20 2017 Wu & Lin

Pacific-Basin Finance Journal 43 (2017) 151–172
Contents lists available at ScienceDirect
Pacific-Basin Finance Journal

journal homepage: www.elsevier.com/locate/pacfin
The impact of media coverage on investor trading behavior and MARK

stock returns
Chen-Hui Wua,⁎, Chan-Jane Linb
a
Department of Accounting and Information Technology, National Chung Cheng University, Taiwan, ROC
b
Department of Accounting, National Taiwan University, Taiwan, ROC
AR TI CLE I NF O AB S T R A CT
JEL classification: The investor recognition hypothesis suggests that, when a firm is recognized by investors through
G12 the mass media, this helps to increase the firm's investor base and causes reactions in stock prices.
G14 Using a comprehensive database of media coverage on Taiwanese listed firms, this paper applies
Keywords: textual analysis to the news content and classifies it into ten news categories. Among them, seven
Buy-sell trading news categories can be identified as conveying a positive or negative tone and the remaining
Investor type three are news announcements. The empirical results show that investor trading behavior is
Textual analysis affected not only by the quantity, but also the quality, of news announcements. Different types of
News categories
investors have different responses to the tone of media coverage. The trading behavior of foreign
institutional investors is consistent with the tone of media coverage. The results are robust after
accounting for the effect of several firm risks, characteristics and investors' unobservable private
information. In addition, the majority of the positive or negative media coverage classified in this
paper is significantly related to the same sign of abnormal returns.
1. Introduction
A securities market is efficient if the stock prices can fully reflect all available information circulated in the market. However, in
financial markets characterized by incomplete information, the news media becomes one of the major instruments for disseminating
information to investors. The media releases timely information, in the form of newspapers or online news, to a broader readership
which helps investors derive the fundamental value of a firm, especially during a period of scant corporate disclosure.
The motivation of this paper is twofold. First, a substantial body of literature in the field of finance and accounting examines the
effect of quantitative data such as accounting numbers on the stock market (see, e.g., Kothari, 2001 for a review). In addition to these
quantifiable data available in the market, firms frequently publish mandatory news announcements and business press agencies also
report firm-specific news events through the news media. While many papers investigate market reactions to quantitative accounting
information, relatively few examine how investors react to news events that are disseminated through the media coverage.1 Second,
most news announcements and firm-specific news items are presented in the form of qualitative descriptions. The reactions of
different types of investors to the tone of media coverage that impound into stock prices should be of great interest to researchers,
practitioners, and policy-makers.
When a firm is recognized by investors through a newspaper or other mass media story, related news about the firm will increase
its investor base and cause reactions in its stock price (Merton, 1987). Given the important influence of the media in financial
⁎
Corresponding author at: National Chung Cheng University, No. 168, Sec. 1, University Rd., Min-Hsiung Township, Chia-yi County 621, Taiwan, ROC.
E-mail address: [email protected] (C.-H. Wu).
1
The media coverage in this paper includes both mandatory news announcements and news appearing in the media or on websites.
http://dx.doi.org/10.1016/j.pacfin.2017.04.001
Received 20 August 2014; Received in revised form 23 February 2017; Accepted 1 April 2017
Available online 05 April 2017
0927-538X/ © 2017 Elsevier B.V. All rights reserved.
C.-H. Wu, C.-J. Lin Pacific-Basin Finance Journal 43 (2017) 151–172
markets, this paper attempts to contribute to the growing literature on this strand of research by analyzing how the trading behavior
of all-inclusive investor types relates to the tone of media coverage. Most theoretical models consider only two kinds of agents in the
market, namely, sophisticated and unsophisticated investors. Empirical studies normally regard institutional investors as
sophisticated investors and domestic or retail investors as unsophisticated ones. Some of them use two types of investors and others
use three types of investors, with this being due to data availability and the purpose of the research.
Past theoretical and empirical studies generally treat institutional investors as sophisticated and individual investors as
unsophisticated (e.g., Grossman and Stiglitz, 1980; Kyle, 1985; Wang, 1993; Grinblatt and Keloharju, 2001; Barber and Odean,
2008; Barber et al., 2009). The comprehensive dataset from the Taiwanese stock market provides a unique opportunity to examine
the buy-sell trading activities of the three types of institutional investors (foreign institutions, investment trust companies, and broker
dealers) and two types of individual investors (margin account holders and non-margin account holders).2 Analyzing the trading
behavior of all types of investors enables us to obtain a comprehensive understanding of the interactions among participants in the
stock market. In particular, in a market where one party has more or better information than others, it is important to study the
dissemination of news information in the market and the degree of information asymmetry among all types of investors. Thus, the
first purpose of this paper is to attempt to shed light on the interplay among all-inclusive types of investors in responding to the
different news announcements.
Recently, several studies have focused on the textual analysis of media content.3 For example, Tetlock (2007) finds that the tone
reported by the media is linked to the investor sentiment and predicts movements in several indicators of stock market activity.
Tetlock et al. (2008) demonstrate that the content of media coverage provides predictions of a firm's fundamentals. Lu et al. (2013)
improve early warning models of financial distress by incorporating information content distilled from linguistic analysis. The second
purpose of this paper is to extend the literature on the textual analysis of media content by applying tone analysis to classify the news
into ten news categories.4 We then investigate how different types of investors react to these news categories.
This paper presents a textual analysis approach to determine the tone of media coverage. The news items are categorized
according to the nature of their content, in which we are able to classify them into ten news categories – three categories of positive
news (i.e., positive revenue news, positive earnings news, and positive investment news), four categories of negative news (i.e.,
negative general news, negative revenue news, negative earnings news, and negative pledge news), and three news announcements
(i.e., dividend news, share transfer news, and share repurchase news).
The empirical evidence presented in this paper suggests that the buy-sell imbalance of sophisticated investors is more consistent
with the positive or negative tone of different news categories, whereas unsophisticated investors exhibit reverse trading behavior.
Specifically, among institutional investors, foreign institutions are better informed, indicating that their buy-sell activities are
positively related to the tone of the media coverage, whereas individual investors trade in the opposite direction to the tone of the
news announced. This empirical finding is consistent with studies on Asia-Pacific countries which indicate that foreign institutional
investors have informational advantages over other types of investors, for example, in the case of Japan (He and Shen, 2014), Korea
(Bae et al., 2011), Taiwan (Barber et al., 2009), and Thailand (Phansatan et al., 2012).5
As to the market reaction test, this paper finds that both the positive and negative tone of media coverage are significantly and
positively associated with abnormal returns. The only exception is the positive investment news announced by listed firms that
frequent announcements of gains on the disposal of assets are perceived negatively by the market, albeit such perceptions are
insignificant.6 The evidence is consistent with that in previous studies whereby the tone of news releases affects the stock market
(e.g., Henry, 2008; Tetlock, 2007; Tetlock et al., 2008).
This paper adds to a growing literature that explores investors' trading behavior, but it differs from the prior literature in two
ways. First, with the availability of comprehensive data, we investigate the trading behavior of all types of investors in the stock
market. Second, we classify news releases into ten news categories and perform a textual analysis of the news content. The empirical
results support the view that institutional investors, especially foreign institutions, are sophisticated investors, because their trading
behavior is consistent with the tone of the media coverage, while non-margin account holders exhibit trading behavior that is
opposite to the tone of the media coverage.
This paper also contributes to the literature on the influence of media coverage on the stock market and the textual analysis of
news content. In a survey conducted by Loughran and McDonald (2016), the authors point out that textual analysis used in
2
According to Taiwan Stock Exchange statistics, individual investors predominate in the Taiwan stock market. The aggregate purchases and sales of domestic
individual investors accounted for roughly 80% of total purchases and sales in 2001 and for 60% in 2014. Since foreign institutions (6% and 24%) and domestic
institutions (10% and 17%) had lower percentages of trading value, they may be representative of sophisticated investors.
3
Textual analysis, in some form, resides across many disciplines under various aliases such as computational linguistics, information retrieval, or content analysis
(Loughran and McDonald, 2016). To be consistent with the terminology commonly used in most of the finance literature, we use textual analysis throughout the paper,
although the term “content analysis” may also appear in this paper because of its original use in the reviewed literature according to the purposes of their studies (e.g.,
Tetlock, 2007; Tetlock et al., 2008; Kothari et al., 2009; Henry and Leone, 2016).
4
A variety of methodologies in studying textual analysis are discussed in Loughran and McDonald (2016). The method used in this paper is based on the tone
analysis of media coverage. Thus, the construction of an index to measure the readability of the news does not accommodate the purposes of our research. The reasons
for this are: (1) most indexes are computed by sentence length or syllable-related measures in which case these indexes are unsuitable for measuring Chinese sentences;
(2) most indexes lack a consensus regarding how closely the readability index measures reflect the actual comprehension process; (3) the standard format of media
coverage may exhibit no discernible indexes as featured by the majority of accounting disclosures (see Loughran and McDonald, 2014 for more discussion).
5
By contrast, other studies have found that foreign investors underperform domestic investors, for example, Indonesia (Dvorak, 2005; Agarwal et al., 2009) and
international portfolio flows (Brennan and Cao, 1997).
6
One possible explanation is that the gain on disposal arises from the non-core operations representing the transitory profit of the company.
152
accounting and finance is still imprecise relative to quantitative methods. However, there has been a growing body of accounting and
finance research focusing on the textual analysis of financial news and documents in recent years. Furthermore, this paper should be
of considerable interest to both practitioners and policy-makers as a reference for the implementation of the “Mechanism for Halting
Trading during the Information Assessment Period” that was launched on January 16, 2016 in order to improve the information
symmetry and fairness in the stock market.7
This remainder of this paper is organized as follows. Section 2 reviews the related literature. Section 3 describes the measurement
of investor trading behavior, other variables, and the textual analysis of media coverage. Section 4 presents and discusses the main
empirical results. Section 5 concludes.
2. Related literature
Theoretical models suggest that, in a financial market with information asymmetry, informed investors gain from trade at the
expense of uninformed investors (e.g., Grossman and Stiglitz, 1980; Kyle, 1985). Informed investors trade on superior information as
compared with uninformed investors, and thus uninformed investors request a higher risk premium resulting in an increase in price
volatility and a negative autocorrelation in their returns. Wang (1993) posits that uninformed investors may rationally behave like
trend chasers, while informed investors act like contrarians. Below, we review related studies which can be divided into three streams
of research. First, we summarize the past literature on news and investor trading behavior. Second, we describe the related literature
on news and stock returns. Finally, we provide a survey on the textual analysis applied in finance and accounting studies.
2.1. News and investor trading behavior
Due to the abundant resources that institutional investors have in gathering information, it is reasonable to assume that
institutional investors are better informed than individual investors. Barber et al. (2009) investigate the Taiwanese stock market and
show that individual investors have an annual aggregate portfolio return that is 3.8 percentage points lower than the overall average,
whereas institutional investors have an annual performance that is 1.5 percentage points higher. Chen et al. (2015a) examine the
intra- and inter-day market reactions to the public information announced on the Taiwanese MOPS. They however do not analyze the
tone of information releases. Seasholes and Wu (2007) study the Shanghai stock market and find that smart traders, who are active
around attention-grabbing events, earn one-day profits of approximately 1.16% by exploring the informational advantage vis-à-vis
individual investors.
Using retail discount brokerage data, Odean (1998) finds that investors are reluctant to realize their losses by selling winner
stocks and holding loser stocks. Nofsinger (2001) reports that individual investors trade actively on good news, whereas institutional
investors buy and sell on both good and bad news. It is also found that bad news travels more slowly than good news in the case of
individual investors (Hong and Stein, 2000). By extracting data from Finland's financial markets, Grinblatt and Keloharju (2000)
show that foreign investors tend to pursue momentum strategies while household investors tend to be contrarians. However, recent
empirical studies show that trades performed by individual investors are profitable. For example, Kaniel et al. (2012) find that intense
buying (selling) by informed individual investors gives rise to large positive (negative) abnormal returns around earnings
announcement dates using US data. In addition, individual trading exhibits predictive ability and is related to future short-horizon
returns (Kaniel et al., 2008).
Recently, there has been an increasing amount of research on the impact of the media on the stock market since the media
provides information on a timely basis and is publicly available at low cost. Bushee et al. (2010) show that the press coverage during
earnings announcement periods reduces bid-ask spreads and improves market depth, indicating the important information
intermediary role played by the business press. It is worth noting that the results found in Bushee et al. (2010) are based on mid-
sized NASDAQ firms that may not be generalizable to samples of large firms with a rich information environment or small, neglected
firms.
Media coverage influences the information disseminated by a firm to the market, which can alleviate information asymmetry
among market participants and affect stock prices. Tetlock (2010) finds that public news helps resolve the problem of asymmetric
information but more in the case of illiquid stocks. News coverage is an attention driver for investors, especially individual investors,
in that it helps them to choose which stocks to buy. Institutional investors face fewer searching problems because they have more
resources to narrow their search. Barber and Odean (2008) employ the Dow Jones Service as a news database and show that individual
investors are net buyers of stocks with high levels of coverage in the news. Moreover, the average buy-sell imbalances are greater for
negative return days than for positive return days.
Past empirical findings regarding the reaction of different types of investors to diverse news depends on the research scope and
purpose of each paper. For example, individual investors exhibit attention-driven buying behavior, whereas institutional investors are
less likely to be influenced by attention-driven purchases (Barber and Odean, 2008). Individual investors lose because of their
aggressive orders. Institutional investors gain from trading. Foreign institutions collect nearly half of the institutional profits (Barber
et al., 2009). Foreign investors tend to be momentum investors, whereas domestic investors, particularly households, are likely to be
7
According to the Taiwan Stock Exchange (TWSE), the “Mechanism for Halting Trading during the Information Assessment Period” permits listed firms to have an
extensive and fair opportunity to deliver their material information, thereby providing investors with the time to digest messages. TWSE press release (2016, January
14) retrieved from http://www.twse.com.tw/en/about/press_room/tsec_news_detail.php?id=18086.
153
Table 1
Summary of relevant literature on different types of investor trading behavior.
Literature Highlights
Andrade et al. (2008) Changes in margin account holdings are the proxy for the non-informational trading imbalances in
[Taiwanese data] the model.
Barber and Odean (2008) Individual investors exhibit attention-driven buying behavior, whereas institutional investors are
[U.S. data] less likely to be influenced by attention-driven purchases.
Barber et al. (2009) Individual investors lose because of their aggressive orders. Institutional investors gain from trading.
[Taiwanese data] Foreign institutions collect nearly half of institutional profits.
Grinblatt and Keloharju (2000) Foreign investors tend to be momentum investors, whereas domestic investors, particularly
[Finnish data] households, are likely to be contrarians. The portfolios of foreign investors seem to outperform the
portfolios of households.
Grinblatt and Keloharju (2001) Sophisticated investors place less weight on past returns in deciding whether the trade is to be a buy
[Finnish data] or sell. By contrast, less sophisticated investors such a households, the general government, and non-
profit institutions are more predisposed to sell than to buy stocks with large past returns.
Hong and Stein (1999) The model characterizes two types of agents: ‘news-watchers’ and ‘momentum’ traders. Each type of
[theory and evidence] agent is only able to process some subset of the available public information.
Kaniel et al. (2008) The trading of individual investors exhibits predictive ability and is related to future short-horizon
[U.S. data] returns.
Kaniel et al. (2012) Informed individual investors' buying (selling) of stocks predicts positive (negative) abnormal
[U.S. data] returns around earnings announcement dates.
Nofsinger (2001) For the firm-specific news releases, individual investors sell on good news but not bad news.
[U.S. Wall Street Journal and macro-economic Institutions buy and sell on both good and bad news. For macroeconomic announcements, both good
announcements] and bad news induce trading by institutions and individuals.
Peress (2014) Newspaper strikes mainly deter retail investors from trading, while professional newswires used by
[France, Greece, Italy, Norway] money managers are less affected.
Shefrin and Statman (1985) The disposition to sell winners too early and hold losers too long. An extension from the behavioral
[theory and evidence] model which includes prospect theory, mental accounting, regret aversion, and self-control.
Tsai (2013) Individual investors are less informed about imminent corporate earnings announcements. Foreign
[Taiwanese data] institutions with superior expertise accrue profits by trading conservatively. Domestic institutions
with better local connections have access to privileged information.
Note: The stock markets or countries studied are in brackets.
contrarians. The portfolios of foreign investors seem to outperform the portfolios of households (Grinblatt and Keloharju, 2000).
Sophisticated investors place less weight on past returns in deciding whether the trade is to be a buy or sell. By contrast, less
sophisticated investors such as households, the general government, and non-profit institutions are more predisposed to sell than to
buy stocks with large past returns (Grinblatt and Keloharju, 2001). Individual investors are less informed about imminent corporate
earnings announcements. Foreign institutions with superior expertise accrue profits by trading conservatively. Domestic institutions
with better local connections have access to privileged information (Tsai, 2013).
Table 1 summarizes the relevant literature reviews on different types of investor trading behavior. This paper extends this strand
of research in which we are able to examine the trading behavior of a comprehensive range of investors, while few papers have been
successful in doing so.
2.2. News and stock returns
A substantial amount of research has focused on the market reaction to the public news initiated by firms or past stock return
patterns. How the market responds to the dissemination of news information is still one of the central issues of financial studies.
Merton (1987) formulates a model of capital market equilibrium with incomplete information where each investor knows only about
a subset of the available securities. Merton posits that media coverage plays an important causal role in affecting the value of a firm.
For example, a large number of investors, who are probably future shareholders, read a newspaper or other mass media story about
the firm or its industry. After evaluation, tentative investors become new shareholders and the value of the firm rises.
Based on the investor recognition hypothesis developed by Merton (1987), Fang and Peress (2009) examine the relationship between
media coverage and a cross-section of stock returns in which they find a ‘no media premium’. Specifically, stocks with no media
coverage earn higher returns than stocks with high media coverage even after controlling for popular risk factors such as the market,
size, the book-to-market ratio, momentum, and liquidity. In an incomplete information market, stocks with no media coverage are
less recognized by investors, and these stocks need to offer higher returns to compensate their holders for being imperfectly
diversified.
Prior studies show that stocks with news are found to exhibit momentum, in which stocks with bad news display a longer negative
drift (Chan, 2003). When there are diverse types of public information on the stock market, political news in comparison with
economic news has a distinct impact on the trading activity (Chan et al., 2001). Although the prior literature finds media coverage to
be related to stock returns, this paper expands the classification of media coverage.
154
2.3. Textual analysis of news
There is a growing body of finance and accounting research that focuses on the textual analysis of financial news and documents.
Loughran and McDonald (2016) provide a comprehensive review of the contemporary textual analysis literature. Several studies use
one of four dictionaries or a combination of them to capture the sentiment tone in media text news and financial disclosures. These
four dictionaries are Henry (2008), Harvard's General Inquirer (GI), Diction, and Loughran and McDonald (2011). However, some
negative words that appeared in the Diction and Harvard psychological dictionary list are typically not negative in a financial context
(Loughran and McDonald, 2015, 2016). Lu et al. (2013) develop a ‘distress intensity of default-corpus’ (DIDC) extracted from Chinese
financial news in predicting the future probability of distress of a firm.
Textual media content can also be used to predict firms' fundamentals (e.g., Tetlock, 2007; Tetlock et al., 2008; Tetlock, 2011).
The effect of media coverage on the stock market may differ across countries. Griffin et al. (2011) find that market reactions to public
news events are considerably stronger for firms in developed markets than for those in emerging markets. Cross-country differences
in market reactions to the information content of financial news are best explained by insider trading and the quality of news
transmission. Jegadeesh and Wu (2013) argue that there are relatively few papers that investigate in detail how investors interpret
descriptive information and whether investors efficiently incorporate that information into stock prices. Thus, this paper also seeks to
contribute to this line of the literature by providing empirical evidence.
3. Data description and variable measurement
This paper collects data on the news coverage of listed firms from the TEJ (Taiwan Economic Journal) database where the sources
of information include news announcements on the MOPS (Market Observation Post System), in several newspapers and on Internet
websites. Most news media companies issue newspapers and provide online news as well. Online news is posted in a timely manner
and reaches a broader readership regardless of their location.
In the early years, listed firms regularly announced information news on media outlets until August 2002, when the Taiwan Stock
Exchange launched the MOPS, a counterpart of the EDGAR (Electronic Data Gathering, Analysis, and Retrieval) system in the United
States. The effectiveness of the MOPS has gradually been improved since then to provide more information to the market. A new
version of the MOPS was introduced in July 2010, and the old version of the MOPS ceased to provide its information retrieval service
in December 2013. Since the sample studied covers the period from January 2001 to December 2014, a number of news
announcements concerning listed firms are subject to the transition from media outlets to the new MOPS era.
According to the Securities and Exchange Act of Taiwan, listed firms are required to publicly announce their financial and
material information. Based on the nature of the media coverage, this paper classifies news announcements into ten news categories.8
The classification of news categories can be divided into two aspects. On the one hand, the identification of those mandatory news
announcements made in the MOPS is straightforward and distinguishable among them, for example, the announcements of revenue
news, earnings news, acquisition or disposal of assets news, shares pledged news, dividend and shareholders' meeting news, transfers
of shares news, and share repurchase news. The aforementioned seven news categories are announced by firms according to the
stipulation of related regulations and they are clearly discernible. On the other hand, news releases on firm's operations, products, or
material events that appear in several newspapers and on Internet websites are classified as general news.
The daily trading activities of various investor types, other financial data and stock returns are also collected from the TEJ
database. This paper conducts the empirical examination on a monthly basis. As argued by Chan (2003), by using monthly data the
results can be compared with momentum studies and the microstructure problems that occur in daily or weekly data can be reduced.
A detailed description of the variables used in this paper is provided in Appendix A.
3.1. Investor trading behavior measures
One of the purposes of this paper is to examine how the trading behavior of different investor types relates to media coverage on a
firm-specific event. The proxy variable for trading behavior is the buy-sell order imbalance from investors. There are five types of
investor that can be identified: foreign institutions, investment trust companies, broker dealers, margin account holders, and non-
margin account holders. The Taiwan Stock Exchange (TWSE) collects and publishes the buy- and sell-trading activities of institutional
investors. Changes in margin accounts can represent individual trading behavior since they are owned by individual investors at local
brokerage companies. Andrade et al. (2008) use changes in margin account holdings from the TWSE as the proxy for the non-
informational trading imbalances in their model.
Recent empirical studies have found that different investor types follow different trading patterns (see Table 1 for a survey). As an
old adage states, “for every buyer there must be a seller”, a positive buy-sell imbalance from institutional investors induces a negative
buy-sell imbalance from individual investors, and vice versa. This paper deepens our understanding of the interplay among all-
inclusive types of investors in responding to the different news announcements.
To examine the trading patterns of investors around media coverage, this paper employs and slightly modifies the net individual
trading measure used in Kaniel et al. (2012) to accommodate the different types of investors. Specifically, the daily net abnormal
8
This paper applies textual analysis to the news content and classifies it into ten news categories. Among them, seven news categories have a positive or negative
tone and the remaining three are news announcements.
155
investor trading patterns are derived from an imbalance measure as follows:

InvNTi, d = OIBDOL i, d − AVGOIBDOL i, d , (1)
where InvNTi , d is the net abnormal investor trading for firm i on day d; OIBDOLi , d is the order imbalance for firm i on day d and is
defined as the buy dollar volume of firm i on day d less the sell dollar volume of firm i on day d, scaled by the average daily dollar
volume in the calendar year of firm i on day d; and AVG_OIBDOLi , d is the firm's average daily dollar volume over the sample period.
D
Then, the daily abnormal net abnormal investor trading is aggregated over the trading days [d, D] on a monthly basis as ∑ InvNTi, k .
k=d
3.2. Abnormal return measures
This paper uses the market model, the Fama-French three-factor model, and the Carhart four-factor model to measure the
abnormal returns. Fama (1998) argues that some empirical findings on the stock market anomalies are the products of deficiencies in
the return measures, and neither the CAPM nor the Fama-French model has complete descriptions of average returns. Thus, this paper
introduces the Carhart four-factor model (Carhart, 1997) to measure the abnormal returns since this model captures the risk factor of
return momentum in addition to other risk factors:
Ri, t − Rf , t = αi, t + bi, t [Rm, t − Rf , t ] + si, t SMB + hi, t HML + mi, t UMD + εi, t , (2)
where Ri , t is the monthly return for firm i; Rm , t is the monthly return on a value-weighted market index; Rf , t is the risk-free rate for
month t; SMB is the difference between the returns on the portfolios of small and big stocks; HML is the difference between the returns
on portfolios of high-B/M and low-B/M stocks; and UMD is the difference between the returns on portfolios of high-momentum and
low-momentum stocks. The procedure used to calculate returns on zero-investment factor-mimicking portfolios for both size and B/M
is similar to Fama and French (1993), and that for the one-year momentum in stock returns is closely related to Carhart (1997). Thus,
the intercept alpha αi , t is the abnormal return for both the Fama-French model and the Carhart model measures.
3.3. Textual analysis of media coverage9
The textual analysis of monthly revenue, quarterly and annual earnings news announcements is straightforward. The release of
monthly revenue or quarterly earnings news announcements can be linguistically neutral in that they solely report the results of firm's
operations, or they can be directional regarding the growth or decline as compared to the prior results. This paper defines revenue
and earnings announcements in which the word ‘growth’ appears as positive news and in which the word ‘decline’ appears as
negative news.10
Listed firms are required to publicly announce acquisitions of assets, such as share investments, debt investments, and plant,
property, and equipment, and provide information on the nature and the amount of the transactions. The disposal of investments
typically represents the financial results of the transactions, e.g., a gain on disposal, in which a gain on disposal suggests ‘positive’
news, whereas a loss on disposal indicates ‘negative’ news. However, in analyzing the content of investment news, listed firms rarely
incur a material loss as a result of news on disposal. Therefore, this paper only tests the impact of positive investment news.
In textual analysis where the financial news is in a Chinese context, Lu et al. (2013) pioneer the construction of a ‘distress intensity
of default-corpus’ (DIDC) extracted from Chinese financial news in predicting the future probability of distress of a firm. The special
terms for the distressed group are considered in this paper as part of a negative word list to gauge the negativity of media coverage.
Not all the special terms are included in the negative word list since several words such as sales, order, or demand, are just operating
terminologies. The linguistic tone of these words depends on whether they connect with words like increase or decrease.
The collection of negative words also complements that of Loughran and McDonald (2011) in that it gathers a more
comprehensive set of negative words. The use of this negative word list is only for general news, and the reasons are twofold.
First, previous studies find little incremental information in the Harvard positive word list and negative words convey more
information than positive words (Tetlock, 2007; Solomon, 2012). Second, there is no positive word list related to financial studies in
Chinese. Fortunately, Lu et al. (2013) provide a list of special terms for distressed firms that this paper employs as a negative word
list. Thus, this paper defines a news article in which at least one negative word appears as negative general news.
Shares pledged by board members and large shareholders of listed firms are required to be publicly announced. Empirical studies
reveal that shares pledged by the aforementioned persons lead to severe agency problems and impair firm performance (Kao et al.,
2004; Kao and Chen, 2007). This paper defines pledged shares as negative news.
The other three types of news announcements are dividend announcements, shares transferred by board members and large
shareholders, and share repurchases. Past theoretical works show that corporate dividends and share repurchases are signals of a
firm's solid financial condition and future earnings; however, the empirical evidence provides mixed results regarding the
effectiveness of the communication device of dividends (see DeAngelo et al., 2008 for a comprehensive review). Since these three
9
The authors would like to thank an anonymous referee for the constructive suggestions that led to this analysis.
10
Although various statistical methods are available for the selection of special terms, we select the critical words referring to the accounting meaning (e.g., gain on
disposal or shares pledged) or based on word lists from prior studies (e.g., extracted from Loughran and McDonald, 2011 and Lu et al., 2013). Since our paper focuses
on the analysis of investor trading behavior in response to the tone of news where human judgement rather than machine learning or information processing is
involved, those computational linguistic studies that apply text categorization or text mining to extract special terms do not fit our data features and research purposes.
156
types of news contain no positive or negative tone, and are just a statement of the facts, this paper counts the number of news articles
released.11
As argued by Chan et al. (2001), distinct kinds of news result in different types of information quality and investors have
perceptual biases. A recent study on the measure of disclosure quality posits that a greater degree of disaggregation of financial data
represents higher disclosure quality. Accordingly, Chen et al. (2015b) count the number of nonmissing financial items indicating
higher disclosure quality. In this paper we count the number of news articles released as a proxy for the quality of news, since the
more media coverage a firm receives, the higher the precision of the exposure acquired, thereby reducing information asymmetry in
the market.12
The classification of media news based on the tone of its content offers investors more information for investment valuation.
However, when there appears to be potentially conflicting news, for example, a firm may have positive earnings news and negative
general news at the same time, the responses of investors to these news items depend on their unobservable private information.
Thus, this paper controls for firm risks, characteristics and also investors' unobservable information processing.
In sum, we are able to extract three news categories of positive news (i.e., positive revenue news, positive earnings news, and
positive investment news), four news categories of negative news (i.e., negative general news, negative revenue news, negative
earnings news, and negative pledge news),13 and finally three news announcements (i.e., dividend news, share transfer news, and
share repurchase news).14 Detailed descriptions and examples of the ten news categories are shown in Appendix B, and the list of
negative words for general news appears in Appendix C.
3.4. Control variables
The trading behaviors of investors are likely to be affected by a number of firm risks and characteristics that this paper intends to
introduce as control variables. For instance, firm size, measured as the natural logarithm of the market value, is positively associated
with the recognition of investors. The book-to-market ratio represents a firm's growth opportunities, indicating that growing firms
capture more investor attention. Firms with higher idiosyncratic risk will have more negative media coverage.
In some cases, investors trade in the market because of liquidity needs rather than due to the influence of information news.
Stocks with significant changes in institutional ownership and recent high stock price momentum always attract the attention of
investors. Institutional ownership can be a proxy variable for the proportion of better-informed traders (Utama and Cready, 1997).
Moreover, momentum is also incorporated to control for the persistent continuation of stock prices in the market.
Prior papers have found that the stock market presents the underreaction and overreaction of stock prices subject to series of good
or bad news. Several cognitive and behavioral explanations are provided, for example, the conservatism and representativeness
heuristic (Barberis et al., 1998), overconfidence (Daniel et al., 1998), limited investor attention (Hirshleifer et al., 2009), and so forth.
All these works point to the existence of investor sentiment in the stock market. In addition, a higher share turnover implies that there
exists a divergence of opinion in the stock market. Thus, this paper introduces share turnover to control for possible influences of
investor sentiment on the quality of news (Baker and Wurgler, 2007).
In a stock market consisting of both sophisticated and unsophisticated investors, diverse investors interpret accounting
information differently so that it affects their trading behavior. Hand (1990) examines a stock market involving both sophisticated
and unsophisticated investors, in which sophisticated investors can properly see through information contained in a firm's financial
statements, whereas unsophisticated investors mechanically interpret a firm's accounting earnings and are misled by the firm's
accounting method and choices. Several empirical studies find that stock prices act as if investors fail to distinguish information
contained in the accrual and cash flow components of current earnings (Sloan, 1996; Xie, 2001). Since this paper examines monthly
revenue and earnings news announcements which are closely related to accounting numbers, the proxy variable of accruals is added
as a control variable as used in Jegadeesh and Wu (2013). For example, firms with large unexpected earnings tend to attract more
media coverage and firms having higher accruals suggests that there is earnings management or a downturn in business conditions.
The empirical evidence suggests that positive local media stories are related to the firm's local media advertising expenditures
(Gurun and Butler, 2012). To control for the possible influences of firms on the quality of news, this paper incorporates advertising
expense in the empirical investigation. The rationale behind this control is that firms with higher advertising expenditure can
establish better relationships with news media. Then, media agencies will tend to adopt a positive tone toward the news reporting.
11
This paper does not explicitly discuss the strength of news as Jegadeesh and Wu (2013) do for the following reasons. First, the media coverage examined in this
paper is classified based on its content into ten news categories, whereas Jegadeesh and Wu (2013) investigate the 10-K documents belonging to only one text category.
Second, those mandatory news announcements studied in this paper usually have standard format and the critical words can indicate whether a news article had a
positive or negative tone. While 10-K documents contain a lot of positive and negative words, it is necessary to compute the strength of various words in conveying a
positive or negative tone. Interestingly, our research results are comparable to Jegadeesh and Wu (2013) in which the majority of the positive or negative tone news
classified in this paper is significantly related to the same sign of abnormal returns.
12
In a market where there are both informed and uninformed investors, a higher level of exposure of a specific firm's news could reduce information asymmetry
between investors. More media coverage improves market efficiency through the increased amount of news.
13
Our news classification rules based on the critical word that appears in the content of news have certain limitations. The classification indeed only reflects the
current status of a specific event with a positive or negative tone, and the interpretation of a positive or negative tone contained in news depends on the investors'
sophistication and judgement.
14
This paper implicitly uses an equal weighting scheme for the tone of news articles. Although more complex weighting techniques are potentially advantageous in
certain contexts, equal-weighted tone measures are generally just as powerful in the context of financial disclosure and capital markets (Henry and Leone, 2016). In
addition, this paper uses critical words to gauge the tone of a news article, and thus counting the frequency of the critical words becomes unnecessary.
157
When firms announce their financial reports, in which revenue, earnings, investment, and other information news are also
revealed to the market, investors must be selective in view of the abundance of information available in the market. Libby et al.
(2002) conduct a laboratory experiment and find that limited attention affects both naïve and sophisticated individual investors, as
well as financial professionals interpreting accounting numbers. Because both individual investors and finance professionals pay
attention to earnings, the business media discuss earnings much more than cash flow and accrual numbers such that investors tend to
neglect the latter information (see Hirshleifer et al., 2011 for a review). Thus, the trading behavior of different types of investors
depends on what information or news attracts their attention and how they interpret it.15
Accordingly, this paper addresses the self-selection of different types of investors regarding several news releases. The trading
behavior of each investor may be affected by possible behavioral biases. The unobservable information processing of investors can be
accommodated to the investor's private information. Thus, we estimate the inverse Mills ratio (Lambda) from the Heckman two-step
procedure and introduce it as a control for potential private information effects. We employ capital intensity as an instrumental
variable that is only included in the first-stage probit model.16 To meet the exclusion restrictions, capital intensity is excluded in the
second-stage regression model. The correlations between capital intensity and net abnormal investor trading are examined in which
they are largely negative and insignificantly different from zero (e.g., the correlation coefficient is − 0.0003 for foreign institutions,
−0.0009 for investment trust companies, −0.0012 for broker dealers, −0.0008 for margin account holders, and 0.0012 for non-
margin account holders, respectively). Detailed definitions of control variables are provided in Appendix A.
4. Empirical results
4.1. Descriptive statistics
4.1.1. News categories by industry

Firms belonging to the same industry generally have similar products or services, technology innovation, and business
environments. The attention of media coverage may, however, differ among industries. For instance, news media focuses more on
technology industries because of product innovations as compared to mature industries. By contrast, firms within the same industry
may be influenced because information about other firms in the same industry is reported by news media.
Table 2 presents the distribution of news categories by industry. The largest share of news articles reported by the media is
concentrated in the electronics sector. In the last column of Table 2, the top five industries most reported by the media are computers
peripheral equipment, semiconductors, optoelectronics, electronic parts and components, and communication and network
industries. The products in these industries are subjected to fluctuating business cycles and demands from international buyers.
Thus, the numbers of negative general news items are also the highest among the various industries. It is worth noting that Table 2
summarizes the numbers of news categories by industry, and so the result is affected by the sizes of firms within an industry.
The last five industries that are the least covered by the media are the glass and ceramic, oil and gas, tourism, paper, and
information service industries. These industries are smaller in size and their products or services are less subject to rapid
technological innovations. By aggregating the total number of news items across industries, the last row in Table 2 shows 122,258
negative general news items, followed by 47,877 positive revenue news items and 33,230 negative revenue news items, respectively.
The least reported news items consist of 4200 share repurchase news items.
The attention of media coverage is generally focused more on emerging and growing industries. The number of news items
reported is associated with the sizes of firms in an industry. Thus, the purpose of Table 3 is to provide the distribution of average news
reported by industry. Evidently, the ranking order in the last column has changed. The top five industries with the highest average
number of news items reported are the automobile, computers peripheral equipment, semiconductors, optoelectronics, and
transportation industries. Compared with other industries, the automobile and transportation industries have on average 3.918
and 1.139 positive investment news items, and 2.233 and 1.253 negative general news items, respectively.
By sorting the average number of news items covered, the industries with the least news reported are the tourism, glass and
ceramic, biochemical and pharmaceutical, oil and gas, and textile industries. The ranking order for the industries with the least
average news reported has changed as compared to that of Table 2. Some of these industries provide services rather than goods to
their customers, such as the tourism industry, and they are less engaged in investment in assets, share transfers or share repurchase
activities. In general, mature industries such as the glass and ceramic industries have changed slowly in terms of product design and
innovation, so they are less attractive to the media coverage. The only exception is the biochemical and pharmaceutical industry,
where firms belonging to this industry are young and some pharmaceutical products are still in their research and development
phase.
15
Henry (2008) finds that the tone of earnings press releases affects investors' reactions and the result is explained by prospect theory. Hirshleifer and Teoh (2009)
provide a detailed discussion on the psychological attraction approach to accounting and disclosure policy. The authors posit that investors have cognitive processing
constraints on aggregated numbers in accounting reports. Unsophisticated investors may focus only on those reports and items that are tailored for them.
16
In the first-stage probit model the dependent variable takes a value of 1 if the stock return is positive, and 0 otherwise, where its independent variables include
positive news items, all control variables listed above as well as capital intensity. The latter variable is obtained from the prior disclosure literature as an instrument
and it is defined as the ratio of property, plant, and equipment to total assets (Larcker and Rusticus, 2010), and in the second-stage regression model we exclude it. We
also test other instrumental variables such as one-year sales growth, the operating margin, and the length of the operating cycle. The empirical results are qualitatively
and largely similar to those for capital intensity.
158
C.-H. Wu, C.-J. Lin
Table 2
Distribution of total news categories by industry.
Industry Positive Positive Positive Negative Negative Negative Negative Dividend Share Share Total
revenue news earnings news investment news general news revenue news earnings news pledge news news transfer repurchase [Rank]
news news
1. Cement 406 258 518 1567 313 219 288 282 237 42 4130 [22]
2. Food 965 722 481 2137 446 535 216 735 650 46 6933 [17]
3. Plastic 1457 987 666 4965 786 799 341 850 523 84 11,458 [13]
4. Textile 2672 1844 872 3198 2551 1450 389 1568 957 225 15,726 [8]
5. Electrical machinery 2564 1702 500 3768 1717 1121 503 1316 710 169 14,070 [10]
6. Electric equipment 902 545 393 1329 660 467 130 487 458 120 5491 [18]
7. Chemical 1577 1138 481 2264 803 911 160 968 804 109 9215 [14]
8. Biochemical & pharmaceutical 994 641 426 978 469 409 82 574 272 40 4885 [20]
9. Glass & ceramic 232 151 37 252 156 121 86 150 148 6 1339 [27]
10. Paper 401 218 468 1409 348 215 58 241 304 51 3713 [24]
11. Iron & steel 2196 1558 1554 6048 1424 1097 586 1032 625 136 16,256 [7]
12. Rubber 682 494 70 1340 326 314 251 391 885 53 4806 [21]
159
13. Automobile 295 362 3111 1773 308 274 216 197 470 16 7022 [16]
14. Semiconductors 4079 2224 2958 19,011 2604 1745 757 1809 1923 447 37,557 [2]
15. Computers peripheral equipment 3748 2526 6764 15,149 3000 1833 405 1853 2405 508 38,191 [1]
16. Optoelectronics 4119 2501 2161 17,601 3011 1989 678 1623 1428 400 35,511 [3]
17. Communication & network 2463 1580 1121 9177 1618 1371 433 1204 1362 283 20,612 [5]
18. Electronic parts & components 5125 3546 1839 9514 3029 2808 604 2707 2232 570 31,974 [4]
19. Electronic distribution 1598 1132 381 2545 883 757 198 799 749 162 9204 [15]
20. Information service 846 443 305 659 568 389 41 449 223 105 4028 [23]
21. Other electronic 2185 1439 601 4218 1628 1174 532 1129 1171 215 14,292 [9]
22. Construction 3660 1345 405 4238 3411 957 634 1545 1199 148 17,542 [6]
23. Transportation 1009 647 3416 3756 821 609 283 702 772 43 12,058 [12]
24. Tourism 474 253 165 790 247 167 112 336 178 21 2743 [25]
25. Trade & wholesale 407 411 1289 1174 393 275 314 422 177 83 4945 [19]
26. Oil & gas 595 230 305 650 179 191 89 347 71 29 2686 [26]
27. Other 2226 1868 916 2748 1531 1217 507 1412 1262 89 13,776 [11]
Total 47,877 30,765 32,203 122,258 33,230 23,414 8893 25,128 22,195 4200 350,163
Note: The sample period extends from January 2001 to December 2014. The last column Rank is sorted in descending order. Detailed news categories definitions can be found in Appendix A.
C.-H. Wu, C.-J. Lin
Table 3
Distribution of average news reported by industry.
Industry Positive Positive Positive Negative Negative Negative Negative Dividend Share Share Total
revenue news earnings news investment news general news revenue news earnings news pledge news news transfer repurchase news [Rank]
news
1. Cement 0.345 0.219 0.440 1.332 0.266 0.186 0.245 0.240 0.202 0.036 3.512 [7]
2. Food 0.288 0.215 0.144 0.638 0.133 0.160 0.064 0.219 0.194 0.014 2.068 [22]
3. Plastic 0.407 0.276 0.186 1.388 0.220 0.223 0.095 0.238 0.146 0.023 3.202 [9]
4. Textile 0.351 0.242 0.114 0.420 0.335 0.190 0.051 0.206 0.126 0.030 2.064 [23]
5. Electrical machinery 0.441 0.293 0.086 0.648 0.295 0.193 0.087 0.226 0.122 0.029 2.421 [18]
6. Electric equipment 0.403 0.243 0.176 0.594 0.295 0.209 0.058 0.218 0.205 0.054 2.452 [17]
7. Chemical 0.405 0.292 0.123 0.581 0.206 0.234 0.041 0.249 0.206 0.028 2.366 [19]
8. Biochemical & pharmaceutical 0.414 0.267 0.178 0.408 0.195 0.170 0.034 0.239 0.113 0.017 2.035 [25]
9. Glass & ceramic 0.338 0.220 0.054 0.367 0.227 0.176 0.125 0.218 0.215 0.009 1.949 [26]
10. Paper 0.341 0.185 0.398 1.198 0.296 0.183 0.049 0.205 0.259 0.043 3.157 [10]
11. Iron & steel 0.470 0.333 0.333 1.294 0.305 0.235 0.125 0.221 0.134 0.029 3.479 [8]
12. Rubber 0.424 0.307 0.044 0.833 0.203 0.195 0.156 0.243 0.550 0.033 2.987 [12]
160
13. Automobile 0.372 0.456 3.918 2.233 0.388 0.345 0.272 0.248 0.592 0.020 8.844 [1]
14. Semiconductors 0.477 0.260 0.346 2.222 0.304 0.204 0.089 0.211 0.225 0.052 4.391 [3]
15. Computers peripheral equipment 0.450 0.303 0.812 1.818 0.360 0.220 0.049 0.222 0.289 0.061 4.584 [2]
16. Optoelectronics 0.500 0.303 0.262 2.135 0.365 0.241 0.082 0.197 0.173 0.049 4.307 [4]
17. Communication & network 0.459 0.295 0.209 1.711 0.302 0.256 0.081 0.225 0.254 0.053 3.843 [6]
18. Electronic parts & components 0.449 0.311 0.161 0.834 0.266 0.246 0.053 0.237 0.196 0.050 2.804 [15]
19. Electronic distribution 0.487 0.345 0.116 0.776 0.269 0.231 0.060 0.244 0.228 0.049 2.806 [14]
20. Information service 0.437 0.229 0.157 0.340 0.293 0.201 0.021 0.232 0.115 0.054 2.078 [21]
21. Other electronic 0.468 0.308 0.129 0.904 0.349 0.252 0.114 0.242 0.251 0.046 3.063 [11]
22. Construction 0.522 0.192 0.058 0.604 0.487 0.137 0.090 0.220 0.171 0.021 2.502 [16]
23. Transportation 0.337 0.216 1.139 1.253 0.274 0.203 0.094 0.234 0.258 0.014 4.022 [5]
24. Tourism 0.331 0.177 0.115 0.551 0.172 0.117 0.078 0.234 0.124 0.015 1.914 [27]
25. Trade & wholesale 0.239 0.241 0.757 0.689 0.231 0.161 0.184 0.248 0.104 0.049 2.904 [13]
26. Oil & gas 0.455 0.176 0.233 0.497 0.137 0.146 0.068 0.265 0.054 0.022 2.052 [24]
27. Other 0.372 0.312 0.153 0.459 0.256 0.203 0.085 0.236 0.211 0.015 2.303 [20]
Note: The sample period extends from January 2001 to December 2014. The last column Rank is sorted in descending order. Detailed news categories definitions can be found in Appendix A.
Table 4
Distribution of news categories by year.
Year Positive Positive Positive Negative Negative Negative Negative Dividend Share Share Total
revenue earnings investment general revenue earnings pledge news transfer repurchase [Rank]
news news news news news news news news news
2001 2914 2664 4388 5593 2880 2761 903 2609 984 395 26,091 [6]
2002 3709 2479 6026 5608 1889 1563 631 2837 1272 230 26,244 [5]
2003 3815 2600 3890 10,162 1518 1382 597 3125 1647 281 29,017 [4]
2004 4394 2761 3893 10,050 1220 884 593 3156 1983 513 29,447 [2]
2005 3190 1840 2836 8018 2048 1437 487 3047 1914 317 25,134 [8]
2006 3288 1881 1624 7538 1815 1001 520 1003 2047 310 21,027 [12]
2007 3669 1940 1981 10,587 1917 864 602 1096 2637 195 25,488 [7]
2008 3410 1087 1777 13,489 2310 1165 1208 1183 2730 749 29,108 [3]
2009 1684 1428 1281 13,994 5119 2310 503 1028 2103 177 29,627 [1]
2010 5540 2849 1276 7594 1650 1080 484 1146 1659 104 23,382 [10]
2011 3579 2255 1099 9179 2492 2224 826 1294 1458 412 24,818 [9]
2012 2504 2174 921 8145 3419 3304 506 1201 697 253 23,124 [11]
2013 2787 2040 831 6510 2835 1667 570 1165 590 124 19,119 [13]
2014 3394 2767 380 5791 2118 1772 463 1238 474 140 18,537 [14]
Total 47,877 30,765 32,203 122,258 33,230 23,414 8893 25,128 22,195 4200 350,163
Note: The sample period extends from January 2001 to December 2014. Detailed news categories definitions can be found in Appendix A.
4.1.2. News categories by year

A firm's business operations may be affected by a market-wide event that further influences the quantity and quality of news
information provided. For example, when a global financial crisis that originated in the United States with a failure in the lending
market broke out in 2008–2009, it rapidly spread across all economic sectors. It also had an impact on the worldwide equity market,
and many countries encountered sharp equity market crashes, which were even more severe than in the United States (Bekaert et al.,
2014). Thus, the purpose of this subsection is to provide the number of news categories by year and to shed light on the amount of
media coverage between the non-crisis period and crisis period.
Table 4 presents the distribution of different types of news categories by year. The highest amount of media coverage appears in
2009 with a total of 29,627 news items. Evidently, the two highest numbers of negative general news items occurred during the
period of the financial crisis in 2008 and 2009, in which the government decided to distribute consumer vouchers amounting to NT$
3600 (about US$115) for every qualified citizen. Because of the downturn in the economy, consumer vouchers were used in the hope
of improving consumers' purchasing power. As a result, the number of negative revenue news items (5119) reached its peak in 2009,
and the number of negative earnings news items was also high.
Due to the financial crisis, there was a credit squeeze for borrowers seeking loans from banks. Board members or large
shareholders of listed firms that usually pledged their shares to borrow money would pledge more shares during a credit squeeze. In
2008, the number of news announcements regarding the shares pledged (1208) of listed companies was almost twice the number for
such media coverage in other years. News of share transfers and share repurchases remained at a high level of media coverage as well.
By contrast, the numbers of news items related to revenue, earnings, and investment with a positive tone reflected a low level of
media coverage during the financial crisis period.
4.2. Correlation between variables
In a stock market where there exists heterogeneity in investor beliefs, the diffusion of news information into the market may
trigger different investor trading behavior. It is predicted that news announcements with a positive tone will induce buying behavior,
whereas those announcements with a negative tone will result in selling behavior on the part of sophisticated investors. An increase
in trading volume suggests that sophisticated investors with better information processing buy (sell) stocks, while unsophisticated
investors sell (buy) stocks.
The first part of Table 5 provides the correlation coefficients between the news categories and the net abnormal investor trading of
the five investor types. It is expected that the trading behavior of investors with better information processing will be positively
correlated with the tone of news announcements. Specifically, positive news results in stock being bought and negative news leads to
stock being sold.
Among the three kinds of institutional investors, the trading behavior of foreign institutions is consistent with the tone of each
news category. For example, the coefficients of a positive tone in revenue news and earnings news exhibit statistically significant
positive correlations, while the coefficients of a negative tone in general news, revenue news, earnings news, and pledge news exhibit
statistically significant negative correlations. The trading behaviors of investment trust companies and broker dealers are less
correlated with the tone of the news reported.
Institutional investors generally use advanced information technology to select stocks. By contrast, individual investors have
limited attention to choose which stock to buy (Barber and Odean, 2008). The majority of margin account holders are composed of
individual investors. Table 5 shows that the trading behavior of margin account holders is insignificantly correlated with the tone of
the news. The non-margin account holders exhibit the opposite sign of correlation with the tone of the news. In sum, this result finds
161
Table 5
Correlation between news categories, investor trading behavior and abnormal returns.
News categories Net abnormal investor trading Abnormal returns
Foreign Investment trust Broker dealer Margin Non-margin Market model Fama-French Carhart Alpha
institutions account account Alpha
Pos. revenue news 0.020⁎⁎⁎ 0.002 − 0.009⁎⁎⁎ 0.000 − 0.010⁎⁎⁎ 0.061⁎⁎⁎ 0.066⁎⁎⁎ 0.036⁎⁎⁎
Pos. earnings news 0.017⁎⁎⁎ 0.006⁎ 0.002 0.001 − 0.011⁎⁎⁎ 0.038⁎⁎⁎ 0.053⁎⁎⁎ 0.029⁎⁎⁎
Pos. investment 0.000 − 0.000 − 0.001 0.002 − 0.002 − 0.003 − 0.003 0.001
news
Neg. general news − 0.010⁎⁎⁎ 0.003 0.001 0.000 0.004 − 0.006⁎⁎ − 0.001 − 0.000
Neg. revenue news − 0.009⁎⁎⁎ 0.000 0.006⁎ 0.001 0.003 − 0.017⁎⁎⁎ − 0.028⁎⁎⁎ − 0.016⁎⁎⁎
Neg. earnings news − 0.014⁎⁎⁎ − 0.009⁎⁎⁎ − 0.002 0.005 0.006⁎ − 0.043⁎⁎⁎ − 0.034⁎⁎⁎ − 0.014⁎⁎⁎
Neg. pledge news − 0.016⁎⁎⁎ − 0.004 − 0.003 − 0.001 0.010⁎⁎⁎ − 0.017⁎⁎⁎ − 0.010⁎⁎⁎ − 0.004
Dividend news 0.013⁎⁎⁎ − 0.002 0.011⁎⁎⁎ − 0.015⁎⁎⁎ 0.005 0.041⁎⁎⁎ 0.046⁎⁎⁎ 0.028⁎⁎⁎
Share transfer news 0.006⁎⁎ 0.011⁎⁎⁎ 0.000 0.003 − 0.009⁎⁎⁎ 0.027⁎⁎⁎ 0.019⁎⁎⁎ 0.008⁎⁎⁎
Share repurchase − 0.019⁎⁎⁎ − 0.008⁎⁎ 0.001 − 0.017⁎⁎⁎ 0.026⁎⁎⁎ − 0.012⁎⁎⁎ 0.004 0.003
news
Note: The sample period extends from January 2001 to December 2014. Detailed news categories definitions can be found in Appendix A.
⁎
Denotes significance at the 10% level.
⁎⁎
⁎⁎⁎
support for the view that individual investors are less attentive to the news and foreign institutional investors have superior ability in
terms of processing news of different categories.
The second part of Table 5 presents the correlation coefficients between media coverage and the abnormal returns. If the tones of
different news categories are fairly measured, then the three measures of abnormal returns used in this paper should have the same
sign of the coefficient to the tone of the news content. In other words, positive news generates positive abnormal returns and negative
news causes negative abnormal returns. For negative general news, the three measures of abnormal returns exhibit a negative
correlation with only the market model reaching statistical significance. The announcements of revenue, earnings, and pledged shares
news with both a positive and negative tone are mostly consistent and significantly correlated with the sign of the abnormal returns.
Finally, the announcements of dividend news generate statistically significant correlation with abnormal returns indicating that the
distribution of dividends is perceived as good news in the market.
4.3. News categories and investor trading behavior
4.3.1. Methodology and specification

The relationship between investor trading behavior and news categories is preliminarily examined based on the correlation
between the variables. While other factors may affect the trading behavior of different types of investors, these possible factors are
introduced to estimate the following multivariate regression:
InvNTi, t = a + b1 Pos Re venueNewsi, t + b 2 PosEarningsNewsi, t

+ b3 PosInvestmentNewsi, t + b4 NegGeneralNewsi, t + b5 Neg Re venueNewsi, t
+ b6 NegEarningsNewsi, t + b 7 NegPledgeNewsi, t + b8 DividendNewsi, t
+ b 9 ShareTransferNewsi, t + b10 ShareRepurchaseNewsi, t + bϕ Controls
+ bφ Month YearFixedEffect + bγ IndustryFixedEffect + εi, t . (3)
The dependent variable is the net abnormal investor trading InvNTi , t. The variables of interest are the ten types of news categories
and they are defined as the natural logarithm of one plus the number of articles for each news category. The control variables are firm
size, the book-to-market ratio, idiosyncratic risk, standard unexpected earnings, illiquidity, institutional ownership, momentum,
trading turnover, accruals, advertising expense, and investors' unobservable private information lambda. Descriptive statistics of
these control variables are provided in Table 6.
In the analyses in the prior subsection, the empirical evidence shows that news categories vary from industry to industry (see
Tables 2 and 3), and each year also has a different amount of media coverage especially during the global financial crisis period (see
Table 4). Thus, Eq. (3) includes month, year and industry fixed effects.17 This means that each month, year, or industry may have a
differential impact on the different investor trading behavior.18 The standard errors are adjusted for heteroskedasticity and clustered
17
One caveat deserves mention here. While we include month and year dummies in the regression analyses, these procedures are insufficient to remove the
seasonality when the seasonality is stochastic, especially in Chinese society where many festivals are based on the lunar calendar. We thank the anonymous referee for
providing this valuable comment.
18
For example, listed firms are required to publicly announce the results of their monthly revenue and quarterly financial statements. This implies that for each
month different types of investors may have different responses to the news announcements.
162
Table 6
Descriptive statistics.
Variable Obs Mean Std. Min. Max.
Market value of equity (Log) 111,227 8.521 1.448 3.466 15.114

Book-to-market 110,552 0.948 0.699 0.005 30.851
Idiosyncratic risk 111,227 0.125 0.081 0.004 3.099
Standard unexpected earnings 109,897 − 0.002 0.177 −10.938 12.821
Illiquidity 110,818 2.747 16.572 0.000 839.186
Institutional ownership 111,227 0.390 0.224 0.000 1.000
Momentum 107,217 0.158 0.619 −0.947 21.143
Trading turnover 111,227 0.167 0.217 0.000 2.911
Accruals 110,367 − 0.004 0.068 −0.969 1.024
Advertising expense 111,227 4.384 4.845 0.000 16.649
Lambda 100,351 0.797 0.251 0.000 5.705
Note: This table provides descriptive statistics for the variables used in this paper. The sample period extends from January 2001 to December 2014. The definitions of
variables can be found in Appendix A.
by firm, allowing each firm to have its differential effect on investor trading behavior.19
4.3.2. Results
According to the investor recognition hypothesis and the attention-grabbing hypothesis (Merton, 1987; Barber and Odean, 2008),
the trading behaviors of investors are affected by the news information that attracts their attention. Unlike institutional investors that
have a complete set of information technology with which to search and analyze stocks that interest them, individual investors rely
more on public information because of its availability and low cost of retrieval. Thus, media coverage is a primary source of public
information.
The reverse relationship between the trading behaviors of different investor types is deduced from an adding-up constraint.
Specifically, each buy transaction coincides with a sell transaction (Grinblatt and Keloharju, 2000). Thus, if institutional investors
become net buyers when particular news is released, individual investors become net sellers exhibiting the reverse trading behavior.
Table 7 presents the empirical results of the relationship between investor trading behavior and news categories after accounting
for the effects of several firm risks, characteristics and investors' unobservable private information. Panel A of Table 7 reports the
trading behavior of five types of investors under the full sample. Among the three types of institutional investors, the buy-sell
imbalances of foreign institutions are significantly and positively associated with positive revenue news, positive earnings news,
announcements of dividends and share transfer news, but significantly and negatively related to negative general news, negative
earnings news, negative pledge news, and announcements of share repurchases. Although the trading behaviors of investment trust
companies and broker dealers in part coincide with the tone of the news categories, the trading behaviors of foreign institutions are
substantially consistent with the tone of the news coverage.
By contrast, the buy-sell imbalances of margin account and non-margin account holders show the opposite trading behaviors to
those of institutional investors, of which the majority are composed of domestic and retail investors. For example, the buy-sell
imbalances of foreign institutions to positive earnings news (coefficient = 0.092, t-value = 3.71) and to negative earnings news
(coefficient = − 0.059, t-value = − 2.39) are significantly consistent with the tone of earnings news, whereas the buy-sell
imbalances of margin account holders to positive earnings news (coefficient = −0.160, t-value = −3.62) exhibit the opposite
trading behavior to the tone of earnings news releases.
It is observed that the coefficient of negative general news is negatively related to the trading behavior of the three types of
institutional investors and margin account holders. In particular, foreign institutions are the only type of investors who become
significant net sellers on the release of negative general news (coefficient = −0.046, t-value = −3.09). When all investors sell
stocks in the face of negative media coverage on general news, non-margin account holders are the only ones who buy stocks
(coefficient = 0.092, t-value = 3.50).
Previous empirical findings on shares pledged by board members and large shareholders exhibit impairment to firm performance.
Evidently, the shares pledged of listed firms are viewed as negative news to the market. The three types of institutional investors
avoid holding stocks with pledged shares news announcements. Foreign institutions (coefficient = −0.168, t-value = − 5.13) and
investment trust companies (coefficient = −0.036, t-value = − 2.36) exhibit a significant selling of stock. However, non-margin
account holders are the only type of investor that become net buyers of stocks (coefficient = 0.210, t-value = 5.59).
After introducing the month/year and industry fixed effects and other control variables, there are no significant responses on the
part of all types of investors to the positive investment news except for investment trust companies. One possible explanation is that
investors are not attentive to the frequent announcements of gains on the disposal of assets. In regard to the share transfer news
announcements, foreign institutions (coefficient = 0.072, t-value = 2.63) and investment trust companies (coefficient = 0.033, t-
value = 3.46) become net buyers, and non-margin account holders become net sellers of stock (coefficient = −0.145, t-
19
Petersen (2009) suggests that when both a firm and a time effect are present in the data, researchers can address one of them parametrically (e.g., by including
time dummies) and then estimate standard errors clustered on the other dimension.
163
Table 7
News categories and investor trading behavior.
Variables Foreign institutions Investment trust Broker dealer Margin account Non-margin account
Panel A: Full sample 2001–2014

News categories
Pos. revenue news 0.063 − 0.008 − 0.011 −0.096 0.053
(2.90)⁎⁎⁎ (−0.70) (−2.08)⁎⁎ (− 1.63) (0.82)
Pos. earnings news 0.092 − 0.004 0.011 −0.160 0.062
(3.71)⁎⁎⁎ (−0.28) (1.78)⁎ (− 3.62)⁎⁎⁎ (1.11)
Pos. investment news − 0.012 0.016 0.001 0.018 − 0.024
(−0.48) (1.95)⁎ (0.38) (1.43) (− 0.90)
Neg. general news − 0.046 − 0.003 − 0.002 −0.041 0.092
(−3.09)⁎⁎⁎ (−0.44) (−0.87) (− 1.65) (3.50)⁎⁎⁎
Neg. revenue news 0.025 0.000 0.006 −0.002 − 0.029
(1.05) (0.01) (1.11) (− 0.08) (− 0.81)
Neg. earnings news − 0.059 − 0.026 − 0.003 0.082 0.006
(−2.39)⁎⁎ (−2.13)⁎⁎ (−0.48) (1.01) (0.07)
Neg. pledge news − 0.168 − 0.036 − 0.009 0.003 0.210
(−5.13)⁎⁎⁎ (−2.36)⁎⁎ (−1.27) (0.12) (5.59)⁎⁎⁎
Dividend news 0.048 − 0.002 0.025 −0.140 0.069
(2.15)⁎⁎ (−0.15) (3.57)⁎⁎⁎ (− 5.20)⁎⁎⁎ (1.92)⁎
Share transfer news 0.072 0.033 0.006 0.034 − 0.145
(2.63)⁎⁎⁎ (3.46)⁎⁎⁎ (1.43) (1.31) (− 3.73)⁎⁎⁎
Share repurchase news − 0.217 − 0.044 0.015 −0.354 0.600
(−4.62)⁎⁎⁎ (−2.40)⁎⁎ (1.05) (− 5.52)⁎⁎⁎ (7.63)⁎⁎⁎
Control variables
Firm size 0.043 − 0.005 0.002 −0.016 − 0.025
(6.17)⁎⁎⁎ (−1.39) (1.17) (− 1.05) (− 1.43)
Book-to-market − 0.070 0.018 − 0.004 0.217 − 0.161
(−3.73)⁎⁎⁎ (1.34) (−0.85) (2.52)⁎⁎ (− 1.77)⁎
Idiosyncratic risk − 0.434 0.019 0.080 0.230 0.103
(−3.58)⁎⁎⁎ (0.35) (2.40)⁎⁎ (0.84) (0.34)
SUE 0.007 − 0.003 0.003 −0.150 0.143
(0.29) (−0.24) (0.67) (− 1.03) (0.93)
Illiquidity 0.002 0.000 − 0.000 0.000 − 0.002
(2.31)⁎⁎ (4.35)⁎⁎⁎ (−0.40) (0.01) (− 2.32)⁎⁎
Institutional ownership − 0.203 − 0.001 − 0.002 0.135 0.071
(−6.63)⁎⁎⁎ (−0.07) (−0.34) (2.01)⁎⁎ (0.89)
Momentum 0.047 − 0.038 − 0.008 0.253 − 0.253
(1.70)⁎ (−1.82)⁎ (−1.28) (3.58)⁎⁎⁎ (− 3.04)⁎⁎⁎
Trading turnover 0.061 0.317 0.060 −0.350 − 0.088
(0.71) (3.99)⁎⁎⁎ (2.47)⁎⁎ (− 1.48) (− 0.33)
Accruals 0.006 − 0.083 0.025 0.206 − 0.154
(0.05) (−1.66)⁎ (0.87) (1.66)⁎ (− 0.85)
Advertising expense − 0.001 0.000 − 0.001 0.003 − 0.001
(−0.82) (0.03) (−1.75)⁎ (2.47)⁎⁎ (− 0.39)
Lambda 0.215 − 0.173 0.007 −1.516 1.466
(1.76)⁎ (−1.66)⁎ (0.20) (− 4.30)⁎⁎⁎ (3.68)⁎⁎⁎
Constant − 0.261 0.041 − 0.057 1.297 − 1.020
(−1.77)⁎ (0.31) (−0.94) (2.32)⁎⁎ (− 1.64)
Month/year fixed effect Yes Yes Yes Yes Yes
Industry fixed effect Yes Yes Yes Yes Yes
Observations 100,351 100,351 100,351 100,351 100,351
R-squared 0.007 0.011 0.002 0.005 0.009
Panel B: Sample excluding the global financial crisis 2008–2009

News categories
Pos. revenue news 0.072 − 0.006 − 0.012 −0.112 0.059
(2.99)⁎⁎⁎ (−0.49) (−2.21)⁎⁎ (− 1.64) (0.80)
Pos. earnings news 0.103 0.019 0.013 −0.194 0.058
⁎⁎⁎
(3.65) (1.12) (1.91)⁎ (− 3.39)⁎⁎⁎ (0.84)
Pos. investment news − 0.024 0.014 0.002 0.027 − 0.019
(−0.95) (1.60) (0.62) (1.96)⁎⁎ (− 0.69)
Neg. general news − 0.030 − 0.000 − 0.003 −0.042 0.075
(−1.72)⁎ (−0.02) (−1.14) (− 1.44) (2.50)⁎⁎
Neg. revenue news 0.024 0.001 0.003 0.010 − 0.038
(0.94) (0.06) (0.46) (0.36) (− 1.06)
Neg. earnings news − 0.068 − 0.028 − 0.002 0.103 − 0.006
(−2.57)⁎⁎⁎ (−2.19)⁎⁎ (−0.22) (1.09) (− 0.06)
Neg. pledge news − 0.139 − 0.030 − 0.007 0.004 0.172
(−4.15)⁎⁎⁎ (−1.70)⁎ (−0.84) (0.15) (3.85)⁎⁎⁎
(continued on next page)
164
Table 7 (continued)
Variables Foreign institutions Investment trust Broker dealer Margin account Non-margin account
Dividend news 0.076 − 0.002 0.024 −0.141 0.042

(3.10)⁎⁎⁎ (−0.15) (3.25)⁎⁎⁎ (− 4.63)⁎⁎⁎ (1.07)
Share transfer news 0.090 0.035 0.006 0.038 − 0.169
(2.99)⁎⁎⁎ (2.97)⁎⁎⁎ (1.10) (1.16) (− 3.68)⁎⁎⁎
Share repurchase news − 0.199 − 0.026 0.009 −0.346 0.562
(−3.85)⁎⁎⁎ (−1.25) (0.54) (− 4.54)⁎⁎⁎ (6.07)⁎⁎⁎
Constant − 0.352 − 0.079 − 0.041 1.620 − 1.148
(−1.99)⁎⁎ (−0.63) (−0.72) (2.53)⁎⁎ (− 1.61)
Control variables Yes Yes Yes Yes Yes
Month/year fixed effect Yes Yes Yes Yes Yes
Industry fixed effect Yes Yes Yes Yes Yes
Observations 85,253 85,253 85,253 85,253 85,253
R-squared 0.007 0.012 0.002 0.004 0.007
Note: This table reports the impact of different news categories on investor trading behavior but excluding January data. The dependent variable is the net abnormal
investor trading of the five investor types. The sample period extends from January 2001 to December 2014. T-statistics are based on standard errors adjusted for
heteroskedasticity and clustered by firm as shown in the parentheses. The definitions of variables can be found in Appendix A.
⁎
⁎⁎
⁎⁎⁎
value = − 3.73). Since there are no detailed data on who receive shares transferred by board members and large shareholders, the
empirical evidence shows that these institutions are net buyers.
Foreign institutions, investment trust companies, and margin account holders dislike listed firms conducting share repurchases.
The regression coefficients for these three types of investors are significantly negative, whereas that of non-margin account holders is
significantly positive (coefficient = 0.600, t-value = 7.63). Finally, foreign institutions and broker dealers favor dividend news
showing significantly positive trading behavior, while margin account holders become net sellers (coefficient = −0.140, t-
value = − 5.20).
Panel B of Table 7 presents the regression results of excluding the global financial crisis 2008–2009 observations.20 The empirical
findings are qualitatively similar to those of the full sample in Panel A of Table 7, but with some exceptions that deserve to be
mentioned. First, the trading behavior of foreign institutions associated with negative general news becomes marginally significant.
This suggests that the significant negative coefficient on negative general news (shown in Panel A) is highlighted by the sample
covering the financial crisis period. Once observations from the years 2008 to 2009 are removed from the sample, the statistical
significance diminishes as shown in Panel B. Second, the trading behavior of investment trust companies becomes insignificantly
related to share repurchase news.
In sum, Table 7 examines the relationship between investor trading behavior and news categories by introducing several firm
risks, characteristics and investors' unobservable private information as control variables. In terms of the textual analysis on news
categories that enables us to distinguish between a positive and negative tone, the trading behavior of foreign institutional investors
is found to be consistent with the tone of media coverage, whereas margin account holders and non-margin account holders are likely
to exhibit the opposite trading behavior. The trading activities of investment trust companies coincide in part with the tone of media
coverage, whereas broker dealers have no eminent systematic trading behavior regarding the news categories. This result in part
lends support to the notion that institutional investors, especially foreign institutions, are better informed and have superior
information processing. If some types of investors pursue buy-sell trading strategies triggered by the tone of media coverage, other
types of investors must follow the opposite trading strategies.21 As argued by Grinblatt and Keloharju (2000), if some investors are
winners, others must be losers.
4.4. News categories and stock returns
4.4.1. Methodology and specification

The relationship between stock returns and news categories is also examined in this section. The signs of the correlation
coefficients in Table 5 show that the relationships between the tone of the news categories and abnormal returns are generally
consistent with expectations. That is, positive news generates positive abnormal returns and negative news causes negative abnormal
returns. In the next test, the relationship between stock returns and news categories is examined after accounting for the effect of
control variables in the following multivariate regression:
20
The authors thank an anonymous referee for providing this insight.
21
We conduct sensitivity analyses by removing all of the January or February data to evaluate whether the empirical results are affected by the Chinese New Year
season. Untabulated results are qualitatively and largely similar to those of Table 7.
165
ARi, t = a + b1 Pos Re venueNewsi, t + b 2 PosEarningsNewsi, t + b3 PosInvestmentNewsi, t

+ b4 NegGeneralNewsi, t + b5 Neg Re venueNewsi, t + b6 NegEarningsNewsi, t
+ b 7 NegPledgeNewsi, t + b8 DividendNewsi, t + b 9 ShareTransferNewsi, t
+ b10 ShareRepurchaseNewsi, t + bϕ Controls + bφ Month YearFixedEffect
+ bγ IndustryFixedEffect + εi, t . (4)
The dependent variable ARi , t is the abnormal return. The ten types of news categories are measured as the natural logarithm of
one plus the number of articles in each news category. The control variables, provided in Table 6, are firm size, the book-to-market
ratio, idiosyncratic risk, standard unexpected earnings, illiquidity, institutional ownership, momentum, trading turnover, accruals,
and advertising expense. As in Eq. (3), month, year, and industry fixed effects are introduced in the multivariate regression (4).
Standard errors are adjusted for heteroskedasticity and clustered by firm.
4.4.2. Results
The performance of stock returns related to an information release can be interpreted as good news or bad news. Based on the
textual analysis of different news categories employed in this paper, it is predicted that positive news causes positive abnormal stock
returns, while negative news leads to negative abnormal stock returns. The preliminary correlation analysis in Table 5 confirms this
relationship prediction.
Table 8 presents the empirical results of the relationship between three measures of abnormal returns and news categories after
accounting for the effects of several firm risks and characteristics. Panel A of Table 8 shows that those news categories classified as
positive (e.g., revenue news and earnings news) are significantly and positively associated with abnormal returns. At the same time,
the majority of news categories classified as negative (e.g., general news, revenue news, earnings news, and pledge news) are
significantly and negatively related to abnormal returns. The results are robust across the three measures of abnormal returns.
The stock market does not perceive positive investment news to be substantial. By scrutinizing the original data, in the statements
of positive investment news appear the words ‘gain on disposal’, but the amount of the gain is small due to the disposal of frequently
traded short-term investments. This fact explains the insignificant stock return reaction to the positive investment news
announcements.
When listed firms announce share repurchases or the distribution of dividends to the market, managers are sending signals of
future earnings prospects. Thus, these two types of news categories are significantly and positively related to the three measures of
abnormal returns as shown in Panel A of Table 8, in which the empirical evidence finds support for the signaling theory.
Panel B of Table 8 presents the regression results of excluding the global financial crisis 2008–2009 observations. The empirical
findings are qualitatively similar to those of the full sample in Panel A of Table 8. There is only one exception to the positive revenue
news, in which the coefficient turns to be insignificant under the market model measure after excluding the financial crisis years. In
sum, this paper applies textual analysis to different types of news categories, and those news announcements that can be distilled into
a positive tone or negative tone are significantly and positively associated with the abnormal returns. In other words, the textual
analysis of news circulated in the stock market has an impact on stock returns.
We conduct sensitivity analyses by removing the entire January or February data to evaluate whether the empirical results are
affected by the Chinese New Year season. The untabulated results appear to be qualitatively and largely similar to those of Table 8.
The results are stable and robust under abnormal returns measured using the Fama-French Alpha and Carhart Alpha. However, the
results for abnormal returns based on the market model are more sensitive to the season effect. Specifically, after removing February
data, the regression coefficients of positive revenue news become statistically insignificant. One possible explanation is that the
market model does not account for the size, value, and momentum risk factors in the market and it is more susceptible to the season
effect.
4.4.3. Sensitivity analysis

Even though we incorporate investors' unobservable private information as a control variable, the interpretation of a diverse
source of media coverage could be affected by the concurrent conflicting news.22 We list all potentially conflicting news as follows:
(1) Pos. Revenue News and Neg. General News; (2) Pos. Earnings News and Neg. General News; (3) Pos. Investment News and Neg.
General News; (4) Pos. Revenue News and Neg. Earnings News; (5) Pos. Earnings News and Neg. Revenue News; (6) Pos. Investment
News and Neg. Revenue News; (7) Pos. Revenue News and Neg. Pledge News; (8) Pos. Investment News and Neg. Earnings News. In
untabulated analyses, we replicate the empirical analyses of Tables 7 and 8 by excluding those observations with conflicting news
from the sample and the results are qualitatively similar. Collectively, the possible conflicting news does not change our main
inferences.
5. Conclusion
The investigation of the effect of media coverage on the stock market is important since media news improves transparency and
reduces information asymmetry. Although institutional investors are considered to be sophisticated traders in the stock market,
media news reported about a firm could influence the trading behavior of both institutional and individual investors.
22
The authors thank the anonymous referee for providing this insightful comment.
166
Table 8
News categories and stock returns.
Market model Fama-French Alpha Carhart Alpha
Panel A: Full sample 2001–2014

News categories
Pos. revenue news 0.002 0.008 0.008
(1.79)⁎ (7.18)⁎⁎⁎ (3.04)⁎⁎⁎
Pos. earnings news 0.015 0.016 0.014
(11.03)⁎⁎⁎ (13.63)⁎⁎⁎ (11.22)⁎⁎⁎
Pos. investment news − 0.001 −0.000 − 0.000
(− 1.20) (− 0.38) (− 0.39)
Neg. general news − 0.004 −0.003 − 0.002
(− 5.13)⁎⁎⁎ (− 3.17)⁎⁎⁎ (− 2.76)⁎⁎⁎
Neg. revenue news − 0.011 −0.011 − 0.013
(− 8.48)⁎⁎⁎ (− 8.65)⁎⁎⁎ (− 5.12)⁎⁎⁎
Neg. earnings news − 0.011 −0.012 − 0.010
(− 7.91)⁎⁎⁎ (− 8.82)⁎⁎⁎ (− 6.72)⁎⁎⁎
Neg. pledge news − 0.017 −0.011 − 0.010
(− 7.81)⁎⁎⁎ (− 5.70)⁎⁎⁎ (− 5.04)⁎⁎⁎
Dividend news 0.013 0.012 0.010
(10.10)⁎⁎⁎ (8.91)⁎⁎⁎ (6.31)⁎⁎⁎
Share transfer news 0.006 0.003 0.003
(4.25)⁎⁎⁎ (2.16)⁎⁎ (2.64)⁎⁎⁎
Share repurchase news 0.006 0.014 0.013
(2.39)⁎⁎ (5.67)⁎⁎⁎ (4.50)⁎⁎⁎
Control variables
Firm size 0.002 0.000 0.001
(3.01)⁎⁎⁎ (0.86) (0.93)
Book-to-market − 0.020 −0.018 − 0.017
(− 15.09)⁎⁎⁎ (− 15.13)⁎⁎⁎ (− 10.74)⁎⁎⁎
Idiosyncratic risk 0.233 0.156 0.104
(17.14)⁎⁎⁎ (14.38)⁎⁎⁎ (3.49)⁎⁎⁎
SUE 0.013 0.015 0.019
(3.15)⁎⁎⁎ (3.69)⁎⁎⁎ (3.98)⁎⁎⁎
Illiquidity 0.000 0.000 0.000
(0.25) (0.47) (0.45)
Institutional ownership 0.011 0.007 0.002
(3.42)⁎⁎⁎ (2.72)⁎⁎⁎ (0.42)
Momentum − 0.034 −0.026 − 0.027
(− 20.30)⁎⁎⁎ (− 19.43)⁎⁎⁎ (− 5.83)⁎⁎⁎
Trading turnover 0.147 0.087 0.091
(36.09)⁎⁎⁎ (25.97)⁎⁎⁎ (14.25)⁎⁎⁎
Accruals − 0.010 0.005 0.008
(− 1.28) (0.56) (0.92)
Advertising expense − 0.000 −0.000 − 0.000
(− 3.19)⁎⁎⁎ (− 3.85)⁎⁎⁎ (− 3.90)⁎⁎⁎
Constant − 0.054 −0.028 − 0.012
(− 6.68)⁎⁎⁎ (− 4.45)⁎⁎⁎ (− 1.77)⁎
Month/year fixed effect Yes Yes Yes
Industry fixed effect Yes Yes Yes
Observations 106,124 106,068 106,068
R-squared 0.117 0.058 0.023
Panel B: Sample excluding the global financial crisis 2008–2009

News categories
Pos. revenue news 0.001 0.007 0.008
(0.87) (6.30)⁎⁎⁎ (2.93)⁎⁎⁎
Pos. earnings news 0.017 0.016 0.015
(11.97)⁎⁎⁎ (13.10)⁎⁎⁎ (10.98)⁎⁎⁎
Pos. investment news − 0.002 −0.000 − 0.001
(− 1.57) (− 0.46) (− 0.71)
Neg. general news − 0.004 −0.003 − 0.002
(− 3.89)⁎⁎⁎ (− 3.43)⁎⁎⁎ (− 2.70)⁎⁎⁎
Neg. revenue news − 0.011 −0.012 − 0.014
(− 7.95)⁎⁎⁎ (− 8.84)⁎⁎⁎ (− 4.87)⁎⁎⁎
Neg. earnings news − 0.011 −0.012 − 0.011
(− 6.99)⁎⁎⁎ (− 8.33)⁎⁎⁎ (− 6.30)⁎⁎⁎
Neg. pledge news − 0.016 −0.010 − 0.009
(− 7.07)⁎⁎⁎ (− 4.98)⁎⁎⁎ (− 4.20)⁎⁎⁎
Dividend news 0.009 0.010 0.010
(6.91)⁎⁎⁎ (7.47)⁎⁎⁎ (6.17)⁎⁎⁎
(continued on next page)
167
Table 8 (continued)
Market model Fama-French Alpha Carhart Alpha
Share transfer news 0.007 0.004 0.005

(4.28)⁎⁎⁎ (2.87)⁎⁎⁎ (3.40)⁎⁎⁎
Share repurchase news 0.008 0.015 0.015
(2.86)⁎⁎⁎ (5.78)⁎⁎⁎ (4.55)⁎⁎⁎
Constant − 0.062 −0.030 − 0.022
(− 7.24)⁎⁎⁎ (− 4.59)⁎⁎⁎ (− 3.06)⁎⁎⁎
Control variables Yes Yes Yes
Month/year fixed effect Yes Yes Yes
Industry fixed effect Yes Yes Yes
Observations 90,259 90,221 90,221
R-squared 0.127 0.062 0.024
Note: This table reports the abnormal returns calculated from the market model, Fama-French three-factor model, and Carhart four-factor model. The sample period
extends from January 2001 to December 2014. T-statistics are based on standard errors adjusted for heteroskedasticity and clustered by firm as shown in the
parentheses. The definitions of variables can be found in Appendix A.
⁎
⁎⁎
⁎⁎⁎
This paper contributes to the literature on the tone analysis of media coverage and its influence on investor trading behavior and
stock returns. Based on different news categories, this paper is able to classify seven of them into positive and negative tone news, as
well as three types of news announcements. The last three are dividend news, share transfer news, and share repurchase news.
The empirical results show that investor trading behavior is affected not only by the quantity, but also the quality, of news
announcements. The trading behavior of foreign institutional investors is consistent with the tone of media coverage. When the news
appears with a positive (negative) tone, such investors become net buyers (sellers). Non-margin account holders exhibit the opposite
trading behavior. The evidence suggests that foreign institutional investors have superior information processing to the other types of
investors. The results are robust after accounting for several firm risks, characteristics and investors' unobservable private
information. Moreover, the majority of positive and negative news announcements classified in this paper have the same direction
and significant effects on abnormal returns.
The increasing importance of textual analysis to news information is evident by its influence on the stock market. For investors in
the market, a better understanding of the content of media coverage generates better investment decisions on aggregate. For firms'
managers, an appropriate release of a firm's specific events through media coverage can reduce information asymmetry. For policy-
makers, the awareness of the role of the media can improve stock market efficiency. This paper is only able to distill a negative tone
from general news. Thus, constructing a complete list of positive words for general news opens a promising avenue for future
research.
Acknowledgements
We would like to thank Jun-Koo Kang (Editor) and an anonymous referee for their valuable comments and suggestions that
substantially improved the paper. Wu acknowledges financial support from the Ministry of Science of Technology in Taiwan (Grant
number: NSC 101-2410-H-259-021).
Appendix A. Variable definitions
Variables Definitions
News categories
Revenue news Monthly revenue announcements on the MOPS. (According to Article 36, paragraph 3, of the Taiwan
Securities and Exchange Act, listed companies are required to file information on sales revenue for the
preceding month by the tenth day of each calendar month.)
Earnings news Firm's quarterly and annual financial reports, earnings per share, and net income announcements. The
sources of earnings news are derived from the MOPS. (According to Article 36, paragraphs 1 and 2, of
the Taiwan Securities and Exchange Act, listed companies are required to file and publicly announce
their quarterly and annual financial reports.)
Investment news Announcement news on the firm's acquisition or disposal of assets such as share investments, debt
investments, and plant, property, and equipment on the MOPS. (According to Article 36-1 of the Taiwan
Securities and Exchange Act, listed companies are required to file and announce the acquisition or
disposal of assets.)
General news News on the firm's operations, products, or material events that appear on media coverage. (News
coverage of listed firms from the TEJ database where the sources of information include several
168
newspapers and Internet websites such as the Commercial Times, Economic Daily News, Digitimes, www.
investor.com.tw, and MoneyDJ.)
Pledged shares news Shares pledged by directors, supervisors and largest shareholders announcements on the MOPS.
(According to Article 25 of the Taiwan Securities and Exchange Act, when the shares are pledged by
directors, supervisors, managerial officers, and large shareholders, the company shall publicly announce
such pledges within five days of their formation.)
Dividend news News on the firm's dividend and shareholders' meeting announcements on the MOPS.
Share transfer news Transfer of shares by directors and supervisors announcements on the MOPS. (According to Articles 22-2
and 25 of the Taiwan Securities and Exchange Act, the transfer of stocks and shares held by directors,
supervisors, managerial officers, and shareholders holding more than 10% of the total shares of the
company shall be filed and public announcements made of such information.)
Share repurchase news Announcement news on the repurchase of shares on the MOPS. (According to Article 28-2 of the Taiwan
Securities and Exchange Act, the company shall publicly announce the procedure, price, quantity,
method, and conversion method of a share repurchase.)
Investor trading and stock returns
Net abnormal investor The difference between order imbalance in dollar volume and the average order imbalance in dollar
trading volume over the sample period calculated for different investor types.
Market model Abnormal returns calculated using the market model.
Fama-French Alpha Daily alphas calculated from regressing the time series of zero-investment factor-mimicking portfolio
returns on the Fama and French (1993) three-factor model, and converting to monthly abnormal returns
by multiplying the mean alpha by the number of trading days in each month.
Carhart Alpha Daily alphas calculated from regressing the time series of zero-investment factor-mimicking portfolio
returns on the Carhart (1997) four-factor model, and converting to monthly abnormal returns by
multiplying the mean alpha by the number of trading days in each month.
Control variables
Firm size Natural logarithm of market capitalization.
Book-to-market Book value of stockholders' equity divided by the market value of equity.
Idiosyncratic risk Standard deviation of monthly stock returns for the previous 12 months.
Standard unexpected Using the most current quarterly EPS reported minus EPS from four quarters ago, deflated by the price
earnings per share unadjusted for stock and cash dividends at the end of the month.
Illiquidity Monthly average illiquidity measure constructed in the spirit of Amihud (2002), which gives the
absolute price change per dollar of daily trading volume.
Institutional ownership Number of shares held by institutional investors.
Momentum Cumulative raw return for the previous 12 months.
Trading turnover Trading volume divided by the number of shares outstanding.
Accruals Accruals are computed as in Sloan (1996), that is, the one-year change in current assets excluding cash
and cash equivalents minus the change in current liabilities excluding debt in current liabilities and
taxes payable minus depreciation and amortization, divided by average total assets.
Advertising expense Log (1 + advertising expense).
Lambda The inverse Mills ratio estimated from the Heckman two-step procedure. In the first-stage probit model
the dependent variable takes a value of 1 if the stock return is positive, and 0 otherwise, where its
independent variables include positive news items, all control variables listed above as well as capital
intensity. The latter variable is an instrumental variable to meet the exclusion restrictions, in which in
the second-stage regression model excludes capital intensity.
Appendix B. Ten news categories descriptions
A. Positive revenue news

Listed firms are required to publicly announce their monthly sales revenue. A typical example of positive revenue news
announcement is as follows:
• 2012/08/12 TSMC announced July revenue totaling NT$ 47,924,170 thousand, single month amount YoY growth of 37.25%,
accumulated revenue from January through July in 2012 totaling NT$ 278,673,453 thousand, with accumulated amount YoY
growth of 13.60%.
The distinction of a positive revenue news announcement is based on the critical word growth.
B. Positive earnings news
An example of a typical positive earnings news announcement can be found in the following excerpt:
169
• 2012/05/15 ASUS announced consolidated financial reports for the period of January through March 2012 totaling NT$
584,136 million of consolidated income before income taxes, and a strong YoY growth of 50.01% as compared to last year, for
the period of January through March 2012 totaling NT$ 500,516 million of net income, earnings before income taxes per share
of NT$7.67, and earnings after income taxes per share of NT$6.65.
The distinction of a positive earnings news announcement is based on the critical word growth.
C. Positive investment news
The announcement of a firm's acquisition of assets provides information on the nature and the amount of the transaction.
Examples are given by:
• 2012/06/26 Uni-President on 2012/06/26 buys 47,747 thousand shares in the Yuanta Wan Tai Money Market Fund, the
transaction amount being NT$ 7000 million.
• 2012/08/16 Uni-President on 2012/08/16 sells 47,747 thousand shares in the Yuanta Wan Tai Money Market Fund, the
transaction amount being NT$ 7007.3 million, the gain on disposal being NT$ 7.3 million.
The disposal of investment news typically presents the financial result of the transaction, e.g., the gain on disposal, in which a
gain on disposal suggests ‘positive’ news, whereas a loss on disposal indicates ‘negative’ news. However, in analyzing the
investment news, listed firms rarely have a material loss on disposal news.
D. Negative general news
Examples of media coverage with respect to a firm's operations, products, or material events are:
• 2011/02/24 The Secretary General Wang Lung-Chie of Tatung Company said yesterday that Tatung Company has finalized the
filing of capital changes registration due to a capital reduction. The net worth will increase from NT$5.98 to NT$14.20.
• 2011/06/15 Since the fundamental has changed from bullish to conservative, Macquarie downgrades HTC's rating.
The list of negative words is mainly extracted from Lu et al. (2013) and Loughran and McDonald (2011) as shown in Appendix C.
E. Negative revenue news
Listed firms are required to publicly announce their monthly sales revenue. A typical example of negative revenue news
announcement is as follows:
• 2012/10/09 CSC announced September revenue totaling NT$ 15,200,826 thousand, a single month amount YoY decline of
− 25.83%, accumulated revenue from January through September in 2012 totaling NT$ 160,817,367 thousand, with
accumulated amount YoY decline of −11.64%.
The distinction of a negative revenue news announcement is based on the critical word decline.
F. Negative earnings news
An example of a typical negative earnings news announcement can be found in the following excerpt:
• 2014/05/15 China Airlines announced consolidated financial reports for the period of January through March 2014 totaling
NT$ − 2512 million of consolidated income before income taxes, and a serious YoY decline of 139.15% as compared to last
year, for the period of January through March 2014 totaling NT$ − 2727 million of net income, consolidated earnings before
income taxes per share of NT$-0.48, and earnings after income taxes per share of NT$-0.52.
The distinction of a negative earnings news announcement is based on the critical word decline.
G. Negative pledge news
Examples of announcements on shares pledged by directors, supervisors, managerial officers, and large shareholders are as
follows:
• 2012/04/27 The Heng Qiang Investment Co. Chairman of TCC terminates 11,500 pledged shares on 2012/04/27, with
accumulated pledged shares totaling 13,600 thousand shares in Taiwan Finance Co., accounting for 19.02% of shareholdings.
• 2012/06/22 The Heng Qiang Investment Co. Chairman of TCC pledges 6000 thousand shares on 2012/06/22, with
accumulated pledged shares totaling 31,100 thousand shares in China Bills Finance Co., accounting for 43.50% of
shareholdings.
Pledged shares news provides two kinds of information, the first one is the ‘shares pledged’ and the second one is the ‘termination
of pledged shares’. This paper considers shares pledged as ‘negative’ news due to the possible impairment to corporate
governance. However, the ‘termination of pledged shares’ is not considered to be ‘positive’ news because several terminations only
account for a small fraction of pledged shares.
H. Dividend news
Examples of dividend and shareholders' meeting announcements are:
• 2012/07/13 Hon-Hai announced the ex-dividend date on 2012/08/10, the last registration date being 2012/08/13, the cash
dividend per share NT$ 1.50, and the payment date 2012/09/12.
• 2004/03/15 The Board of Directors of TCC decided on 2004/03/15 that the annual shareholders' meeting of 2003 will take
place on 2004/06/11, with the distribution of dividends per share being NT$ 0.70.
I. Share transfer news
The transfer of shares by directors, supervisors, managerial officers, and large shareholders of a listed company usually has the
170
following statement:
• 2012/07/04 Managerial officer Chang Gwo-Yang files to transfer 61 thousand shares, accounting for 18.1% of his original
shareholdings of 339 thousand shares.
J. Share repurchase news
Examples of the announcement on the buyback of treasury shares are as follows:
• 2012/10/03 The Ritek Board decided on 2012/10/02 to anticipate the repurchase of 30,000 thousand treasury shares from
2012/10/01 through 2012/12/02, accounting for 1.13% of the firm's total outstanding shares, the expected buyback price
ranging from NT$ 3.10–4.10, and the expected upper limit amount of the repurchase being NT$ 14.1million.
• 2012/10/19 Ritek from 2012/10/01 through 2012/10/19 buys back treasury shares amounting to 30,000 thousand shares,
the total amount of repurchases being NT$ 96.46 million, and accumulated treasury shares being 32,500 thousand shares,
accounting for 1.23% of the firm's total outstanding shares.
Appendix C. List of negative words extracted from Loughran and McDonald (2011) and Lu et al. (2013)
Adverse, adversely (pu-li) Difficult (kun-nan) Negative, negatively
Against (ti-kang) Directors (tun-chien-shih) (fu, fou-ting)

Bounced check (tiao-piao) Discontinued (ting-yeh) Nonperforming (pu-liang)
Capital financing (hsien-tseng) Dispose (chu-fen) Not enough (pu-tsu)
Capital formation (tseng-tzu) Doubtful (i-lu) Not good enough (pu-chia)
Capital-reducing (chien-tzu) Downgrade (tiao-chiang) Over-fall (tieh-po)
Claims (so-pei) Downward price limit (tieh-ting) Plaintiffs (yuan-kao)
Closing, closed (kuan-pi) Enlargement (kuo-ta) Pledge (chih-ya)
Complaint (kung-kao) Exposed (pu) Plunge (chung-tso)
Critical (wei-chi) Failure (shih-pai) Refund (tui-piao)
Cum-rights (tien-chuan) Force (po-shih) Restated (chung-pien)
Damages (pei-chang) Impairment, impaired (chien-sun) Restructuring (chung-tsu)
Decline, declined, declines Limitations (hsien-chih) Termination, terminated
(hsia-chiang, shuai-tui, chu-chueh) Litigation (su-sung) (chung-chih, chieh-shu)
Default (wei-yueh) Loss, losses (kui-sun) Transformation (chuan-hsing)
Defendants (pei-kao) M & A (kou-ping) Unable (wu-fa)
Delays (yen-chih) Margin buying (jung-tzu) Volatility (po-tung-hsing)
Note: The Chinese language transcribed using Roman spelling is presented in parentheses.
References
Agarwal, Sumit, Faircloth, Sheri, Liu, Chunlin, Rhee, S.G., 2009. Why do foreign investors underperform domestic investors in trading activities? Evidence from
Indonesia. J. Financ. Mark. 12 (1), 32–53.
Amihud, Yakov, 2002. Illiquidity and stock returns: cross-section and time-series effects. J. Financ. Mark. 5 (1), 31–56.
Andrade, Sandro C., Chang, Charles, Seasholes, Mark S., 2008. Trading imbalances, predictable reversals, and cross-stock price pressure. J. Financ. Econ. 88 (2),
406–423.
Bae, Sung C., Min, Jae Hoon, Jung, Sunbong, 2011. Trading behavior, performance, and stock preferences of foreigners, local institutions, and individual investors:
evidence from the Korean stock market. Asia-Pac. J. Financ. Stud. 40 (2), 199–239.
Baker, Malcom, Wurgler, Jeffrey, 2007. Investor sentiment in the stock market. J. Econ. Perspect. 21 (2), 129–151.
Barber, Brad M., Odean, Terrance, 2008. All that glitters: the effect of attention and news on the buying behavior of individual and institutional investors. Rev. Financ.
Stud. 21 (2), 785–818.
Barber, Brad M., Lee, Yi-Tsung, Liu, Yu-Jane, Odean, Terrance, 2009. Just how much do individual investors lose by trading. Rev. Financ. Stud. 22 (2), 609–632.
Barberis, Nicholas, Shleifer, Andrei, Vishny, Robert, 1998. A model of investor sentiment. J. Financ. Econ. 49, 307–343.
Bekaert, Greert, Ehrmann, Michael, Fratzscher, Marcel, Mehl, Arnaud, 2014. The global crisis and equity market contagion. J. Financ. 69 (6), 2597–2649.
Brennan, Michael, Cao, Henry H., 1997. International portfolio flows. J. Financ. 52 (5), 1851–1880.
Bushee, Brian J., Core, John E., Guay, Wayne, Hamm, Sophia J.W., 2010. The role of the business press as an information intermediary. J. Account. Res. 48 (1), 1–19.
Carhart, Mark M., 1997. On persistence in mutual fund performance. J. Financ. 52 (1), 57–82.
Chan, Wesley S., 2003. Stock price reaction to news and no-news: drift and reversal after headlines. J. Financ. Econ. 70 (2), 223–260.
Chan, Yue-Cheong, Chui, Andy C.W., Kwok, Chuck C.Y., 2001. The impact of salient political and economic news on the trading activity. Pac. Basin Financ. J. 9 (3),
195–217.
Chen, Ruey-Shii, Hung, Pi-Hsia, Tsai, You-Da, 2015a. How does the public information of market observation post system affect stock return? Evidence from intra- and
inter-day analyses in Taiwan Stock Exchange. Review of Securities and Futures Market 27 (2), 125–184.
Chen, Shuping, Miao, Bin, Shevlin, Terry, 2015b. A new measure of disclosure quality: the level of disaggregation of accounting data in annual reports. J. Account. Res.
53 (5), 1017–1054.
Daniel, Kent, Hirshleifer, David, Subramanyam, Avanidhar, 1998. Investor psychology and security market under- and overreactions. J. Financ. 53 (6), 1839–1885.
DeAngelo, Harry, DeAngelo, Linda, Skinner, Douglas J., 2008. Corporate payout policy. Foundations and Trends in Finance 3 (2–3), 95–287.
171
Dvorak, Tomas, 2005. Do domestic investors have an informational advantage? Evidence from Indonesia. J. Financ. 60 (2), 817–839.
Fama, Eugene F., 1998. Market efficiency, long-term returns, and behavioral finance. J. Financ. Econ. 49 (3), 283–306.
Fama, Eugene F., French, Kenneth R., 1993. Common risk factors in the returns on stocks and bonds. J. Financ. Econ. 33 (1), 3–56.
Fang, Lily, Peress, Joel, 2009. Media coverage and the cross-section of stock returns. J. Financ. 64 (5), 2023–2052.
Griffin, John M., Hirschey, Nicholas H., Kelly, Patrick J., 2011. How important is the financial media in global markets? Rev. Financ. Stud. 24 (12), 3941–3992.
Grinblatt, Mark, Keloharju, Matti, 2000. The investment behavior and performance of various investor types: a study of Finland's unique data set. J. Financ. Econ. 55
(1), 43–67.
Grinblatt, Mark, Keloharju, Matti, 2001. Who makes investors trade? J. Financ. 56 (2), 589–616.
Grossman, Sanford J., Stiglitz, Joseph E., 1980. On the impossibility of informationally efficient markets. Am. Econ. Rev. 70 (3), 393–408.
Gurun, Umit G., Butler, Alexander W., 2012. Don't believe the hype: local media slant, local advertising, and firm value. J. Financ. 67 (2), 561–598.
Hand, John R.M., 1990. A test of the extended functional fixation hypothesis. Account. Rev. 65 (4), 740–763.
He, Wen, Shen, Jiangfeng, 2014. Do foreign investors improve informational efficiency of stock prices? Evidence from Japan. Pac. Basin Financ. J. 27, 32–48.
Henry, Elaine, 2008. Are investors influenced by how earnings press releases are written? J. Bus. Commun. 45 (4), 363–407.
Henry, Elaine, Leone, Andrew J., 2016. Measuring qualitative information in capital markets research: comparison of alternative methodologies to measure disclosure.
Account. Rev. 91 (1), 153–178.
Hirshleifer, David, Teoh, Siew Hong, 2009. The psychological attraction approach to accounting and disclosure policy. Contemp. Account. Res. 26 (4), 1067–1090.
Hirshleifer, David, Lim, Sonya S., Teoh, Siew Hong, 2009. Driven to distraction: extraneous events and underreaction to earnings news. J. Financ. 64 (5), 2289–2325.
Hirshleifer, David, Lim, Sonya S., Teoh, Siew Hong, 2011. Limited investor attention and stock market misreactions to accounting information. Rev Asset Pric Stud 1
(1), 35–73.
Hong, Harrison, Stein, Jeremy C., 1999. A unified theory of underreaction, momentum trading and overreaction in asset markets. J. Financ. 54 (6), 2143–2184.
Hong, Harrison, Stein, Jeremy C., 2000. Bad news travels slowly: size, analyst coverage, and the profitability of momentum strategies. J. Financ. 55 (1), 265–295.
Jegadeesh, Narasimhan, Wu, Di, 2013. Word power: a new approach for content analysis. J. Financ. Econ. 110 (3), 712–729.
Kaniel, Ron, Saar, Gideon, Titman, Sheridan D., 2008. Individual investor trading and stock returns. J. Financ. 63 (1), 273–310.
Kaniel, Ron, Liu, Shuming, Saar, Gideon, Titman, Sheridan D., 2012. Individual investor trading and return patterns around earnings announcements. J. Financ. 67 (2),
639–680.
Kao, Lanfen, Chen, Anlin, 2007. Directors' share collateralization, earnings management and firm performance. Taiwan Accounting Review 6 (2), 153–172.
Kao, Lanfen, Chiou, Jeng-Ren, Chen, Anlin, 2004. The agency problems, firm performance and monitoring mechanisms: the evidence from collateralized shares in
Taiwan. Corp. Gov. 12 (3), 389–402.
Kothari, S.P., 2001. Capital markets research in accounting. J. Account. Econ. 31, 105–231.
Kothari, S.P., Xu, Li, Short, James E., 2009. The effect of disclosures by management, analysts, and business press on cost of capital, return volatility and analyst
forecasts: a study using content analysis. Account. Rev. 84 (5), 1639–1670.
Kyle, Albert S., 1985. Continuous auctions and insider trading. Econometrica 53 (6), 1315–1336.
Larcker, David F., Rusticus, Tjomme O., 2010. On the instrumental variables in accounting research. J. Account. Econ. 49, 186–205.
Libby, Robert, Bloomfield, Robert, Nelson, Mark W., 2002. Experimental research in financial accounting. Acc. Organ. Soc. 27 (8), 775–810.
Loughran, Tim, McDonald, Bill, 2011. When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. J. Financ. 66 (1), 35–65.
Loughran, Tim, McDonald, Bill, 2014. Measuring readability in financial disclosures. J. Financ. 69 (4), 1643–1671.
Loughran, Tim, McDonald, Bill, 2015. The use of word lists in textual analysis. J. Behav. Finance 16 (1), 1–11.
Loughran, Tim, McDonald, Bill, 2016. Textual analysis in accounting and finance: a survey. J. Account. Res. 54 (4), 1187–1230.
Lu, Yang-Chen, Shen, Chung-Hua, Wei, Yu-Chen, 2013. Revisiting early warning signals of corporate credit default using linguistic analysis. Pac. Basin Financ. J. 24,
1–21.
Merton, Robert C., 1987. A simple model of capital market equilibrium with incomplete information. J. Financ. 42 (3), 483–510.
Nofsinger, John R., 2001. The impact of public information on investors. J. Bank. Financ. 25 (7), 1339–1366.
Odean, Terrance, 1998. Are investors reluctant to realize their losses? J. Financ. 53 (5), 1775–1798.
Peress, Joel, 2014. The media and the diffusion of information in financial markets: evidence from newspaper strikes. J. Financ. 69 (5), 2007–2043.
Petersen, Mitchell A., 2009. Estimating standard errors in finance panel data sets: comparing approaches. Rev. Financ. Stud. 22 (1), 435–480.
Phansatan, Suwipa, Powell, John G., Tanthanongsakkun, Suparatana, Treepongkaruna, Sirimon, 2012. Investor type trading behavior and trade performance: evidence
from the Thai stock market. Pac. Basin Financ. J. 20, 1–23.
Seasholes, Mark S., Wu, Guojun, 2007. Predictable behavior, profits, and attention. J. Empir. Financ. 14 (5), 590–610.
Shefrin, Hersh, Statman, Meir, 1985. The disposition to sell winners too early and ride losers too long: theory and evidence. J. Financ. 40 (3), 777–790.
Sloan, Richard G., 1996. Do stock prices fully reflect information in accruals and cash flows about future earnings? Account. Rev. 71 (3), 289–315.
Solomon, David H., 2012. Selective publicity and stock prices. J. Financ. 67 (2), 599–637.
Tetlock, Paul C., 2007. Giving content to investor sentiment: the role of media in the stock market. J. Financ. 62 (3), 1139–1168.
Tetlock, Paul C., 2010. Does public financial news resolve asymmetric information? Rev. Financ. Stud. 23 (9), 3520–3557.
Tetlock, Paul C., 2011. All the news that's fit to reprint: do investors react to stale information? Rev. Financ. Stud. 24 (5), 1481–1512.
Tetlock, Paul C., Saar-Tsechansky, Maytal, Macskassy, Sofus, 2008. More than words: quantifying language to measure firms' fundamentals. J. Financ. 63 (3),
1437–1467.
Tsai, Shih-Chuan, 2013. Investors' information advantage and order choices in an order-driven market. Pac. Basin Financ. J. 21 (1), 932–951.
Utama, Siddharta, Cready, William M., 1997. Institutional ownership, differential predisclosure precision and trading volume at announcement dates. J. Account.
Econ. 24 (2), 129–150.
Wang, Jiang, 1993. A model of intertemporal asset prices under asymmetric information. Rev. Econ. Stud. 60, 249–282.
Xie, Hong, 2001. The mispricing of abnormal accruals. Account. Rev. 76 (3), 357–373.
172

20 2017 Wu & Lin

Uploaded by

Copyright:

Available Formats

20 2017 Wu & Lin

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

20 2017 Wu & Lin

Uploaded by

Copyright:

Available Formats

Pacific-Basin Finance Journal 43 (2017) 151–172

Contents lists available at ScienceDirect

Paciﬁc-Basin Finance Journal

The impact of media coverage on investor trading behavior and MARK

2.1. News and investor trading behavior

Note: The stock markets or countries studied are in brackets.

2.2. News and stock returns

2.3. Textual analysis of news

3. Data description and variable measurement

3.1. Investor trading behavior measures

investor trading patterns are derived from an imbalance measure as follows:

3.2. Abnormal return measures

3.3. Textual analysis of media coverage9

3.4. Control variables

4.1. Descriptive statistics

4.1.1. News categories by industry

4.1.2. News categories by year

4.2. Correlation between variables

News categories Net abnormal investor trading Abnormal returns

4.3. News categories and investor trading behavior

4.3.1. Methodology and speciﬁcation

InvNTi, t = a + b1 Pos Re venueNewsi, t + b 2 PosEarningsNewsi, t

Variable Obs Mean Std. Min. Max.

Market value of equity (Log) 111,227 8.521 1.448 3.466 15.114

Panel A: Full sample 2001–2014

Panel B: Sample excluding the global ﬁnancial crisis 2008–2009

Dividend news 0.076 − 0.002 0.024 −0.141 0.042

4.4. News categories and stock returns

4.4.1. Methodology and speciﬁcation

ARi, t = a + b1 Pos Re venueNewsi, t + b 2 PosEarningsNewsi, t + b3 PosInvestmentNewsi, t

4.4.3. Sensitivity analysis

Market model Fama-French Alpha Carhart Alpha

Panel A: Full sample 2001–2014

Panel B: Sample excluding the global ﬁnancial crisis 2008–2009

Market model Fama-French Alpha Carhart Alpha

Share transfer news 0.007 0.004 0.005

Appendix A. Variable deﬁnitions

Appendix B. Ten news categories descriptions

A. Positive revenue news

Adverse, adversely (pu-li) Diﬃcult (kun-nan) Negative, negatively

Against (ti-kang) Directors (tun-chien-shih) (fu, fou-ting)

You might also like