Media Coverage and Abnormal Trading Volume

Abnormal Trading Volume

Master’s Thesis 30 credits

Department of Business Studies Uppsala University

Spring Semester of 2017

Date of Submission: 2017-05-30

Erik Fahlman Eric Pettersson

Supervisor: Adri de Ridder

Abstract

In this study, we examine the media coverage effect on abnormal trading volume using two frameworks: divergence of opinion and information asymmetry. Our sample consists of 420 Swedish small cap firms and includes 125 000 articles from over 800 Swedish media sources.

Utilising three definitions of abnormal trading volume, we find that media coverage of a stock leads to positive abnormal trading volume for that stock. Media coverage on any given day increases divergence of opinion and decreases information asymmetry among investors.

The media coverage effect is dependent on both the frequency and source of coverage. We find that sources with broad information dissemination have greater impact on investor behaviour than sources with narrow dissemination and that information dissemination rather than production is more impactful for firms with low investor recognition. Further, the media coverage effect on divergence of opinion is stronger for firms with high investor recognition.

However, the media coverage effect on information asymmetry is not stronger for firms with high investor recognition but instead, decreases with firm size. Thus, our results suggest an important distinction between the two frameworks.

Keywords: media coverage, abnormal trading volume, divergence of opinion,

information asymmetry, investor recognition

Acknowledgements

We would like to express gratitude to our supervisor Adri de Ridder for providing support

and constructive feedback throughout the writing process. We would also like to express our

gratitude to Peter Benson for introducing us to the subject and giving insights into the

functions of media. Finally, we would like to thank Aktietorget for their support in the data

collection process.

Table of Contents

1 Introduction 1

2 Literature Review 3

2.1 Media Coverage and Stock Returns 3

2.2 Media Coverage and Trading Volume 4

2.3 Information Production and Dissemination 5

2.4 Divergence of Opinion 6

2.5 Information Asymmetry 6

2.6 Summary and Hypotheses 7

3 Research Methodology 8

3.1 Abnormal Trading Volume 8

3.1.1 Adjusted Trading Volume 9

3.1.2 Market Adjusted Trading Volume 9

3.1.3 Standardised Unexplained Trading Volume 10

3.1.4 Abnormal Trading Volume Comparison 11

3.2 Media Coverage 12

3.2.1 Keyword Construction 12

3.2.2 News Day 13

3.2.3 News Frequency 13

3.2.4 Source Categorisation 14

3.2.5 Level of Media Coverage 14

3.3 Control Variables 15

4 Data and Descriptive Statistics 16

4.1 Data 16

4.2 Descriptive Statistics 17

5 Empirical Results and Analysis 19

5.1 Correlation Analysis 19

5.2 Univariate Analysis 21

5.3 Multiple Linear Regression Analysis 24

5.3.1 Multiple Linear Regression on Coverage Quartiles 27

5.4 Robustness 28

6 Conclusion 29

References 31

Appendix A – Datastream Mnemonics 36

Appendix B – Regression Coefficient Confidence Intervals 37

Appendix C – Coverage Quartile Regressions 38

1

1 Introduction

Information today is more available than ever. Mass media outlets are the main channels of information distribution to broad audiences and media coverage now comes in many different shapes and forms, ranging from traditional print media to online-only content. Smartphones, tablets and portable computers make an ever growing stream of information more readily accessible to all parts of society, including financial markets. A key assumption for the efficient market hypothesis to be valid is that the price of a stock continuously reflects all publicly available information (Fama, 1970). The theory assumes complete and equal access to information among investors, highlighting the importance of information distribution channels to investors in financial markets.

Media’s role related to information in financial markets can be illustrated using two prominent theoretical frameworks in behavioural finance, divergence of opinion (Miller, 1977) and information asymmetry (Merton, 1987). The divergence of opinion framework argues that media coverage can increase divergence of opinion among investors due to differential preferences and interpretations of information whilst the information asymmetry framework suggests that media coverage should reduce information asymmetry among investors and improve investor recognition of stocks. These frameworks have important implications for both firm costs of capital and transaction costs experienced by investors.

Regarding cost of capital, low information asymmetry and high divergence of opinion should increase stock prices through a larger pool of potential investors and, due to for example short selling constraints, optimistic investors can influence prices to a larger degree than pessimistic investors. Regarding transaction costs, low information asymmetry should decrease bid-ask spreads, whereas high divergence of opinion should have the opposite effect.

Empirical findings in prior literature support a media coverage effect in financial markets as proposed by the information asymmetry and divergence of opinion frameworks.

Fang and Peress (2009) find that media coverage increases investor recognition, causing firms with low media coverage to exhibit higher information costs and a “no-media premium”. Similarly, Barber and Odean (2008) find that media coverage increases investor recognition among individual investors leading to increased trading volume and Chae (2005) finds that trading volume increases as information asymmetry between investors decreases.

Holthausen and Verrecchia (1990) support the notion that information dissemination

2

increases divergence of opinion among investors leading to increased trading volume, which is also supported by the findings of Garfinkel and Sokobin (2006).

The purpose of this study is to evaluate how media coverage affects investor behaviour in the form of abnormal trading volume. Based on hypotheses derived from the divergence of opinion and information asymmetry frameworks, we test whether media coverage leads to positive abnormal trading volume for Swedish small cap firms. As small- cap firms are characterised by low media coverage and thus high information asymmetry, the information dissemination role of media coverage is hypothesised to influence investor behaviour to a greater degree in line with Tetlock (2011). Trading volume is of particular interest due to its connection to information costs and ability to capture causal impacts of media coverage suggested by both the divergence of opinion and information asymmetry frameworks. In order to test this, we conduct a cross-sectional analysis and gather over 125 000 articles from over 800 Swedish media sources and estimate three different definitions of abnormal trading volume on a daily basis.

We find that media coverage leads to positive abnormal trading volume. Our findings suggest a news day effect which increases divergence of opinion and decreases information asymmetry and that these effects are dependent on both the frequency and source of media coverage. Furthermore, trading volume in firms with relatively lower investor recognition reacts more strongly to media coverage in widespread sources. We also find that the importance of information dissemination rather than production increases for firms with lower investor recognition. Further, trading related to the divergence of opinion effect is more prominent in firms with higher investor recognition. However, abnormal trading related to information asymmetry is not stronger for high investor recognition firms but instead, is stronger for small firms. Thus, we find an important distinction between the two frameworks.

The methods used are univariate tests and multiple regressions and results are statistically robust and valid after controlling for variables that in prior literature has been identified as driving trade.

This study contributes to the growing body of literature on media’s role in financial

markets by utilising two frameworks on the mechanisms of how media coverage influences

investors in financial markets. Further, our results extend the findings of recent studies, which

have largely been based on Anglo-Saxon markets, to the Swedish context with special

emphasis on small cap firms in which the role of media coverage can be argued to be stronger

than larger, more closely monitored companies. Thus, this study can be used for interpreting

prior findings on media effects in financial markets and results extend the scope of the

3

literature to a Swedish firm and media context, as well as relatively smaller and illiquid stocks where information flow mechanisms differ from prior literature.

The remainder of the thesis is organised as follows. Section 2 reviews prior literature on the effects of media coverage in financial markets, describes the divergence of opinion and information asymmetry frameworks as well as specifies our research hypotheses. Section 3 outlines our research methodology. Section 4 presents data and descriptive statistics. Section 5 contains our empirical results and analysis. Section 6 concludes the study and suggests future areas of research.

2 Literature Review

2.1 Media Coverage and Stock Returns

While the relationship between information flows and financial markets has for long been of interest to researchers, the direct relationship between media coverage and its effects on financial markets has more recently garnered attention from academia. One main topic of study is the effect of media coverage on stock returns. Fang and Peress (2009) show that there is a return premium for stocks with low media coverage. They find that this effect is particularly strong for stocks normally associated with a high degree of information asymmetry, for instance, small stocks and stocks with a high degree of individual ownership.

Similarly, Gaa (2008) proposes a neglected firm premium effect suggesting that asymmetric investor attention can an explanation. This asymmetric investor attention can be attributed to a bias in media coverage towards negative news which generates higher levels of coverage than positive news and are thus incorporated into stock prices more often than positive news.

However, media coverage bias towards negative news is not consistent in the prior literature, as Solomon (2012) instead finds a bias towards positive media coverage related to the use of investor relations firms.

Additionally, the longevity of trends in returns is affected by media coverage. Hillert et

al. (2014) report that media coverage can enhance momentum effects depending on the tone

of the content. Chan (2003) finds that stocks experiencing negative returns following

negative media coverage exhibit more persistent price drifts than those experiencing negative

returns but lacking media coverage. Stocks with negative returns but no media coverage

instead often recuperated their losses. Positive media coverage also has an effect on stock

returns. Solomon (2012) shows that when high attention is given to positive news it increases

4

investor expectations of future returns, increasing prices on a short term basis, with lower future returns as a consequence. Attention also affects the speed at which information is incorporated into prices, supported by Kerl et al. (2014) who find that when media coverage receives high attention, price changes are incorporated into stock prices more quickly.

2.2 Media Coverage and Trading Volume

Another main topic of study is how media coverage affects investor behaviour, as evident by the growing body of literature relating to media coverage and trading volume. The literature has also identified that the effect is more prominent among individual investors than institutional investors.

On a general note, Busse and Green (2002) show that information coming from television sources increases trading volumes. Riordan et al. (2013) find the same effect using newswire messages as a measurement of media coverage and Engelberg and Parsons (2011) report that media coverage in local media spurs local trading specifically. Mitchell and Mulherin (1994) argue that the frequency of media coverage relating to a certain stock leads to increased trading volume for that stock. Tetlock (2010) finds that there is a correlation between absolute returns and trading volume which temporarily increases with media coverage. This effect is stronger for small, illiquid stocks characterised by high information asymmetry. Seasholes and Wu (2007) conclude that there is an investor recognition effect related to media coverage, where individual investors purchase stocks with media coverage that they previously did not own. This is furthered by Sankaraguruswamy et al. (2013) who study the relationship between information asymmetry and frequency of media coverage.

They find that a high frequency of media coverage lowers information asymmetry in the covered stocks which is attributed to uninformed investors acting on the news.

Barber and Odean (2008) show that individual investors react more strongly to media

coverage than institutional investors. They divide trading patterns into buying and selling

components and find that individual investors tend to be net buyers of stocks recently

grabbing their attention, for instance through media coverage. This furthers the finding of Lee

(1992) that small traders are net buyers around earnings announcements. Similarly, Sarkar

and Schwartz (2009) find that information asymmetry causes buyer initiated trades and

divergence of opinion drives two-sided trading. Contrasting the results of Barber and Odean

(2008) on individual investor behaviour, Yuan (2015) finds that individual investors are net

sellers of stocks on high attention days. Griffin et al. (2011) compare the relationship between

5

trading volume and media coverage on developed and undeveloped markets. They find that emerging markets typically have fewer articles, and a weaker reaction to coverage compared to developed markets. The effect of media coverage and trading behaviour extends to the universe of mutual funds, where Kaniel et al. (2005) find that net flows significantly increase with media coverage and particularly for funds with low investor recognition.

2.3 Information Production and Dissemination

When evaluating the effects of media coverage it is important to distinguish between information production and information dissemination. While the distinction between the two functions is clear in principle, it is difficult to assess its implications in practice due to their interconnectedness, requiring specialised research methodologies as highlighted by Fang and Peress (2009) where the distinction between stock news (information production) and mass media coverage (information dissemination) is emphasised. Tetlock (2011) furthers this notion through findings that investors overreact to stale news, i.e. repeated news with relatively low information production value. Several studies suggest methodologies for isolating the news dissemination effect on trading volume, finding significant results.

Solomon (2012) evaluates how the use of investor relations firms affects news dissemination functions and subsequently, financial markets whilst Soltes (2009) examines the impact of varying accessibility to specific news sources and finds a significant effect although the relevant company information remains constant. Peress (2014) investigates the relationship between newspaper strikes and trading, reporting that share turnover drops the day newspapers go on strike.

An alternative approach to evaluating the functions of media coverage is to consider the effect of sentiment rather than dissemination. In this body of literature, psychological aspects of media content rather than dissemination is analysed by, for example, classifying coverage as positive or negative. Tetlock (2007) finds support for negative market sentiment in media coverage pressuring returns downwards before a reversal to fundamentals effect. Dougal et al. (2012) advance this by utilising assumptions on the exogenous nature of a columnist’s writing style in finding a causal relationship between media coverage and investor behaviour.

Tetlock et al. (2008) argue that negative sentiment in media coverage can capture otherwise

hard to quantify fundamentals of a firm. Leinweber and Sisk (2011) use a long-short strategy

based on extreme news sentiment and find that it generates positive alphas for small stocks

increasing with the breadth of available news items. Kothari et al. (2009) find that favourable

6

disclosures in business press lower the firm's cost of capital as well as lowers return volatility.

An opposite effect is shown for unfavourable disclosures.

2.4 Divergence of Opinion

A key theoretical concept describing media’s role in financial markets is through divergence of opinion. Miller (1977) outlines a framework on how divergence of opinion among potential investors for a particular stock affects trading behaviour and pricing. This is in contrast to the traditional Capital Asset Pricing Model (CAPM) (Sharpe, 1964; Lintner, 1965) which assumes homogeneous expectations on returns among investors. Miller (1977) illustrates this by disseminating the uncertainty component of expected returns into risk and divergence of opinion, whereas CAPM does not make a distinction between uncertainty and risk. The framework thus presents several implications related to stock pricing and trading, for instance, that stock prices are set by the most optimistic investors and not by the consensus (average) leading to, for example, potential overpricing.

Applying this in a media coverage context, there are several potential interactions. As media coverage is associated with information flows in financial markets, increased media coverage makes investors more frequently re-evaluate expectations on future performance.

Thus, increased media coverage could theoretically increase the frequency of changes in the divergence of opinion component of stock uncertainty leading to increased financial markets activity (Holthausen and Verrecchia, 1990). Atiase et al. (2016) further divide the divergence of opinion effect with regards to information releases into three components: divergent prior beliefs, divergent interpretation and consequent divergence of opinion level. As such, the effect of media coverage in financial markets within the divergence of opinion framework is not solely determined by the actual information content of the news article, but rather a product of the information content as well as investor specific prior beliefs, interpretation and analysis.

2.5 Information Asymmetry

An alternative theoretical concept is the role of information asymmetry. Merton (1987) also

presents a framework challenging the common CAPM (Sharpe, 1964; Lintner, 1965), but

instead highlights the importance of asymmetric information considerations. As CAPM

assumes complete and equal access to information among investors, the information

asymmetry framework rather highlights implications in cases of asymmetric distribution of

7

information. Apart from potential differences in access to price moving information, it can also be access to fundamental information such as the mere existence of a stock. This is commonly referred to as the investor recognition hypothesis. The implications of the hypothesis are that investors invest only in companies which they have existing information and knowledge. Or simply put, an investor cannot invest in a company of which he or she is not aware exists. Also, the information asymmetry framework is conditional on all investors interpreting available information similarly. Investors share homogenous beliefs and the framework is thus fundamentally different from the divergence of opinion framework.

Thus, following the information asymmetry framework, media can be argued to have an important role in financial markets due to its direct influence on investor attention and recognition. In practice, this means that media coverage does not have to include any element of information production, rather dissemination of information to new investors can reduce information asymmetry. This notion is supported by the prior literature, for example, Fang and Peress (2009) documenting a significant neglected firm premium and Chae (2005) finds an inverse relation between information asymmetry and abnormal trading volume.

2.6 Summary and Hypotheses

Considering implications from both the divergence of opinion and information asymmetry frameworks, it is suggested that media coverage should increase trading volume for a number of reasons. First, in the information asymmetry framework, media coverage makes information more broadly available to investors, which reduces information asymmetry and increases investor recognition of stocks. Second, in the divergence of opinion framework, media coverage increases divergence of opinion among investors who might act on their newly updated beliefs. These effects are consistent regardless of information production and dissemination considerations for media coverage and lead to increased trading volume. Thus, our first hypothesis is:

H1: Media coverage of a stock leads to positive abnormal trading volume for that stock.

Further, the information asymmetry framework suggests that the mere spread of information

increases the investor recognition effect. The divergence of opinion framework suggests that

broader dissemination leads to higher opinion divergence among investors. This indicates that

8

the breadth of information dissemination can influence investor behaviour, leading to our second hypothesis:

H2: Abnormal trading volume for a stock increases with the frequency of media coverage for that stock.

Similarly, the impact of information dissemination is dependent on the source of media coverage. Where sources with broader dissemination have a larger impact on information asymmetry and divergence of opinion related trading volume than sources with low dissemination. Therefore, our third hypothesis is:

H3: Media coverage in sources with broad dissemination increase abnormal trading volume more than sources with narrow dissemination.

3 Research Methodology

3.1 Abnormal Trading Volume

Prior literature identifies two main drivers of trading volume: liquidity and information trading (Admati and Pfleiderer, 1988). Liquidity based trading is trading that occurs due to pure liquidity reasons whilst information-based trading is trading that occurs because of information reaching investors. As the purpose of this study is to evaluate the media coverage effect on trading volume, only information-based trading is of interest. This requires making adjustments to control for liquidity trading, isolating abnormal trading volume associated with information-based trading. The methodology thus avoids several issues associated with using unadjusted or unscaled trading volume and is particularly useful for cross-sectional analysis (e.g. Lee and Swaminathan, 2000; Gebhardt et al., 2001; Chae, 2005; Atiase et al., 2016). Unadjusted and unscaled trading would, for example, be systematically higher for firms with relatively higher amounts of liquidity trading. In essence, the methodology aims to adjust trading volume for different systematic variations in trading volume not related to media coverage, in line with the comprehensive trading volume review by Tkac (1999).

In defining abnormal trading volume, this study follows the methodology outlined by

Garfinkel and Sokobin (2006). We use three different definitions of unexplained trading

9

volume: adjusted trading volume (AdjTV), market adjusted trading volume (MATV) and standardised unexplained trading volume (SUTV).

3.1.1 Adjusted Trading Volume

The first definition of abnormal trading volume, AdjTV, is estimated as follows:

( )

∑ [(

)]

(1)

Where is the change in adjusted trading volume for firm i in day t, is the trading volume for the stock of firm i in day t measured in number of shares, is the total number of shares outstanding for firm i in day t, and c is the number of days included in the historical control period. Following the information asymmetry and divergence of opinion frameworks, high values of AdjTV can be interpreted as decreased information asymmetry.

3.1.2 Market Adjusted Trading Volume

The second definition of abnormal trading, MATV builds on the same premises as adjusted AdjTV. However, this definition accounts for potential macroeconomic events increasing trading volume across the market. The adjustment for market-related trading activity is not wholly embraced by prior literature, but several studies argue for its inclusion (e.g. Tkac, 1999; Garfinkel, 2009). It is not the goal of this study to assert which methodology is the most appropriate, hence both are included.

[(

) (

)]

∑ [(

) (

)]

(2)

Where is the change in market adjusted trading volume for firm i in day t, is

the trading volume for the stock of firm i in day t measured in number of shares, is

the total number of shares outstanding for firm i in day t, is the trading volume for all

stocks listed on the associated exchange mkt in day t measured in number of shares,

is the total number of shares outstanding for all stocks listed on the associated

exchange mkt in day t, and c is the number of days included in the historical control period.

10

As MATV builds on similar premises as AdjTV, interpretation is similar where no distinction is made between the asymmetric information and divergence of opinion effects.

3.1.3 Standardised Unexplained Trading Volume

The third definition of abnormal trading is standardised unexplained trading volume (SUTV) which is identified by Garfinkel (2009) to be a suitable proxy for divergence of opinion. It applies a fundamentally different methodology, although still based on the assumption of liquidity and information trading. However, SUTV is designed to not only control for a liquidity effect on trading volume but also a consensus effect of information trading. Instead, it is designed to capture the potential effect on trading volume which is due to divergence of opinion among investors. It mirrors a traditional market model approach associated with estimating abnormal returns, but with the adaptation that trading volume is the variable of interest as outlined by Crabbe and Post (1994) and further developed by Garfinkel and Sokobin (2006).

(3)

Where is the standardised unexplained trading volume for firm i in day t, is the unexplained trading volume for firm i in day t, and is the standard deviation of the residuals from the expected trading volume regression (equation 5). Unexplained trading volume is estimated as follows:

[ ( )] (4)

Where is the unexplained trading volume for firm i in day t, is the natural

log of trading volume for firm i in day t measured in number of shares, and [ ( )] is

the expected value of the natural log of trading volume for firm i in day t estimated using the

following regression:

11

[ ( )] [ ] [ ] (5)

Where [ ( )] is the expected value of the natural log of trading volume for firm i in day t and is the total return for shares of firm i in day t. The superscripts (plus and minus) denote whether returns for firm i in day t are positive or negative. This distinction is made because of an empirical regularity in which trading volume is more sensitive to positive returns than negative returns (Garfinkel, 2009). We use the natural log of trading volume across all SUTV calculations in order to mitigate skewness concerns.

3.1.4 Abnormal Trading Volume Comparison

Whilst AdjTV, MATV and to a certain extent SUTV are based on similar theoretical assumptions, SUTV differs substantially in its interpretation. SUTV is not designed to capture all information trading, but rather only the trading due to divergence of opinion among investors. As such, it is in theory closely related to the divergence of opinion framework. In order to control for if these theoretical differences are also present in our data, we present bivariate correlations in Table 1 below.

Table 1

Abnormal Trading Volume Correlation

AdjTV MATV

MATV 0.757**

(0.00)

SUTV 0.295**

(0.00)

0.200**

(0.00)

Table 1 presents the Pearson-correlation coefficients (p-values in italics) for the definitions of abnormal trading volume.

AdjTV is the change in adjusted trading volume, MATV is the change in market adjusted trading volume and SUTV is the standardised unexplained trading volume. All values are winsorized at the 1

^st

and 99

^th

percentiles. * and ** denote statistical significance at the 5% and 1% levels respectively.

Table 1 presents the bivariate correlations of the different abnormal trading volume

definitions for our sample. All variables are winsorized at the 1st and 99th percentiles. As

suggested by the theoretical assumptions outlined above, AdjTV and MATV are more strongly

correlated than they are to SUTV. Thus, we find preliminary empirical support for the notion

that SUTV proxy only for part of the variation (divergence of opinion) in abnormal trading

volume whereas AdjTV and MATV also capture a broader informedness effect. This

highlights the potential sensitivity of any conclusion to the definition of abnormal trading

12

volume as well as the importance of adapting control variables to the chosen abnormal trading volume definition.

3.2 Media Coverage

In defining media coverage several definitions are outlined in the prior literature are utilised.

By not limiting our study to one specific media coverage definition, it is possible to assess each of the specified research hypotheses and adds robustness to results. All media coverage definitions outlined below are based on the same media coverage archive. However, as prior literature suggests that media coverage is biased towards a subset of media attention grabbing firms, we make several adjustments to each of the definitions in order not to skew our results.

3.2.1 Keyword Construction

A media coverage archive is compiled by constructing firm-specific keywords, which are then searched for in headlines and ingresses of all articles included in the Retriever Research Media Archive. Keyword construction is necessary as Retriever Research does not index firms to related articles, and thus we instead have to rely on firm-specific keywords used by the media. Our methodology follows that of Kaniel et al. (2005), Fang and Peress (2009), Bank et al. (2011) and Hu et al. (2013). Fundamentally, these keywords aim to find the shortest possible unique identifier for the company. We thus remove common identifiers of the firm’s legal form, such as “AB” (“Aktiebolag”) as well as common add-ons to names such as “International” and “Holding”. In order to control for aliasing and name changes, we check for potential acronyms for firm names and utilise the Retriever Research function of related keywords. Further, all firms are checked for name changes using the Swedish Tax Agency (“Skatteverket”) record on public companies and historical firm names within the sample period are included in the keyword construction.

However, some keywords such as those which include a geographical location, word

in either Swedish or English, or personal name, risk rendering false positives. We therefore

extensively manually quality control and trim observations whose firm names indicate an

increased risk for sampling error. We also exclude articles related to market-wide activity, for

example, daily business calendars listing all firm events, as these are not related to firm-

specific media coverage. Limiting our search criteria to headlines and ingresses effectively

works as a relevancy tool, where the firm is more likely to be central to the article compared

to searching within the full article text. Although sampling error cannot be fully eliminated,

13

our manual quality controls aimed at minimising false positives appear sufficient as extensive spot checks support high accuracy in the media coverage archive.

3.2.2 News Day

Utilising the media coverage archive, we construct a News Day dummy variable which controls for media coverage for any given firm on a daily basis. The News Day variable takes the value of 1 if there is at least one news article for the firm during that day and 0 if not. This follows the methodology of Tetlock (2010) and Sankaraguruswamy et al. (2013). News days are matched to the corresponding trading day using the date and time of publication. If a news day occurs on a non-trading day, the news day is matched to the following trading day.

The News Day variable is designed to mitigate the effect of counting the same news event more than once, and thus captures the existence of media coverage rather than broadness of coverage. Thus, the effect of media coverage being biased towards high coverage stocks is mitigated and is related to the first research hypothesis (H1).

3.2.3 News Frequency

We use a news frequency definition in line with findings in prior literature (e.g. Mitchell and Mulherin, 1994; Kaniel et al., 2005; Fang and Peress, 2009; Gong and Gul, 2011). In order to mitigate skewness concerns due to biases in media coverage, we evaluate the news frequency effect following the methodology of Hillert et al. (2014):

( ∑

) (6)

Where is the frequency of media coverage for firm i in day t and

is the total number of articles for firm i in day t. News Frequency is matched to the

following trading day if the news event occurs on outside of trading hours. News Frequency,

in contrast to the News Day variable, allows for analysis of whether the media coverage effect

on abnormal trading volume is related to the breadth of information dissemination, i.e. the

second research hypothesis (H2).

14

3.2.4 Source Categorisation

We categorise each news article into one of six categories based on the source in which it was published The source categories are: a) Business, b) National, c) Newswire, d) Local, e) Industry and f) Other. Business include sources limited to covering business and stock news, National include major nationwide Swedish media outlets, Newswire include specialised newswires, Local include sources with limited geographical coverage, Industry includes industry specific media (including associated unions and interest groups) and Other include all other sources (e.g. public sector, lifestyle magazines, uncategorised online media etc.).

This builds on prior literature utilising different sources of media coverage, where for example Engelberg and Parsons (2011) evaluate local media sources and Dougal et al. (2012) focus their study on the Wall Street Journal. We follow prior outlined News Day methodology for each category of news and create dummy variables for each source category. These variables are related to the third research hypothesis (H3), as Business, National and Newswire are assumed to have broader dissemination among potential investors than Local, Industry and Other.

3.2.5 Level of Media Coverage

In order to ensure that results are not driven by a few attention-grabbing firms, we construct a media coverage ranking for all firms included in our sample defined as:

∑

∑ (7)

Where is the level of media coverage for firm i, is the total

number of articles for firm i during the sample period and is the total number of

News Days for firm i during the sample period. Scaling the number of articles by news days

ensures that the level of media coverage is not biased against firms with an IPO-date during

the sample period. We then rank all firms on the level of media coverage and divide them

into three groups: Low Coverage, Medium Coverage and High Coverage. Low Coverage

includes firms in the bottom quartile (Q1) of media coverage whilst High Coverage includes

firms in the top quartile (Q4) of media coverage. The Medium Coverage group includes all

other firms (Q2 and Q3). The High Coverage firms generate on average a higher number of

articles per news day showcasing a systematic difference in information dissemination

15

between the groups, indicating different levels of information asymmetry as well as investor recognition.

3.3 Control Variables

In order to control for potential factors influencing abnormal trading volume beyond media coverage, we include several control variables identified by prior literature. We use the natural log of market value of equity as a firm size proxy as it is shown to influence both media coverage (Fang and Peress, 2009) and trading volume (Tetlock, 2010). Market-to- Book (M/B) ratio is an important firm characteristic which can be inflated by extensive media coverage (Chen et al., 2013; Hillert et al., 2014). It is estimated as the market value of equity divided by book value of equity. We do not estimate M/B for firms with negative book value of equity. As ownership structures can affect trading volumes (Tkac 1999), we include three ownership variables: Individual Ownership, Foreign Ownership and Ownership Concentration. Each of the variables is calculated as the percentage of shares held by each investor group, where Ownership Concentration defined as the aggregated holdings of the five largest shareholders.

For AdjTV and MATV observations, two separate return variables are included:

positive absolute return and negative absolute return. Absolute price changes are found to influence trading volume (Chae, 2005; Statman et al., 2006). Positive and negative returns are separated because of trading volume being differentially sensitive to them respectively (Garfinkel, 2009). To complement the absolute return variables, we include a variable for the standard deviation of returns (Std. Dev.) for a 20 trading day control period prior to the observation. As SUTV is standardised on absolute returns, we do not include them in the multiple regressions as it would skew any inferences made.

Further, illiquidity is a key characteristic of interest for both media coverage and abnormal trading volume. Similar to, for example, Fang and Peress (2009) and Hillert et al.

(2014) we estimate the Amihud illiquidity ratio (IlliQ) to control for this (Amihud, 2002). We

define the Amihud illiquidity ratio as the average daily illiquidity ratio for the same 20 day

control period as used for standard deviation of returns estimation. IlliQ is not estimated for

firms whose return index is rounded to zero. IlliQ is estimated as follows:

16

∑ | |

⁄ (8)

Where is the Amihud (2002) illiquidity ratio for firm i in day t estimated over 20 day control period. | | is the absolute return for firm i in day t and is the total value of shares traded measured in MSEK for firm i in day t. Finally, we include two calendar related variables influencing trading volume. Mitchell and Mulherin (1994) find that market activity is directly related to the day of the week, we therefore include weekday dummy variables. Additionally, prior literature finds that filing dates are associated with high trading volume (e.g. Kim and Verrecchia, 1994; Solomon, 2012). We collect all quarterly and annual filing dates for each firm within our sample period and create a dummy variable for whether the firm has released any financial report during that day.

4 Data and Descriptive Statistics

4.1 Data

Our sample consists of Swedish small cap firms listed on Aktietorget, First North and OMXS Small Cap during a two-year sample period from 2015-01-01 to 2016-12-31. The full sample period constitutes 505 trading days. We exclude firms with an IPO-date later than 2016-10-01 due to our reliance on control periods in estimating abnormal trading volume. We also exclude preferred stocks and international cross-listings. An additional four firms are excluded from the sample as they lack any media coverage during the sample period, bringing the total sample to 420 firms. For cases of firms with multiple share classes, only the most frequently traded share class is included. Out of these, nine firms exhibit partially missing financial data (due to suspended trading, changing listing or issuance of new share class) and are therefore only included for part of the sample period.

For each firm, we collect two types of data on a daily basis: financial data and media

coverage. Financial data is gathered from Thomson Reuters Eikon with some exceptions. All

relevant Thomson Reuters mnemonics are presented in Appendix A. Report dates for firms

listed on Aktietorget are collected directly from the exchange. In cases where the 2016

Annual Report had not been filed in time for our data collection, we manually collect book

value of equity from the latest available quarterly filing. All ownership data is collected from

17

the Swedish Central Securities Registrar (Euroclear Sweden). Media Coverage data is collected using Retriever Research Media Archive, a database covering 3610 Swedish news sources ranging from online to traditional print media, in accordance with the research methodology outlined in Section 3.2.1.

4.2 Descriptive Statistics

Table 2 presents descriptive statistics of firm characteristics. All values presented are estimated by first averaging daily firm observations and then estimating statistics across firms. A general conclusion from the statistics is that although our sample is limited to small cap firms, there are large differences in firm characteristics, e.g. market value of equity which ranges from 4.5 to 8 668 MSEK and M/B which ranges from 0 to 809.6 respectively. Most extreme observations with regards to firm size are listed on Aktietorget, which is an exchange that exists outside of the NASDAQ sphere. It is also noticeable that the average firm size in our sample is generally smaller than used in prior literature.

Table 2

Descriptive Statistics of Firm Characteristics

Sample Mean Median Std. Dev. Min Max

Market Value 420 338.5 154.5 634.2 4.5 8668.3

Trading Volume 420 316.9 30.7 1918.3 0.1 24876.0

Number of Shares Outstanding 420 51485 14909 174921 444 2871484

M/B 413 14.0 3.5 52.7 0.0 809.6

Number of Owners 420 2358 1483 2772 40 20515

Individual Ownership (%) 420 0.37 0.34 0.20 0.01 0.92

Foreign Ownership (%) 420 0.16 0.09 0.19 0.00 0.89

Ownership Concentration (%) 420 0.54 0.53 0.17 0.10 0.99

IlliQ 417 3.87 0.78 8.05 0.00 62.42

Table 2 presents descriptive statistics of firm characteristics. The figures are calculated by averaging daily observations on a firm basis over the sample period. Market Value is measured in MSEK, Trading Volume and Number of Shares Outstanding in thousands of shares.

M/B is measured as the market value of outstanding equity divided by the book value of common equity. Firms with negative common equity are excluded. Individual Ownership (%) and Foreign Ownership (%) represent the percentage of shares held by individual and foreign investors respectively. Ownership Concentration (%) is the percentage of shares owned by the five largest shareholders. IlliQ is the Amihud measure of illiquidity estimated as the average measure during a control period consisting of 20 trading days. Firms whose return index is rounded to zero at any point in time are excluded.

The large variation in trading volume highlights the importance of considering the distinction

between liquidity and information-based trading. Average IlliQ is around ten times higher

than presented by Amihud (2002), which is reasonable given that firm average market value

of equity is less than a tenth of the average of the US sample in Amihud (2002). This

supports the notion that illiquidity is a key characteristic of our sample.

18

Table 3

Descriptive Statistics of Media Coverage

Sum Mean Median Std. Dev. Min Max

All News 125075 297.8 228 280.7 4 3095

Business News 18540 44.1 18.5 61.0 0 459

National News 6178 14.7 8 22.0 0 267

Newswire News 78789 187.6 163 134.1 4 968

Local News 9439 22.5 4 78.1 0 1291

Industry News 7293 17.4 5 33.6 0 357

Other News 4836 11.5 7 16.6 0 153

News Days 25242 60.1 55 39.6 1 290

Level of Media Coverage 4.5 4.4 1.4 1.2 10.7

Table 3 presents descriptive statistics of media coverage on a firm basis. All News is measured as number of articles and contains all articles in our media coverage archive. Business, National, Newswire, Local, Industry and Other News is measured in number of articles categorised by source category. News Day is the number of firm days containing any type of news article. Level of Media Coverage is the average number of articles published per News Day on a firm basis. The Sum column represents the total figure for the full sample whilst all other columns are presented on a firm basis.

Table 3 presents descriptive statistics of media coverage in our sample. The first column sums the total number of articles and news days across our sample, whereas the remaining descriptives are aggregated on a firm basis before being compared across the sample. Thus, our sample includes a total of over 125 000 articles distributed over 25 242 news days. It includes over 800 different sources, ranging from online to traditional print media, with the most frequent sources belonging to the newswire category (over 78 000 Newswire articles).

On a firm basis, the average firm in our sample has 297.8 articles distributed over 60 news days, with most frequent articles in newswire and business press and more sporadic coverage in other source categories. The highest number of articles experienced by a firm is 3095, whereas the median is 228. The number of articles is thus skewed by a few media coverage grabbing firms, with the most extreme observation constituting 2.5% of all articles. Using News Day as a proxy for media coverage instead of unadjusted number of articles mitigates this problem to some extent, with the same observation constituting 1.1% of all news days.

Further, we note that the coverage level varies across sample firms, where the lowest

coverage firm on average renders 1.4 articles per news day but the highest coverage firm

enjoys 10.7 articles on average per news day.

19

5 Empirical Results and Analysis

5.1 Correlation Analysis

In order to identify potential biases in media coverage, we construct a correlation matrix presented in Table 4 between News Days and by prior literature identified relevant firm characteristics. In line with prior literature (e.g. Fang and Peress, 2009), we find media coverage to be biased towards larger firms in our sample. It is also biased towards growth firms rather than value firms, as outlined by M/B. However, the M/B effect might stem from firm size, as they are correlated. Fang and Peress (2009) find a similar pattern in their US sample. Foreign Ownership is significantly correlated to News Day, however, it difficult to draw inferences on causality between the two variables. While we find no correlation between Individual Ownership and News Day, high Ownership Concentration is negatively correlated to News Day which suggests increased information asymmetry between majority shareholders and financial markets.

News Day is also correlated to days with high values of absolute return. This indicates that media coverage generates changes in consensus among investors, highlighting the interrelation between media coverage and information asymmetry. The relatively stronger correlation between positive absolute returns and negative absolute returns supports a differential sensitivity to positive and negative coverage. Std. Dev. is negatively correlated with News Day indicating less coverage of volatile firms, which could be explained by the firm size effect.

IlliQ is negatively correlated to News Day. This is in line with prior literature on the

existence of neglected firms, whom exhibit higher return premiums and are more price

sensitive than firms present with high media coverage. It could also be partially influenced by

the firm size effect, as smaller firms exhibit higher illiquidity. We find that media coverage

varies depending on weekday, which given findings by prior literature that trading volume

varies by the weekday supports the importance of controlling for day of the week effects in

our analysis. The day of the week effect is also present in absolute returns. Finally, report

days are more likely to generate media coverage than non-report days.

E. F a hl m an a n d E. P et te rss on | M a st er T h es is | U p psa la U ni ve rsi ty ( 2 01 7 ) 20

T a ble 4 C or relatio n Ma tr ix N ew s Da y Lo g( M V ) M/ B In d. Ow n. Fo r. Ow n. Ow n. C on c Lo g| R + | Lo g| R -| St d. D ev . Il liQ Mo n Tu es We d Th ur s Lo g( M V ) 0. 10 3* * M/ B 0. 01 5* * 0. 18 5* * In d. O w n. -0. 00 2 -0. 23 9* * -0. 08 1* * Fo r. O w n. 0. 02 3* * 0. 23 6* * 0. 10 7* * -0. 40 8* * Ow n. C on c. -0. 06 3* * -0. 03 3* * 0. 03 4* * -0. 54 4* * 0. 08 1* * Lo g| R + | 0. 10 9* * -0. 10 4* * -0. 01 6* * 0. 04 4* * -0. 02 2* * -0. 00 5* Lo g| R -| 0. 03 7* * -0. 16 5* * -0. 02 8* * 0. 05 4* * -0. 01 6* * -0. 01 4* * -0. 25 9* * St d. D ev . -0. 00 6* * -0. 46 7* * -0. 07 5* * 0. 13 5* * -0. 07 0* * 0. 00 7* * 0. 16 5* * 0. 22 2* * Il liQ -0. 05 3* * -0. 41 0* * -0. 05 8* * -0. 00 7* * -0. 04 4* * 0. 14 2* * 0. 08 0* * 0. 08 0* * 0. 39 9* * Mo n 0. 00 5* -0. 00 2 0. 00 0 0. 00 0 0. 00 0 0. 00 1 0. 00 0 0. 04 0* * 0. 00 2 0. 00 2 Tu es -0. 01 2* * 0. 00 0 0. 00 0 0. 00 0 0. 00 0 0. 00 0 -0. 01 0* * 0. 01 1* * 0. 00 1 0. 00 1 -0. 25 1* * We d -0. 00 4 0. 00 0 0. 00 0 0. 00 0 0. 00 0 0. 00 1 0. 00 2 -0. 00 8* * 0. 00 0 0. 00 0 -0. 25 1* * -0. 25 7* * Th ur s 0. 00 6* 0. 00 0 0. 00 0 0. 00 0 0. 00 0 0. 00 0 -0. 00 1 -0. 01 5* * -0. 00 1 -0. 00 1 -0. 24 7* * -0. 25 2* * -0. 25 2* * Fr i 0. 00 5* 0. 00 2 0. 00 0 0. 00 0 0. 00 0 -0. 00 1 0. 00 9* * -0. 02 8* * -0. 00 2 -0. 00 2 -0. 24 4* * -0. 25 0* * -0. 25 0* * -0. 24 5* Re po rt 0. 27 9* * 0. 02 6* * 0. 00 8* * -0. 00 9* * 0. 01 7* * -0. 00 7* 0. 01 6* * 0. 06 7* * -0. 01 5* * -0. 00 8* * -0. 02 1* * -0. 01 7* * 0. 00 1 0. 01 1* T ab le 4 p re se nt s Pe ar so n- co rr el at io n co ef fi ci en ts fo r va ri ab le s w hi ch p re vi ou s lit er at ur e su gg est a s d ri ve rs of me di a co ve ra ge . N ew s D ay is de fi ne d as a d ay c on ta in in g an y ty pe o f ar ti cl e on a f ir m b asi s. L og (M lo g of m ar ke t v al ue o f equi ty me asu re d in M SEK . M /B is t he mark et v al ue o f eq ui ty d iv id ed b y th e bo ok v al ue o f co mm on e qu ity . In d. O w n. a nd Fo r. O w n. ar e th e pe rc en ta ge o f sh ar es he ld by ind iv idu al a nd re sp ec ti ve ly . O w n. C on c. i s th e pe rc en ta ge o f sh ar es he ld b y th e fi ve l ar ge st sh ar eh ol de rs. L og |R + | a nd L og |R -| ar e th e na tu ra l lo g of 1 + a bso lu te r et ur n fo r po si ti ve a nd n eg at iv e re tu rn s re sp ec ti ve ly . St d. De de vi at io n of r et ur ns du ri ng a c on tr ol p er io d co nsi st in g of 20 tr adi ng d ay s. I lli Q i s the A mi hu d me asu re o f ill iq ui di ty e st imat ed a s th e av er ag e me asu re dur ing a c on tr ol p er io d co nsi st in g of 20 tr adi ng da ys . M T hu rs a nd F ri a re d umm y va ri ab le s w it h th e va lu e 1 fo r co rr esp on di ng w ee kd ay a nd 0 if n ot . R ep or t i s a d umm y va ri ab le ta ki ng t he v al ue 1 if th e fi rm r el ea se d a fi na nc ia l r ep or t o n th e da y of o bse rv at io n an d 0 if * and ** d en ot e st at is tic al si gn if ic an ce a t t he 5 % an d 1% le ve ls re sp ec ti ve ly .

21

5.2 Univariate Analysis

Table 5

Univariate Analysis of News Day Effect on Abnormal Trading Volume Panel A – Full Sample

AdjTV MATV SUTV

News Day 0.0016**

(0.00)

0.0016**

(0.00)

0.3358**

(0.00)

Non-News Day -0.0006**

(0.00)

-0.0005**

(0.00)

-0.0703**

(0.00)

Difference 0.00213**

(0.00)

0.00211**

(0.00)

0.40603**

(0.00) Panel B – Level of Coverage Quartiles

Low Coverage Medium Coverage High Coverage High – Low

MATV SUTV MATV SUTV MATV SUTV MATV SUTV

News Day 0.0015**

(0.00)

0.2949**

(0.00)

0.0016**

(0.00)

0.3282**

(0.00)

0.0016**

(0.00)

0.3650**

(0.00)

0.00012 0.39

0.07008**

(0.00) Non-News Day -0.0004**

(0.00)

-0.0517**

(0.00)

-0.0005**

(0.00)

-0.0645**

(0.00)

-0.0006**

(0.00)

-0.0979**

(0.00)

-0.00019**

(0.00)

-0.04624**

(0.00)

Difference 0.00192**

(0.00)

0.34658**

(0.00)

0.00212**

(0.00)

0.39271**

(0.00)

0.00223**

(0.00)

0.46291**

(0.00) Panel C – Firm Size Quartiles

Small Firms Medium Firms Large Firms Large – Small

MATV SUTV MATV SUTV MATV SUTV MATV SUTV

News Day Non-News Day

0.0023**

(0.00) -0.0008**

(0.00)

0.3239**

(0.00) -0.0500**

(0.00)

0.0017**

(0.00) -0.0004**

(0.00)

0.3674**

(0.00) -0.0713**

(0.00)

0.0011**

(0.00) -0.0004**

(0.00)

0.3009**

(0.00) -0.0901**

(0.00)

-0.00121**

(0.00) 0.00042**

(0.00)

-0.02302 0.26 -0.04010**

(0.00)

Difference 0.00312**

(0.00)

0.37392**

(0.00)

0.00210**

(0.00)

0.43876**

(0.00)

0.00149**

(0.00)

0.39100**

(0.00)

Table 5 presents univariate analysis (t-tests) of the news day effect on abnormal trading volume. AdjTV is adjusted trading volume, MATV is market adjusted trading volume and SUTV is standardised unexplained trading volume. All values are winsorized at the 1

^st

and 99

^th

percentiles. News Day is defined as a day containing any type of article on a firm basis.

Panel A tests whether mean abnormal trading volume is significantly different from 0 on news days and non-news days respectively.

Difference tests whether there are significant differences in mean abnormal trading volume between news days and non-news days.

Panel B sorts our sample into quartiles based on Level of Coverage, defined as the average number of articles per news day for each firm. Low Coverage is the lowest quartile (<25%), Medium Coverage is the two middle quartiles (25%-75%) and High Coverage is the highest quartile (>75%). For each group, we test if mean abnormal trading volume is significantly different from 0 on news days and non- news days respectively. Difference tests whether there are significant differences in mean abnormal trading volume between news days and non-news days. High – Low tests for differences in mean abnormal trading volume between the High and Low Coverage quartiles on news days and non-news days respectively.

Panel C sorts our sample into quartiles based on firm size, defined as the average market value of equity for each firm during the sample period. Small Firms is the lowest quartile (<25%), Medium Firms is the two middle quartiles (25%-75%) and Large Firms is the highest quartile (>75%). For each group, we test if mean abnormal trading volume is significantly different from 0 on news days and non-news days respectively. Difference tests whether there are significant differences in mean abnormal trading volume between news days and non-news days. Large – Small tests for differences in mean abnormal trading volume between Large and Small Firms on news days and non-news days respectively.

* and ** denote statistical significance at the 5% and 1% levels respectively (p-values in italics).

Table 5 contains a set of t-tests aimed to investigate the effect news days have on abnormal

trading volume. Panel A presents tests based on the full sample for all three definitions of

abnormal trading volume. Panel B is based on MATV and SUTV splitting firms into quartiles

based on level of media coverage. Panel C similarly splits firms into quartiles bases on size.

22

From Panel A, a couple of interesting things can be noted. First, on days with news, all tests are positive and statistically significant at the 1% level. Also, on non-news days all definitions are negative and statistically significant. This suggests that abnormal trading volume tends to be negative on non-news days and that media coverage is a key driver of abnormal trading volume. Further evidence of this effect is that tests comparing volumes for news days and non-news days all are positive and statistically significant at the 1% level.

Thus, we find a news day effect on abnormal trading volume in the sample, which supports the first research hypothesis (H1). The results are consistent for all abnormal trading volume definitions, indicating that media coverage both reduces information asymmetry and increases divergence of opinion.

In Panel B, the sample is split into quartiles based on their level of news coverage and the same tests as in Panel A are conducted. All quartiles show the same pattern as for the full sample with news days being positive and significant and non-news days being negative and significant at the 1% level. However, when looking at the difference between the high and low coverage groups we see that regarding MATV, there is no difference between the groups.

This suggests that the media effect is of equal size regardless of how much media coverage a firm receives on average. Thus, media coverage reduces information asymmetry among investors regardless of prior levels of information asymmetry. However, when focusing on the effects of divergence of opinion, there is a statistically significant difference between the low and high coverage group. This can be explained by divergence of opinion being an exponential function of available information, and that increasing available information for firms with already high levels of information will increase divergence of opinion trading relatively more. This is in line with prior literature arguing that high media coverage increases divergence of opinion among investors (e.g. Holthausen and Verrecchia, 1990;

Garfinkel, 2009).

In Panel C the sample is split into quartiles based on firm size, as one key consideration in our study is the potential irregularity that media coverage is biased towards large firms. We find that the media coverage effect on SUTV does not significantly differ across firm size quartiles. However, as it does vary significantly between levels of media coverage (Panel B), suggesting that the divergence of opinion effect is greater for high than low coverage firms.

Regarding MATV, there is a statistically significant difference between size quartiles where

the media coverage effect is greater for small than large firms. This suggests that media

coverage reduces information asymmetry relatively more for small than large firms. Thus, the

differing effects on divergence of opinion and information asymmetry suggest that the media

23

coverage effect related to divergence of opinion increases with levels of coverage and that the media coverage effect related to information asymmetry decreases with size.

For non-news days there is a significant, and negative, difference between the high and low coverage quartiles (Panel B) for both MATV and SUTV. Panel C also presents significant differences between the quartiles, but the distinction that MATV is greater for large than small firms. This supports our findings that although firm size and level of coverage are correlated, they capture different aspects of a firm’s information environment with regards to information asymmetry. These aspects could be explained by other factors than firm size, presented in Table 6.

Table 6

Level of Coverage Quartiles Descriptives

Quartile MV Foreign Ownership Ownership Concentration IlliQ

1 199.1 0.15 0.59 5.75

2 244.1 0.12 0.54 5.33

3 313.8 0.14 0.55 2.98

4 597.0 0.24 0.49 1.38

Table 6 presents average values of firm characteristics belonging to each Level of Coverage quartile as used in Table 5, Panel B. Level of Coverage is defined as the average number of articles per news day for each firm. MV is market value of equity measured in MSEK.

Foreign Ownership (%) represents the percentage of shares held foreign investors. Ownership Concentration (%) is the percentage of shares held by the five largest shareholders. IlliQ is the Amihud measure of illiquidity estimated as the average measure during a control period consisting of 20 trading days.

Table 6 presents average values of firm characteristics related to a firms’ information environment. We find differences in all variables between the level of coverage quartiles.

Firms in the Low Coverage quartile generally have higher ownership concentration, lower

foreign ownership and higher illiquidity in addition to previously suggested lower size when

compared to firms in the High Coverage quartile. Higher Foreign Ownership indicates higher

investor recognition whilst lower Ownership Concentration indicates lower information

asymmetry among shareholder. Further, IlliQ suggests notable differences in illiquidity

between the level of coverage quartiles indicating higher information costs for investors and

higher cost of capital for firms in the Low Coverage quartile. Thus, these variables highlight

information asymmetry effects not fully captured by firm size showcasing the importance of

including additional control variables when making inferences on the media coverage effect

on abnormal trading volume.