigate whether investors overweight the earnings surprise of the
early-announcing peer with the highest media co-coverage when updating their expectations about
the firm’s current-quarter earnings. Specifically, I seek evidence of price reversal around the firm’s
earnings announcement by testing whether the earnings surprise of the highest co-coverage peer
is negatively associated with the firm’s announcement return in the same quarter.
After examining the effects of media co-coverage on investors’ processing of peer
information, I move on to investigate whether and how investors use high co-coverage peers as
benchmarks in firm evaluation. Humans tend to make decisions based on gains and losses vis-à-vis
a benchmark (i.e., a reference point) rather than the final outcome (Kahneman and Tversky 1979).
Thus, the benchmark is critical in the evaluation process because it determines what is considered a
“gain” or a “loss”, or more generally, what is “good” or “bad.” Prior research on investors’ evaluation
of firm performance suggests that the market evaluates a firm more favorably if its performance
exceeds, or “beats”, certain benchmarks, such as analysts’ forecasts (e.g., Kasznik and McNichols
3
2002; Kirk, Reppenhagen, and Tuckeer 2014). I propose that, if investors pay more attention to
industry peers with higher media co-coverage, they are also likely to use such peers as benchmarks.
However, as high co-coverage peers are not necessarily the most comparable peers, doing so may
bias their evaluations.
I test whether investors use high co-coverage peers as benchmarks in the setting of earnings
announcements. Specifically, I investigate whether investors evaluate a firm’s sales growth
performance by comparing it to the early-announcing peer with the highest media co-coverage. I
focus on sales growth because it is an easy-to-calculate performance metric that is relevant to the
investors and comparable across firms. If investors benchmark a firm against the highest co-
coverage peer, I expect the firm’s earnings announcement return will be more positive if its sales
growth beats the highest co-coverage peer. However, if on average, the highest co-coverage peer
is not the most comparable peer, using it as a benchmark may bias investors’ reaction to the firm’s
earnings announcement. If the bias is gradually corrected in the post-announcement period, a
firm’s post-announcement return will be more negative if its sales growth is higher than the highest
co-coverage peer.
I test my hypotheses using a sample of approximately 30,674 quarterly earnings
announcements made between 2003 and 2012 by the U.S. firms covered by RavenPack. The
sample period is limited by the availability of data necessary to calculate the media co-coverage
measure. To isolate the effects of context-driven co-coverage, I employ a two-step process. The
first step is to eliminate the confounding effects of common economic events (e.g., the initiation
of a joint venture) on media co-coverage and investors’ choice of peers through sample selection.
Specifically, I only keep in the analysis non-press-release news articles for which an underlying
4

economic event cannot be identified.
6
In addition, I only include firms that do not play a key role
in the news story and are instead just passively mentioned in the article.
7
These two filters can
reasonably ensure that any documented effects are not entirely attributable to economic events that
involve both the firm and the co-covered peer.
In the second step, I use a quarterly cross-sectional regression to purge the effects of other
factors that may simultaneously affect media co-coverage and investors’ peer selection.
Specifically, in each quarter, I regress the raw co-coverage variable (the number of times both firm
i and industry peer j are mentioned in the same article, scaled by the number of articles that mention
firm i) on the following variables: (1) peer firm’s own salience, i.e., its overall media coverage; (2)
peer designation by other market participants, including the frequency with which a peer is
mentioned in the firm’s press releases, analyst co-coverage, and the EDGAR website co-searches;
(3) firm similarity, i.e., similarity of firms’ business descriptions in 10-K filings (Hoberg and
Phillips 2010, 2016) and return synchronicity; (4) economic relatedness along the supply chain;
and (5) unobservable peer characteristics that remain constant across time, captured by peer fixed
effects.
8
Context-driven co-coverage is measured as the error term of this regression.

6
I rely on the RavenPack database for event classification of news articles. News articles with an unidentifiable
underlying event are mostly editorials and commentaries on the performance of the overall market or specific sectors
and companies or reports on non-economic events such as industry conferences. Appendix B provides examples of
news articles for which the underlying economic event cannot be identified.
7
RavenPack assigns a relevance score to each firm mentioned in a story. The score ranges from 0 to 100, and a higher
score indicates higher prominence of the firm in the news. I include in the analysis only firms with a relevance score
less than 50.
8
A potential issue with this measure is that firms with higher (lower) raw co-coverage may have lower (higher)
context-driven co-coverage. For example, suppose firm A’s raw co-coverage with peer B is 50% and their context-
driven co-coverage is -10%. Firm A’s raw co-coverage with peer C is 20%, and the context-driven co-coverage is
10%. In this case, although firm A has higher context-driven co-coverage with peer C than with peer B, investors are
likely to find peer B more salient than peer C. However, empirically this issue does not significantly bias the context-
driven co-coverage measure—the observations with context-driven co-coverage in the highest tercile and raw co-
coverage in the lowest tercile or with context-driven co-coverage in the lowest tercile and raw co-coverage in the
highest tercile constitute only 1.2% of the whole sample.
5
Using this context-driven media co-coverage measure, I find empirical results consistent
with my hypotheses. First, investors attend more to the earnings news of peers with higher media
co-coverage—a firm’s stock price reaction to an early-announcing peer’s earnings surprise is more
positive when the peer has higher context-driven media co-coverage. Moving from the bottom to
the top decile of media co-coverage increases the sensitivity of the non-announcing firm’s prices
(one-day market-adjusted return) to the early-announcing peer’s earnings news by 142%. Second,
investors appear to overweight the surprise of the peer with the highest media co-coverage as a
signal of the firm’s current-quarter earnings—the earnings surprise of the highest co-coverage peer
is negatively associated with the firm’s own announcement return in the same quarter, consistent
with a price reversal of a prior overreaction to the surprise of the highest co-coverage peer. A long-
short strategy based on the surprise of the highest co-coverage peer generates a one-day four-factor
(market, SMB, HML, and momentum) alpha of 0.437 basis points, or a 43.7% abnormal annual
return.
9
Overall, these findings lend support to my hypotheses that investors pay more attention,
perhaps even excessive attention, to high media co-coverage peers.
After documenting that media co-coverage can indeed affect the way investors process peer
information, I move on to provide evidence consistent with investors using the highest co-coverage
peers as a benchmark in firm evaluation. Specifically, I find that a firm’s earnings announcement
return is more positive if its sales growth exceeds the early-announcing peer with the highest media
co-coverage, even if beating the highest co-coverage peer is not a signal of better future
performance. In addition, the positive return to beating the highest co-coverage peer reverses in
the post-announcement period, indicating a price correction. These results are robust to the
inclusion of a battery of control variables that have been shown to affect stock returns, including
9
In my sample, this strategy can be implemented for an average of 100 trading days per year.
6

but not limited to, the earnings news of other industry peers, the firm’s own earnings performance,
its exposure to systematic risk, and other firm characteristics as well as alternative research designs.
Taken together, empirical evidence suggests that media co-coverage can influence investors’ peer
selection, which in turn affects their evaluation of firm performance and ultimately market prices.
This study contributes to the literature in several ways. First, it adds to the research on peer
selection. Prior studies have mainly focused on developing technologies to select the “most
appropriate” peers for benchmarking (e.g., Fan and Lang 2000; Bhojraj and Lee 2002; Hoberg and
Phillips 2010, 2016; Kaustia and Rantala 2013). However, relatively little is known about which
peers investors actually use. One exception is a recent study by Lee, Ma, and Wang (2015) that
infers financial statement users’ peer selection from EDGAR co-search traffic. They show that
EDGAR users’ “collective wisdom”—as revealed by their co-search patterns—can be used to
identify peers that dominate other industry classification schemes (e.g., six-digit GICS) in
grouping fundamentally similar firms. This paper complements their findings by suggesting that
investors may not always use the most comparable peers due to the influence of the financial media,
and that such suboptimal peer selection may lead to stock price distortions. Second, this paper adds
to the emerging literature on the role of media in financial markets. While previous studies have
primarily focused on media coverage and tend to consider each firm in isolation, this study
examines the effects of co-coverage and suggests that it can shape the way investors choose peers.
Finally, this study adds to the research on firms’ meeting or beating various benchmarks (e.g.,
Burgstahler and Dichev 1997; Degeorge, Patel, and Zeckhauser 1999; Kasznik and McNichols 2002).
Prior research has mainly focused on the effects of beating analysts’ forecasts, the firm’s own past
earnings, and the break-even point. This study suggests that investors may also benchmark a firm
7

against salient peers with high media co-coverage and that doing so may result in temporary
mispricing.
2. LITERATURE REVIEW AND HYPOTHESES DEVELOPMENT
2.1. The Role of Media in Financial Markets
The financial media serves as an important information intermediary in capital markets.
Prior studies suggest that media coverage catches investors’ attention and could significantly affect
their trading. For example, Engelberg and Parsons (2011) exploit a setting in which investors have
access to different media coverage of the same information by examining the effects of local
newspaper coverage on local investor trading. Peress (2014) shows that investors’ trading activities,
as measured by trading volume, return dispersion, and volatility, decreased significantly during a
newspaper strike.
However, there is mixed evidence regarding how the media affects the price discovery
process. On one hand, media coverage could alleviate information frictions by disseminating
information to a broader audience. Fang and Peress (2009) find that firms with high media
coverage have lower cost of equity because of higher transparency. Coverage in the media is also
shown to reduce information asymmetry (Tetlock 2010) and facilitate the incorporation of
information in stock prices (Huberman and Regev 2001; Peress 2008). On the other hand, media
coverage may exacerbate investor biases and lead to price distortions. For example, several studies
have shown that the tone of media news could contribute to investor sentiment and overconfidence,
resulting in systematic mispricing in the stock market (e.g., Tetlock 2007; Gurun and Butler 2012;
Engelberg, Sasseville, and Williams 2012; Hillert, Jacobs, and Muller 2014).
Besides documenting the effects of media coverage on the trading and pricing of stocks,
prior research also provided some preliminary evidence on the determinants of individual firms’
8

coverage. For example, Solomon and Soltes (2011) find that firm size, industry, the timing and
characteristics of firm disclosures, the management’s public relations efforts, and the objectives of
the media can all affect journalists’ decisions about whether to cover a firm. Specifically, they find
that larger firms, firms in consumer-focused industries, and those offering more contacts to
journalists receive more coverage. Firm disclosures that are released during business hours, convey
more surprises, and are more negative attract more media attention, though the effects are stronger
for print newspapers than newswires. Overall, extant literature suggests that the media can have a
significant effect on the stock market, and that firms’ media coverage may be affected by both
journalists’ incentives and managers’ efforts.
2.2. Media Co-coverage and Investors’ Peer Selection
Although prior research has mainly focused on individual firms’ media coverage, the press
also frequently co-covers multiple firms, especially those in the same industry, in the same news
article. Co-coverage of industry peers can occur in a variety of contexts, not all of which require
the co-mentioned firms to be the most comparable peers to each other or be involved in the same
economic event (e.g., mergers and acquisitions or lawsuits). For example, the media routinely
publish commentaries on the performance of the overall market or specific sectors and firms,
editorials with in-depth coverage of the careers of notable entrepreneurs, and reports of non-
economic events such as conferences. In these instances, co-coverage likely depends more on the
context of the story than the firms’ economic comparability. Therefore such “context-driven” co-
coverage may not always involve the most comparable peers to a firm.
Nevertheless, context-driven co-coverage may still affect investors’ peer selection. Due to
bounded rationality and limited resources (e.g., time) (e.g., Simon 1972; Kahneman 2011),
investors may not be able to identify and follow every relevant peer for a firm. Instead, they are
9
likely to focus on the salient peers as a “satisficing” strategy—a heuristic that could produce a
reasonably relevant peer without imposing the high search and analysis costs. As co-appearance
in the same news article can increase a peer’s salience and highlight its association with the firm,
investors are likely to pay more attention to peers with higher media co-coverage. Furthermore, as
humans tend to make judgments and decisions based on a small set of salient information that they
could easily retrieve at the time of decision-making rather than all relevant information, their
judgments and decisions often reflect an overweighting of salient signals—a phenomenon termed
“what you see is all there is” (Kahneman 2011). Since information from high co-coverage peers is
salient, they are likely to over-extrapolate signals from such peers when updating expectations about a
firm.
To examine whether media co-coverage indeed affects investors’ processing of peer
information, I need a way to gauge the extent to which investors regard a firm as a peer. As industry
peers’ earnings news can be a relevant signal for investors to revise their expectations about the
non-announcing firm (e.g., Foster 1981; Han and Wild 1990; Freeman and Tse 1992), and that on
average, good news for the peer has positive implications for the firm (e.g., Kim, Lacina and Park
2008, Wang 2014), I infer investors’ de facto choice of peers through a firm’s market price reaction
to early-announcing peers’ earnings announcements.
10 , 11
Specifically, if investors pay more
attention to the earnings news of high co-coverage peers and if, on average, the information
transfer between the firm and the peer is positive, a non-announcing firm’s stock price should react
10
An alternative method is to examine whether media co-coverage affects the likelihood of a peer being “co-searched”
with the firm on the EDGAR website (e.g., Lee et al. 2015). However, EDGAR users are only a fraction of financial
statement users and tend to be more sophisticated, while I’m interested in the broader market. Thus, this method may
not be suitable for this study.
11
Prior research has documented both positive and negative information transfers among same industry firms, with
the former driven by industry commonalities and the latter by competitive threats (e.g., Kim, Lacina, and Park 2008;
Koo, Wu, and Yeung 2017). However, the average reaction of a non-announcing firm to an early-announcing industry
peer’s earnings news is positive (e.g., Kim, Lacina, and Park 2008; Wang 2014), suggesting that industry commonality
is the dominant factor in determining investors’ reactions to industry peers’ earnings news.
10

more positively to the earnings surprise of peers with higher media co-coverage. Therefore I
formulate the following hypothesis:
H1a: Ceteris paribus, a firm’ s price reaction to early-announcing peers’ earnings surprise is more
positive when the peer has higher media co-coverage with the firm.
If investors over-extrapolate signals from high co-coverage peers, they are likely to
overweight the earnings surprise of such peers when updating expectations about the non-
announcing firm. As early-announcing peers’ surprises are more relevant as a signal for the firm’s
current quarter earnings than its future earnings, I expect the bias to pertain mostly to the firm’s
current quarter earnings. When the firm’s current quarter earnings are realized, the biased
expectation is likely to be corrected, resulting in a price reversal on the firm’s own announcement
day. Based on these arguments, I formulate the following hypothesis:
H1b: Ceteris paribus, the relationship between a firm’ s earnings announcement return in the same
quarter and the earnings surprise of the early-announcing peer is more negative when the peer
has higher media co-coverage with the firm.
However, context-driven co-coverage may not necessarily affect investors’ reaction to
peers’ earnings news. First, it is likely that investors identify peers based on similarity in business
activities and other economically relevant characteristics, rather than co-appearance in news
articles. Second, to the extent that retail investors, who may not have sufficient knowledge and
resources to conduct a more complete analysis to identify the “best” peers, are more likely to rely
on media co-coverage to select peers than institutional investors, the cognitive bias of a subset of
investors may not be able to significantly distort stock prices in the presence of arbitrage.
Nevertheless, arbitrage can be costly. Market frictions, such as transaction costs, liquidity, and
uncertainty about potential returns, could prevent professional arbitrageurs from fully exploiting
11

the mispricing (e.g., Pontiff 1996; Shleifer and Vishny 1997; Mitchell, Pulvino, and Stafford 2002;
Mashruwala, Rajgopal, and Shevlin 2006). Thus media co-coverage may significantly influence
investors’ peer selection and market prices.
2.3. High Media Co-coverage Peers as Benchmarks
After examining the effects of media co-coverage on investors’ peer selection, I move on
to investigate whether and how investors use high co-coverage peers as benchmarks in firm
evaluation. More often than not, people make evaluations in relative terms—an apple is small and
a deer is fast, but only by comparison with other fruits and animals. Similarly, investors can only
tell whether a firm’s performance is “good” or “bad” in the context of peers—a sales growth of
50% may indicate great performance if a peer’s sales only increased by 5% but not so great
performance if a peer had 200% growth. Prospect theory formalizes this idea by suggesting that
humans make decisions based on “gains” and “losses” vis-à-vis a benchmark, rather than based on
the final outcome (Kahneman and Tversky 1979). Thus, the benchmark is critical in the evaluation
process because it determines what is considered a “gain” or a “loss” (or more generally, “good”
or “bad”).
Prior research on investors’ evaluation of firm performance suggests that the market
evaluates a firm more favorably if its performance beats certain benchmarks. For example, the
price reaction to a firm’s earnings announcement is more positive if its earnings are higher than
analysts’ forecasts, its own past earnings, or the break-even point (e.g., Burgstahler and Dichev
1997; Degeorge, Patel, and Zeckhauser 1999; Kasznik and McNichols 2002; Kirk, Reppenhagen,
and Tuckeer 2014).
I propose that, if investors attend more to industry peers with higher media co-coverage,
they may also use such peers as benchmarks when evaluating a firm’s performance. However, as
12

high co-coverage peers may not necessarily be the most comparable ones, doing so may bias their
evaluations. I examine whether investors use high co-coverage peers as benchmarks in the setting
of earnings announcement. Specifically, I investigate whether investors evaluate a firm’s sales
growth performance (sales in quarter q over sales in q-4) by comparing it to the early-announcing
peer with the highest media co-coverage. I examine sales growth for three reasons. First, it is a
relevant performance metric to investors, as prior research finds that investors react positively to
a firm’s revenue surprise over earnings announcement (Ertimur, Livnat, and Martikainen 2003).
Second, there is time-series variation in whether a firm beats the sales growth of high media co-
coverage peer—the first-order autocorrelation of beating the early-announcing peer with highest
media co-coverage is only 0.18. Third, sales growth is comparable across firms—the definition of
“sales” is consistent across firms, and it is a model-free measure that is easy to calculate (i.e., there
is no need to make assumptions about expectations, etc.). If investors use the highest co-coverage
peer as a benchmark, they are likely to react more positively to the firm’s announcement if its sales
growth is higher than that of the highest co-coverage peer. Therefore I hypothesize the following:
H2a: Ceteris paribus, investors’ price reaction to a firm’ s earnings announcement is more positive
if its sales growth beats that of the early-announcing peer with the highest media co-coverage.
If benchmarking against the highest co-coverage peers biases investors’ reactions to a
firm’s earnings announcements, and if the bias is gradually corrected in the post-announcement
period, I formulate the following hypothesis:
H2b: Ceteris paribus, a firm’ s post-announcement return is more negative if its sales growth beats
that of the early-announcing peer with the highest media co-coverage.
13

3. SAMPLE AND RESEARCH DESIGN
3.1. Data
I use the RavenPack database to collect firms’ media coverage from 2003 to 2012. This
sample spans a period where data necessary to calculate the media co-coverage measure is
available. RavenPack covers news articles from the The Wall Street Journal, Barron’ s,
MarketWatch, and Dow Jones Newswires, which are major financial media in the United States.
These outlets are likely to reach a wide audience, as the press mainly targets the general public,
while finance professionals rely primarily on newswires (Solomon and Soltes 2011). This allows
me to examine the potential effects of media co-coverage on the broader market, rather than on a
subset of firms with predominantly individual or institutional investors.
I merge media co-coverage data with firms’ earnings announcements, which are obtained
from the I/B/E/S details and actuals file. I restrict the analysis to firms in the same two-digit SIC
industry to ensure that there is non-zero probability for investors to consider them as peers to begin
with. Early-announcing peers are required to announce earnings at least one trading day before the
announcement of the firm in question. I further require firms to have (1) fiscal quarter-ends in
March, June, September, or December, to ensure that their performance (e.g., earnings and sales)
is measured over the same period and is comparable across firms (e.g., Jennings, Seo, and Soliman
2015), and (2) non-missing values on key variables (e.g., total assets, earnings before extraordinary
items, book value of equity, share prices, and historical SIC codes). Return data are from CRSP,
accounting variables and SIC classifications are from Compustat, and institutional ownership data
are from Thomson Reuters. The final sample consists of approximately 30,674 firm-quarter
observations.
14

3.2. Context-Driven Media Co-coverage
3.2.1. Purging the Confounding Effects of Common Economic Events
Media co-coverage is endogenous and can be correlated with other factors that affect
investors’ peer selection. To address this issue and isolate the effects of “context-driven” co-
coverage, I employ a two-step process. In the first step, I attempt to eliminate the effects of
common economic events on media co-coverage through sample selection. This step is important
because investors may pay more attention to an industry peer not because of co-coverage but
because the co-covered firms are involved in the same economic event, such as mergers and
acquisitions. To do this, I pre-process the raw RavenPack data in four major steps, which are
summarized in Table 1.
First, to focus on the effects of media per se, I exclude company-initiated press releases
from the analysis. This reduces the number of unique news articles by 16%.
Second, to ensure that investors’ peer selection is not driven by two firms’ involvement in
the same economic event, I include in the analysis only articles for which an underlying economic
event cannot be identified by RavenPack.
12
This allows me to alleviate the concern that the
relationship between media co-coverage and investors’ peer selection is confounded by shared
economic events. News articles with an unidentifiable underlying event are mostly editorials and
commentaries on the performance of the market or specific sectors and companies, careers of
notable entrepreneurs, or coverage of non-economic events such as conferences.
13
This step
further reduced the sample of unique news articles by 34%.

12
RavenPack classifies news stories into over 2,000 event categories using a proprietary algorithm. Some examples
of the event categories include mergers and acquisitions, analyst ratings, credit ratings, lawsuits, and product- or
service-related activities.
13
Appendix B provides examples of news articles with unidentifiable underlying event.
15

Third, to further alleviate the concern that two firms may be regarded as peers due to their
common exposure to an economic event, rather than media co-coverage, I only include firms that
are not central to a news story and are instead just passively mentioned in the article. RavenPack
assigns a relevance score, which ranges from 0 to 100, to each firm mentioned in an article, with
a higher score indicating higher relevance of the firm to the news story. The score reflects the role
a firm plays in the news story, and is affected by the position in the text where the firm is first
mentioned (e.g., headline, first paragraph, etc.), the number of times it appears in the article, and
the total number of entities mentioned in the story. RavenPack considers a score higher than 75 as
“significantly relevant.” To be conservative, I set a lower cut-off point and include only firms
whose relevance score is less than 50.
Finally, I require a firm to have non-missing two-digit SIC code. These four filters result
in a sample of 8,361 unique firms, which are used in the second step to calculate a pairwise measure
of context-driven media co-coverage.
3.2.2. Purging the Effects of Other Confounding Variables
In the second step, I use a quarterly cross-sectional regression to purge the effects of other
confounding factors that may affect media co-coverage and investors’ peer selection. I consider
five such factors. The first is peer salience—investors may pay more attention to an industry peer
because it is highly salient in itself rather than because of media co-coverage. The second factor is
peer designation by other market participants—the media may simply follow others (e.g.,
managers, analysts, and sophisticated financial statement users) when deciding co-coverage firms,
and therefore any documented effects of media co-coverage may be attributed to those “opinion
leaders” rather than the media per se. The third factor is firm similarity. The fourth is two firms’
economic relatedness along the supply chain. I control for these two factors because investors may
16

consider two firms as peers due to their relatedness in business fundamentals, rather than co-
appearance in the same news article. Finally, I include peer fixed effects to control for other
unobservable peer characteristics that may affect a firm’s co-coverage and investors’ propensity to
consider it as a peer.
Specifically, in each quarter, I estimate the following regression:
PERC_COMENTIONi,j,q = β0 + β1*MEDCO_PEERj,q + β2*PRCOi,j,q-1 + β3*ANLYCOi,j,q-1
+ β4*EDGARCOi,j,q-1 + β5*HPSCOREi,j,q-1 + β6*RETSYNCHi,j,q-1 + β7*CUSTi,j,q-1 +
β8*CUST_PEERi,j,q-1 + β9*SUPPi,j,q-1 + β10*SUPP_PEERi,j,q-1 + Peer Fixed Effects + εi,j,q, (1)
where PERC_COMENTIONi,j,q is the number of times both firm i and industry peer j are
mentioned in the same article over quarter q (scaled by the number of articles that mention firm i).
MEDCO_PEERj,q is peer j’s media coverage, calculated as the logarithm of one plus the number
of articles mentioning j in quarter q. PRCOi,j,q-1 is the frequency with which peer j is mentioned in
firm i’s press releases in quarter q-1 (scaled by the total number of press releases issued by firm i).
ANLYCOi,j,q-1 is the extent of analyst co-coverage, calculated as the number of analysts issuing
forecasts for both firm i and peer j over quarter q-1 (scaled by the total number of analysts issuing
forecasts for firm i). EDGARCOi,j,q-1 is the frequency with which sophisticated financial statement
users search for peer j on the EDGAR website after they search for firm i in quarter q-1.
14

HPSCOREi,j,q-1 is the similarity between firm i and peer j’s business descriptions in 10-K filings
(Hoberg and Phillips 2010, 2016). RETSYNCHi,j,q-1 is firm i and peer j’s return synchronicity—an
overall measure of similarity that encompasses multiple dimensions, such as size and growth
potential. CUSTi,j,q-1, CUST_PEERi,j,q-1, SUPPi,j,q-1, and SUPP_PEERi,j,q-1 are the importance of
firm i’s industry to firm j’s industry as either a customer or a supplier and vice versa. As firm-level

14
Data on EDGAR search traffic is available from 2003 to 2013.
17

supplier-customer data is not available for all firms, I measure supply-chain relatedness between
two firms using industry-level input-output (IO) tables prepared by the Bureau of Economic
Analysis (BEA).
15
All variables are standardized by subtracting the sample average from the raw
value and scaling the difference by sample standard deviation. Detailed variable definitions are
available in Appendix A. Context-driven media co-coverage (MEDIACO) is measured as the error
term (εi,j,q) from regression (1).
3.3. Research Design
3.3.1. Media Co-coverage and Investors’ Peer Selection
To test whether investors pay more attention to the news of peers with higher media co-
coverage (H1a), I estimate the following regression:
RET1di,j,q = β0 + β1*SURP_EARLYj,q + β2*RANK_MEDIACOi,j,q-1 + β3*SURP_EARLYj,q *
RANK_MEDIACOi,j,q-1 + Controls + Firm Fixed Effects + Year Fixed Effects + εi,j,q, (2)
where RET1di,j,q is firm i’s one-day market-adjusted return on early-announcing peer j’s
announcement day. I use a one-day window to mitigate the concern that the firm’s return may be
affected by peers’ earnings announcements on adjacent days. SURP_EARLYj,q is peer j’s earnings
surprise, calculated as the difference between j’s realized earnings and the consensus analysts’
forecasts, deflated by j’s price three days before the announcement. RANK_MEDIACOi,j,q-1 is the
decile ranking of j’s context-driven media co-coverage with firm i. If a firm’s price reaction to an
early-announcing peer’ surprise is stronger when the peer has higher media co-coverage with the
firm (H1a), β3 is expected to be significantly positive.

15
To ensure the best mapping between the individual firms and industry data, I use detailed-level IO tables for
benchmark year 2002 and 2007. The IO tables account for all producers and purchasers in the U.S. economy, and the
detailed level tables are updated every five years from 1982 to 2007. On average, the tables contain input-output data
for 428 industries.
18

I include the following control variables in regression (2). The first is the average earnings
surprise of all other two-digit SIC peers that report earnings on the same day (AVG_SURP_PEER).
The second is firm i’s characteristics (SIZE, BM, and total accruals ACC). Finally, I control for
momentum, measured as i’s market-adjusted return over the past six months (RET6). Firm and
year fixed effects are included to control for unobservable firm- or time-invariant factors. Standard
errors are double-clustered by firm and quarter (Petersen 2009). Detailed variable definitions are
available in Appendix A.
To test whether investors overweight the earnings news of high co-coverage peers (H1b),
I estimate the following regression:
RETi,q = β0 + β1*SURP_EARLY_TOPj,q + β2*SURP_EARLY_OTHERq + Controls + Firm
Fixed Effects + Year Fixed Effects + εi,q, (3)
where RETi,q is firm i’s two-day market-adjusted buy-and-hold excess return from the day
of to the day after the earnings announcement [0, +1]. SURP_EARLY_TOPj,q is the earnings
surprise of the early-announcing peer with the highest media co-coverage.
SURP_EARLY_OTHERq is the average earnings surprise of all other early-announcing peers. If
investors overweight the earnings surprise of the highest co-coverage peer when updating their
expectation about firm i’s current-quarter earnings and if the biased expectation is corrected when
the firm reports its realized earnings, β1 is expected to be significantly negative.
I include the following groups of control variables in regression (3). First, to control for
the effects of same-day earnings announcements on firm i’s announcement return, I include the
average of the earnings surprises of all industry peers that report earnings on the same day as firm
i (AVG_SURP_PEER).
19

Second, to control for the effects of firm i’s own earnings performance on its announcement
return, I include its earnings surprise (SURP) and whether its earnings meet or beat various
benchmarks, such as analysts’ forecasts (both consensus and individual forecasts, following Kirk,
Reppenhagen and Tucker (2014)), its own past earnings (earnings in q-4), and the break-even point
(MEET_ANLY, PERC_MEET_ANLY, MEET_SEASONAL, and LOSS).
Third, I control for various firm characteristics that have been shown to affect expected
returns, either because they are associated with systematic risk or mispricing. These variables
include profitability (ROA), sales growth (SALESGR), total accruals (ACC), LEVERAGE, SIZE,
book-to-market ratio (BM), and momentum (RET6).
Fourth, to control for the effects of the firm’s past earnings surprise on its announcement
return, i.e., post-earnings announcement drift (Bernard and Thomas 1989), I include the firm’s
earnings surprise in quarter q-1 and q-4.
Fifth, to control for the effects of market frictions on firms’ expected returns (e.g., Hou and
Moskowitz 2005), I include share turnover (TURNOVERq) and the Amihud measure of illiquidity
(ILLIQq) (Amihud 2002).
Sixth, to control for the effects of management guidance on investors’ earnings
expectations and announcement returns (e.g., Atiase, Li, Supattarakul, and Tse 2005; Seybert and
Yang 2011), I include whether the firm routinely issues management guidance (GUIDEq) and
whether it issues guidance in conjunction with the earnings announcement (BUNDLE_GUIDEq).
Finally, to control for the effects of reporting timeliness on firms’ announcement returns
(e.g., Hall, Sunder, and Sunder 2015), I include the logarithm of reporting lag, measured as the
number of days between the end of the fiscal quarter and its earnings announcement day
(REPORTING_LAGq). Detailed variable definitions are available in Appendix A.
20

3.3.2. High Media Co-coverage Peers as Benchmarks
To test whether investors benchmark a firm against the highest co-coverage peer when
evaluating its sales growth on earnings announcement (H2a), I estimate the following regression:
RETi,q = β0 + β1*BEAT_MEDIAi,q + β2*BEAT_LARGESTi,q + β3*BEAT_LARGEST_INDi,q
+ β4*BEAT_SIMILARi,q + β5*BEAT_RECENTi,q + Controls + Firm Fixed Effects + Year Fixed
Effects + εi,q, (4)
where BEAT_MEDIAi,q is an indicator variable equal to one if firm i’s sales growth is higher
than the early-announcing peer with highest media co-coverage. Compared with other research
designs (e.g., directly including peers’ sales growth in the regression or taking the difference
between the firm and the peer’s sales growth), using indicator variables allows easier interpretation
of results, as it helps to alleviate multicollinearity. If investors condition their response to a firm’s
earnings announcement on whether the firm beats the highest co-coverage peer, β1 is expected to
be significantly positive.
Besides BEAT_MEDIA, I also include four additional variables to examine whether
investors use other peers as benchmarks. First, prior research suggests that investors may
benchmark against the largest peer (e.g., Hartzmark and Shue 2016). Thus, I include whether the
firm beats the sales growth of the largest firm among all early-announcing firms
(BEAT_LARGEST), and whether it beats the largest firm among all early-announcing industry
peers in the regression (BEAT_LARGEST_IND). Second, if investors select peers based on
similarity, they are likely to benchmark the firm against its most similar early-announcing peer.
Therefore, I control for whether the firm beats the sales growth of the early-announcing peer with
highest return synchronicity (BEAT_SIMILAR). Finally, investors may use the most recent early
announcers as a benchmark (e.g., Hartzmark and Shue 2016). To examine this possibility, I include
21

whether the firm beats the average sales growth of all firms that announced earnings one day before
(i.e., on day t-1) in the regression (BEAT_RECENT). Control variables include all controls in
regression (3) and two additional variables that capture the extent of information transfers—the
average of all early-announcing industry peers’ announcement returns (AVG_RETEOq) and the
average of the firm’s reactions to early announcers’ announcements (AVG_RETLEq) (e.g.,
Ramnath 2002; Thomas and Zhang 2008).
To test whether there is a return reversal to beating the highest co-coverage peer (H2b), I
estimate the following regression:
POSTRETi,q = β 0 + β1*BEAT_MEDIAi,q + β2*BEAT_LARGESTi,q +
β3*BEAT_LARGEST_INDi,q + β4*BEAT_SIMILARi,q + β5*BEAT_RECENTi,q + Controls + Firm
Fixed Effects + Year Fixed Effects + εi,q, (5)
where POSTRETi,q is firm i’s market-adjusted buy-and-hold return in the post-
announcement period, starting from two days after the announcement over three consecutive one-
month windows [t+2, t+23], [t+24, t+44], and [t+45, t+65]. If the initial reaction to beating the
highest co-coverage peer contains an overreaction component, β1 is expected to be significantly
negative. Control variables are the same as those in regression (4).
4. EMPIRICAL RESULTS
4.1. Descriptive Statistics
Table 2 presents the average coefficients, t-statistics, and adjusted R-squared of the
quarterly regression used to calculate context-driven media co-coverage (regression (1)). Several
observations can be made. First, peer salience has a strong and statistically significant effect on
media co-coverage—the average coefficient on MEDCO_PEER is 0.250 with an average t-statistic
of 18.86. This suggests that the more attention a peer receives from the media, the more likely that
22
it will co-appear with a firm in news articles. Second, there is some overlap between media co-
coverage peers and peers identified by other market participants, but the overlap is small. For
example, although the average coefficients on PRCO and EDGARCO are all positive, their
magnitudes are at best modest (0.016 and 0.011, respectively). This indicates that journalists do
not simply follow others (e.g., firm managers and sophisticated financial statement users) in
picking co-coverage peers. The coefficient on ANLYCO is negative (-0.003) and the average t-
statistic is -0.73. This implies that there is little overlap between media co-coverage peers and
common-analyst-based peers. Third, firm similarity has a positive effect on media co-coverage,
but the effect is small—the average coefficient on HPSCORE is 0.041 while that on RETSYNCH
is 0.025. Finally, supply-chain relatedness has a minimal effect on media co-coverage.
Table 3 provides descriptive statistics of the co-coverage measure (MEDIACO) and other
main variables. MEDIACO has a mean (median) of 0.00 (-0.08), with a standard deviation of 0.63.
The mean of BEAT_MEDIA is 0.50, and the median is 0. The distributions of the other four
indicator variables (BEAT_LARGEST, BEAT_LARGEST_IND, BEAT_SIMILAR, and
BEAT_RECENT) are similar to that of BEAT_MEDIA. Firms’ announcement return (RET) has a
mean (median) of 0.00 (0.00) with a standard deviation of 0.07. The distributions of other variables
are similar to those in prior studies.
4.2. Media Co-coverage and Investors’ Peer Selection
4.2.1. Reaction to Peers’ Earnings News and Media Co-coverage
Table 4 presents the estimation results for regression (2). In Column (1), the coefficient on
the early-announcing peers’ earnings surprise (SURP_EARLY) is significantly positive at the 10%
level. This shows that, on average, information transfers between the early-announcing peers and
the non-announcing firms are positive, which is consistent with previous studies (e.g., Kim, Lacina,
23
and Park 2008; Wang 2014). The interaction variable between SURP_EARLY and the decile
ranking of media co-coverage is significantly positive, lending support to H1a that investors react
more strongly to the earnings news of peers with higher media co-coverage. Moving from the
bottom to the top decile of media co-coverage increases the sensitivity of non-announcing firms’
prices (one-day market-adjusted return) to early-announcing peers’ earnings news by 142%.
Inferences remain similar in Column (2), where firm and year fixed effects are added to the
regression. Overall, empirical evidence is consistent with investors paying more attention to the
earnings news of peers with higher media co-coverage.
4.2.2. Overweighting of Peers’ Information and Media Co-coverage
Table 5, Panel A, presents the estimation results for regression (3). Consistent with the
prediction in H1b, the coefficient on the earnings surprise of the highest co-coverage peer
(SURP_EARLY_TOP) is negative, and it is statistically significant at the 10% level. The relatively
small magnitude of the coefficient on SURP_EARLY_TOP (-3.162 basis points) is likely to be a
result of the research design, as I only focus on the early-announcing peer with highest media co-
coverage. The coefficient on the average earnings surprise of all other early-announcing peers
(SURP_EARLY_OTHER) is positive, albeit statistically insignificant. These results are consistent
with investors overweighting the surprise of high media co-coverage peers when updating their
expectations about the non-announcing firm’s earnings.
To provide further evidence on whether investors overweight signals from high co-
coverage peers, I develop a trading strategy based on the earnings surprise of the highest co-
coverage peer and examine whether it can generate abnormal returns. Specifically, I first identify
a set of days when at least 10 firms will announce earnings. Second, for each identified
announcement day, I sort the announcing firms into deciles based on SURP_EARLY_TOP. Third,
24

I long (short) announcing firms in the lowest (highest) SURP_EARLY_TOP decile and hold the
position for days t to t+1 beginning at the market open on day t and ending at the market open on
day t+1. Portfolios are value-weighted based on the firms’ market capitalization three-days before
the announcement. I then calculate abnormal returns from this strategy by regressing daily
portfolio returns on the market, size (SMB), book-to-market (HML), and momentum (UMD)
factors. t-statistics (in parentheses) are based on Newey-West standard errors.
Table 5, Panel B, presents the estimation results. This long-short strategy yields a one-day
four-factor alpha of 0.437 basis points and it is statistically significant at the 5% level. In my
sample, this strategy can be implemented for an average of 100 trading days per year, which
generates a 43.7% abnormal annual return. However, whether this strategy can generate positive
abnormal returns in the presence of transaction costs is an empirical question that I do not address
in this paper, as the purpose of this exercise is not to quantify precisely how much trading profits
one can generate. Rather, the aim is to show that the results from the regression analysis in Panel
A are robust to an alternative research design. Overall, empirical evidence lends support to my
prediction that investors over-extrapolate signals from peers with high media co-coverage.
4.3. High Media Co-coverage Peers as Benchmarks
4.3.1. Beating the Highest Co-coverage Peer—A Future Performance Signal?
Before discussing the results on whether investors benchmark a firm against the early-
announcing peer with the highest media co-coverage, I first examine whether beating the sales
growth of the highest co-coverage peer can indeed provide a signal for the firm’s future
performance. Specifically, I test whether beating the highest co-coverage peer is associated with
higher future sales growth by estimating the following regression:
25

SALESGRi,q+n = β 0 + β1*BEAT_MEDIAi,q + β2*BEAT_LARGESTi,q +
β3*BEAT_LARGEST_INDi,q + β4*BEAT_SIMILARi,q + β5*BEAT_RECENTi,q + Controls + Firm
Fixed Effects + Year Fixed Effects + εi,q, (6)
where SALESGRi,q+n is the firm’s sales growth in quarter q+n (n=1, 2, 3, 4). Control
variables include sales growth, ROA, size, book-to-market ratio, and leverage (all measured as of
the end of quarter q). If having sales growth higher than that of the highest co-coverage peer is
positively correlated with future performance, β1 is expected to be significantly positive.
Table 6, Panel A, presents the estimation results. In all specifications, the coefficient on
BEAT_MEDIA is statistically insignificant, indicating that beating the sales growth of the highest
co-coverage peers is not associated with future sales growth. The coefficients on the other four
indicator variables (BEAT_LARGEST, BEAT_LARGEST_IND, BEAT_SIMILAR, and
BEAT_RECENT) are statistically insignificant in most cases, with two exceptions. First, when the
dependent variable is one-quarter-ahead sales growth, the coefficients on BEAT_LARGEST_IND
and BEAT_SIMILAR are negative and marginally significant at the 10% level. Second, when the
dependent variable is two-quarters-ahead sales growth, the coefficient on BEAT_LARGEST_IND
is negative and statistically significant at the 10% level. The negative coefficients may be a result
of mean reversion—exceptional performance (e.g., having stronger growth than industry leaders
or close competitors) tend to be less persistent and thus is likely to mean revert in the future. Taken
together, beating the early-announcing peer with highest media co-coverage is not predictive of
better future performance.
4.3.2. Beating the Highest Co-coverage Peer and Firms’ Announcement Returns
Table 6, Panel B, presents the estimation results for regression (4). In Column (1), the
coefficient on BEAT_MEDIA is 2.241 basis points, and it is statistically significant at the 1% level.
26

This suggests that, even if beating the highest co-coverage peer is not informative about a firm’s
future sales growth, investors still react more positively to the firm’s earnings announcement if its
sales growth is higher than that of the highest co-coverage peer. The coefficients on the other four
indicator variables (BEAT_LARGEST, BEAT_LARGEST_IND, BEAT_SIMILAR, and
BEAT_RECENT) are all statistically insignificant, indicating that investors’ evaluation of firm
performance is less influenced by other peers (e.g., the largest peers, the most similar peers, and
the most recent peers), after controlling for whether the firm beats the highest media co-coverage
peer. Inferences remain similar in Column (2), where firm and year fixed effects are included in
the regression. Overall, the empirical evidence lends support to my prediction that investors
condition their response to a firm’s earnings announcement on whether it beats the highest co-
coverage peer.
4.3.3. Mispricing Associated with Benchmarking Against the Highest Co-coverage Peer
Table 7, Panel A, presents the estimation results for regression (5). In all specifications, the
coefficients on BEAT_MEDIA are significantly negative, consistent with the prediction in H2b that
the positive return to beating the highest co-coverage peer reverses in the post-announcement
period. When the dependent variable is POSTRETt+2, t+23 (POSTRETt+24, t+44 and POSTRETt+45,
t+65), BEAT_MEDIA has a coefficient of -0.613 (-1.035 and -0.271, respectively). This suggests
that the price correction starts within one month after the earnings announcement and lasts for at
least three months. The coefficients on the other four indicator variables (BEAT_LARGEST,
BEAT_LARGEST_IND, BEAT_SIMILAR, and BEAT_RECENT) are all statistically insignificant,
suggesting that beating other peers (e.g., large peers, similar peers, and recent peers) does not have
a significant effect on post-announcement returns.
27

To test whether the return reversal results are robust to alternative research designs, I
examine whether it is possible to earn abnormal returns from a trading strategy based on whether
the firm beats the highest co-coverage peer. Specifically, on a given day t, I form a portfolio that
is long firms that announced earnings from 2 to 23 (or from 24 to 44, or 45 to 65, respectively)
trading days ago and with BEAT_MEDIA equal to 0 and short those with BEAT_MEDIA equal to
1. The portfolios are value-weighted based on the firms’ market capitalization as of the end of t-3.
I then calculate abnormal returns from this strategy by regressing daily portfolio returns on the
market, size (SMB), book-to-market (HML), and momentum (UMD) factors. t-statistics (in
parentheses) are based on Newey-West standard errors.
Table 7, Panel B, presents the estimation results. When the portfolios are based on earnings
announcements 2 to 23 (24 to 44) trading days ago, this strategy generates a significantly positive
daily alpha of 0.027 (0.039) basis points, which is equivalent to an annual abnormal return of
6.75% (9.75%), as this strategy can be implemented on virtually every trading day over the year.
When the portfolios are based on earnings announcements 45 to 54 trading days ago, the alpha is
positive (0.011 basis points) but statistically insignificant. These findings are consistent with the
evidence from the regression analysis in Panel A that the return reversal is concentrated within the
first two months after the earnings announcement. Taken together, the evidence in Table 7 is
consistent with the prediction that benchmarking against the highest co-coverage peer may bias
investors’ reaction to a firm’s earnings announcement.
4.3.4. Cross-sectional Tests
To provide more evidence on the underlying mechanism of investors’ overreaction to a
firm beating the highest co-coverage peer, I conduct three cross-sectional tests. First, if the
mispricing is indeed caused by media co-coverage, I expect the initial reaction to beating the
28

highest co-coverage peer and the subsequent reversal to be stronger when the firm and the peer
share higher co-coverage. Second, as institutional investors have more resources to employ when
analyzing a firm, their peer selection is less likely to be influenced by media co-coverage than
retail investors. Thus, I predict the overreaction to be stronger when the firm has lower institutional
ownership. Finally, if the existence of abnormal return is due to limits to arbitrage, I expect the
mispricing to be more severe when market frictions are higher.
I test these predictions by estimating the following regressions:
RETi,q = β0 + β1*BEAT_MEDIAi,q + β2*COND + β3*BEAT_MEDIAi,q*COND + Controls
+ Firm Fixed Effects + Y ear Fixed Effects + εi,q, (7)
POSTRETi,q = β0 + β1*BEAT_MEDIAi,q + β2*COND + β3*BEAT_MEDIAi,q*COND +
Controls + Firm Fixed Effects + Year Fixed Effects + εi,q, (8)
where the conditioning variable COND is the decile ranking of one of the following three
variables: the media co-coverage between firm i and the early-announcing peer with the highest
co-coverage (MEDIACO), the number of institutional investors in a firm (INSTOWN), and the
Amihud (2002) measure of illiquidity (ILLIQ). Detailed variables definitions are available in
appendix A.
Table 8 presents the estimation results. In Panel A, where the conditioning variable is
MEDIACO, the coefficient on the interaction variable between BEAT_MEDIA and the decile
ranking of MEDIACO is significantly positive (negative) in Column (1) (Column (2)), where the
dependent variable is the firm’s announcement return (post-announcement return
POSTRETt+2,t+23).
16
This suggests that the reaction to beating the highest co-coverage peer and the
subsequent reversal are both stronger when the peer has higher media co-coverage with the firm.

16
I only tabulate the results for POSTRET
t+2, t+23
. Inferences remain similar when I use POSTRET
t+24, t+44
and
POSTRET
t+45, t+65
as the dependent variable.
29

The magnitude of the reaction and the subsequent reversal nearly doubles when the firm moves
from the bottom to the top MEDIACO decile. Thus, evidence in Panel A lends further support that
the documented mispricing is indeed due to co-coverage.
In Panel B, the conditioning variable is the decile ranking of a firm’s institutional
ownership. Consistent with my prediction, the coefficient on the interaction variable between
BEAT_MEDIA and RANK_INSTOWN is significantly negative (positive) in Column (1) (Column
(2)), where the dependent variable is the firm’s announcement return (post-announcement return
POSTRETt+2,t+23). Compared with firms in the top INSTOWN decile, the average effect of beating
the highest co-coverage peer nearly triples for RET, and almost doubles for POSTRETt+2,t+23 if a
firm is in the bottom INSTOWN decile.
In Panel C, where the conditioning variable is the decile ranking of the firm’s illiquidity,
the reaction (reversal) is stronger when the firm is less liquid, as suggested by the positive
(negative) coefficient on the interaction variable between BEAT_MEDIA and RANK_ILLIQ in
Column (1) (Column (2)). Moving from the bottom to the top decile of ILLIQ increases the initial
reaction to beating the highest co-coverage peer and the subsequent reversal by more than four
times. These findings suggest that retail investors are more prone to using the highest co-coverage
peer as a benchmark than institutional investors, and that the mispricing is more severe when
arbitrage is more costly due to higher trading frictions.
4.3.5. Alternative Explanations
An alternative explanation to the higher announcement return when the firm beats the
highest co-coverage peer is that the firm’s higher exposure to systematic risk. Although this
explanation does not easily account for the subsequent return reversal, I nevertheless examine this
possibility by regressing firms’ announcement returns (RET) on the four risk factors (market excess
30

return, size, book-to-market, and momentum) as well as the interactions between these four factors
and BEAT_MEDIA (with firm and day fixed effects). If firms that beat the highest co-coverage
peer have higher risk exposure, the coefficients on the interaction terms are expected to be
significantly positive. Untabulated results suggest that this is not the case, as the interaction terms
are all statistically insignificant. Thus, exposure to systematic risks is unlikely to explain the results.
Another alternative explanation is related to liquidity. If firms that beat the highest co-
coverage peer have lower liquidity when they announce earnings, their expected returns are likely
to be higher. To rule out this possibility, I regress firms’ two-day announcement window [0, +1]
turnover and Amihud (2002) illiquidity on BEAT_MEDIA, along with firm and day fixed effects.
Untabulated results show that BEAT_MEDIA is not significantly associated with firms’
announcement window liquidity.
Finally, it is possible that the observed positive reaction to beating the highest co-coverage
peer and the subsequent reversal are due to revenue manipulation. If a firm manipulates its revenue
upward (e.g., through premature recognition of sales), it is likely to beat the sales growth of its
peers, leading to a more positive reaction to its earnings announcement. However, as the market
gradually observes the firm’s true performance, the return reverses in the future. To account for
this possibility, I estimate a logit model by regressing an indicator variable equal to one if the firm
has a restatement due to revenue recognition issues over the next four quarters after the
announcement on BEAT_MEDIA and control variables (size, book-to-market, and total accruals).
17

Untabulated results suggest that beating the sales growth of high media co-coverage peer is not
associated with future restatements.

17
Restatement data is obtained from Audit Analytics.
31

5. ROBUSTNESS CHECKS
I perform several additional analyses to ensure the robustness of my results. First, it is
possible that firms that are co-covered in articles with unidentifiable economic events overlap with
those that are co-covered in articles with identifiable economic events, and it is the latter co-
coverage that drives the results. To address this concern, I exclude firm-pairs that are not only co-
mentioned in articles with unidentifiable economic events and with a relevance score less than 50
but also co-mentioned in articles with identifiable economic events and with a relevance score
greater than or equal to 50 over the same quarter. This filter only eliminates 0.6% of the firm-pairs
used in the main analysis, suggesting that there is no significant overlap between context-driven
co-coverage and co-coverage attributable to identifiable economic events. Untabulated results
suggest that inferences from the main analysis remain similar using this subsample of firm-pairs.
Second, I use alternative definitions of returns. To address the concern that market-adjusted
returns contain risk-based return movements and do not capture firm-specific returns well, I
calculate characteristic-adjusted returns (i.e., raw returns in excess of market return and the return
on a portfolio of stocks with similar characteristics, including size, book-to-market, and
momentum). Inferences are similar when I use this measure.
Third, I repeat the analysis using a three-day [-1, +1] window around earnings
announcement dates, and the results are qualitatively similar.
Fourth, I consider alternative definitions of industry peers. Specifically, I define industry
peers using two-digit NAICS codes, and results (untabulated) are qualitatively similar to those
reported in the main analysis.
32

Finally, I examine whether the results are sensitive to the inclusion of alternative fixed
effects in the regression. Untabulated results suggest that inferences are similar when I use firm
and day (rather than year) fixed effects.
6. CONCLUSION
This paper provides preliminary evidence that investors’ peer selection may be affected by
media co-coverage. Specifically, I find that investors not only react more strongly to the earnings
news of early-announcing industry peers with higher media co-coverage, but also overweight the
surprise of the highest co-coverage peer when updating expectations about the non-announcing
firm’s earnings in the same quarter. In addition, I find that investors suboptimally benchmark a
firm against the highest co-coverage peer when reacting to its earnings announcement. Specifically,
although beating the sales growth of the highest co-coverage peer is not a signal of better future
performance, investors react more favorably to a firm’s earnings announcement if its sales growth
is higher than that of the highest co-coverage peer. The positive return to beating the highest co-
coverage peer reverses in the post-announcement period, suggesting that the initial positive
reaction contains an overreaction. These results are robust to the inclusion of a battery of control
variables and alternative research designs. To the best of my knowledge, this paper is the first to
examine the effects of media co-coverage on investors’ use of peers in the stock market.
33

REFERENCES
Amihud, Y. 2002. Illiquidity and stock returns: cross-section and time-series effects. Journal of
Financial Markets 5 (1): 31–56.
Atiase, R. K., Li, H., Supattarakul, S., and Tse, S. Market reaction to multiple contemporaneous
earnings signals: Earnings announcements and future earnings guidance. Review of
Accounting Studies 10 (4): 497–525.
Barber, B., and T. Odean. 2008. All that glitters: The effect of attention and news on the buying
behavior of individual and institutional investors. Review of Financial Studies 21: 785–818.
Bernard, V . L., and Thomas, J., K. 1989. Post-earnings announcement drift: Delayed price response
or risk premium? Journal of Accounting Research 27: 1–36.
Bhojraj, S., and Lee, C.M.C. 2002. Who is my peer? A valuation-based approach to the selection
of comparable firms. Journal of Accounting Research 40 (2): 407–439.
Browning, E.S. (2002, November 26). Industrials, Nasdaq gain modestly. The Wall Street Journal.
Burgstahler, D., and Dichev, I. 1997. Earnings management to avoid earnings decreases and losses.
Journal of Accounting and Economics 24 (1): 99–126.
Crovitz, L.G. (2008, June 2). Information age: Face-to-facebook. The Wall Street Journal.
DeAngelo, L. 1990. Equity valuation and corporate control. Accounting Review 65: 93–112.
De Franco, G., Hope, O.-K., and Larocque, S. 2015. Analysts’ choice of peer companies. Review
of Accounting Studies 20 (1): 82–109.
Degeorge, F., Patel, J., and Zeckhauser, R. 1999. Earnings management to exceed thresholds.
Journal of Business 72 (1): 1–33.
Engelberg, J. E., and Parsons, C. A. 2011. The causal impact of media in financial markets. Journal
of Finance 66 (1): 67–97.
Engelberg, J. E., Sasseville, C., and Williams, J. 2012. Market madness? The case of mad money.
Management Science 58 (2): 351–364.
Ertimur, Y., Livnat, J., and Martikainen, M. 2003. Differential market reactions to revenue and
expense surprises. Review of Accounting Studies 8 (2-3): 185–211.
Fan, J., and Lang, L. 2000. The measurement of relatedness: an application to corporate
diversification. Journal of Business 73: 629–660.
Fang, L., and Peress, J. 2009. Media coverage and the cross-section of stock returns. Journal of
Finance 64 (5): 2023–2052.
Foster, G. 1981. Intra-industry information transfers associated with earnings releases. Journal of
Accounting and Economics 3: 201–232.
Freeman, R., and Tse, S. 1992. An earnings prediction approach to examining inter-company
information transfer. Journal of Accounting and Economics 15: 509–523.
Grullon, G., Kanatas, G., and Weston, J.P. 2004. Advertising, breadth of ownership, and liquidity.
Review of Financial Studies 17 (2): 439–461.
Gurun, U.G., and Butler, A. W. 2012. Don’t believe the hype: Local media slant, local advertising,
and firm value. Journal of Finance 67 (2): 561–598.
Hall, C., Sunder, J., and Sunder, S. 2015. Earnings seasons: Analysis of inferred and announced
news. Working paper, Drexel University, University of Arizona, and University of Arizona.
Han, J., and J. Wild. 1990. Unexpected earnings and intra-industry information transfer: Further
evidence. Journal of Accounting Research 28: 211–219.
Hartzmark, S.M., and Shue, K. 2016. A tough act to follow: Contrast effects in financial markets.
Working paper, University of Chicago and Yale School of Management.
34

Hillert, A., Jacobs, H., and Muller, S. 2014. Media makes momentum. Review of Financial Studies
27 (12): 3467–3501.
Hoberg, G., and Phillips, G. 2010. Product market synergies and competition in mergers and
acquisitions: A text-based analysis. Review of Financial Studies 23(10): 3773–3811.
Hoberg, G., and Phillips, G. 2016. Text-based network industries and endogenous product
differentiation. Journal of Political Economy 124 (5): 1423–1465.
Hou, K., and Moskowitz, T. J. 2005. Market frictions, price delay, and the cross-section of expected
returns. Review of Financial Studies 18 (3): 981–1020.
Hubermand, G., and Regev, T. 2001. Contagious speculation and a cure for cancer: A nonevent
that made stock prices soar. Journal of Finance 56 (1): 387–396.
Jennings, J., Seo, H., and Soliman, M. 2015. Relative Performance Ranking: Insights into
Competitive Advantage. Working paper, Washington University, National University of
Singapore, and University of Southern California.
Kahneman, D., and Tversky, A. 1979. Prospect theory: An analysis of decision under risk.
Econometrica 47 (2): 263–292.
Kahneman, D. 2011. Thinking, fast and slow. New York: Farrar, Straus and Giroux.
Kaplan, S. N., and R. S. Ruback. 1995. The valuation of cash flow forecasts: An empirical analysis.
Journal of Finance 50: 1059–1093.
Kaustia, M. and Rantala, V. 2013. Common-analyst-based method for defining peer firms.
Working paper. Aalto University School of Business and University of Miami.
Kasznik, R., and McNichols, M. F. 2002. Does meeting earnings expectations matter? Evidence
from analyst forecast revisions and share prices. Journal of Accounting Research 40 (3):
727–759.
Kim, Y., Lacina, M., and Park, M.S. 2008. Positive and Negative Information Transfers from
Management Forecasts. Journal of Accounting Research 46 (4): 885–908.
Kirk, M.P., Reppenhagen, D.A., and Tucker, J.W. 2014. Meeting individual analyst expectations.
The Accounting Review 89 (6): 2203–2231.
Koo, D.S., Wu, J.J., and Yeung, P.E. 2017. Earnings attribution and information transfers.
Contemporary Accounting Research 34 (3): 1547–1579.
Lee, C.M.C., Ma, P., and Wang, C.Y.C. 2015. Search-based peer firms: Aggregating investor
perceptions through internet co-searches. Journal of Financial Economics 116 (2): 410–
431.
Lou, D. 2013. Attracting investor attention through advertising. Review of Financial Studies 27
(6): 1797–1829.
Mashruwala, C., Rajgopal, S., and Shevlin, T. 2006. Why is the accrual anomaly not arbitraged
away? The role of idiosyncratic risk and transaction costs. Journal of Accounting and
Economics 42: 3–33.
Mitchell, M., Pulvino, T., and Stafford, E. 2002. Limited arbitrage in equity markets. Journal of
Finance 57 (2): 551–584.
Paleari, S., Signori, A., and Vismara, S. 2014. How do underwriters select peers when valuing
IPOs? Financial Management 43 (4): 731–755.
Petersen, M. A., 2009. Estimating standard errors in finance panel data sets: comparing approaches.
Review of Financial Studies 22 (1): 435–480.
Peress, J. 2008. Media coverage and investors’ attention to earnings announcements. Working
paper, INSEAD.
35

Peress, J. 2014. The media and the diffusion of information in financial markets: Evidence from
newspaper strikes. Journal of Finance 69 (5): 2007–2043.
Pontiff, J. 1996. Costly arbitrage: Evidence from closed-end funds. Quarterly Journal of
Economics 111 (4): 1135–1151.
Ramnath, S. 2002. Investor and analyst reactions to earnings announcements of related firms: An
empirical analysis. Journal of Accounting Research 40: 1351–1376.
Seybert, N., and Y ang, H. I. The party’s over: The role of earnings guidance in resolving sentiment-
driven overvaluation. Management Science 58 (2): 308–319.
Simon, H.A. 1972. Theories of bounded rationality. Decision and organization, 1 (1): 161–176.
Shleifer, A., and Vishny, R. W. 1997. The limits of arbitrage. Journal of Finance 52 (1): 35–55.
Solomon, D., and Soltes, E. 2011. The determinants of coverage in the business press. Working
paper. University of Southern California and Harvard Business School.
Tetlock, P.C. 2007. Giving content to investor sentiment: The role of media in the stock market.
Journal of Finance 62 (3): 1139–1168.
Tetlock, P.C. 2010. Does public financial news resolve asymmetric information? Review of
Financial Studies 23 (9): 3520–3557.
Tetlock, P . C. 2011. All the news that’s fit to reprint: Do investors react to stale information? Review
of Financial Studies 24 (5): 1481–1512.
Thomas, J., and Zhang, F. 2008. Overreaction to intra-industry information transfers? Journal of
Accounting Research 46 (4): 909–940.
Wang, C. 2014. Accounting standards harmonization and financial statement comparability:
Evidence from transnational information transfer. Journal of Accounting Research 52 (4):
955–992.

36

Appendix A. Variable Definitions
Variable Definition
Variables used in media
co-coverage regression:

MEDIACO
i,j,q
Context-driven media co-coverage, calculated as the error term (ε
i,j,q
) from the
following cross-sectional regression (estimated quarterly):
PERC_COMENTION
i,j,q
= β
0
+ β
1
*MEDCO_PEER
j,q
+ β
2
*PRCO
i,j,q-1
+
β
3
*ANLYCO
i,j,q-1
+ β
4
*EDGARCO
i,j,q-1
+ β
5
*HPSCORE
i,j,q-1
+
β
6
*RETSYNCH
i,j,q-1
+ β
7
*CUST
i,j,q-1
+ β
8
*CUST_PEER
i,j,q-1
+ β
9
*SUPP
i,j,q-1
+
β
10
*SUPP_PEER
i,j,q-1
+ Firm Fixed Effects + Peer Fixed Effects + ε
i,j,q
.
PERC_COMENTION
i,j,q
The number of times both firm i and peer j are mentioned in the same article
over quarter q, scaled by the total number of articles that mention firm i.
MEDCO_PEER
j,q
The logarithm of (1 + the number of articles mentioning peer j in quarter q).
PRCO
i,j,q
The number of times industry peer j is mentioned in the press releases of firm
i over quarter q (scaled by the total number of press releases issued by firm i).
ANLYCO
i,j,q
The number of analysts issuing forecasts for both firm i and peer j over quarter
q (scaled by the total number of analysts issuing forecasts for firm i).
EDGARCO
i,j,q
The number of times EDGAR users search for peer j after they search for firm
i over quarter q (scaled by the total number of searches for firm i). Following
Lee et al. (2015), I only include in the analysis IP addresses that searched for
at least two but no more than 50 unique firms on a given day.
HPSCORE
i,j,q
Pairwise product similarity score calculated using firms’ business descriptions
in 10-K filings, developed by Hoberg and Phillips (2010, 2016). This measure
is calculated on a yearly basis.
RETSYNCH
i,j,q
Return synchronicity between firm i and peer j over quarter q. Specifically, I
first estimate a market regression by regressing firm i’s daily returns on daily
value-weighted CRSP return over quarter q and obtain the R-squared
(RSQ_MKT). Second, I estimate the full regression by regressing firm i’s daily
returns on daily value-weighted CRSP return and peer j’s daily return over
quarter q and obtain the R-squared (RSQ_FULL). RETSYNCH is calculated as
(RSQ_FULL-RSQ_MKT)/(1-RSQ_MKT).
CUST
i,j,q
The importance of firm i’s industry to peer j’s industry as a customer,
calculated using the Input-Output (IO) tables prepared by the BEA.
Specifically, it is the value of output produced by j’s industry that is purchased
by i’s industry, divided by the total output produced by j’s industry.
CUST_PEER
i,j,q
The importance of peer j’s industry to firm i’s industry as a customer,
calculated as the value of output produced by i’s industry that is purchased by
j’s industry, divided by the total output produced by i’s industry.
SUPP
i,j,q
The importance of firm i’s industry to peer j’s industry as a supplier.
Specifically, it is the value of input used by j’s industry that is supplied by i’s
industry, divided by the total value of input used by j’s industry.
SUPP_PEER
i,j,q
The importance of peer j’s industry to firm i’s industry as a supplier.
Specifically, it is the value of input used by i’s industry that is supplied by j’s
industry, divided by the total value of input used by i’s industry.
Variables used in main
analysis:

ACC
i,q
The total accruals of firm i as of the end of quarter.
AVG_SURP_PEER
i,q
The average earnings surprise of all two-digit SIC peers (excluding peer j) that
report earnings on the same day as peer j.
37

AVG_RETEO
i,q
The average of the market-adjusted two-day announcement returns of early-
announcing industry peers associated with firm i.
AVG_RETLE
i,q
The average of firm i’s two-day market-adjusted returns on days when early-
announcing industry peers report earnings for quarter q.
BEAT_MEDIA
i,q
An indicator variable equal to one if firm i’s sales growth in quarter q is higher
than the early-announcing peer with the highest context-driven media co-
coverage.
BEAT_LARGEST
i,q
An indicator variable equal to one if firm i’s sales growth in quarter q is higher
than the largest firm among all early-announcing firms.
BEAT_LARGEST_IND
i,q
An indicator variable equal to one if firm i’s sales growth in quarter q is higher
than the largest firm among all early-announcing two-digit SIC industry peers.
BEAT_SIMILAR
i,q
An indicator variable equal to one if firm i’s sales growth in quarter q is higher
than the early-announcing industry peer with the highest return synchronicity.
BEAT_RECENT
i,q
An indicator variable equal to one if firm i’s sales growth in quarter q is higher
than the average sales growth of all firms that announced earnings one day
before (i.e., on day t-1).
BM
i,q
The book-to-market ratio of firm i as of the end of the quarter.
BUNDLE_GUIDE
i,q
An indicator variable equal to one if the management issues guidance in
conjunction with the earnings announcement in quarter q and zero otherwise.
GUIDE
i,q
An indicator variable equal to one if firm i has issued at least one guidance
over the past three months and zero otherwise.
ILLIQ
i,q
The average of daily Amihud (2002) illiquidity (absolute return divided by
dollar trading volume) over the past three months.
INSTOWN
i,q
The number of institutional investors in firm i as of the end of quarter q.
Missing values for INSTOWN are set to 0.
LEVERAGE
i,q
Firm i’s long-term debt scaled by lagged total assets.
LOSS
i,q
An indicator variable equal to one if firm i’s actual total earnings are less than
zero and zero otherwise.
MEET_ANLY
i,q
An indicator variable equal to one if the firm’s earnings surprise (SURP
i,q
)

is
no less than zero and zero otherwise.
MEET_SEASONAL
i,q
An indicator variable equal to one if firm i’s actual total earnings (i.e., actual
earnings per share times number of shares outstanding) in quarter q are no less
than its actual total earnings in q-4 and zero otherwise.
PERC_MEET_ANLY
i,q
The percentage of individual analysts’ forecasts that firm i meets or beats.
POSTRET
i,q
Firm i’s market-adjusted buy-and-hold return in post-announcement period,
starting from two days after the announcement over three consecutive one-
month windows [t+2, t+23], [t+24, t+44], and [t+45, t+65].
RET
i,q
Firm i’s two-day market-adjusted buy-and-hold excess return from the day of
to the day after the earnings announcement [0, +1].
RET1d
i,j,q
Firm i’s one-day market-adjusted return on early-announcing peer j’s
announcement day.
RET6
i,q
The market-adjusted buy-and-hold returns of firm i over the past six months.
SALESGR
i,q
The sales growth of firm i in quarter q, measured as firm i’s net sales in quarter
q divided by its net sales in quarter q-4.
SIZE
i,q
The natural logarithm of firm i’s market capitalization, measured as of the end
of the quarter.
SURP
i,q
Firm i’s earnings surprise in quarter q.
SURP_EARLY
j,q
Peer j’s earnings surprises, calculated as the difference between j’s realized
earnings and the consensus analysts’ forecasts, deflated by j’s price three days
before the announcement.
38

SURP_EARLY_TOP
j,q
The earnings surprise of the early-announcing peer with highest context-
driven media co-coverage.
SURP_EARLY_OTHER
q
The average earnings surprise of all two-digit SIC early-announcing industry
peers (excluding the peer with highest context-driven media co-coverage).
REPORTING_LAG
i,q
The number of days between fiscal quarter-end and the earnings
announcement day.
ROA
i,q
Firm i’s quarterly return-on-assets, calculated as the firm’s earnings before
extraordinary items scaled by lagged total assets.
TURNOVER
i,q
The average of daily turnover (volume divided by total number of shares
outstanding) over the past three months.

39

Appendix B. Examples of News Articles with Unidentifiable Economic Events
1. “Heard on the Street” (The Wall Street Journal, 2012-12-03, by Miriam Gottfried)
“… Netflix had 38% margins on the basis of earnings before interest, taxes, depreciation
and amortization in 2011. That is about the same as HBO, according to SNL Kagan estimates—
although that doesn’t include shared overhead benefits that it reaps from being part of Time
Warner.”

2. “Buying Binge Shows Baker Fickleness of Fashion” (The Wall Street Journal, 2009-02-02, by
Peter Lattman and Rachel Dodes)
“Richard Baker caught the fashion bug at the wrong time. The charismatic 43-year-old
scion of strip-mall magnate Robert Baker, Mr. Baker bulldozed his way into the retail business
over the past three years…Mr. Baker, who lives in Greenwich, Conn., learned the retail trade from
the vantage point of a landlord. He and his father run closely held National Retail & Development
Corp., which owns some 20 million square feet of shopping centers with tenants such as Wal-
Mart and Kohl’s.”

3. “Information Age: Face-to-Facebook” (The Wall Street Journal, 2008-06-02, by L. Gordon
Crovitz)
“… The D: All Things Digital conference, run by the Journal's Walt Mossberg and Kara
Swisher, is a dramatic example of how even the most digitally minded people still thrive on human
contact. This conference, which draws speakers such as Microsoft Chairman Bill Gates
and Yahoo CEO Jerry Y ang, attracts a sold-out audience of some 600 tech entrepreneurs, investors
and futurists who, when they are not at conferences, are at the front lines of the digital revolution.”

4. “Industrials, Nasdaq Gain Modestly” (The Wall Street Journal, 2002-11-26, by E. S. Browning)
“…With once-spurned stocks such as Juniper Networks and Ciena surging, the Nasdaq
index advanced 0.90% ...”

5. “New Economy vs. Old in First Reader Contest” (The Wall Street Journal, 2000-10-01, by
Solijane Martinez)
“Burton Litwin, a tax attorney who lives in Arlington Heights, Ill., decided on Extreme
Networks, a maker of computer-network switches that trades on Nasdaq. ‘It’s like a little Cisco,’
he says. ‘I’ve been reading up about it and recently invested in it.’ Mr. Litwin, who does his
research via online financial sites, has been investing for five years.”
40

Table 1. RavenPack Data-Filtering Steps
No. of unique news articles No. of unique firms mentioned in the article
(1) Raw sample from 2003–2012 28,150,422 35,423
(2) Keep if the article is not press
release initiated by the firm.
23,632,639 34,526
(3) Keep if underlying economic
event cannot be identified by
RavenPack.
15,674,511 33,663
(4) Keep if a firm’s relevance
score is less than 50.
N/A 31,778
(5) Keep if a firm has nonmissing
two-digit SIC codes.
N/A 8,361

Table 1 describes the filtering steps used to process the RavenPack data and the number of observations remaining
after each step.

41

Table 2. Summary Statistics for Media Co-coverage Regression
Avg. Coefficient Avg. t-stat.
MEDCO_PEER 0.250 18.86
PRCO 0.016 5.81
ANLYCO -0.003 -0.73
EDGARCO 0.011 5.48
HPSCORE 0.041 3.40
RETSYNCH 0.025 1.96
CUST 0.002 0.64
CUST_PEER 0.002 0.36
SUPP -0.002 -0.22
SUPP_PEER 0.000 -0.14

Peer FE Yes
Average Adj. RSQ: 0.19
Average Observations: 140,417

Table 2 reports summary statistics for the media co-coverage regression. Specifically, I report the average coefficients,
t-statistics and adjusted R-squared from the following regression (estimated quarterly): PERC_COMENTION
i,j,q
= β
0

+ β
1
*MEDCO_PEER
j,q
+ β
2
*PRCO
i,j,q-1
+ β
3
*ANLYCO
i,j,q-1
+ β
4
*EDGARCO
i,j,q-1
+ β
5
*HPSCORE
i,j,q-1
+
β
6
*RETSYNCH
i,j,q-1
+ β
7
*CUST
i,j,q-1
+ β
8
*CUST_PEER
i,j,q-1
+ β
9
*SUPP
i,j,q-1
+ β
10
*SUPP_PEER
i,j,q-1
+ Peer Fixed
Effects + ε
i,j,q
. t-statistics are based on standard errors clustered by firm and peer. Detailed variable definitions are
available in Appendix A.

42

Table 3. Descriptive Statistics
Mean Std. P75 Median P25 N
MEDIACO 0.00 0.63 0.18 -0.08 -0.31 750,795
BEAT_MEDIA

0.50 0.50 1 0 0 30,674
BEAT_LARGEST 0.50 0.52 1 0 0 30,674
BEAT_LARGEST_IND 0.51 0.48 1 1 0 30,674
BEAT_SIMILAR 0.51 0.50 1 1 0 30,674
BEAT_RECENT 0.50 0.53 1 0 0 30,674
RET 0.00 0.07 0.04 0.00 -0.03 30,674
POSTRETt+2, t+23 0.00 0.11 0.05 0.00 -0.05 30,674
POSTRETt+24, t+44 0.01 0.11 0.05 0.00 -0.05 30,674
POSTRETt+45, t+65 0.00 0.12 0.05 0.00 -0.07 30,674
ROA 0.021 0.019 0.030 0.022 0.011 30,674
SALESGR 1.15 0.27 1.21 1.10 1.01 30,674
SIZE 7.72 1.48 8.58 7.64 6.72 30,674
BM 0.61 0.66 0.71 0.50 0.30 30,674
REPORTING_LAG 31.50 1.36 38.09 30.88 25.03 30,674

Table 3 presents descriptive statistics of the main variables over the sample period from 2003 to 2012. All continuous
variables are winsorized at 1% and 99% (except MEDIACO). Detailed variable definitions are available in Appendix
A.

43

Table 4. Reaction to Early-Announcing Peers’ Earnings News and Media Co-coverage
RET1d
(1) (2)
SURP_EARLY 0.042* 0.039*
(1.69) (1.78)
RANK_MEDIACO 0.000 0.000
(0.23) (-0.52)
SURP_EARLY*RANK_MEDIACO 0.006** 0.005**
(2.35) (2.23)
CONTROLS Yes Yes

Firm FE No Yes
Year FE No Yes
Observations 750,795 750,795
Adj. RSQ 0.01 0.04

Table 4 provides evidence on whether investors’ reaction to early-announcing peer’s earnings news is stronger when
the peer has higher context-driven media co-coverage with the firm. Control variables include AVG_SURP_PEER,
SIZE, BM, ACC, and RET6. t-statistics (in parentheses) are based on standard errors double-clustered by firm and
quarter. All variables are defined in Appendix A, and all continuous variables are winsorized at 1% and 99% (except
MEDIACO). *, **, and *** represent significance levels at 10%, 5%, and 1%, respectively.

44

Table 5. Overweighting High Co-coverage Peers’ Information
Panel A. Regression Analysis
Predicted sign RET
q
(in basis point)
SURP_EARLY_TOP - -3.162*
(-1.83)
SURP_EARLY_OTHER 0.052
(0.94)

CONTROLS Yes

Firm FE Yes
Year FE Yes
Observations 30,674
Adj. RSQ 0.19

Panel B. Trading Strategy
[t, t+1]

Daily Alpha (in basis point) 0.437**
(2.08)
β
Mkt
0.003
(0.66)
β
SMB
0.004
(0.43)
β
HML
0.012
(0.53)
β
UMD
0.069
(1.26)

Observations 1,004

Table 5 provides evidence on whether investors overweight the earnings news of the early-announcing peer with the
highest context-driven media co-coverage when updating expectations about the non-announcing firm’s current
quarter earnings. Panel A reports estimation results from the regression analysis. Control variables include
AVG_SURP_PEER, SURP, MEET_ANLY, PERC_MEET_ANLY, MEET_SEASONAL, LOSS, ROA, SALESGR, ACC,
LEVERAGE in q-1, SIZE in q-1, BM

in q-1, RET6, SURP

in q-1, SURP

in q-4, TURNOVER, ILLIQ, GUIDE,

BUNDLE_GUIDE, and REPORTING_LAG. t-statistics (in parentheses) are based on standard errors double-clustered
by firm and quarter. Panel B reports the results of a trading strategy based on SURP_EARLY_TOP. Specifically, I first
identify a set of days when at least 10 firms will announce earnings. Second, for each identified announcement day, I
sort the announcing firms into deciles based on SURP_EARLY_TOP. Third, I long (short) announcing firms in the
lowest (highest) SURP_EARLY_TOP decile and hold the position for days t to t+1 beginning at the market open on
day t and ending at the market open on day t+1. Portfolios are value-weighted based on the firms’ market capitalization
three-days before the announcement. I then calculate abnormal returns from this strategy by regressing daily portfolio
returns on the market, size (SMB), book-to-market (HML), and momentum (UMD) factors. t-statistics (in parentheses)
are based on Newey-West standard errors. All variables are defined in Appendix A, and all continuous variables are
winsorized at 1% and 99% (except PERC_MEET_ANLY). *, **, and *** represent significance levels at 10%, 5%,
and 1%, respectively.

45

Table 6. High Media Co-coverage Peers as Benchmarks
Panel A. Beating the Highest Co-coverage Peer and Future Sales Growth
Dependent variable is: SALESGR
q+1
SALESGR
q+2
SALESGR
q+3
SALESGR
q+4

(1) (2) (3) (4)
BEAT_MEDIAq -0.001 0.000 0.001 0.001
(-1.06) (-0.06) (1.09) (1.11)
BEAT_LARGEST
q
0.003 0.001 0.002 0.002
(1.46) (1.02) (1.52) (1.27)
BEAT_LARGEST_IND
q
-0.005* -0.006* -0.005 -0.006
(-1.74) (-1.75) (-1.12) (-1.29)
BEAT_SIMILAR
q
-0.003* -0.004 -0.004 -0.002
(-1.83) (-1.63) (-1.57) (-0.78)
BEAT_RECENT
q
0.000 0.000 0.001 0.000
(0.34) (0.29) (1.09) (0.52)
CONTROLS Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
Observations 30,674 29,813 29,471 29,290
Adj. RSQ 0.68 0.68 0.68 0.68

Panel B. Beating the Highest Co-coverage Peer and Earnings Announcement Return
RET
q
(in basis points)
Predicted signs (1) (2)
BEAT_MEDIAq + 2.241*** 2.177**
(3.94) (2.50)
BEAT_LARGEST
q
0.912 0.725
(0.99) (1.53)
BEAT_LARGEST_IND
q
1.427 0.914
(1.06) (0.55)
BEAT_SIMILAR
q
0.432 0.541
(0.05) (0.58)
BEAT_RECENT
q
0.296 0.145
(0.90) (0.72)
CONTROLS Yes Yes

Firm FE No Yes
Year FE No Yes
Observations 30,674 30,674
Adj. RSQ 0.15 0.26

Table 6 provides evidence on whether investors use the early-announcing peer with the highest context-driven media
co-coverage as a benchmark when evaluating a firm’s sales growth performance. Panel A reports estimation results
on whether beating the sales growth of the highest co-coverage peer is associated with higher future sales growth.
Control variables include SALESGR, ROA, SIZE, BM, and LEVERAGE. Panel B presents estimation results on whether
investors’ reaction to a firm’s earnings announcement is more positive if its sales growth is higher than that of the
highest co-coverage peer and other peers. Control variables include AVG_SURP_PEER, SURP, MEET_ANLY,
PERC_MEET_ANLY, MEET_SEASONAL, LOSS, ROA, SALESGR, ACC, LEVERAGE in q-1, SIZE in q-1, BM

in q-1,
RET6, SURP

in q-1, SURP

in q-4, TURNOVER, ILLIQ, GUIDE,

BUNDLE_GUIDE, REPORTING_LAG,
AVG_RETEO, and AVG_RETLE. t-statistics (in parentheses) are based on standard errors double-clustered by firm
and quarter. All variables are defined in Appendix A, and all continuous variables are winsorized at 1% and 99%
(except PERC_MEET_ANLY). *, **, and *** represent significance levels at 10%, 5%, and 1%, respectively.
46

Table 7. Beating the Highest Co-coverage Peer and Future Returns
Panel A. Regression Analysis
Dependent variable (in basis points) is: POSTRET
t+2, t+23
POSTRET
t+24, t+44
POSTRET
t+45, t+65
Predicted signs (1) (2) (3)
BEAT_MEDIAq - -0.613** -1.035** -0.271*
(-2.07) (-2.16) (-1.95)
BEAT_LARGEST
q
0.015 -0.004 0.001
(0.69) (-0.20) (0.67)
BEAT_LARGEST_IND
q
-0.019 0.015 0.003
(-0.91) (0.58) (0.55)
BEAT_SIMILAR
q
0.016 0.018 0.008
(1.08) (0.43) (0.48)
BEAT_RECENT
q
0.017 0.013 0.007
(0.34) (0.37) (0.53)
CONTROLS Yes Yes Yes

Firm FE Yes Yes Yes
Year FE Yes Yes Yes
Observations 30,674 30,674 30,674
Adj. RSQ 0.14 0.14 0.14

Panel B. Trading Strategy
Earnings announcements over [t-2, t-23] [t-24, t-44] [t-45, t-65]
(1) (2) (3)
Daily Alpha (in basis point) 0.027* 0.039** 0.011
(1.76) (1.98) (1.34)
β
Mkt
0.003 0.001 0.001
(0.47) (0.27) (0.14)
β
SMB
0.006 0.005 0.002
(0.95) (1.06) (0.23)
β
HML
-0.016 0.003 -0.005
(-1.51) (0.50) (-0.41)
β
UMD
0.008 0.004 0.002
(0.63) (0.78) (0.05)

Observations 2,486 2,486 2,486

Table 7 provides evidence on whether the positive return to beating the sales growth of the highest co-coverage peer
reverses in post-announcement period. Panel A presents the estimation results of the regression analysis. Control
variables include AVG_SURP_PEER, SURP, MEET_ANLY, PERC_MEET_ANLY, MEET_SEASONAL, LOSS, ROA,
SALESGR, ACC, LEVERAGE in q-1, SIZE in q-1, BM

in q-1, RET6, SURP

in q-1, SURP

in q-4, TURNOVER, ILLIQ,
GUIDE,

BUNDLE_GUIDE, REPORTING_LAG, AVG_RETEO, and AVG_RETLE. t-statistics (in parentheses) are
47
based on standard errors double-clustered by firm and quarter. Panel B reports the results of a trading strategy based
on BEAT_MEDIA. For example, in Column 1, on a given day t, I form a portfolio that is long firms that announced
earnings from 2 to 23 trading days ago and with BEAT_MEDIA equal to 0 and short those with BEAT_MEDIA equal
to 1. The portfolios are value-weighted based on the firms’ market capitalization as of the end of t-3. I then calculate
abnormal returns from this strategy by regressing daily portfolio returns on the market, size (SMB), book-to-market
(HML), and momentum (UMD) factors. t-statistics (in parentheses) are based on Newey-West standard errors. All
variables are defined in Appendix A, and all continuous variables are winsorized at 1% and 99% (except
PERC_MEET_ANLY). *, **, and *** represent significance levels at 10%, 5%, and 1%, respectively.
48

Table 8. Cross-Sectional Analysis
Panel A. Media Co-coverage
Dependent variable (in basis points): RET
q
POSTRET
t+2, t+23

(1) (2)
BEAT_MEDIAq 1.128** -0.357*
(2.12) (-1.87)
RANK_MEDIACO
q-1
0.000 0.003
(0.76) (0.34)
BEAT_MEDIAq* RANK_MEDIACOq-1 0.216* -0.072**
(1.88) (2.03)
CONTROLS Yes Yes

Firm FE Yes Yes
Year FE Yes Yes
Observations 30,674 30,674
Adj. RSQ 0.26 0.15

Panel B. Institutional Ownership
Dependent variable (in basis points):
RET
q
POSTRET
t+2, t+23

(1) (2)
BEAT_MEDIAq 3.162** -1.049**
(2.54) (-2.37)
RANK_INSTOWN
q-1
0.000 0.002
(0.35) (0.23)
BEAT_MEDIAq* RANK_INSTOWNq-1 -0.251* 0.068**
(1.78) (2.07)
CONTROLS Yes Yes

Firm FE Yes Yes
Year FE Yes Yes
Observations 30,674 30,674
Adj. RSQ 0.26 0.15

Panel C. Market Frictions
Dependent variable (in basis points): RET
q
POSTRET
t+2, t+23

(1) (2)
BEAT_MEDIAq 0.781* -0.189
(1.69) (-1.43)
RANK_ILLIQ
q-1
0.160*** 0.231
(3.59) (1.57)
BEAT_MEDIAq* RANK_ILLIQq-1 0.389* -0.112**
49
(1.78) (2.14)
CONTROLS Yes Yes
Firm FE Yes Yes
Year FE Yes Yes
Observations 30,674 30,674
Adj. RSQ 0.26 0.15
Table 8 provides evidence on whether overreaction to beating the highest co-coverage peer varies cross-sectionally
by context-driven media co-coverage (Panel A), institutional ownership (Panel B), and market frictions (Panel C).
Control variables include AVG_SURP_PEER, SURP, MEET_ANLY, PERC_MEET_ANLY, MEET_SEASONAL,
LOSS, ROA, SALESGR, ACC, LEVERAGE in q-1, SIZE in q-1, and BM

in q-1, RET6, SURP

in q-1, SURP

in q-4,
TURNOVER, ILLIQ, GUIDE,

BUNDLE_GUIDE, REPORTING_LAG, AVG_RETEO, and AVG_RETLE. t-statistics
(in parentheses) are based on standard errors double-clustered by firm and quarter. All variables are defined in
Appendix A, and all continuous variables are winsorized at 1% and 99% (except PERC_MEET_ANLY). *, **, and
*** represent significance levels at 10%, 5%, and 1%, respectively.
50

Abstract (if available)

Abstract Benchmarking with industry peers is ubiquitous in financial markets, yet relatively little is known about investors’ peer selection process. This paper examines media co‐coverage as a factor in peer selection. An industry peer’s frequent co‐appearance with a firm in media news articles can increase the peer’s salience and highlight its association with the firm. In the presence of information gathering and processing frictions, increased salience may cause investors to pay more attention to, and even overweight signals from high media co‐coverage peers. The empirical evidence is consistent with this conjecture. First, investors attend more to information from high co‐coverage peers—a non‐announcing firm’s stock price reacts more positively to the earnings surprise of the early‐announcing industry peer with higher media co‐coverage. Second, investors overweight signals from high co‐coverage peers—earnings surprise of the highest co‐coverage peer negatively predicts the firm’s announcement return in the same quarter, which is consistent with a correction of the initial overreaction to the highest co‐coverage peer’s earnings news. In addition, I find that media co‐coverage distorts investors’ benchmarks—investors overreact to a firm’s earnings announcement if its sales growth exceeds the early‐announcing peer with the highest media co‐coverage. By documenting the economic consequences of investors’ (suboptimal) peer selection, this paper contributes to prior literature that has mainly focused on developing technologies that identify the most appropriate peers.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on information and financial economics

PDF

Changing fundamental analysis in the new economy: the case of DuPont analysis and STEM firms

PDF

Essays on the effect of cognitive constraints on financial decision-making

Asset Metadata

Creator Xia, Jingjing (author)

Core Title “What you see is all there is”: The effects of media co‐coverage on investors’ peer selection

School Leventhal School of Accounting

Degree Doctor of Philosophy

Degree Program Business Administration

Publication Date 03/09/2018

Defense Date 02/10/2018

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag benchmarking,earnings announcements,industry peers,market efficiency,media,OAI-PMH Harvest

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Bonner, Sarah (committee chair), Soliman, Mark (committee chair), Ahern, Kenneth (committee member), Ogneva, Maria (committee member)

Creator Email jingjinx@usc.edu,xiajingjing11@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-484263

Unique identifier UC11268465

Identifier etd-XiaJingjin-6090.pdf (filename),usctheses-c40-484263 (legacy record id)

Legacy Identifier etd-XiaJingjin-6090.pdf

Dmrecord 484263

Document Type Dissertation

Rights Xia, Jingjing

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA