Monitoring with the Media: Social Media as a Corporate Governance Mechanism

Transcription

1 Wesleyan University The Honors College Monitoring with the Media: Social Media as a Corporate Governance Mechanism by Joli Holmes Class of 2017 A thesis (or essay) submitted to the faculty of Wesleyan University in partial fulfillment of the requirements for the Degree of Bachelor of Arts with Departmental Honors in Economics Middletown, Connecticut April, 2017

2 Table of Contents Acknowledgements... i Abstract... ii 1 Introduction Related Literature Motivations for Governance Mechanisms Solutions to Agency Problems: Corporate Governance Mechanisms Data Sample Summary Graphs and Statistics Methodology, Models, and Hypotheses Methodology Models Hypotheses and Research Questions Results Complete Sample OLS Regressions Complete Sample OLS Regressions with Interactions Segmented Sample OLS Regressions Simultaneous with Instrumental Variable Regressions Conclusion...97 References...99 Appendix A: Summary and Regression Tables Exploratory and Descriptive Tables Complete Sample OLS Regression Tables Complete Sample OLS with Interactions Regression Tables Segmented Sample OLS Regression Tables Simultaneous with Instrumental Variable Regression Tables Appendix B: Figures Twitter Media Market Analysis Media Market Figures Appendix C: Models Bhagat and Bolton Models Media Models Instrumental Variable Models Hypotheses

3 Acknowledgements First and foremost, I would to express my gratitude for having the opportunity to work with Abigail Hornstein throughout this process. Her unlimited generosity and mentorship from day one made this idea feasible and fun. Her guidance as my advisor and at the university has grown my confidence in Economics and acutely demonstrated the importance of female leadership. Thank you for supporting all of my ideas, and making thesis writing not only interesting, but fun. This work would not have been possible without the endless support of Manolis Kaparakis. He gave me an opportunity two years ago which opened my eyes and changed my path at Wesleyan entirely. I would also like to thank Logan Dancey for taking me on as a research assistant two years ago. It was a shot in the dark for both of us, but I think has paid off. It was his influence that gave this project depth and allowed the original research question to take form. I would also like to give a huge thanks to Oona Wallace, who read every page of this thesis and helped me turn vague ideas into constructive thoughts. 49 Home thank you for seeing me through; you guys are my rocks. Walker you ve granted me the serenity to accept the things I cannot change, Casey the courage to change the things I can, and Becca, the wisdom to know the difference. Walker you are a foundation to us all. Casey you are an academic inspiration. And Becca your willingness to take everything in stride is beyond admiral, thanks for seeing me through all the late nights. And special thanks to Mira for being the one of the best friends Wesleyan could offer. To the QAC tutors, faculty, and staff, the community you foster is what makes academic center feel like home. Grace and Carlo you two are the OGs, Asie and Tiffany, you guys make it all worth it. To my parents and family none of my dreams would be possible without your endless love and support. Wesleyan has been the opportunity of my lifetime and I will always be grateful that you re the ones that got me here. Special thanks to Doug Holmes, for introducing me to a few vague concepts which have defined my areas of interest within Economics and at Wesleyan at large. i

4 Abstract Of recent, social media has become an increasingly important method of communication as an aggregator and distributor of news and information. Social media plays a corporate governance role by shaping the reputational capital of managers and directors, and by shaping shareholders informational environments. This research uses a unique, social media dataset created from fifteen different media organizations Facebook pages over the period I provide evidence that there is a statistically significant relationship between social media and governance and that this relationship affects the financial performance of the firm. Additionally, I show these results are robust to media biases and endogeneity issues. ii

5 1 Introduction Recently, the media has started to catch the attention of corporate governance researchers as a potential corporate governance mechanism; a mechanism that helps to align managerial interests with those of shareholders. Corporate governance is an important sub-category of corporate finance because governance mechanisms help to reduce agency costs, presumably increasing the financial performance of the firm. There are five broad governance mechanisms which governance research generally focuses on: the board of directors, block shareholders, executive compensation, corporate bylaws, and outside analysts. However, past research also shows that managerial decision-making is affected by sentiment and frequency of media reporting. This indicates there might also be a role for media in governance research. This research questions if social media can play that role. The analysis combines a social media dataset constructed from 410,000 Facebook posts collected over a five-year period, 2011 to 2015, with financial data from Compustat, CRSP, and additional sources. The goals of this research are four-fold. First, this research seeks to add depth to an existing literature that looks at the relationship between the corporate governance role of the media and how this affects the financial performance of the firm. Second, the research adds breadth to the literature by analyzing an untraditional media source. While most other analyses have used content from print media, social media data, from Facebook, will be the source of analysis. Third, this research considers how media sources outside of mainstream financial media may have differing corporate governance effects on the financial performance of the firm. The fourth goal of this research is address endogeneity issues, common to corporate governance research. The models will be estimated with as a simultaneous system of equations with instrumental variables. 1

6 Prior literature finds that there is a significant and economically important relationship between the frequency and sentiment of media coverage and governing decisions at the firm, although the mechanisms for this are not fully understood. Theory suggests the media acts as an accountability mechanism to shareholders in two ways. First, the media may act as an informational intermediary by informing shareholders. More informed shareholders have greater incentives to pressure agents to make changes at the firm. Second, the media shapes the reputational environments of both managers and directors. Pressure from the media encourages managers and directors to act in the best interest of shareholders by affecting future managerial labor market decisions. These theories have been tested empirically in several recent papers. Lui and McConnell (2013) research how management responds to media coverage of controversial acquisitions. They find that increased frequency and negative sentiment of reporting increases the probability of abandonment of the acquisition. Dyck et al. (2008) research institutional investor involvement in corporate governance in Russia over the period of 1999 to They find hedge funds help to increase shareholder value by amplifying media coverage of corporate governance violations which leads to a higher probability of violation reversal. Farrell and Whidbee (2002) study the effects of media coverage on forced CEO turnover. Their research suggests that increased scrutiny by the media leads to a higher probability of forced CEO turnover. Joe et al. (2009) look at the relationship between board effectiveness and the media. They suggest media coverage of ineffective boards informs shareholders. They find that firms targeted by the media take corrective measures and increase shareholder value. Corporate governance is of interest to corporate finance research because it s assumed that governance changes result in changes in the financial performance of the 2

7 firm. The focus of the above papers is media s impact on governance decisions, not media s impact on financial performance through governance changes at the firm. The first goal of this research is to examine the media s impact on financial performance through changes in governance and ownership at the firm. The second goal of this research is to add breadth to previous research by analyzing social media. Most previous studies that examine the corporate governance role of the media use content from traditional print media sources, mainly the New York Times and Wall Street Journal. However, recent research by the Pew Research Center reveals that 62 percent of adults receive news and information through social media platforms, such as Facebook, Twitter, and Reddit (Gottfried and Shearer, 2016). The popularity, uptake, and pervasive use of social media has changed the ways in which people receive information. The recent adoption of social media platforms as a form of receiving news and information leaves a relatively new area of content to research. Social media has affected consumers information environment in three ways: accessibility to news, content interaction, and the speed with which consumers receive news. Prior to the development of social media platforms, users received daily content through paid news subscriptions. Now users can subscribe to a news organization by liking the news organization s social media page. Paywalls still exist, but followers of a social media page receive headlines, blurbs, and images through their social media newsfeed for free. Content can reach broader audiences at little or no additional cost to the news organization. 1 The number of actual, paying subscribers of a news organization 1 For example, there are 1.2 million digital subscribers to the New York Times and between 600,000 to a 1.1 million print subscribers, depending on the day of the week (Ember 2016). The subscriptions are not exclusive. In 2013, the Wall Street Journal had 2.4 million subscribers, of which 900,000 were digital subscribers (Stynes, 2013). 3

8 pales in comparison to how many people follow the news organizations social media pages. 2 The second way social media has changed the consumer s information environment is giving the consumer the ability to interact with the content they receive. Consumers can comment and express approval or concern about the content of the news article, which is then received by other consumers and the news organization itself. The interactive component of social media gives consumers a way to engage with the firm and express their interest in the firm s actions. The development of social media pages has also increased the speed with which users receive content from multiple sources. Social media pages are updated in real time several times a day or even hourly, and at pre-scheduled intervals. The third goal of this research is to provide greater insight into how different news organizations contribute to and influence the media-firm relationship. Prior literature has mostly focused on analyzing content from two sources, the New York Times and Wall Street Journal. The New York Times and Wall Street Journal are the two most credible sources of financial and business information in the United States, making them worthy of analysis. However, only considering these two sources may bias conclusions. News organizations choose how frequently they will report on an event, and how negatively or positively they will respond to that event. One obvious bias that has received a lot of attention is the political leaning of news organizations. The New York 2 The New York Times has 13 million followers on Facebook and 31 million followers on Twitter (Table 1). The Wall Street Journal has 5.3 million followers on Facebook and 12.7 million followers on Twitter (Table 1) 4

9 Times is thought to lean left, while the Wall Street Journal is thought lean center or right, which may result in differing financial coverage and sentiment. 3 Another strength of this analysis versus that of prior studies is the greater breadth of the data analyzed, in terms of number of media organizations and number of articles per organization. For this research 410,000 articles were collected and analyzed from 15 sources. Of the articles collected, 78 percent of the articles came from financial or mainstream news sources other than the New York Times and Wall Street Journal. There are many other financially adept news organizations and tt s likely that stakeholders of the firm will source information from more than those two sources. Additionally, other sources can disseminate and amplify original content to a broader group of media consumers. 4 This research will compare how media that fall into different categories affect the corporate governance role of the media. The four categories are: financial versus non-financial, mainstream versus non-mainstream, paywall-protected versus freely accessible, and right-leaning versus left-leaning. The last goal of this research is to apply frequently used econometric modeling techniques that circumvent endogeneity problems, such as instrumental variables, to this research. Without using research methodologies that can account for potential endogeneity issues, models suffer from bias and inconsistency. This research follows a standard thesis format and progresses as following. The literature review is Chapter 2. Next is Chapter 3, the models, methodologies, and hypotheses section. Chapter 4 summarizes the data sample and discusses how the dataset was created. Results are discussed in Chapter 5 and concluded in Chapter 6. Summary 3 See Table 2, Appendix A, for different political biases 4 See Figure 1, to see the overlap in media markets of the New York Times and Wall Street Journal compared to other news organizations 5

10 tables and regression results are found in Appendix A, figures are found in Appendix B, and models are found in Appendix C. 6

11 2 Related Literature 2.1 Motivations for Governance Mechanisms Neoclassical economics assumes the goal of a firm is to be a profit maximizer, or exit the marketplace. Thus, existing firms maximize shareholder utility because they are also profit maximizers. However, contracting issues and agency costs can inhibit the neoclassical model s assumptions and results. Different mechanisms, known as corporate governance mechanisms, have been devised to help counteract agency issues and contracting problems. Corporate governance has become an area of interest to financial economists because of two problems that regularly occur in organizations: agency costs resulting from divert shareholder and managerial interests, and contracting issues. The common structure of separated ownership and control in public firms leads to a system where the firm may be operationally efficient, but pay the implicit price of agency costs. Theory also predicts that contracting issues will occur, given that shareholders cannot predict and account for every state of nature that a manager will face Contracting If shareholders had complete information regarding the CEO's activities and the firm's investment opportunities, they could design a contract specifying and enforcing the managerial action to be taken in each state of the world. Managerial actions and investment opportunities are not, however, perfectly observable by shareholders; indeed, shareholders do not often know what actions the CEO can take or which of these actions will increase shareholder wealth. - Jensen and Murphy, 1990 Contracts are created to minimize conflicts of interest that can arise when there is more than one person associated with any decision in an organization. Public firms, the unit of analysis in this research, are often owned by tens of thousands or hundreds of 7

12 thousands of individuals. As part-owners of the firm, shareholders are entitled to give input about decisions regarding the firm. However, it s impossible for hundreds of thousands or tens of thousands of individuals with differing agendas to reach consensus and make efficient decisions regarding a firm. To circumvent this issue, firms delegate management to specific individuals, managers and directors, through contracts. Contracts specify a decision allocation structure, or management structure, and the allocation of residual rights of control, or ownership structure. (Fama and Jensen, 1983). Specifying management and ownership structure helps avoid future conflicts of interest between the different actors in the firm. Hart (1995) argues that in order for contracts to effectively mitigate conflicts of interest between different actors, they need to be comprehensive. Comprehensive contracts establish the set of all actors in the firm, the set of actions available to the actors, and the set of outcomes from these actions, given all possible states of nature. A comprehensive contract reduces the necessity of making residual decisions, or decisions made after the inception of the firm (Hart, 1995). Incomplete contracts leave room for actions and decisions not specified at the initiation of the contract, which can lead to an increased likelihood that conflicts of interest, known as agency costs, will occur. Agency costs can arise when the utility maximization prospects of one agent conflict with the prospects of another. Although, comprehensive contracting is desirable, it s extremely unlikely, as described in the quote by Jensen and Murphy (1990) at the beginning of this section. Williamson (1973) suggests that there are two groups of factors, which he terms human and transactional factors, which affect the completion and comprehensibility of contracting. 8

13 Human factors include bounded rationality, which refers to limits on the capacity of human retrieval and processing of information; opportunism, which refers to an interest in individual gains over others; and atmosphere, referring to human values over pecuniary gains. The second factor that Williamson (1973) says may result in contracting issues is transactional factors. Transactional factors include uncertainty and information impactedness. Williamson describes information impactedness as part informational asymmetry and part informational equity. The different factors, both human and transactional, decrease the likelihood that contracts will be comprehensive and complete, and increase the probability of conflicts of interest occurring. Corporate governance is necessary because corporate governance mechanisms help counteract contracting issues that will inherently occur from incomplete contracting Agency Costs Agency theory is concerned with resolving two problems that can occur in agency relationships. The first is the agency problem that arises when (a) the desires or goals of the principal and agent conflict and (b) it is difficult or expensive for the principal to verify what the agent is actually doing. -Eisenhardt, 1989 The black box model of the firm theorizes that managers of a firm (agents), will pursue an agenda that maximizes the utility of their investors (principals), but opportunism and other agency costs often impede this goal. It is well documented that managers don t always maximize shareholder utility, but instead maximize their own utilities. This can be achieved through the maximization of tangible assets, salary and perquisites, and intangible assets, power and status (Mueller, 1972). Divergent interests between shareholders and managers are known in the literature as agency costs. 9

14 Corporate finance literature frequently uses agent opportunism as an example of agency costs, but agency costs can arise from other, often unavoidable, problems such as informational asymmetry, moral hazard (the agent incurring risks taken by the principal), differing risk preferences, or contracting issues (Eisenhardt, 1989). Agency costs accrue from managerial decisions that may benefit the manager, but may hurt the firm s operations in the long-run. Agency costs are both tangible and intangible. An example of a tangible agency cost is when managers over-compensate themselves or receive more perquisites than are generally acceptable. One example of intangible agency costs is when a manager increases the size of the firm with the goal of accruing greater personal power or status, but from the shareholders perspective the associated benefits are less than the associated costs. Shleifer and Vishny (1986) suggest that the most dangerous of agency costs that may result is managerial entrenchment. Entrenchment occurs when shareholders cannot oust a manager they believe is performing poorly or is no longer qualified to manage the firm. Managerial entrenchment issues run the risk of causing long-term problems for the firm. Managerial entrenchment is the most dangerous of agency costs, but all agency problems are important to minimize. Unaddressed agency costs create inefficiencies in capital markets, which at the extreme may result in a market failure. Agency costs are inherent to organizations defined by a principal-agent relationship, as is the case with all public firms. In 1932, Berle and Means coined the phrase separation of ownership and control to describe this relationship. Separated ownership is very common across public firms and organizations, but this control structure inherently results in agency costs. The control structure may be more efficient than having a large group of shareholders, with competing interests, trying to make 10

15 decisions on behalf of the firm. However, separated ownership and control frequently results in two problems described by the quote in the beginning of this paragraph: conflicts of interest and costs from monitoring. Fama and Jensen (1983) explain the theory behind separated ownership and control. Fama and Jensen (1983) postulate that optimal control structure of an organization is determined by two organizational characteristics: the concentration of knowledge necessary to run the organization and the concentration of ownership. Fama and Jensen refer to these two characteristics as complexity and openness, respectively. They define noncomplex organizations as those where specific information relevant to decisions is concentrated in one or a few agents. (Specific information is detailed information that is costly to transfer among agents) (Fama and Jensen, 1983). Smaller organizations, which can efficiently disseminate specific information, often tend to be noncomplex, while larger organizations, which may struggle to efficiently disseminate information, tend to be complex (Fama and Jensen, 1983). If the organization is non-complex, then it isn t necessary to have more than a few key individuals lead the organization. The other characteristic that Fama and Jensen suggest determines optimal control structure is the openness of the firm. Closed organizations are generally smaller and have residual claims that are largely restricted to the internal decision agents (Fama & Jensen 1983). Conversely, open organizations are often larger and the residual 5 claimants are not the internal decision agents, but rather the principals. 5 Fama and Jensen (1983) define residual claimants as the people who bear the risk of managerial decisions. They write, The residual risk the risk of the difference between the stochastic inflows of resources and promised payments to agents is borne by those who contract for the rights to net cash flows. We call these agents the residual claimants or residual risk bearers. 11

16 In general, public firms fall under the category of open and noncomplex. They are noncomplex because the knowledge necessary to the survival and success of the firm is concentrated in the hands of a few managers and is thus efficient. They are open because ownership is diffuse, across shareholders. This implies that for public firms it is efficient for decision-making and control to be allocated to a few agents, while ownership can be spread diffusely. The agency literature suggests that a trade-off exists between the presence of agency costs and operational efficiency. In most organizations, operational efficiency is favored over the reduction or elimination of agency costs, resulting in separation of ownership and control. Given the prevalence of problems that can cause agency costs to arise (e.g. information asymmetry, risk-aversion, self-interest etc.), it is likely that agency costs will occur. Agency costs and contracting motivate the study and development of a robust corporate governance literature. 12

17 2.2 Solutions to Agency Problems: Corporate Governance Mechanisms Corporate governance deals with the ways in which suppliers of finance to corporations assure themselves a return on their investment. How do suppliers of finance get managers to return some of the profits to them? How do they make sure that managers do not steal the capital they supply or invest it in bad projects? How do suppliers of finance control managers? -Shleifer and Vishny, 1997 Corporate governance mechanisms minimize agency costs and contracting issues by aligning managerial interests with those of shareholders. As suggested by the Shleifer and Vishny quote at the beginning of this sub-section, without governance mechanisms in place, suppliers of finance shareholders, risk being taken advantage of by managers. Traditional corporate governance research has focused primarily on five different mechanisms: the board of directors, shareholders, corporate bylaws, executive compensation, and external analyst coverage. This section will present a review of relevant past research on the board of directors and shareholders. Then this section introduces how media may also be thought of as a governance mechanism and the implications. Note that there are three other broad areas of governance research: corporate bylaws, outside analyst coverage, and executive compensation. These areas will not be discussed beyond this introduction as data limitations restrict the use of these measures in the subsequent empirical analysis section. 6 6 Note this isn t problematic from an estimation perspective. Corporate bylaws are a fixed characteristic of the firm. Using panel data estimation techniques, differences in bylaws across firms would be treated as a fixed effect, circumventing omitted variable bias. Outside analysts may be an important as analysts monitor firms and sell this information. However, analyst reports are reviewed by large shareholders with means to purchase extra information. I argue that the independent media under analysis in this research can be seen as a public substitute for private analyst monitoring. Vega (2006) shows that information affects financial performance, and this occurs whether the information is private (e.g., from analysts) or public (e.g., from media). Private analysts are known to focus their attention on companies that receive greater news coverage, analyst coverage is known to affect the firm s performance (Chang et al., 2006; Yu, 2008; and He and Tian, 2013), just as media coverage also affects firm s performance (Fang 13

18 2.2.1 Board of Directors Probably the most widely discussed question regarding boards is, does having more outside directors increase corporate performance? -Hermalin and Weisbach, 2003 As part-owners of the firm, many shareholders have an interest in decisions regarding the firm. However, shareholders of a public firm are characterized by diffuse ownership and free-rider problems, making it inefficient for shareholders to make decisions collectively regarding the firm. The most commonly adopted solution is to delegate a third-party to oversee management and represent shareholders the board of directors. The board can protect shareholders by reducing asymmetrical information, decreasing agency costs, and increasing firm value. The black box theory of the firm assumes that directors, like managers, will be profit maximizers and always act in the best interest of the firm, and thereby the best interest of the firm s shareholders. It is well documented that this is not always the case. The board and shareholders also share a principal-agent relationship, which can result in additional conflicts of interest, agency costs, between the board and shareholders. Hermalin and Weisbach (2003) write that the board may have incentives to align their interests with corporate management instead of shareholders. They write, The CEO has an incentive to capture the board, so as to ensure that he can keep his job and increase the other benefits he derives from being CEO (Hermalin and Weisbach, 2003). Hermalin and Weisbach (2003) suggest that agency costs between the board and and Peress, 2009). Executive compensation is often tied to the financial performance of the firm, to encourage managers to align their incentives with those of shareholders. However, it s likely that large shareholders like hedge funds and mutual funds, who monitor management, pay attention to corporate decisions such as executive compensation, and factor executive compensation structure into their holding decisions. Thus, hedge fund and mutual fund ownership are substituted for executive compensation. 14

19 shareholders stem from the board s relationship with the firm s CEO. Coles et al. (2014) find empirical results consistent with the idea that CEOs do attempt to capture the board. They define a new variable, co-option, which is the percent of the board composed of directors elected by the current CEO and find that co-option and board monitoring are negatively associated. Thus, a good proxy for board strength may be the degree of independence the board has from the CEO. Board independence is measured in two different ways: using aggregated governance indices and individual indicators. Governance indices are a combination of board characteristics and other measurable governance characteristics. Governance indices can capture greater variation across firms and industries than would be observed using individual indicators. However, indices rely on a weighing system and are considered more subjective. Each researcher chooses which weights they feel are appropriate for each indicator, thereby giving greater importance to some indicators over others (Bhagat and Bolton, 2008). The most famous governance index, known as the G-index, was created by Gompers et al. (2003) using 24 governance indicators. The indicators used in the G-index were given equal weights, which may or may not be appropriate given specific firm or industry circumstances. Because of this subjectivity, recent literature favors the usage of individual indicators (e.g. board independence) over indices (such as the G-index or the sub-components thereof). Additionally, Bhagat et Bolton (2008) find no strong association between governance indices and corporate performance. Indicators of board independence that appear consistently across the literature are board composition (the ratio of outsiders to insiders), board size, and CEO duality (when the CEO is also the chairman of the board). Multiple empirical studies use these 15

20 indicators, which signals these measures are easily observable, and these measures contribute significantly to the strength of the board. Insiders are directors that are currently or recently employed by the firm, and outsiders are board members with extensive knowledge of the firm s industry, but who have no underlying ties to the firm itself. Outside directors are believed to have fewer conflicts of interest than insiders and are assumed to engage in a larger monitoring role than insiders (Weisbach, 1988). If it becomes necessary to remove a manager due to poor decisions affecting the value of the firm, outsider directors are expected to lead this movement, often out of concern for their own reputations (Fama and Jensen, 1983; Weisbach, 1988). Insiders, with greater ties to managers and whose careers often depend on a manager continuing their tenure, are expected to have fewer incentives to remove incumbent managers (Weisbach, 1988). Recent literature takes the same perspective. Similarly, Coles et al. (2014) confirm that independents are expected to play a larger monitoring role than inside directors. Hermalin and Weisbach (2003) push back against the generalization that outsiders play a larger monitoring role. They suggest that while this may be the case, the incentives are unclear. However, Fama and Jensen (1983), Fama (1980), and Lowenstein (1996), suggest that from a reputational standpoint all directors have an incentive to monitor and engage on behalf of investors, because they re re-elected annually. Theoretical literature on board independence predicts that firms with a greater number of outside directors will result in lower agency costs, and therefore report better financial performance. Empirical evidence suggests that boards with more independents are better at executing particular tasks, such as replacing underperforming CEOs (Bebchuk and Weisbach, 2010), but there is mixed empirical evidence that supports the 16

21 hypothesis that a greater number of independents on the board results in better financial performance (Weisbach, 1988). Multiple studies that looked at both accounting measures and non-accounting measures of financial performance failed to show any statistically significant association between financial performance and board composition (Hermalin and Weisbach, 2003). Hermalin and Weisbach (2003) suggest two possible explanations for this result.: model misspecification or endogeneity issues. They suggest that actual board composition doesn t affect financial performance of the firm, but changes in board composition (e.g., adding more outsiders to the board) could signal a greater interest in shareholders, thus affecting the firm s financial performance. Boone et al. s (2007) empirical results support Hermalin and Weisbach s hypothesis that changes in board structure are more indicative of financial performance. Boone et al. (2007) write, results indicate that board size and composition vary across firms and change over time to accommodate specific growth, monitoring, and managerial characteristics of the firm. Coles et al. (2014) offer a different explanation of why there of isn t a statistically significant relationship between financial performance and percent of independents. They argue that traditional measures of board independence can be refined. Using an alternative measure of independence, co-option, their results show a negative relationship between co-option and monitoring. This suggests there is a statistically significant and economically important relationship between board independence and financial performance. Board size is also an important factor to control for in governance research. The board monitors and advises managers, so theoretically a larger board has more monitoring power. However, the literature suggests that there is a tradeoff between 17

22 additional benefits from monitoring and efficiency costs (Boone et al., 2007). Empirical results suggest that board size is negatively associated with firm profitability (Hermalin and Weisbach, 2003), implying that costs from efficiency are greater than benefits from additional monitoring. However, Hermalin and Weisbach s (2003) research also suggests that there may be an endogeneity issue and the resulting negative association may be a spurious relationship driven by other factors. Board duality exists when the CEO of the firm is also the chairman of the board. Economic theory hypothesizes that the presence of a dual CEO-chairman increases agency costs. Thus, firms with more effective boards have separated the two positions and exhibit fewer agency costs, resulting in better corporate performance (Fama and Jensen, 1983). Fama and Jensen (1983) write, the board is not an effective device for decision control unless it limits the decision discretion of individual top managers. The board is the top-level court of appeals of the internal agent market. However, an outside chairman may also impose agency costs on the firm (Brickley et al., 1997). Separating the two positions generates transaction costs from the transfer of information (Brickley et al., 1997). Trade-offs introduced by the presence of an outside chairman could be resolved if the chairman has a large investment in the firm (Brickley et al., 1997). Fama and Jensen (1983) indicate in the quote at the beginning of this sub-section that the positions of CEO and chairman should be separate in order for the board to be effective. However, Fama and Jensen write, that to accomplish and achieve effective separation of top-level decision management and control, we expect the board of a large open corporation to include several of the organization s top managers. Empirical research by Bhagat and Bolton (2008) suggests that separating the two positions leads to highly statistically significant increases in operating performance. 18

23 There are clear costs from a dual CEO-chairman, but the theoretical literature also suggests that there are benefits to having some insiders on the board. Whether that comes in the form of a dual chairman-ceo is debatable Shareholders Shareholder participation in the governance of the firm has become an increasingly important governance mechanism as other more traditional governance mechanisms, such as takeover markets, have weakened over time (Smith, 1996; Jefferis and Bhagat, 2002). Shareholder participation can reduce agency costs by making shareholders interests better known to management and thus reducing informational asymmetry. Karpoff et al. (1996) write a central tenet of shareholder activism holds that shareholder proposals ameliorate the shareholder-manager agency conflict and pressure managers to adopt value-increasing policies. As part-owners of the firm, shareholders participate in the firm in several ways. First, if shareholders are unsatisfied they can practice the Wall Street Rule or selling their shares and exiting the firm (Hirschman, 1970). Second, they can participate in governance at the firm, what Hirschman (1970) terms voice. The largest shareholders are often institutional investors. Historic evidence shows that when institutions were unsatisfied with management they generally followed what was called the Wall Street Rule. However, recently institutional behavior has changed (Coffee and Palia, 2016). Specifically, institutions have begun acting as if there is a third means to participate in the firm: activism. Over the last few decades, the term activist investor has entered the literature as investors have begun to try to effect meaningful change within existing firms. The number of activist filings has increased substantially over the last two decades. However, 19

24 Edmans (2009) show that block shareholders can still improve firm value even if they only participate in governance by exiting the firm. Additionally, Parrino et al. (2003) show that institutional investors who exit the firm increase the probability of forced CEO turnover. Voice has become an increasingly important activism tool as the number of shares held by institutional investors has increased, but exit is still an important strategy. There are two mechanisms through which shareholders can voice opinions about the governance of the firm: proxy voting and shareholder resolutions. Shareholders vote to elect board members, confirm or deny executive officer compensation, and pass or refuse independent auditor firm choices. Shareholder resolutions address issues that aren t regularly discussed or voted on, but may be pertinent to the firm s business practices. Recent literature suggests that voice is an important governance mechanism utilized by institutional investors. In their survey of institutional investors, McCaherty et al. (2016) write 63% of respondents state that in the past five years they have engaged in direct discussions with management, and 45% state that they have had private discussions with a company s board outside of management s presence. In theory, shareholder oversight and monitoring should be an effective way to reduce agency costs; however, many shareholders choose not to exercise shareholder rights. Collectively, the social benefits from monitoring are much greater than the private costs, but because each shareholder internalizes the private costs, many will choose not to monitor (Jensen and Meckling, 1976). If a shareholder chooses not to monitor, they still receive benefits from the monitoring undertaken by other shareholders, which creates free-riders. The pervasive free-rider problem of shareholder activism was 20

25 identified by Berle and Means in 1932 (Bebchuk et al., 2009). Free-riders result in the under-provision of monitoring, which fails to reduce agency costs. Free-rider problems are a well-recognized issue within the study of economics at large. The common solution to the free-rider problem is that public goods, goods which are non-excludable and non-rival in consumption, are often maintained by a governing body or an institution with oversight responsibilities. In the case of public companies, institutional oversight is provided by block shareholders or outside analysts. Monitoring by these actors helps to alleviate the under-provision of monitoring, but monitoring will always be under-provided because of free-riders (Shleifer and Vishny, 1986). Although, most governance literature suggests that shareholder activism benefits the firm, some contest this notion. Activism from certain shareholders, such as hedge funds, may have negative effects on the firm (Klein and Zur, 2009; Coffee and Palia, 2016). Additionally, some researchers contend that managers know what is best for the firm and shareholder interference through governance negatively effects the firm (Karpoff et al., 1996). The next section will focus on which types of shareholder activism benefit the firm and why certain types of shareholders choose to engage with the firm Block Shareholder as a Governance Mechanism A block shareholder can be any investor that holds a large fraction of the firm s shares. That is, a block shareholder could be as small as an individual person or as big as a larger organization such as a hedge or mutual fund (Edmans and Holderness, 2016). Block shareholders are typically defined in the literature as a shareholder who owns at least five percent of the company. The presence of block shareholders is believed to indicate that a firm has stronger internal governance for two reasons. First, the presence of a large block 21

26 shareholder signals a greater level of monitoring and engagement, increasing the value of the firm (Shleifer and Vishny, 1986). Second, historically, large shareholders have helped to facilitate third-party takeovers in the case where other governance mechanisms fail (Shleifer and Vishny, 1986). Takeovers can be important mechanisms for retaking corporate control, as an acquirer will install new management, thereby improving the value of the firm and increasing shareholder utility. The presence of a large shareholder attracts third-parties increasing the likelihood of a takeover (Barber, 2007). Academic interest in the impact on firm performance of institutional block shareholders has grown significantly in recent years as the proportion of equity held by institutions has increased. Holderness (2009) found that 96 percent of public companies had at least one block shareholder using a representative sample of 357 public companies in the United States. Concurrently, institutional activism has increased and the strength of the takeover market has fallen. Large block shareholders are expected to play an important monitoring role and some academics suggest that block shareholders act as a bellwether for other investors (Tricker, 1998). However, different types of block shareholders face internal incentives to engage with a firm, consistent with their individual investors characteristics, and should not be viewed as a universal solution. Like all shareholders, institutional block shareholders will engage in monitoring activities up until the point where marginal benefits from exercising shareholder rights exceeds marginal costs. Marginal benefits are dependent on profits received from dividends, stock premiums, and investors preferences. 22

27 Çelik and Isaksson (2013) suggest that there are several different factors that determine whether institutional investors will participate in engagement. The level of institutional engagement is affected by the institution s purpose, liability structure, investment strategy, portfolio structure, fee structure, political or social objectives, and regulatory environment (Çelik and Isaksson, 2013). The next two sections will briefly discuss how different types of institutional investors, hedge and mutual funds, effect the financial performance of the firm. The literature acknowledges that there are differences in institutional investors, with regards to their usage of activist methods, agendas, and success rates. However, the literature leads back to an idea established by Shleifer and Vishny (1986), that firms with more concentrated ownership have higher levels of shareholder monitoring and engagement. The broad literature suggests that there is an association between shareholder engagement and the size of investment stake. However, Çelik and Isaksson (2015) revise this idea, suggesting that the act of ownership alone is positively associated with monitoring, not the proportion of shares. They write, incentives for ownership engagement is not a function of share ownership itself. They result from the business model and are beyond the reach of public policy. Çelik and Isaksson s comment suggests that the proportion of shares may not matter as much as other factors. McCarhey et al. s (2016) research supports Çelik and Isaksson s hypothesis. Their empirical results show there is not statistical significant relationship between assets under management (size of the institution) and their dependent variable, engagement, which is measured as an index. Thus, there is a shift in the literature away from the traditional measure which captures the intensity of shareholder ownership (magnitude of block ownership stake) to the newer measure which captures the extensiveness of shareholder 23

28 ownership (count of investors who have block ownership stakes). This research follows the more recent literature and focus on the extensive margin of institutional block ownership Hedge Funds In recent years, the most important players in the activism landscape have been activist hedge funds. Weisbach, 2010 Hedge funds have a greater incentive and ability to participate in governance at the firm, but this participation may not result in better financial performance in the longrun. There is some disagreement over whether hedge fund activism creates or destroys value, but history suggests that hedge funds are the most successful activist investor in the institutional setting. Other institutions have had limited success increasing stock prices (Brav et al., 2008). Most hedge funds share some common characteristics even as there is no universal definition of the term hedge fund. First, they re privately owned by a small number of shareholders who meet minimum financial requirements (e.g., net worth of at least $1 millon) and are managed by professional managers. Second, managers often own a large share of the fund. Third, hedge funds hope to effect large changes at firms through whatever technique seems appropriate. Finally, hedge funds often hold large investment stakes in a portfolio firm, both relative to the size of the firm and relative to the size of the hedge fund s portfolio itself. This is succinctly described by McCahery (2016), hedge funds have particularly strong incentives to engage, can take concentrated portfolio positions, and face low conflicts of interest. The concentrated portfolio holdings of many hedge funds gives hedge funds sufficient leverage over the board to lead the appointment of new directors (Brav et al., 2008) or management. Hermalin and Weisbach (2003), write her power [the large 24

29 shareholders] works through her position on the board or her control of some number of directors. Hedge funds are known to operate with this strategy in mind. They often hold substantial claims with the intention of installing new management or getting the company acquired by another firm (Bebchuk and Weisbach, 2010). Additionally, hedge funds are known for holding claims for a short period of time, making changes to management or corporate structure, and then ditching the shares after, a strategy known as pump and dump (Coffee and Palia, 2016). This strategy has received much criticism but can be profitable for the hedge fund. The hedge funds concentrated holdings make it possible to pressure firms to increase dividends, leverage, and stock buy-backs, and decrease long-term investments like research and development (Coffee and Palia, 2016). Empirical research by Coffee et al. (2016) and Klien and Zur (2009) suggests that capital expenditures and research and development expenses decrease after hedge fund engagement with target companies. Additionally, Klien et al. (2009) show that on average dividend payouts double in the years following hedge fund engagement. Stock prices increase in the short run, but findings are inconclusive for the long-run (Brav et al., 2008; Coffee and Palia, 2016). However, there is also evidence that hedge fund activism may positively benefit the firm. Brav et al. (2008) find that in the year following an announcement of hedge fund engagement or targeting of a firm, total payouts increases by percent due to a reduction in agency problems from reduced free cash flow. At the same time, managerial compensation decreases on average by one million dollars, and CEO turnover increases by 10-percent (Brav et al., 2008). 25

30 Mutual and Public Pension Funds Mutual and public pension funds are treated as a combined group in this analysis as the two types of funds share many commonalities. First, they both serve a heterogeneous set of investors who often meet minimal, if any, financial requirements (e.g., minimum net worth). Second, the investors are usually investing for the long-term (e.g., for retirement) and are often interested in moderate, conservative investments. Finally, both types of funds usually hold diversified portfolios. Public pension funds manage the retirement savings of public employees and retirees. The public sector, as one of the largest employers in the United States, has placed public pension funds into a prominent position among institutional investors. Unlike hedge funds, public pensions serve a more socioeconomically diverse group of investors. Generally, their investors are risk-averse individuals interested in saving for retirement and thus desire moderate and consistent returns. Shareholder preferences have resulted in stricter fiduciary requirements, such as diversification requirements (Brav et al., 2008). Mutual funds 7 are professionally managed funds that pool monies from individual shareholders to invest in a diversified pools of assets. Mutual funds are important to the study of corporate governance because as of 2015 they manage $15.7 million in total assets, accounting for 31-percent of all U.S. corporate equities, and percent of American households invest in mutual funds (ICI, 2016). Pension funds and mutual funds also participate in institutional activism, but they have been criticized for their effectiveness. Brav et al. (2008) blame differences in 7 While there are meaningful differences between mutual and pension funds, there are more commonalities. Accordingly, hereafter we will use the terms mutual funds and mutual and public pension funds interchangeably. 26

31 investor groups, purpose, and organizational structure for their lack of effectiveness. Brav et al. (2008) write, earlier studies show that when institutional investors, particularly mutual funds and pension funds, follow an activist agenda, they do not achieve significant benefits for shareholders. Relative to hedge funds, both have larger, and more diverse investors groups that they serve, in addition to facing strict diversification requirements. Additionally, the passing of recent laws such as the Dodd-Frank act in the 2010, have affected their effectiveness as institutional activists. Coffee et al. (2016) write historically, brokers were permitted to vote shares held in their street name for their clients on routine matters. As a practical matter, this did not significantly affect contested elections for board seats but it did mean that in voting on shareholder proposals or on corporate governance issues, brokers would typically vote the shares held by retail shareholders in favor of management s position. The 2010 law led to a decrease in the number of proxy votes submitted annually by mutual funds, and also by individual investors, since the costs of voting often outweigh the benefits of voting. Although, the literature is fairly pessimistic when it comes to mutual fund activism, research by Del Guercio et al. (2008) on just vote no campaigns suggests there are facets of activism where pension funds and mutual funds are effective. Del Guercio et al. (2008) study 112 just vote no campaigns from 1990 to The typical campaign in their sample is sponsored by a pension fund and very few campaigns in the sample are sponsored by hedge funds. They found that campaign targets had significantly improved operating performance and forced a much higher rate of CEO turnovers. Although, most of the literature seems to side with the view that mutual funds and pension funds aren t the most effective activist investors, there is some evidence that they are successful with certain tools. 27

32 Market for Corporate Control The existence of corporate governance mechanisms signals that the firm also derives benefits from owners access to some rights of control. Without checks on management, providing capital to the firm would be a very risky proposition and firms would face capital access constraints. Nonetheless, most firms still face significant freerider issues, as many shareholders are not active or effective monitors. Most shareholders have access to few options by which to assert control over management. However, the three mechanisms by which shareholders can re-assert control are the proxy fight, the direct purchase of shares to establish a majority, or a merger (Manne, 1965). Historically, the market for corporate control was an important governance mechanism, particularly during the 1980s. However, the academic focus on the market for corporate control has lessened as institutional equity holdings have increased at a faster rate than that of the market itself (Smith 1996; Bhagat and Jefferis, 2002). Bhagat and Jefferis (2002) note that starting in the early 1990s, both the popular and academic commentators started emphasizing the monitoring role of relational [activist] investors. As the volume of takeover activity has decreased significantly since the 1980s, this research will not discuss the market for corporate control in great depth. However, there is one aspect of the market for corporate control, the corporate raider, which was already discussed in an earlier section (Section Block Shareholders). Shleifer and Vishny (1986) discuss the existence of corporate raiders in conjunction with the presence of a large block shareholder. Their research suggests that the presence of a large shareholder can attract raiders who view the company as undervalued. The corporate 28

33 raider is sometimes thought of as the solution to the free-rider problem (Grossman and Hart, 1980). If the raider buys enough stock to control the company, they can reduce agency by installing new management or making other changes as they see fit Media Evidence indicates that the press facilitates earlier public knowledge of a fraud by both original investigative reporting and broadly rebroadcasting information from other intermediaries. Miller, 2006 The existence of independent media has long been recognized as a governance accountability mechanism for both nations and firms. Recently, economists have started to apply theories of media accountability to theories about corporate governance. There are two main components to media accountability. First, the media can act as an informational intermediary through investigate reporting. Second, the media amplifies and spreads existing information to larger, more diverse audiences. The first theory, referred to as the information intermediary theory in this research, suggests that the role of the media is to act as a watchdog and provide new, insightful, and unbiased information to shareholders. Informed shareholders have greater incentives to pressure directors to make changes at the firm, thereby increasing shareholder utility. Bushee et al. (2010) find that the media helps to reduce informational asymmetry problems between investors and firms. Djankov et al. (2003) also recognize the media s importance as an independent source of quality, unbiased information. They study how media ownership should be organized (e.g. corporate vs. government), and the effects of unbiasedness information provided to investors. A second theory, which this research will refer to as the reputational capital theory, suggests that the media plays a governance role by making risky decisions reputationally costly to the managers and directors. Lui and McConnell (2013) find that 29

34 the frequency and sentiment of coverage affect managerial decisions regarding capital allocations, which suggests that managers listen to the media. They interpret this to mean that the media affects governance by changing managerial or directorial reputations, and thereby influencing their future labor market prospects. Similarly, Dyck et al. (2008) and Dyck et al. (2002) write about how reputational costs, as a function of media coverage, change managerial behavior and decision-making at the firm. The reputational costs theory suggests that the corporate governance role of the media is to shape the reputations of managers and directors, encouraging directors and managers to make decisions that align with shareholders interests Informational Intermediary Within economics, one way the media can act as an accountability mechanism is by decreasing agency costs from informational asymmetry. The media provides new information to shareholders which can help overcome free-rider and voter apathy issues that regularly affect how shareholders engage with the firm. Media coverage can thus increase the likelihood of shareholders being active investors. Lowenstein (1999) writes, the potential for adverse publicity compensates for the damage inflicted under the socalled free-rider doctrine, which says that if there are too many shareholders then none of them will bother to become active. Lowenstein (1999) suggests that the media acts as a governance mechanism by informing shareholders, which spurs shareholders to engage with the firm by increasing opportunity costs from inaction. This is important because most of the literature recognizes shareholder engagement as a necessary governance mechanism, indicating that increased shareholder participation benefits the firm. This assumption is supported by many in the literature, but not by all. Gompers et al. (2003) suggest that governing responsibilities should reside with shareholders. They 30

35 write, Corporations are republics. The ultimate responsibility rests with the voters (shareholders). These voters elect representatives (directors) who then delegate most decisions to bureaucrats (managers) (Gompers et al., 2003). Although, Gompers et al. (2003) place a great amount of responsibility on shareholders, other economists such as Berle and Means (1932) felt differently. Lowenstein (1999) paraphrases Berle and Means, [the shareholder] has no direct property rights in the assets of the corporation, and even expressed in terms of control, the shareholder s role is and ought to be, they said, very limited. The literature differs on what the role of the shareholder should be. However, most of the literature sides with the opinion that shareholders can decrease agency costs through monitoring and engagement activities. Therefore, decreasing asymmetric information should benefit the firm. There is still the question of which shareholders the media helps to inform. Given the value of their investments it is likely that institutional shareholders source information from analysts and other resources, and don t rely as heavily on independent media for investigative information. Joe et al. (2009) found that individual investors often respond negatively to media exposure, while institutional investors seemed to anticipate media exposure, suggesting that institutions may have had prior knowledge of a surprise announcement or event. Over the last few decades, the proportion of shares held by small, individual shareholders has decreased significantly from 84 percent in the 1960s to 40 percent in the 2000s (Çelik and Isaksson, 2013). Thus, many argue that from a governance perspective, the only shareholders that matter are institutional investors. Not only has the proportion of institutional investors increased significantly, but their holdings are also also very concentrated. This suggests that institutions not only have the greatest 31

36 influence on firms today, but they are the only investors who can influence firms. This indicates that while it is important to control for institutional investor ownership in empirical estimations, it may not be necessary to control for individual ownership Reputational Capital The reputational capital hypothesis suggests that the media acts as governance mechanism by shaping and maintaining reputations. Lui and McConnell (2013) hypothesize that managers pay attention to the media because it affects their future job prospects in the managerial labor market. The media informs the labor market in two ways: by making managers more widely known and by affecting attitudes towards those managers. Fama (1980) also writes about the managerial labor market as an incentivizing mechanism, arguing that individual participants in the firm, and in particular its managers, face both the discipline and opportunities provided by the markets for their services, both within and outside the firm. Dyck and Zingales (2002) agree that the media affects attitudes in the managerial labor market, but add that the media also affects reputations through an additional channel, society at large. They argue that a manager s decisions are motivated not just by tangible assets such as future wages, but also by consumption of intangible assets such as social capital. The reputational capital hypothesis suggests that decisions are made through a balancing act in which managers weigh private benefits from a decision against the changes in tangible capital and reputational capital, reflected in Lui and McConnell (2013) s equation, [1] Private Benefits Tangible Capital + Reputational Capital. Reputational capital is a function of the frequency and sentiment of reporting. Private benefits, also often called non-pecuniary rewards in the literature, are non-measurable 32

37 benefits such as status and power. Non-pecuniary rewards are often looked upon negatively because they re loosely associated with firm size, rather than firm profitability (Mueller, 1972). Firm size is often positively associated with free cash flow, and free cash flow can exacerbate agency issues between managers and shareholders (Jensen and Meckling, 1976). Thus, the financial economics literature consistently concludes the costs of a size increase outweigh the benefits. If the above model, Equation 1, is indicative of how mangers make decisions, then theory would predict that a decrease in reputational capital is associated with a decrease in the amount of private benefits a manager will be willing to accept, ceteris paribus. Lui and McConnell s (2013) results support their theory. They find that the likelihood of abandonment of a controversial acquisition is positively associated with the frequency and negative sentiment of reporting. The media plays a corporate governance role, helping to align managers interests with those of shareholders, by shaping the accumulation of reputational capital. Dyck et al. (2008) find supporting evidence of the reputational costs hypothesis in their research about hedge fund lobbying efforts in Russia. They hypothesize that managers will maximize private benefits up until the point where the marginal benefit equals the marginal reputational cost multiplied by the probability that a certain critical audience will receive the information plus costs of punishment multiplied by the probability of punishment. This is expressed mathematically as Equation 2: [2] E(Private Benefits) E(Reputation Cost) ρ + E(Punishment) π, where ρ is the probability of information reaching critical audiences and π is the probability of punishment. Dyck et al. (2008) found that engaged hedge funds were 33

38 successful in increasing the frequency coverage of corporate governance violations, which then increased the probability the violations were reversed. Although the reputational capital hypothesis may be persuasive, there are critics of it. One criticism suggests that CEOs who hold long tenures are unlikely to re-enter the managerial market after leaving their current position, and are therefore less affected by reputational costs. There are two rebuttals to this criticism. First, managers often serve in other roles (e.g., academia, consulting, or directorships) after retiring from a firm (Brickley et al., 1999). Using a panel sample of the largest 500 companies by sales, profits, assets, and equity from , Brickley et al. (1999) found that 88 percent of retired CEOs held at least one board position after retirement. Additionally, 16 percent of retired CEOs in Brickley et al. s sample continued as the chairman of board at their own company. These findings suggest that it is prevalent in the CEO community to continue service after retirement, indicating that reputational concerns will persist past the CEO s retirement. Secondly, recent research by Murphy and Zábojník (2007) suggests that there has been a shift in preferences within the managerial labor market to prioritize external hires with prior experience over internal promotions. Their research shows the average number of prior firms a CEO has served at has increased significantly since the 1970s. This suggests that CEOs do change firms, and thus would have in interest in maintaining their reputational capital. Another way the media serves as a governance mechanism is through the directorial labor market (Dyck and Zingales, 2002; Del Guercio et al., 2008). Directors also evaluate reputational costs from decisions made by management at the firm. If under enough pressure board members will negotiate with shareholders to avoid 34

39 reputational damages. Del Guercio et al. write that directors value their reputation as monitors and want to avoid the public embarrassment and associated damage to their reputations. Fama and Jensen (1983) suggest that outsiders on the board of directors have greater incentive to align with shareholders and create managerial pressure Facebook and Social Media Facebook usage and engagement is on the rise, while adoption of other platforms hold steady. Greenwood, Perrin, and Duggan, 2016 Social media has increasingly become an area of interest to social scientists as social media platforms have become a prevalent part of daily life and changed the way people receive news and information 8. Social media platforms, such as Facebook or Twitter, which serve as hosts and aggregators of information. Research from Pew Research Center shows that by a wide margin, Facebook is the most frequently used social media platform. 9 Although news media sites often have larger followings on Twitter than Facebook, people frequent Facebook more than Twitter (Table 1), making Facebook a more interesting source of analysis. Sites such as Facebook and Twitter have changed how people receive and interact with news in three ways. First, social media has created greater accessibility to a variety of news organizations. Instead of subscribing to just one or two print newspapers, social media users can follow unlimited pages and receive headlines, blurbs, and images for free, decreasing the cost of accumulating information from multiple sources. Additionally, content reaches a much broader and more diverse audience when 8 In 2016, Pew researchers found that 44 percent of adults received news from Facebook and 9 percent received news through Twitter (Gottfried & Shearer 2016). Pew generates these findings by randomly sampling adults in the United States every few months. The sample that they used to generate these findings were from 1,500 adults in the United States over March April, In 2016, 79 percent of adults in the United States use Facebook, compared to 24 percent of adults that use Twitter, the next most frequently followed news aggregator (Greenwood et al. 2016). 35

40 shared over social media. All the friends or followers of an individual receive the content when shared, even if they themselves don t subscribe to the social media page. Second, people can interact with the content more. Facebook gives people the option to like and comment on the content 10. This gives consumers the opportunity to share their beliefs, essentially voting about different policies or changes at the firm. Although the people who interact with a Facebook post about a certain company might not all be shareholders, the firm receives information from the aggregate reactions on Facebook about their societal reputation. This helps to inform managers and directors about how the firm is viewed by shareholders. Third, the speed at which news is shared has also increased drastically. Before news aggregators, only subscribers of print media had access to published content and content had to be shared over the phone, , or in-person, a much slower method of communication. Organizations that utilize social media platforms can spread information much more quickly, easily, inexpensively, and broadly than more traditional methods of information dissemination Media Segments The last part of the literature considers how different types of media may influence corporate governance outcomes at a firm. Although, the media are often regarded as independent, unbiased sources of information, it s well established that no institution can be unbiased. Bias comes from what is reported on, how it is reported, when it is reported, and with what frequency it is reported. 10 Facebook has recently added other emotions. This research measures engagement using likes, the original emotion, for which there is data over the whole time period

41 Dyck and Zingales (2003) argue that there are three ways in which reporters intentionally or unintentionally bias content. First, Dyck and Zingales (2003) suggest that journalists will put a positive spin on their content in exchange for the valuable information they divulge to consumers. Second, journalists will positively spin an article, because they don t have a good understanding of the content and therefore can be manipulated by their sources. Third, journalists shape their content in response to what they think consumers want to hear. Their initial results indicate that journalists do appear to be influenced by a company s spin. Previous research that has looked at the relationship between the media and corporate governance has analyzed content produced almost exclusively by the New York Times and Wall Street Journal (Farrell and Whidbee, 2002; Dyck et al., 2008; Lui and McConnell, 2013). It s sensible that these two news organizations, often regarded as the most prominent sources of financial news in the United States, would be the primary source of analysis. However, limiting analysis to these two sources may be restrictive in two ways. First, audiences make judgments based on the information they receive and if that information differs by media source, then audiences will perceive firms differently. Second, other sources amplify and disseminate information to greater and more diverse audiences. A visual analysis of overlap in Twitter shows differences across media markets (Figure 1 -Figure 8). 11 To study how differences among media institutions affect corporate governance outcomes, four segments will be included in this research. The segments are: financial versus non-financial sources, mainstream versus non-mainstream media, paywall versus 11 Twitter was used instead of Facebook due to data restrictions. However, demographically the Facebook users are very similar to Twitter users (Table 3). 37

42 non-paywall sources, and media with different political leanings. These segments are included for several reasons. First, segmentation helps with robustness. If biases do exist with regards to frequency or sentiment of reporting, it s important to understand these differences and factor differences into results and conclusions. If estimation results are similar between two groups, then we may fail to reject the null hypothesis that the segments are different which further validates the results obtained by other researchers using only the most visible media sources. Second, the group of media institutions chosen for this research, listed in Table 4, were chosen for a variety of different reasons. It is possible that corporate governance outcomes differ based on which types of media organizations are reporting on it. For example, if financial media institutions are reporting about a firm, then investors are likely to receive this information. By contrast, mainstream media coverage might be received by a more general audience. Dyck et al. (2008) hypothesize that the media is only important in-so far as in reaches a critical, target audience. However, the impact of media reporting depends on which audience is reached and the likelihood of audience engagement. The receiving audience can be shareholders, directors, or managers, which affects which hypothesis, the informational intermediary theory and reputational capital theory, will be more impactful. If the media informational intermediary theory is more impactful, then it s expected that prominent sources of financial information, or information protected by a paywall, will be more impactful than sources of non-financial information. Miller (2006) finds that the business-oriented press undertakes more original analysis and investigation. However, the non-business press picks up the stories generated by the business press and spreads the stories to a wider audience. The model introduced by Dyck, et al. (2008), 38

43 presented as Equation 2, suggests that the media that not only informs, but also amplifies the message. This indicates that non-business media should be included in analysis of how media affects corporate actions. If the reputational capital hypothesis is more impactful, it s expected that mainstream media sources, which reach large, diverse audiences, will be more important than niche providers of financial information. Additionally, political ideology becomes an important segment to analyze because the sources included in other analyses are thought to lean left or center. 12 If conservative-leaning media organizations report differently than liberal or center-leaning media organizations, then their audience will receive different information. Research has consistently found that people with liberal ideologies turn to liberal news sources and people that identify as conservative read and watch conservative media sources (Mitchell et al., 2014). Additionally, liberals get their news from multiple sources, whereas conservatives tend to mostly listen to Fox News 13, indicating it may be insightful to include a greater diversity of sources in this analysis. 12 See Table 2. Allsides bias rating and Pew research agree that the New York Times leans left, while the Wall Street Journal leans right. 13 Pew research indicates that 47 percent of conservatives cite Fox News as their main source of news, vs. 10 percent of liberals who cite New York Times as their main news source. Twelve percent of liberals said MSNBC was their main news source, 13 percent said NPR, and 15 percent said CNN (Mitchell et al. 2014). 39

44 3 Data 3.1 Sample The final sample contains a panel data set of 3,357 firm-year observations from January 2011 through December The sample contains 1377 unique firms from 59 industries excluding finance, insurance, and real estate (FIRE companies), which are consistently excluded from corporate finance research. The sample of firms is determined by the media data; that is, all non-fire firms mentioned by media organizations Facebook pages are included in the analysis. The data are cleaned for outliers to reduce measurement error. All variables were windzorized at the 1 st and 99 th percentiles to eliminate outliers. In addition, industries with less than three observations were dropped. Firms with estimated book leverage greater than one were dropped. Firms with Tobin s Q greater than 10 were dropped, consistent with the literature (Elyasiani and Jia, 2010). The resulting panel sample is unbalanced. Variable definitions can be found in Table Media Data Content contributing to this analysis came from the following media organizations: Barrons, Chicago Tribune, CNBC, the Economist, Forbes, Fox News, the Los Angeles Times, Market Watch, New York Times, National Public Radio (NPR), Reuters, USA Today, Washington Post, Wall Street Journal, Yahoo Finance. Sources were chosen based on their position as leaders amongst different segments of the population (Table 4). The social media data were collected and processed through several steps. First, the social media data were collected using Facebook s Application Programmable Interface (API) from January 1 st, 2011 through December 31 st, 2015 from the respective 40

45 source Facebook pages. Next, the full-length articles from each of the different posts were collected using links embedded in the Facebook data. Full-length articles were collected for all news organizations except the Wall Street Journal. The Wall Street Journal subscriber agreement limits collection of full-length articles, so article summaries were collected instead. Facebook data from other large and prominent news sources, such as the Associated Press, Financial Times, and Bloomberg Business News were initially collected, but omitted from the research due to a lack of working hyperlinks. Hyperlinks are a necessary component of this analysis because they allow the scraper to collect text data from each of articles linked to the Facebook post. Without the links the text analysis component is minimal, therefore those three sources were omitted. The text from each article was then tagged using a text parser, Open Calais 14, maintained by Thomson Reuters. Articles are tagged to identify which firms are mentioned in each article. After tagging, the article text was stemmed and cleaned of stop words, common practice in textual analysis. After processing, sentiment analysis was used to gauge tone within each of the articles. The sentiment analysis applied here counts the number of positive and negative words in a document, and then subtracts, to get an overall sentiment score. Sentiment scores depend on the dictionary used to classify words as positive or negative. While sentiment dictionaries like the Harvard-IV-4 are commonly used in sentiment analysis (Tetlock et al., 2008), this research will use a financial dictionary created by Loughran and McDonald. Loughran and McDonald argue that dictionaries such as Harvard-IV-4 don t capture financial sentiment well 15 ; 74 percent of words We find that almost three-fourths (73.8%) of the negative word counts according to the Harvard list are attributable to words that are typically not negative in a financial context. Words such as tax, costs, capital, board, liability, foreign, and vice are on the Harvard list (Loughran & McDonald 2011). 41

46 classified as negative in the Harvard-IV-4 typically aren t negative in a financial context (Loughran and McDonald, 2011). The dictionary was created using a sample of 10-Ks filed over the period of This dictionary is used by Loughran and McDonald in their own work and is also used in other research such as Lui and McConnell (2013). After applying sentiment analysis to the tagged text, the data were aggregated by year and by company Governance, Financial, and Board Data Data on firm fundamentals, finances, and other control variables were obtained through the Compustat database, available from Research Insight, and the CRSP database. Compustat contains financial statement data for all firms in the US and CRSP contains equity market data. Data about the board were obtained from several different sources. For years , the governance data came from the ISS database (formerly Riskmetrics) from the Wharton Research Data Services. My advisor, Abigail Hornstein, accessed the data from prior collaborative research at NYU Stern. Economics professor, Reda Moursli, supplied governance data from the year 2015, from Orbis. This dataset contained all previous and current board members through the year Board data for the year 2014 were imputed from the years 2013 and If a board member was appointed before the year 2015 and served through 2015, they were assumed to be a director in Directors appointed in 2015 were assumed not to have served in Directors that were appointed prior to 2014, but were categorized as previous directors were search individually to identify their status in Because director data for 2014 was imputed from 2015, outliers in the 2014 director data were manually checked using Morningstar.com and changed if there were differences. 42

47 Institutional ownership data were obtained from Osiris, an international database on public-company financials. All data were aggregated by firm by year, such that variables are a measure of the total number of funds invested in the firm each year. 3.2 Summary Graphs and Statistics The subset contains about 20-percent of the Compustat dataset, a dataset used almost ubiquitously across public firm research. The distribution of industries is similar to Compustat with the exception of a few outliers (Table 6). Industries such as Chemical Products (SIC: 28), Metal Mining (SIC: 10), Electronic and Electrical Equipment (SIC: 36), and Water Transportation (SIC: 44) are the most under-represented. Some over-represented industries are: Eating and Drink Places (SIC: 58), Apparel and Accessory (SIC: 56), Food Products (SIC:20) and General Merchandise (SIC: 53). These results are consistent with expectations of which groups of firms the media mentions frequently and those it doesn t. 16 Variables where we do see differences in summary statistics are research, development and advertising expenses, size of the firm, and volatility (Table 7). All other variables used in the analysis have similar distributions to that of the full Compustat dataset, or have a slightly narrower distribution. 16 Note that over-representations and under-representations of certain industries don t change variable distributions dramatically (Table 7). 43

48 4 Methodology, Models, and Hypotheses This section translates the theoretical literature and the empirical goals outlined in the introduction and earlier chapters into models backed with econometrically appropriate methodologies. The empirical goals of this research are three-fold. First, this research seeks to answer the question is there a statistically significant and economically important relationship between financial performance and social media? This question is addressed in the first lines of analysis: estimation of the complete sample by ordinary least squares (OLS). The second goal of this research is to explore more specifically how media affects financial performance. Does the relationship between financial performance and media coverage manifest by making actions reputationally costly for managers or directors, or persuading shareholders to exit the firm? This area of inquiry is explored using segmented sample OLS regression and complete sample OLS with interactions. The third goal of this research is to estimate the financial performance social media relationship through potentially more appropriate modeling methods than OLS. Corporate governance research frequently suffers from endogeneity concerns from simultaneity bias. The common solution to this problem is to estimate the models as a simultaneous system with instrumental variables. The line of inquiry is explored in the third line of analysis: Simultaneous Models with Instrumental Variables. This chapter is structured in three parts: methodology, models, and research questions and hypotheses. First, the methodology section explains why several different methodologies are used and how they help answer the empirical goals outlines above. Then the modeling section explains how the baseline models are constructed, which 44

49 variables are included in the models, and why the variables were included. Finally, research questions and hypotheses are outlined in the last part of this section. 4.1 Methodology The empirical literature and theoretical literature motivates four distinct types of empirical inquiry. First, the models will be estimated using standard panel data estimation techniques OLS and least square dummy variables (LSDV). Second, interaction terms are included and the LSDV models are re-estimated. Third, the models are re-estimated after segmenting the dataset, to test for media bias and differing corporate governance effects. Fourth, the models are estimated as a system with instrumental variables, acknowledging there may be endogeneity issues using standard panel data estimation techniques. All the models will be estimated with lagged independent variables. Lagging the dependent variables affects the models in two ways. First, it s expected that the lagged models will fit the data better 17 ; changes in governance or ownership structure aren t immediate. Second, lagging the variables reduces endogeneity concerns Complete Sample OLS Regressions Econometrics literature suggests that estimation techniques that account for unobserved industry-specific effects will be the most appropriate techniques to use. Controlling for unobserved industry-specific effects is important because industryspecific effects are likely correlated with the independent variables. The econometrics literature suggests a fixed effect model with fixed effects transformation (FE) or dummy variables (LSDV) to control for industry-specific 17 See Board of Directors for further discussion about how differencing and lagging fits the data better. 45

50 effects. 18 The FE and LSDV are similar approaches in that both models seek to account for unobserved heterogeneity, which can bias standard errors. The key difference is that LSDV controls for industry by including dummy variables. The fixed effect model controls for heterogeneity bias by clustering standard errors around industry. As there are a large number of within-groups, using clustered standard errors can increase the standard error, indicating the LSDV approach is most likely the best approach for this research. 19 All models are estimated with robust standard errors Complete Sample OLS Regressions with Interactions The main question of interest to this research is how financial performance changes in response to media sentiment and coverage. However, the literature indicates that media may not directly affect the financial performance of the firm, but instead affects governing decisions or ownership at the firm. The complete sample OLS regressions from the previous section consider the effects of media on governance, media on ownership, and media on financial performance, but not how effects of media motivate changes in governance or ownership, which cause changes in financial performance. To test this, interaction terms are introduced into the regressions. The methodology is replicated from the complete sample OLS regression section, but only regressions of interest, the financial performance regressions are reported. 18 The random effects estimation technique assumes unobserved firm-specific effects are uncorrelated with other explanatory variables in the model and the pooled OLS estimation could introduce heterogeneity bias which would result in biased and inconsistent estimators. The literature suggests, that fixed effects and least squares dummy variables will be preferred to both these techniques. 19 Originally, the models were estimated with random effects models and fixed effects, but as expected by the econometrics literature, the LSDV models fit the data best. Only OLS and LSDV models are reported in the results. 46

51 4.1.3 Segmented Sample OLS Regression The third part of the empirical analysis looks at how different types of coverage affect financial performance, ownership, governance, capital structure, and media coverage. Previous literature has primarily looked at news articles from the New York Times and Wall Street Journal. However, all media organizations have biases, which may affect how organizations report and what they report on. 20 The segments analyzed are: financial versus non-financial media, mainstream versus non-mainstream media, paywall vs. non-paywall media, and political leanings of media. 21 Summary statistics for the segments are reported in Table 12 through Table 15. Several steps are taken to test if bias affects coverage. First, variation in the data are explored using visualization techniques (Figure 1 - Figure 8). After the visual analysis, simple statistical tests: Chi-square tests 22, ANOVA 23, and t-tests are run to motivate the following regression analysis. These tests are reported in Table 16 and Table 17. After the visual and exploratory statistical analysis, regressions are run to test if there are coefficient differences between the segments. Several different analyses can be used to test for coefficient differences between two or more groups. The first option is to estimate two or more groups independently and compare slope and significance changes. A second option, the preferred option, is to pool the groups together and compare slope differences by including interaction terms. However, the models become too complex with the addition of fifteen interaction terms and rank problems occur. To avoid the problems described above, each segmentation will instead be estimated using 20 For example, see Table 2 which shows political bias in media institutions. 21 See Table 8 - Table 11 for individual media organization segment categorization. 22 The Chi-square test is used to test if there are statistical differences between two categorical variables. 23 The t-test is used to test for significant differences in the means between two groups, and ANOVA was used to test the significance of segments with more than two groups. 47

52 separate regressions, and the coefficients from separate regressions will be compared independently of each other. F-tests are used to test for joint significant of the media variables across the different segments Simultaneous with Instrumental Variable Regression Models The question of interest is: does financial performance change with changes in social media coverage and sentiment. However, it is likely that past firm performance or industry performance informs future media coverage about firms, resulting in joint determinism. This is an endogeneity issue. Corporate governance research frequently suffers from endogeneity issues. Without adapting estimation techniques to account for endogeneity issues, models suffer from inconsistency and simultaneity bias. Endogeneity issues are partially addressed by lagging the independent variables. However, there is still a question of causality, that lagging variables does not solve. The problem is circumvented in the literature 24 by estimating endogenous variables as simultaneous system with instrumental variables (Cho 1998; Bhagat and Jefferis, 2002; Bhagat and Bolton, 2008). Estimating the system using instruments helps to determine the causality of the media and may effectively mitigate endogeneity issues. Following the econometric literature, instrumental variables were selected based on two characteristics: high correlation with the endogenous variables of interest and lack of correlation with the error term. 24 Bhagat and Jefferis s (2002) research suggests that while governance variables affect financial performance, performance may also affect changes in governance structure. Their research suggests other variables: capital structure and ownership may also be endogenous. Cho (1998) suggest corporate ownership, financial performance, and corporate investment are endogenous. Bhagat and Bolton (2008) suggest financial performance, governance, ownership, and capital structure are endogenous, and suggest a four-equation simultaneous regression framework to circumvent the issue. 48

53 The literature indicates that several techniques can be used to estimate jointsystems including: seemingly unrelated regressions (SUR), two-stage least squares (2SLS) and three-stage least squares (3SLS). SUR is used as an estimation technique when all independent variables are exogenous, which indicates that SUR isn t the correct jointsystem estimation technique for this research. 2SLS and 3SLS with instrumental variables are commonly used when the system contains endogenous variables. 3SLS is generally preferred to 2SLS when estimating a system with more than two equations because it is a more efficient estimation technique (Wooldridge, 2010). To examine the relevance of selected instruments, several statistical tests are employed. First, F-tests are verified as significant in the first stage regression (Rubin & Smith 2009). Then the Hausman specification test is performed, along with the Sanderson-Windmeijer test for under-specification and Sanderson-Windmeijer weak identification test. IV estimators can make estimates worse if they re weakly correlated with the endogenous variables in question (Wooldridge, 2013). Additionally, all 2SLS and 3SLS models were estimated with robust standard errors. Instrumental variable definitions are in Table 18 and summary statistics are reported in Table 19. Instrumental variable tests are reported in Table 22. Ideally, the simultaneous regressions would also include interaction terms to test to see how media interacts with governance and ownership variables, replicated from earlier methodologies, section and As we were unable to identify additional instruments for which the data was also available, we conclude that such models are beyond the scope of this analysis. 49

54 Financial Performance Instrumental Variables Bhagat and Bolton (2008) suggest that treasury stock may instrument for financial performance. However, treasury stock has a very low correlation with both ROA and Tobin s Q in this sample. Regressing treasury stock on ROA and Tobin s Q doesn t show statistical significant and the coefficient is of little practical importance. This indicates that treasury stock isn t a strong instrument to use with this dataset. Alternatively, Arino et al. (2010) suggest that the social performance of a firm, defined as the quality of relationships between a firm and its stakeholders is highly associated with the type of industry the firm is in. Similarly, social performance is most likely very related to the frequency of media coverage of a firm. Acknowledging that financial performance is dependent on the type of industry a firm is in, Arino et al. (2010) address endogeneity concerns by re-defining their financial performance variables. They redefine their instruments as the value relative to the average of the industry for a given year (Garcia-Castro et al., 2010). Demeaning the financial performance variables results in financial performance variables that are independent of industry-specific shocks or events that affect the financial performance of a firm Governance Instrumental Variables Knyazeva et al. (2013) suggest that the market for board members affects the level of board independence. They proxy for board independence by looking at the nonfinancial local director pool located within 60 miles of the firm. The governance data used for this research aren t as detailed as the data in Knyazeva et al. s study. However, this research proxies for local director pool by looking at the average number of board 25 Demeaning dependent variables is a controversial instrumental variable approach. Gormley and Matsa (2014) suggest using demeaned dependent variables may lead to biased and inconsistent estimates. However, the instruments are being used as independent variables, so this criticism doesn t apply. 50

55 members that serve on a board at another firm. If the local director pool is constrained by regional factors, as suggested by Knyazeva et. al (2013), then it s likely that directors who are regionally restricted serve on more than one board. There is a positive and statistically significant association, 0.04, between percent of independents and average number of directorships per firm per year, indicating that the percent of independents increases with the number of directorships served (Table 20). The observed positive relationship is consistent with Knyazeva et al. s (2013) theory of director independence. The positive association suggests that regions with small local director pools rely more heavily on independent directors to fulfill investor- or federally-specified independence regulations. Bhagat and Bolton (2008) suggest two instruments for governance: the number of active CEOs on a board and the percent of board ownership. The governance data used in this research aren t as specific as the data used in Bhagat and Bolton s research, so constructing a variable that includes the number of active CEOs is not possible for this research. Similarly, the governance data on board ownership is not consistent across years. Although, the data on exact board ownership percentages is inaccessible, there are data on which board members are also shareholders. Aggregating the data to the firm level, a variable is constructed which represents the percent of board members that are also shareholders of the firm. Bhagat and Bolton (2008) suggest that board ownership indicates stronger governance, thought to be correlated with the number of independents, because board members with an investment in the firm are more likely to provide good oversight and monitoring. The statistically significant association between percent of independents and percent of shareholders is.13, indicating that there is a 51

56 positive and economically important relationship between number of independents and percent of shareholders (Table 20) Ownership Instrumental Variables One instrument that has been used to proxy for ownership in the past is stock market index inclusion. Arino et al. (2010) write firms listed in SP500 are supposed to have higher exposure to investors, media, activists, etc. and therefore, they are expected to have higher visibility. They operationalize this observation into an instrument by creating a dummy variable that equals 1 if the firm is in the S&P 500 and 0 otherwise. S&P 500 make-up is more significantly and strongly correlated with mutual fund ownership than with hedge fund ownership (Table 20). In their study on institutional ownership, stability, and corporate performance, Elysiani and Jia (2010) suggest four different instruments to proxy for institutional ownership. They suggest instrumenting ownership using different measures of stability as institutional investors are most likely attracted to firms that are stable and liquid (Rubin and Smith, 2009; Elyasiani and Jia, 2010). The instruments that they use to measure stability are daily stock turnover, dividend yield, a dummy variable indicating positive earnings, and the number of analysts following the institutional owners. 26 The other three variables they suggest daily stock turnover, dividend yield, and positive earnings are used as instruments Capital Structure Instrumental Variables Altman s Z-score is used to instrument for capital structure as the formula is widely used to predict the likelihood of a firm going bankrupt. Typically, a high Z-score 26The number of analysts is used in the literature as an instrument for institutional ownership by multiple researchers (e.g. Cornett et al., 2007; Elyasiani and Jia, 2010), but unfortunately the data isn t available for this research. 52

57 (i.e., a score greater than 2.99) indicates that the firm is not at risk of bankruptcy (Ross et al., 2010). A Z-score between 1.81 and 2.99 indicates there is some risk of bankruptcy and a score less than 1.81 indicates that the firm is in distress (Ross et al., 2010). Table 20 shows the slope coefficients between the original dependent variables and identified instrumental variables. There is a negative relationship between Z-score and leverage which is consistent which what the literature would predict. A firm with higher leverage faces greater risk from bankruptcy and would most likely have a lower Z-score Media Instrumental Variables This research introduces media coverage as a fifth equation into the system of equations, which makes a fifth instrument necessary. Dyck et al. (2008) create an instrument which they coin the natural newsworthiness of a firm. They define the natural newsworthiness of a firm to be the number of references in the Wall Street Journal and Financial Times during the six-month period prior to their sample period and prior to the default that they examine in their research. This research doesn t consider a pre and post-period like the Dyck et al. (2008) paper. However, extrapolating from the natural newsworthiness idea, a similar instrument can be created to proxy for natural newsworthiness. In the media, certain industries are covered more regularly than other industries. Retail and technology are examples of two industries that receive more media coverage, while industries related to transportation or pharmaceutical research generally appear less frequently in the news (Table 6). Differences in coverage are likely due to industryspecific attributes, such as performance. It s likely that industries which perform on the tails are covered more frequently than firms which perform closer to the average. Natural newsworthiness is most likely driven by the firm itself and the industry the firm 53

58 is in. Demeaning the media variables creates instruments that are independent of industry-specific events or shocks that may affect frequency of industry coverage. 54

59 4.2 Models The modeling section contains two sub-sections: Bhagat and Bolton models and the Media Models. The Bhagat and Bolton (2008) models are reviewed because the models estimated in this research, referred to as media models, are based off the of them Media Models includes the models in equation form, and an explanation of why different dependent and independent variables included in the models Bhagat and Bolton Models Bhagat and Bolton create a system with four equations to model determinants of financial performance, governance, ownership, and capital structure using a sample from The relationship of interest is the association between corporate governance and financial performance, but they use a system of equations to estimate this relationship. 27 Bhagat and Bolton (2008) measure financial performance using return on assets (ROA), calculated as operating income before depreciation divided by total assets. They model financial performance as a function of governance, ownership, capital structure, and controls. Governance is measured using governance indices and individual indicators such as CEO-chair duality. Ownership is measured as the percent of the firm owned by the CEO. Capital structure is a measure of leverage calculated as (long-term debt plus current debt) divided by total assets. Bhagat and Bolton (2008) include controls of research, development, and advertising expenses (RDA), size of the firm, and return volatility. RDA expenses are the sum of research, development, and advertising expenses. Volatility is calculated as the 27 See section Simultaneous Models with Instrumental Variables 55

60 standard deviation of the monthly stock return from the previous five years. Size of the firm is proxied for using the natural log of assets. The Bhagat and Bolton (2008) models are replicated below in Equation 3 - Equation 6. The X vector represents the control variables: firm size, research, development and advertising expenses; and return volatility. [3] Performance = f 1 (Governance + Ownership + Capital Structure X, ε) [4] Governance = f 2 (Performance + Ownership + Capital Structure X, ε) [5] CEO Ownership = f 3 (Performance + Governance + Capital Structure X, ε) [6] Capital Structure = f 4 (Performance + Governance + Ownership X, ε) Media Models The media models build off the Bhagat and Bolton models introduced in the previous section. The media models include social media variables: total coverage, Facebook shares, positive sentiment, and negative sentiment. For consistency, the variables used in this research are the same variables recommended by Bhagat and Bolton, except for a few changes due to data constraints Dependent Variables The dependent variable of interest to this research is financial performance. Financial performance is often measured using data from financial statements, accounting-based measures, in conjunction with market data, market-based measures. Many researchers that study the relationship between corporate governance and financial performance use Tobin s Q, a market-based measure of performance. Tobin s Q is the ratio of a firm s market value to the firm s replacement value. A Tobin s Q value greater than one indicates the firm is overvalued, while a value less than one indicates the converse. 56

61 Although, Tobin s Q is frequently used in the literature, there are limitations. Firms with higher than average intangible assets have higher Tobin s Q values (Bhagat and Jefferis, 2002), which interferes with causal interpretations. Thus, researchers have suggested other market-based measures, such as stock returns. However, these measures suffer from anticipation issues (Bhagat and Jefferis, 2002). Accounting-based measures of financial performance, including return on sales (Karpoff et al., 1996) and return on assets (Bhagat and Bolton, 2008), have also been suggested as alternatives to market-based measures of performance. Accounting-based indicators don t suffer from market moods or anticipation problems, but also fail to reflect true economic costs or benefits. For robustness measures, this research will consider both market-based measures and accounting measures of performance, Tobin s Q and ROA, respectively. ROA is equal to net income divided by total assets, which is a measure for return on invested capital. A high return on assets indicates that the firm is good at converting investments into net income Independent Variables Independent variables of interest to this research are governance variables, institutional ownership, capital structure, and media coverage. Bhagat and Bolton measure governance using a series of difference governance indices and individual indicators. Individual indicators are favored by the literature, so this research measures governance with three indicators: percent of directors who are independent, CEO-chair duality, and board size. 28 Although there are other variables that could encompass 28 See section Board of Directors for reference. 57

62 governance, these are the variables consistently used in the literature. 29 Board size is logged, consistent with the literature (Bhagat and Bolton, 2008). Institutional ownership data used in this analysis are limited to mutual funds and hedge funds. The two different types of ownership make for an interesting comparison because of their differing investment strategies. 30 The ownership variables are counts of the number of funds invested at the firm in a given-year. Counts are used because the literature suggests engagement is not so much a function of shareholder ownership concentration, but rather shareholder ownership along (Çelik and Isaksson, 2013). The literature suggests that changes in fund ownership proxy for exit and captures institutional investors who are voting with their feet. The media literature, presented earlier in Section 2.2.3, suggests there are two media characteristics of interest: sentiment and frequency of reporting. Sentiment is measured with two variables, positive sentiment and negative sentiment, which are the average sentiment over year. Frequency of reporting is also measured using two variables: Facebook shares and total coverage. Facebook shares is the average number of shares articles mentioning the firm received in a year. Total coverage is the total number of times a firm was mentioned in a year by the media. These two variables capture the intensive margin and extensive margin of coverage, respectively. Summary statistics (Table 7) show that all the media variables are right-skewed, so they are logged in the estimations to reduce the range of values. This research looks at two measures of coverage to examine the difference between media-driven content and consumer-driven content. The total coverage a firm 29 Coles et al. s work (2014) refines the definition of independent director. However, this research will stick to traditional definitions of board independence due to data limitations. 30 See section Shareholders for reference 58

63 receives in a year is determined by media institutions, who choose what to report on and when to report. Facebook shares is a measure of consumers responses to that content. Because of this Facebook shares is expected be a more reactionary response to changes at the firm, and total coverage is expected to represent more regular coverage. The literature also suggests including as control variables firm size, capital structure, research and development expenses, advertising expenses, and industry performance to account for heterogeneity among firms (Bhagat and Bolton, 2008). Firm size is important to control for because size may be associated with institutional investment (Smith, 1996). Smith (1996) writes, If larger firms comprise a larger percentage of an institution s investment portfolio (perhaps due to indexing strategies), the expected benefits may be larger from targeting these firms since the private gain to the activist, if targeting is successful, is larger. Additionally, larger firms are expected to attract more media attention than smaller firms. Size of the firm is measured using property, plant, and equipment, commonly used to proxy for firm size in corporate finance research. Property, plant, and equipment is logged to narrow the range of values. Bhagat and Bolton (2008) also include leverage in their models. Firms that issue debt signal a reduction in agency costs, because managers are more accountable to their actions (Grossman and Hart, 1980). Debt increases the probability of bankruptcy, which increases the probability that managers will lose their jobs and perquisites. Although issuing debt signals accountability to shareholders, taking on too much debt can put the firm at risk of bankruptcy. The ratio used for leverage in this paper is the sum of longterm debt and current liabilities divided by total assets. This measure of leverage is also known as book leverage, and it is consistent with the leverage ratio used by Bhagat and Bolton (2008) and Bebchuk, Cohen, and Ferrel (2009). 59

64 Expense variables, such as advertising and research and development, are included in the analysis to control for existing agency costs. Managers that devote significant resources to research and development expenses decrease free cash flow resources, and signal long-term thinking. Decreased advertising or research and development expenses may be signs of an entrenched manager (Morck et al., 1988; Bhagat and Jefferis, 2002). Advertising and research, and development are combined into a single variable, referred to as RDA in the estimations. Firms not reporting advertising expenses, or research and development expenses, are assumed to have spent zero, and the variable is coded as such. This is consistent with the treatment of RDA in Bhagat and Bolton (2008). Another important factor to control for is industry performance. Macroeconomic conditions may favor some industries over others. Additionally, the media is likely to report on some industries more than other industries. For example, retail and technology are commonly reported on, but other industries such as transportation or manufacturing are less likely to be reported on (Table 6). Industry performance is controlled for by the inclusion of industry dummy variables in the estimations. Variable definitions can be found in Table Complete sample OLS Media Models The first line of analysis considers how financial performance is affected by media coverage. To test if there are endogeneity concerns, all potential endogenous variables are estimated as a function of other potential endogenous variables. The vector of control variables, X, is the size of the firm, research, development, and advertising expenses, and return volatility, as in Bhagat and Bolton (2008) and others. The models are presented as Equations 7-11, below: 60

65 [7] Performance = f 1 (Governance + Ownership + Capital Structure + Media X, ε) [8] Governance = f 2 (Performance + Ownership + Capital Structure + Media X, ε) [9] CEO Ownership = f 3 (Performance + Governance + Capital Structure + Media X, ε) [10] Capital Structure = f 4 (Performance + Governance + Ownership + Media X, ε) [11] Media = f 5 (Performance + Governance + Ownership + Capital Structure X, ε) Complete sample OLS Media Models with Interactions The second line of analysis considers whether financial performance is also affected indirectly by media coverage. Thus, the independent variables included in the model include interactions between media and governance and between media and ownership. The literature suggests that financial performance is affected by media coverage in two ways. First, media educates shareholders who have greater incentives to monitor and engage with management or exit the firm. Second, the media shapes managers and directors reputational environments, resulting in governance changes at the firm. To test these theories, media is interacted with governance characteristics and ownership variables. Equation 7 is thus rewritten as Equation 12: [12] Performance = f 1 (Governance + Media + Media Governance + Media Ownership X, ε) Segmented Sample OLS Media Models The segmented regression also utilizes Models 7-11, to test for media bias and differing governance effects, but does so with segmented data. Please see Section 4.1.3, earlier, for a full discussion of why it is necessary to estimate these equations multiple times for subsets of the data vs. using the entire dataset. 61

66 Simultaneous with Instrumental Variable Models The third line of analysis in this section looks at how different instruments may help solve endogeneity issues that are often present in corporate governance research. Instrumental variables were explained in Section 4.1.4, and the definitions are provided in Table 18. The instrumental variable models are represented in equation form below: [13] Performance = f 1 (G + O + C + M X, ε), where performance is instrumented as either industry-adjusted Tobin s Q or industryadjusted ROA. [14] Governance = f 2 (P + O + C + M X, ε), where governance is instrumented as either percent shareholder or average number of directorships. [15] Institutional Ownership = f 3 (P + G + C + M X, ε), such that institutional ownership is instrumented as daily stock turnover, dividend yield, or positive earnings. [16] Capital Structure = f 4 (P + G + O + M X, ε), with Altman s Z-score used as an instrument for capital structure. Finally, [17] Media = f 5 (P + G + O + C X, ε), where media is instrumented as industry-adjusted total coverage or industry-adjusted Facebook shares. 62

67 4.3 Hypotheses and Research Questions This section outlines how the media is expected to affect other governance mechanisms, shareholders and the board, which in turn affects the financial performance of the firm. A flow-chart diagram outlines this process in Figure 9. The literature suggests that a firm s financial performance can be affected through three governance avenues. First, changes in governing decisions can be motivated by changes in reputational capital. Second, changes in governing decisions through institutional engagement can affect the financial performance of the firm. Third, changes in ownership, such as shareholders exiting the firm can affect stock prices. The first relationship has been studied extensively by Dyck et al. (2008) and Lui and McConnell (2013). They suggest that frequency and sentiment of coverage affect reputational capital, which can be tested empirically by regressing financial performance on interaction terms of media and governance variables. Extrapolating from the Dyck et. al (2008) and Lui and McConnell (2013) papers, a new model can be devised that updates earlier models presented in Equation 1 and 2 to incorporate the effects of media on decision-making on financial performance: [18] Financial Performance = f(governance X) and Governance = f(reputational Capital c + Reputational Capital s X), where c is the probability of information reaching critical audiences and s is the sentiment of coverage The second theoretical relationship is supported by a narrower empirical literature due to the reduced availability of the necessary data. There is limited data on which institutional investors engage with management, how they choose to engage with management, and what the outcomes of that engagement are. Empirical evidence suggests that the media helps to inform the investors, which helps to reduce information 63

68 asymmetry (Bushee et al., 2010). Lowenstein (1999) argues that decreased informational asymmetry, increases opportunity costs and spurs shareholders to action. However, Lowenstein s theory cannot be tested empirically because there is limited data on institutional engagement. The question of how media affects institutional engagement, which in turn affects governance, which then affects financial performance is a much more complicated question. This question could be tested using a three-way interaction between the media, institutional engagement, and governance, but is beyond the scope of this research due to data availability considerations. The third relationship is supported by research from Lowenstein (1999), Djankov (2003), Bushee (2010), and others. All of these papers suggest that the media helps reduce informational asymmetry, spurring shareholders to engage with managers. One way shareholders can engage is to vote with their feet and exit the firm. This effect can be tested by interacting the ownership variables, hedge and mutual funds, with the media variables. It is not obvious from the literature how the media variables will affect ownership. However, it is likely that institutional investors are already informed and thus total coverage will not be as important as the other media variables. However, the total number of Facebook shares is a more a measure of consumer-driven content (i.e., aggregate public engagement with the firm), and so Facebook shares may affect ownership. Additionally, the tone of the coverage may also affect how shareholders interact with the firm. Specifically, behavioral theories suggest that a larger quantity of articles with extreme sentiment (i.e., most negative or positive) may encourage additional engagement with managers or may spur shareholders to enter or exit the marketplace. 64

69 From these hypotheses, a third model, which can be interpreted as an updated version of the models presented earlier as Equations 1 and 2, can be constructed: [19] Financial Performance = f(ownership X) and Ownership = f(informational Environment c + Informational Environment s X), where c is the probability of information reaching critical audiences and s is sentiment The literature suggests that changes in governance may be affected through two avenues: changes in reputational capital or institutional engagement. Although, this research cannot directly test the effects of changes in institutional investors informational environments on governance and financial performance, the segmented sample OLS regressions presented in this thesis help to inform about how changes in the information set available to institutional investors (i.e., media coverage) may affect governance and performance. The literature suggests that if the information intermediary hypothesis is important then firms that produce quality investigative reporting should be included in the analysis. This suggests that more traditional sources of business and finance information will explain better the observed variation in corporate performance and governance. Additionally, media sources that have paywalls might engage in more and higher quality investigative reporting (e.g., the New York Times or Wall Street Journal vs. Reuters and USA Today; see Table 8 and Table 10). The reputational capital hypothesis is tested by exploring whether mainstream media differ from non-mainstream media. That is, more mainstream media sources might not engage in as much investigative reporting as specialized or non-mainstream media sources, as mainstream media often amplify content from other prominent 65

70 sources of financial information (e.g., New York Times vs. Forbes; see Table 9). If the observed variation in corporate performance and governance can be better explained with the mainstream media subset than with the non-mainstream media subset, that provides evidence in support of the reputational costs hypothesis. To summarize, the literature review and methodology thus pose three large theoretical and empirical research questions and hypotheses: Question 1: Is there a statistically significant and economically important relationship between financial performance and social media coverage? Hypothesis 1: There is a statistically significant and economically important relationship between social media coverage and the financial performance of the firm. This will be tested through estimation of Equations Question 2: Does the relationship between financial performance and media coverage manifest by making actions reputationally costly for managers or directors, or persuading shareholders to exit the firm? Hypothesis 2: There is a statistically significant and economically important relationship between the media variables and governance variable: percent of independents. Additionally, there are coefficient differences on the media variables in the segmented regressions: financial, mainstream, and paywall media. Mainstream media and nonpaywall media, and non-financial media will fit the governance model better. This will be tested through estimation of equation 8 31 in the segmented sample OLS regressions. Question 3: Is there evidence of endogenous dependent variables? Do the instrumental variables models appear to better explain the data than the OLS models? Hypothesis 3: Financial performance, ownership, board characteristics, leverage, and media are endogenous and estimating the regressions as a simultaneous system is the appropriate methodology. This will be tested by equations Equation is of particular interest, because equation 8 looks at the relationship between governance and the media. The literature predicts that both the informational intermediary hypothesis and the reputational capital hypothesis will affect governance. 66

71 5 Results As stated in the last chapter the empirical goals of this research are three-fold. First, this research seeks provide evidence if favor or against the first hypothesis: that there is a statistically significant and economically important relationship between media coverage and the financial performance of the firm. The first two lines of empirical analysis, estimation by complete sample OLS and estimation by complete sample OLS with interaction terms, provide support in favor of this hypothesis. A full discussion of results is presented in this chapter and references to Table 23 - Table 36, the regression tables. The second goal of this research is to explore more specifically the mechanisms by which media plays a corporate governance role. This area of inquiry is explored using complete sample OLS with interactions and segmented sample OLS regression. Results from the complete sample OLS with interactions and the segmentation models provide support in evidence of hypothesis 2, that the reputational costs hypothesis is more persuasive than the informational intermediary hypothesis, consistent with what the literature would predict. The third goal of this research is to see if the earlier results are robust to tests that control for potential endogeneity concerns. Results from the first line of analysis, complete sample OLS, indicate that there are potential endogeneity problems, which validates concerns and motivates re-estimation of the models using alternative estimation techniques, instrumental variables. 5.1 Complete Sample OLS Regressions Complete Sample OLS Regressions are reported in Table 23 through Table 30. Model 1 in the first column of each regression table in this section is an approximate 67

72 replication of the Bhagat and Bolton (2008) with OLS. Model 2 in the second column of is the approximate replication of Bhagat and Bolton (2008) using the LSDV approach. The successive models include the media variables singly and then jointly, with all odd numbered models using OLS and all even numbered models using LSDV for the industry fixed effects. Positive score and negative score are estimated independently both to highlight the idiosyncratic effects of positive or negative sentiment and to avoid potential multicollinearity. Later, in Models 7-10 both types of sentiment are included simultaneously. Robust standard errors are reported for each model to account for possible heteroscedasticity and autocorrelation Financial Performance Results show there a statistically significant relationship between total coverage and Facebook shares, and financial performance (Table 23 and Table 24). The media variables positive score, negative score, Facebook shares, and total coverage are jointly significant in both models of financial performance. Although, the estimates indicate total coverage is significantly statistically related with Tobin s Q, and Facebook shares is statistically significantly associated with ROA, the coefficients are small. For example, a one-percent increase in total coverage is associated with a.001-unit increase in Tobin s Q. The summary statistics (Table 7) show that the media variables are all very rightskewed, which is why they appear in logged form in the regressions. The skewness also indicates that it may be helpful to interpret the variables as a standard deviation change, rather than a percentage change. From the summary statistics, which report the media variables in levels, standard deviation unit changes can be computed to reflect percentage changes. For example, the standard deviation of the variable negative score is A one-standard deviation 68

73 unit change from the mean is then equivalent to 95-percent change, or the standard deviation (27.16) divided by the mean of negative score (28.57). Similarly, a one-standard deviation change in positive score corresponds to a 90 percent change in score units. A one-standard deviation change in Facebook shares corresponds to a 311 percent change in shares, and a one-standard deviation change in total coverage corresponds to a 680 percent change in total coverage. 32 Returning now to the original example from the prior paragraph looking at the association between Tobin s Q and total coverage, a onestandard deviation unit increase in coverage is associated with a.62 unit increase in Tobin s Q. We note that this is economically large as the estimated Tobin s Q in our dataset has a mean and standard deviation of 1.67 and 1.40, respectively. Other estimated coefficients for variables have the expected signs and statistical significance, indicating the models are consistent with the literature. In both the Tobin s Q and ROA model (Table 23 and Table 24) leverage is consistently negative and statistically significant, which is consistent with Bhagat and Bolton (2008). Governance variables included in this in analysis are duality, board size, and percent of independents. Duality and percent of independents are statistically insignificant for Tobin s Q and ROA models, but board size is statistically significant in the ROA model. Ownership variables for hedge and mutual funds aren t very significant across the different financial performance models and aren t of much practical importance given their relatively small coefficients. The R-squared for the two financial performance models increased between 0.5 and 1 percent with the addition of the media variables (Models 9 and 10 vs. Models 1-2) 32 From Table 7, percent changes are calculated as the standard deviation divided by the mean. Negative score = 27.16/28.57 =.95. Positive score = 13.04/14.66 =.90. Facebook shares = / = 3.11 and total coverage = /33.09 =

74 and the media variables are individually and jointly significant in both financial performance models. Although, the magnitudes of the coefficients appear small, the relationship between performance and media is economically important if changes in frequency of coverage are large enough. 33 Overall, the financial performance models are consistent with the literature, including Bhagat and Bolton (2008) Governance Board independence is statistically significantly affected by the media variables positive score, negative score, and Facebook shares (Table 25). The inclusion of media variables increased the percent of explained variation, R-squared, by 2.5 percent, indicating a better model fit. The F-test indicates that the media variables are jointly significant, and the signs of the media variables are as expected. For example, the percent of independents is negatively associated with positive score and positively associated with negative score. As the sentiment about a firm decreases the number of independents increases the following year, which could indicate that management may be listening to the media. Again, interpreting coefficients with standard-deviation unit changes yields more interpretable results, due to the small magnitude of estimated coefficients on all the significant media variables. A one-standard-deviation decrease in positive sentiment is associated with a 2-percent increase of percent of independents (.02) increase of percent of independents. 34 A one-standard deviation decrease in Facebook shares is associated with a 4-percent (.04) increase in independents at the firm. Interestingly the sentiment 33 Standard deviation-unit changes are frequently observed in the data. From the data, a one-standard deviation-unit change is observed for positive score 250 times and for negative score 325 times. A one-standard deviation-unit change is observed 80 times for total coverage and 190 times for Facebook shares. 34 Calculated as (e ) *.95 =.02 70

75 variables only become significant after controlling for the frequency of reporting, which indicates that it is important to control for reporting frequency. The theoretical literature suggests that an increase in independents reduces agency costs, increasing shareholder value, although there is mixed empirical evidence to supports this theory. The results in Table 25 indicate that board independence increases following negative sentiment but the reverse occurs following increased shares on Facebook. Other variables that are statistically significantly associated with board independence are Tobin s Q, ROA, duality, board size, and mutual fund ownership. The consistently observed statistical significance of both Tobin s Q and ROA across the models suggests that there may be endogeneity issues. Board independence is significantly and negatively associated with Tobin s Q, and significantly and positively associated with ROA. This suggests that Tobin s Q and ROA capture different aspects of financial performance which affect governance decisions differently, consistent with the rationale for using market and accounting measures of performance. From a market perspective, a Tobin s Q value greater than one indicates the firm may be overvalued. Thus, if independent board members do have a positive effect on the firm, then a negative relationship between Tobin s Q and percent of independents is expected. And, from an accounting perspective, the associated positive sign on ROA is also consistent with the literature. In Table 25, the governance variables duality and board size are statistically significant and positive. Boards with a CEO-chairman duality might be trying to balance what could be seen as conflicts of interest with additional independent board members. 71

76 Similarly, board independence increases board size, which indicates that most firms adding directors are adding independent directors as opposed to insiders. Interestingly, independence is statistically significantly associated with mutual funds ownership, but hedge fund ownership isn t. Results suggest board independence is inversely related to mutual fund ownership. A one-percent (.01) decrease in board independence is associated with an increase of five mutual funds. This could indicate that mutual funds are participating in similar strategies to hedge funds by trying to place leverage on boards. Additionally, the statistical significance of Tobin s Q indicates that endogeneity issues may be present Ownership We now use the count of hedge fund and mutual fund owners in a firm as separate dependent variables, with results for hedge fund ownership reported in Table 26 and mutual fund ownership in Table 27. Although, none of the media variables are statistically significantly associated with hedge fund ownership, coefficient signs on these and other variables are consistent with what the literature would predict. Hedge fund ownership is statistically significantly associated with leverage, board size, RDA, and volatility. Leverage is positively associated with hedge fund ownership, suggesting that hedge funds are interested in investing with firms that are already taking greater risks. Board size is negatively associated with hedge fund ownership, indicating that hedge funds invest in firms with small board sizes. This is consistent with the theory that hedge funds will often try to flip firms by placing their own employees on a board because an individual director would have more voting power on a smaller board, ceteris paribus. The signage on financial performance variables, although not statistically significant, is consistent with the literature. For example, Tobin s Q is positively 72

77 associated with hedge fund ownership and ROA is negatively associated with hedge fund ownership. This is consistent with the idea that hedge funds are interested in investing with firms that are underperformers and could be improved upon. Results indicate that media helps to explain mutual fund ownership. Both Facebook shares and total coverage are both statistically significantly associated with mutual fund ownership. The media variables are jointly significant in the mutual fund estimation. Additionally, the R-squared in the mutual fund model is much higher, 5.5 percent (27.2-percent in Model 10 versus 21.7-percent in Model 2), with the addition of the media variables, suggesting a better model fit. The variable Facebook shares is positively associated with mutual fund ownership, while the variable total coverage is negatively associated with mutual fund ownership. To quantify this result: a one-standard unit increase in Facebook shares is associated with a five-unit increase in mutual fund owners; and a one-standard unit increase in total coverage is associated with a decrease a six-unit decrease in mutual fund owners. The ownership variables capture changes in fund ownership, or exit, of individual firms in their portfolios. This set of results thus indicates that consumerdriven coverage (Facebook shares) results in increased mutual fund interest, but mediadriven coverage (total coverage) encourages mutual fund exit. The informational intermediary literature suggests that the corporate governance role of the media is to provide new information to institutional owners. If this is correct than increases in frequency of coverage should have little effect given that owners are already aware of changes at the firm. The significance of both frequency variables indicates that this isn t what s happening, which suggests there is merit to the 73

78 informational intermediary hypothesis, or there are other unexplored mechanisms by which coverage affects changes in institutional ownership. In the mutual fund model, financial performance variables, governance variables, and capital structure variables are very statistically significant. Mutual fund ownership is negatively associated with Tobin s Q and positively associated with ROA, which is consistent with the literature as well. Where hedge funds seek to invest in firms that may be undervalued by the market, mutual funds generally are more interested in firms that generate consistently positive returns. Mutual fund ownership is also statistically significant and negatively associated with both governance variables, percent of independents and board size. This suggests that mutual funds, similar to hedge funds, are interested in investing with boards that are more insider-dominated and are smaller. The consistent significance of board independence in the mutual fund estimation suggests that endogeneity issues may be present Capital Structure Results for the capital structure model are reported in Table 28. Not surprisingly, the media variables aren t very economically important in this model even as both Facebook shares and total coverage have statistically significant estimated coefficients and are also jointly statistically significant. The addition of the media variables doesn t increase the overall R-squared, or model fit. Leverage is positively associated with Facebook shares and negatively associated with total coverage, which may suggest that lenders are more comfortable with firms that enjoy widespread social awareness (e.g., Facebook coverage) but that too much coverage might be viewed cynically. The other variables in the model are consistent with what the literature would predict: leverage is significantly associated negatively with Tobin s Q, and positively with 74

79 both mutual and hedge fund ownership. Leverage is also significantly associated with board size and firm size. Similar to results from the last few models, the statistically significant association between Tobin s Q, leverage, and hedge fund ownership and leverage indicates that endogeneity issues may be present Media The media models are estimated to inform concerns regarding endogeneity. Research by Dyck et al. (2008) suggests that institutional investors, particularly hedge funds, can affect the frequency of media coverage. Mutual fund ownership is positively associated with total coverage and Facebook shares (Table 29), but hedge fund ownership is negatively associated with Facebook shares (Table 30). Post-estimation results also indicate that institutional investors are important to include in the media models as hedge and mutual fund ownership are jointly significant in both models. The significance of institutional ownership and other endogenous variables in question (financial performance, governance, and leverage), also signals that there are potential unresolved endogeneity issues. Tobin s Q is positively related to total coverage and Facebook shares, although it is only statistically significant across the total coverage models, indicating that overvalued companies might be covered by the media more than undervalued companies. ROA is negative and statistically insignificant across the models, indicating that less profitable companies are covered more by the media. Tobin s Q and ROA are jointly significant in the total coverage model (Table 29), but not in the Facebook shares model (Table 30). This indicates media organizations coverage of particular companies is affected by their past financial performance. However, this same trend is not observed for consumer-driven media coverage, Facebook shares. 75

80 Percent independent is the only variable that is consistently statistically significant in these models. Percent of independents is negative, indicating firms with higher insider ownership are more consistently covered by the media, both intensively and extensively Complete Sample OLS Take-aways The regressions showed three things of importance. First, the models are generally consistent with the literature. This indicates that the statistical significant relationships explored in this section may be causal and not spurious. Second, there are statistically significant relationships between financial performance, mutual funds, percent of independents, and the media. Financial performance is statistically significantly related to both media variables, total coverage and Facebook shares (Tables 23-24). Board independence is also statistically significantly related to the media variables: negative score, positive score, and Facebook shares (Table 25). Finally, mutual fund ownership is strongly affected by Facebook shares and total coverage (Table 27), although hedge fund performance (Table 26) is not associated with any of the media variables. Additionally, the results are also economically important. Third, the statistical significance of commonly endogenous regressors across the models indicates that there might be endogeneity problems. 76

81 5.2 Complete Sample OLS Regressions with Interactions Results from the previous section indicate there is a statistically significant relationship between social media and ownership, and between social media and governance. The next step in this research is to ask how does social media affect financial performance? The complete sample OLS regression tables with interactions, are reported with Tobin s Q as a dependent variable (Table 31), and with ROA as the dependent variable (Table 32). Hedge funds, mutual funds, and percent of independents were chosen as the subjects of interaction terms, because the literature suggests that effects from the media will pass through these channels. Moreover, shareholders or board members are expected to be the ones who might implement governance changes at the firm, presumably affecting subsequent changes in financial performance. The interactions are reported in separate regressions to avoid multicollinearity issues. The first set of regressions in Table 31, Models 1 and 2, compare how interactions between negative score and hedge funds, mutual funds, and percent of independents affects the financial performance of the firm. Model 1, Table 31, is replicated from Table 23. Model 2 indicates that negative sentiment doesn t have a large or statistically significant effect on Tobin s Q. The R-squared in Model 2 isn t significantly different from the R-squared in Model 1, and the interaction terms aren t jointly significant. The second set of regressions in Table 31, Models 3 and 4, compare how interactions between positive score and hedge funds, mutual funds, and percent of independents affects the financial performance of the firm. The third regression is replicated from Table 23. Results from Model 4 indicate that positive score, interacted with mutual, hedge, and percent of independents doesn t have a large or economically important relationship with financial performance. The interaction terms are not 77

82 statistically significant and or jointly significant. There isn t strong statistical evidence that sentiment of reporting affects the financial performance of the firm through either channel: ownership or changes in governance. The next set of models in Table 31, Model 5 through model 7, considers how frequency of coverage, total coverage and Facebook shares, effects the financial performance of the firm. Model 5 is replicated from Table 23. Model 6 and 7 build off Model 5, by considering how hedge funds, mutual funds, and percent of independents interact with frequency of reporting or sharing. Model 6 looks at the relationship between financial performance, ownership, governance, and Facebook shares. Results from Model 6 show that there is a statistically significant relationship between Facebook shares and percent of independents. A one-unit standard-deviation increase in Facebook shares, ceteris paribus, results in a unit increase in the percent of independents. Ceteris paribus, a one-standard-deviation increase in Facebook shares and a one-unit increase in the percent of independents results in a 3.31-unit increase 36 in Tobin s Q. Model 7 interacts ownership and governance variables with total coverage. However, none of the interaction terms are statistically significant and the interaction terms aren t jointly significant, indicating they don t explain additional variation in Tobin s Q. Results from Models 5 through 7 suggest that there is a statistically significant relationship between financial performance, governance, and consumer-driven media coverage. An examination of how ROA is affected by the direct and indirect impact of media coverage is reported in Table 32. Similar to the results with dependent variable Tobin s Q, neither the positive or negative sentiment scores, interacted with percent of 35 (e ) * 3.11 = The 3.11 (a one-standard deviation unit change from the mean in levels) comes from an earlier calculation in section Financial Performance =

83 independents, mutual funds, or hedge fnds is highly statistically significant. Additionally, none of the interaction terms are jointly significant. This indicates there isn t a strong statistical relationship between ROA, sentiment, and governance or shareholders. The results shown in Table 31 and 32 indicate that the sentiment variables, positive score and negative score, don t have much effect on the financial performance of the firm. There are two reasons why this might be the case. First, the literature suggests that sentiment score may not affect hedge or mutual fund ownership because hedge funds and mutual funds are already informed outside of the mainstream media. A second reason why sentiment scores aren t statistically significant, is because the scores are still a rough estimate of sentiment. Sentiment was constructed using the Loughran and McDonald financial dictionary. The number of negative words and number of positive words were counted in each article subtracted to obtain an overall sentiment score. All the words are given the same weight in the score, such that a firm being mentioned profitable once in an article receives the same score as a firm which is mentioned as satisfactory once, as both words are coded as positive financial words. Additionally, interactions between frequency variables, total coverage and Facebook shares, aren t statistically significant or jointly significant in Model 6 or 7. In general, the media variables seem to fit the model with Tobin s Q as an indicator of financial performance better than the model with ROA as an indicator financial performance. However, media variables are jointly significant with both variables of financial performance without the interaction terms (Table 23 and Table 24). This indicates that there is merit to consider governance effects from media on financial performance. In additional to examining the channels, governance and ownership, but which media affects the financial performance of the firm. 79

84 5.3 Segmented Sample OLS Regressions The first goal of this line of inquiry is to investigate whether there is evidence of media bias across the segmentations. This goal is analyzed visually, descriptively, and then through regression analysis. 37 Figure 2 shows the overlap of media markets comparing the unique Twitter users. Although there is significant overlap in users who follow multiple media sites, the Venn diagrams also show that the segments cater to different populations. Figure 4 and Figure 6 show that sentiment score distributions and daily posts do not differ drastically by segment. However, from Figure 8 there appear to be substantial differences between segments among shares per post; Figure 8 shows right-leaning news is shared more per post than the other groups. To test if differences among the segments are statistically significant, ANOVA 38 and t-tests of the means are performed on the media variables: positive score, negative score, total coverage, and Facebook shares. Results are reported in Table 16. In only two cases 39 is the null hypothesis of equal means rejected at a.01 percent significance level, indicating that media segments report differently in terms of sentiment and frequency. Chi-square tests were performed to see if there are significant differences in industry coverage between two or more groups (Table 17). In only one instance, the mainstream vs. non-mainstream media, that the null hypothesis of no difference is rejected. This indicates that all other media segments report on various industries with similar frequencies. The Chi-square results suggests that differences in regression 37 The political leaning regressions are dropped due to limited number of observations in the right-leaning group which could bias estimates. However, the exploratory analyses are kept, as degrees of freedom aren t an issue. This indicates that further research should be undertaken to continue this area of inquiry. 38 An ANOVA is used to compare significant differences in the means when there are more than two groups. In this case the ANOVA was used to test the political leaning segment, which has three groups: left-leaning, right-leaning, and center. 39 The two cases are: negative score tested against financial vs. non-financial media, and total coverage tested against paywall vs. non-paywall media. 80

85 coefficients may stem from differences in sentiment and reporting frequency, not from the type of firm reported on. Results from the t-tests and ANOVA motivate regression analysis, to see if differences in sentiment of reporting or frequency of reporting change coefficient values. Estimation results for the segmented regressions are reported in Table 33 through Table 35. Regression results show that differences across the media variables don t appear to affect coefficients or significance much, which helps to eliminate questions about how biases might affect results. However, the one limitation of this analysis is that coefficient differences cannot be tested empirically. 41 The second goal of this line of inquiry is to help answer the question of which media accountability hypothesis, informational intermediary or reputational capital is more reputable. Results from the complete sample OLS showed that media variables, positive score, negative score, and Facebook shares are statistically significantly associated with future percent of independents (Table 25). However, it s unknown whether changes in governance are driven by changes in reputational capital or engagement by institutional investors. The segmented regressions support the reputational capital hypothesis because the model fit on percent of independents is higher for mainstream media, non-financial media, and non-paywall media. Results from Table 33, which examines differences between financial and non-financial media, indicate that variation in board composition is better explained by non-financial media than financial media (Table 33). In Model 6, the R-squared shows that 25.1 percent of the variation in percent of independents is 41 The empirical way to test coefficient differences is to include interactions. However, as mentioned in the section Segmented Sample OLS regressions, the inclusion of too many interactions results in rank issues. If interactions were included, coefficient estimates could be tested using the Chow test of F-test. 81

86 explained by non-financial media, while only 19.6 percent of the variation in board composition is explained by financial media (Model 5). The media variables are jointly significant in both cases. Results from mainstream media follow a similar pattern. Mainstream media results are reported in Table 34. In Model 6, the R-squared indicates that 27 percent of the variation in board composition is explained by mainstream media, while in Model 5, only 19.3 percent of the variation of independents on the board is explained by nonmainstream media. The media variables are jointly significant in both cases. However, the media variables from the paywall regressions are only jointly significant in the paywall estimation of percent of independents. This suggests that if the reputational capital theory is valid, managers and directors, pay more attention to media that has a paywall. This result doesn t contradict the theory that reputational capital is more persuasive than the informational intermediary theory, but it also doesn t provide evidence in support of the reputational capital theory. Overall, results suggest that the reputational costs hypothesis is more impactful than the information intermediary theory. Additionally, the differences in R-squared don t appear to be driven by differences in the types of firms. The Chi-square test shows that in all cases, expect mainstream media, the type of industries covered didn t vary by segment (Table 17) Financial vs. Non-Financial Regressions Summary statistic tables for financial versus non-financial variables are reported in Table 12. Overall, the summary statistics are very similar between the two. There are slight differences within the media variables. Financial media has a slightly wider positive score range than non-financial media and the average number of Facebook shares is 82

87 larger for financial media than non-financial media. The summary statistics indicate that the population of firms that financial media report similarly to non-financial media, which is supported by the Chi-square test. This indicates that differences between the two populations may be due to differences in how financial media vs. non-financial report, in terms of sentiment or frequency. Financial vs. non-financial regressions are in Table 33. Overall, the financial models don t differ drastically from the non-financial models, although there are small changes. One interesting difference between the two models are the coefficients differences of the media variables in the hedge fund model, Models 9 and 10. In the hedge fund model, positive and negative scores become statistically significant for financial media, but stay statistically insignificant for non-financial data. Interestingly, positive score is negatively associated with hedge fund ownership and negative score is positively associated with hedge fund ownership. This suggests that hedge funds are more interested in investing with firms that are reported on negatively in the media, which makes sense given the hedge fund business model. The variables are statistically significant, but the coefficients are still small. A one-standard deviation unit increase in positivity about the firm is associated with a decrease of.22 hedge funds. Other differences between the models show up in Models 15 and 16, with the dependent variable Facebook shares. The variable Facebook shares is positively associated with Tobin s Q and negatively associated with ROA. These signs are consistent with the LSDV model in Table 30, but the variables are much more statistically significant and the magnitudes of the coefficients have increased. Additionally, ROA is only statistically significant in the financial media model, indicating that a one-unit increase in ROA (.01) is associated with a decrease in 2-percent of 83

88 Facebook shares. This indicates that Facebook users are less likely to share news stories about firms that perform better. Percent of independents is also statistically significant in the financial media model (Model 15), consistent with the earlier LSDV model shown in Table 30. This indicates that firms with a lower number of independents are reported on more by the financial media than by the non-financial media. Firms with greater research and investment are also covered the financial media more than the non-financial media. The R-squared is much higher for the financial media than for the non-financial media. The R-squared from the LSDV model in Table 30 is about 14-percent, but the R-squared jumps to 27-percent when only considering financial media. These results indicate that the information from financial media organizations is shared more when firms aren t performing as well and if they have fewer independents Mainstream vs. Non-Mainstream Regression Tables Summary statistics for mainstream media vs. non-mainstream media are very similar in general (Table 13) but small differences are observed within the media variables. Positive score has a wider range in the non-mainstream media than in the mainstream media, and total coverage is on average higher for mainstream media than for non-mainstream media. The estimation results for the mainstream and non-mainstream media are very similar across the board. The biggest difference that stands out is that total coverage is positive and statistically significantly associated with Tobin s Q. However, this relationship is only statistically significant for mainstream media, which seems appropriate. The R-squared increased from 34-percent in Table 23 to 40-percent in the 84

89 mainstream model in Table 34. The ROA models, Models 3 and 4, are very similar to each other and to the complete sample model shown earlier in Table 24. The results for board independence are very similar when comparing the mainstream model to the non-mainstream model. There are small changes in standard error, and thereby the significance of the different variables, but the signs and magnitudes of estimated coefficients are very similar. The model with mutual fund ownership has interesting results; the R-squared is much higher for mainstream media, 36 percent, than for non-mainstream media, 27 percent. This makes sense as it s likely that there is more variation from the nonmainstream media than the mainstream media. Additionally, there are significant differences in the coefficients between the two groups. The coefficients of Tobin s Q, ROA and board size for mainstream media are double the size of non-mainstream media. Estimates from Model 7, the mainstream model, indicate a one-unit increase in Tobin s Q is associated with a 5-unit decrease in future mutual fund ownership, which makes sense, because firms with larger Tobin s Q are overvalued by the market. A one-unit increase in ROA (.01) is associated with an increase in mutual fund ownership of.7 mutual funds, by Model 7. Board size is also statistically significantly associated with mutual fund owners and has a much larger coefficient magnitude for the mainstream model than the nonmainstream model. A one-percent increase in board size is associated with a.16-unit decrease of mutual fund ownership. Additionally, total coverage is negative and statistically significantly associated with mutual fund ownership in the mainstream media model, but not the nonmainstream model. This suggests that increased coverage is associated with a decrease in 85

90 future mutual fund ownership when the mainstream media reports, but not when more niche media organizations cover a firm. ROA is negative and statistically significantly associated with total coverage, indicating firms with worse financial performance are covered more by the media. It s consistent that mutual fund ownership would be negatively associated with coverage if firms that are being covered perform less. The hedge fund models, Models 9 and 10, and leverage models, Models 11 and 12, are very similar for the two groups. While there are small differences in magnitude of estimated coefficients, most of the significance differences are due to changes in standard errors. The last four models, the media models, exhibited very similar behavior when comparing the mainstream model with the non-mainstream model. The total coverage models, 13 and 14, had small differences in RDA and firm size (property, plant, and equipment). The Facebook shares models, Models 15 and 16, looked very similar with small changes in significance due to differences in the standard error. Overall, the mainstream models looked very similar to the non-mainstream models, which a few small differences. This suggests that the types of firms and reporting methods are similar between the two groups Paywall vs. No-Paywall Regressions Summary statistics for paywall vs. no-paywall are reported in Table 14. Overall, the summary statistics are very similar for paywall media versus non-paywall media. There is a small difference in the number of Facebook shares as non-paywall media is shared slightly more often than paywall media. Moreover, the summary statistics are consistent with the Chi-square test results shown earlier in Table 17, which indicate that the distribution of industries is similar in paywall media and non-paywall media. Thus, 86

91 differences in regression results may be driven by differences in media sentiment or frequency rather than by the types of firms those media sources cover. Overall the regressions looked similar, between paywall and non-paywall media. In the model where Tobin s Q is the dependent variable, there are a few small changes within the media variables, but nothing large. Total coverage, Facebook shares, and percent of independents are statistically significant in the non-paywall media regression, Model 2, but not statistically significant in the paywall media regression. However, the paywall media regression fits the data slightly better because it has a higher R-squared. The ROA model looks very similar across models, with very small variation in magnitude of the estimated coefficients or significance. The governance model looks very similar, with small differences in the media variables. However, the economic significance from the media variables is marginal. The R-squared is higher for the non-paywall media which indicates that the variation in percent of independents is better explained by non-paywall media than paywall media, however the media variables are not jointly significant in the non-paywall media. The mutual fund regressions also appear to be very similar. There are small changes in magnitude of the coefficients between the two models. Most of the variables have relatively similar coefficient magnitudes and significance levels. Additionally, the R- squared is very similar across the two models and similar to the complete sample OLS model. The mutual funds results don t reveal strong evidence in favor one media accountability theory over another. The hedge fund models also look quite similar across models, with small changes on variables like board size. Similar to the mutual fund media regressions, results from 87

92 the hedge fund model don t provide strong evidence in support of one theory over another. The last two regressions of interest are the regressions with media coverage as a dependent variable. Segmenting the regressions by paywall significantly increases the R- squared in both cases. In the lagged, pooled regression, the R-squared is 13.3-percent and 13.9-percent, for total coverage and Facebook shares, respectively. In the new segmented model, the R-squared jumps to 38.5-percent and 24.2-percent for the total coverage paywall media model and Facebook shares media model, respectively. This suggests that more variation in total coverage and Facebook shares is explained when segmenting along paywall vs. non-paywall data. Within these results, the estimations do not differ a lot. Negative score is statistically significant in the Facebook shares model for paywall media, but not in the non-paywall media model. This indicates negative sentiment predicts Facebook shares, but only for companies covered by the paywall media. Overall, the results from the paywall versus non-paywall media look very similar. The Chi-square results in Table 17 indicate that the two types of media cover a similar distribution of industries. Although the t-tests shown in Table 16 indicate paywall companies report differently than non-paywall companies, regression results (Table 35) show that these differences do not significantly alter coefficients. Additionally, there is little evidence to support one media accountability theory over the other, given the similarity between models and changing significance of media variables across models. 88

93 5.4 Simultaneous with Instrumental Variable Regressions Instrumental variable tests, the F-test 42, Hausman test 43, Sanderson-Windmeijer Chi-square test 44, and Sanderson-Windmeijer F-test 45 are reported in Table 21 and Table 22. Results show that the instruments are jointly significant, by the F-test, identified by the Sanderson-Windmeijer Chi-square test, and efficient by the Hausman test. The Sanderson-Windmeijer F-test shows that some of the variables are weakly identified, using a 10-percent significance level of Regression tables are structured as such: Table 36 compares the LSDV estimation results with the instrumental variable estimation results from both the 2SLS and the 3SLS. Overall, the results were similar to the LSDV variable results with some changes in coefficient magnitude and significance, but fewer changes in sign. Consistent with the Bhagat and Bolton (2008) results, after instrumentation most coefficient magnitudes increased compared to the OLS model. A detailed discussion of the results follows Financial Performance Tobin s Q Results indicate that after instrumenting total coverage is positively associated with Tobin s Q in the LSDV model, but is negatively associated with Tobin s Q in both the 2SLS and 3SLS model. None of the other media variables are statistically significantly associated with Tobin s Q in the LSDV models or simultaneous models. The expected 42 The F-test measures the relevance of excluded instruments in the first-stage regression by testing for joint significance of all instruments 43 The Hausman specification test compares the efficiency of estimators. 44 The Sanderson-Windmeijer (SW)Chi-square test tests for under-identification. The null hypothesis of the SW Chisquare test is that the model is under-identified. 45 The Sanderson-Windmeijer (SW) Chi-square test tests for weak instruments. It s standard to use a significance of 10-percent (Baum 2007). 89

94 coefficient on total coverage could have been positive or negative, as it s hypothesized that firms which perform better or worse than average will be covered more by the media. The results indicate that after demeaning, industry effects are removed, there is a negative and statistically significant relationship between total coverage and Tobin s Q. The IV variables suggest that a 10-percent increase in coverage is associated with a.005 decrease in Tobin s Q. This result is consistent with the signage in the model with ROA as the dependent variable. Other variables in the Tobin s Q model also become statistically significant after instrumenting. The governance variables are statistically significant after instrumenting, which is consistent with the literature (e.g., Hermalin & Weisbach 2003). Tobin s Q is positively associated with duality, and negatively associated with board size and percent of independents in the IV models (but not in the LSDV models). The theoretical literature indicates that financial performance should be negatively associated with duality and that positively associated with percent of independents, however empirically there has been little evidence to support either of these two hypotheses. Neither of the ownership variables are statistically significant, which is consistent with the LSDV model, although the magnitudes of the coefficients increase significantly after instrumenting ROA Results indicate that the media variables aren t very statistically significant in the ROA models, which is consistent with the complete sample OLS regressions. However, total coverage is statistically significant and negatively associated with ROA. This suggests that perhaps firms that are being covered by the media result in worse future 90

95 financial performance. Similar to the LSDV model, frequency of coverage is statistically significantly associated with financial performance, which isn t predicted by the literature. Results are different than LSDV and the changes are predicted by the literature after controlling endogeneity. Duality, board size, RDA, mutual fund ownership, and total coverage are highly statistically significant in the IV models. In the complete sample OLS regressions the governance variables often aren t statistically significant. Hermalin and Weisbach (2003) suggest that this may be due to endogeneity issues. The results in the simultaneous regression equations are consistent with this hypothesis. Percent of independents and board size are negatively associated with ROA, while duality is positively associated with ROA. The negative association between board size and ROA is expected by the literature, as efficiency costs from larger boards outweigh additional monitoring benefits. The theoretical literature predicts a positive association between percent of independents and a negative association between ROA and duality, as this signals increased monitoring and decreased agency costs. Both signs in the models contradict the theoretical literature, but the theoretical literature is robustly supported by the empirical literature. The other variable of significance in the ROA model is mutual funds.the coefficient on mutual funds is economically important. Results suggest an increase in ten mutual funds results in a 12-percent decrease in ROA in the 2SLS model, and a 9- percent decrease in ROA in the 3SLS model. The sign is not predicted by the literature, which expects a positive association, due to higher monitoring and engagement, between financial performance and number of mutual funds. 91

96 5.4.2 Governance In general, results from the instrumental regressions look similar to results from the least squares dummy variables regression, however the magnitudes on many of the variables have increase after instrumenting, as have the standard errors. The one group of variables, where this is a fair amount of change is the media variables. In the LSDV model, positive score, negative score, and Facebook shares are statistically significant. However, after instrumenting, only Facebook shares remains significant and total coverage becomes significant. The signs of the sentiment variables remain after instrumenting, but the standard errors increase, resulting in positive score and negative score losing their significance. After instrumenting, Facebook shares is positively associated with percent of independents, and total coverage is negatively associated with percent of independents. This suggests that media-driven content, total coverage, is negatively associated with the future percent of independents on the board, while consumer-driven content, Facebook shares, is positively associated with percent of independents on the board. Tobin s Q, duality, RDA, and mutual fund ownership are also statistically significant across the 2SLS and 3SLS models. Tobin s Q is negatively associated with percent of independents, which is consistent with the LSDV model. ROA is positively associated with percent of independents, but isn t statistically significantly associated with percent of independents, because of increase standard errors in the instrumented estimations. Duality is positively associated with percent of independents, suggesting that the presence of a dual chairman-ceo increases the number of independents at the firm in the future year, which is consistent with results from the LSDV model. Mutual fund ownership is negatively associated with percent of independents, which is the same 92

97 association observed in the LSDV model, but the effect is larger in the instrumented models. A ten-unit increase in mutual funds is associated with a decrease in future independents of 8-percent decrease in the 2SLS model, and a 6-percent decrease in the 3SLS model Ownership Hedge Funds Results from the instrumented hedge fund model indicate that none of the media variables are predictive of future hedge fund ownership, which is consistent with the results from the OLS models in Table 26. All the variables that were statistically significant in the LSDV estimation have consistent signage across the models. Leverage, board size, volatility, and RDA are statistically significantly associated with future hedge fund ownership in the LSDV model. However, only board size remains statistically significant after instrumenting, although the coefficient directionality of these variables remains. Similar to previous results, the coefficients in the IV model increased significantly in magnitude compared to the LSDV estimation coefficients. After instrumenting, percent of independents becomes negatively and statistically significantly associated with future hedge fund ownership. This is consistent with what the literature would predict. A 20-percent change in percent of independence is associated with a 1-unit decrease in the number of hedge funds at the firm. Board size is also statistically significantly and negatively associated with hedge fund ownership. This indicates that an increase in the size of boards discourages hedge fund ownership, which is consistent with what the literature would predict. Interestingly, 93

98 neither indicator of financial performance is statistically significantly associated with hedge fund ownership, which isn t predicted by the literature Mutual Funds Results from the instrumental variable model indicates that media variable Facebook shares is negatively and statistically associated with future mutual fund ownership, which is different than the LSDV model. This suggests that an increase in consumer-driven media content is negatively associated with future mutual fund ownership of the company. Additionally, total coverage isn t statistically significantly associated with mutual fund ownership after instrumenting. Like many of the other regressions, the IV results for the other variables are similar to the results from the LSDV with the major differences being changes in magnitude and significance of estimated coefficients. Leverage, percent of independents, total coverage, and board size are all statistically significant in both the LSDV model and the IV model. Tobin s Q and ROA are statistically significantly associated with mutual fund ownership in the LSDV model, but not in the IV model. Additionally, leverage, percent of independents, and board size change significantly in magnitude. Leverage increases by a factor of five in the 3SLS estimation compared to the LSDV estimation. By IV estimation, a 1-percent unit increase in leverage (.01 increase) is associated with a.65 unit increase in mutual fund ownership. Percent of independents is negatively associated with future mutual fund ownership, such that.01 unit increase in percent of independents (a 1-percent increase) is associated with a decrease in future mutual fund ownership by.854., which is consistent with results from the LSDV model. Board size is also associated with future mutual fund ownership. A 10-percent increase in board ownership is associated with a decrease in mutual fund ownership of.072. Total 94

99 coverage is also statistically significantly associated with mutual fund ownership. A 10- percent increase in coverage is associated with a.023 unit decrease in future mutual fund ownership Capital Structure The results show that the media variables are not statistically significantly related with leverage, which is what is expected from the literature. In the LSDV model, hedge fund ownership, mutual fund ownership, Facebook shares, Tobin s Q, and board size are all statistically significant that the 95-percent level. After instrumenting, mutual funds and ROA are statistically significant. ROA is positively associated with leverage in the 2SLS and 3SLS model, but negatively associated with leverage in the LSDV model. The opposing signs on Tobin s Q and ROA are consistent with other results, and after instrumenting the signs are consistent with what the literature would expect. A ten-percent decrease in ROA is associated with a.03-unit decrease in leverage Media Facebook Shares In the IV model, hedge fund ownership and firm size (property, plant, and equipment), RDA, and percent of independents are statistically significantly associated with future Facebook shares. In the LSDV model, hedge fund ownership, mutual fund ownership, percent of independents, positive score, and property, plant, and equipment (ppe), statistically significantly predict future Facebook shares. Signage is consistent for the variables that are statistically significant across the models. Consistent with the literature the number of hedge funds and mutual funds are statistically significantly associated with coverage. The signs differ from the Dyck et al. (2008) study, but Dyck et 95

100 al. looked at Russia and a specific hedge fund campaign, which suggests their results may be more anecdotal. Additionally, it s possible that hedge fund involvement can increase specific types of media coverage while decreasing overall coverage about the firm. Past size of the firm, measured by log(ppe) and RDA are also both positively associated with Facebook shares. This is unsurprising, as consumers are more likely to share articles about firms they know more about, which is likely to be correlated with size of the firm and how much advertising (RDA) the firm buys Total Coverage Results show that similar to the regression with Facebook shares as the dependent variable, hedge fund ownership, mutual fund ownership, RDA, and firm size (property, plant, and equipment) are statistically significantly associated with future coverage of the firm. The signs are consistent across the variables in the LSDV and IV models. Interestingly, variables such as ROA becomes statistically insignificant after instrumentation. Tobin s Q is statistically significantly and negatively associated with total coverage. It s surprising that financial performance variables aren t consistently significant across the IV models. The variables used to instrument for Tobin s Q and ROA could be stronger, as indicated in Table 21, which is one potential reason why financial performance isn t consistently statistically significant across the models. Hedge fund ownership is also negatively associated with total coverage. This suggests that hedge funds may invest with firms that don t have large media presences. Mutual fund ownership however, is positively associated with total coverage, suggesting that mutual funds invest with firms that have a larger media presence. 96

101 6 Conclusion Social media platforms have changed how Americans receive news and information. The literature suggests that the media plays a corporate governance role as an investigative information intermediary and by amplifying information about the firm. Empirical results show evidence in support of the hypothesis that social media plays a corporate governance role. Results from Table 25 and Table 27 suggest there is a statistically significant relationship between governance and social media, and mutual fund ownership and social media. Negative sentiment is associated with an increase in future board independence and increases in consumer-driven content, Facebook shares, results in a decrease in future board independence (Table 25). Consumer-driven media coverage, is associated with an increase in the number of mutual funds at the firm, indicating that consumers perhaps do vote with their feet. Companies that are talked about more on social media, encourage investment from mutual funds. Results from Table 31 indicate that there is a statistically significant relationship between governance, social media, and future financial performance. Results from Table 33 - Table 35 indicate that results are robust to media biases and that the reputational capital hypothesis is more persuasive than the informational intermediary hypotheses. Results from the simultaneous models suggest that media frequency variables, total coverage and Facebook shares, are important predictors of future board independence. Facebook shares is positively associated with percent of independents, indicating that consumer-driven content increases future independence on boards, but total coverage decreases future board independence. Facebook shares is negatively and statistically significantly associated with mutual fund ownership, indicating that companies that are talked about more on social media, discourage investment. 97

102 The results indicate that consumer-driven content through engagement and media coverage are important to the governance of the firm. The effects are relatively small, however, so there have to be fairly large changes in coverage or shares to have an economically important impact. The results also indicate the media is perhaps more important as a mechanism for creating reputational costs, rather than as an informant to shareholders. This thesis also leaves much room for further research. First, interactions testing the relationship between the effects of social media on governance and ownership, and that effect on the financial performance were not estimated in the instrumental regressions. Additional variables needed to be identified as potential instruments. Second, this research explores only two of at least three channels by which social media could affect firm performance and governance due to data limitations. Third, this research only looks at the effects of social media on one governance variable, percent of directors who are independent. However, there are other indicators and definitions of board independence that could be used. Fourth, if current trends are an indicator of future trends, social media will continue to be an important source of information to many consumers. The relative newness of social media data and interactive features offers unique, pioneering opportunities to study social behavior. 98

103 References ICI., Investment Company Fact Book: A Review of Trends and Activities in the U.S. Investment Company Industry. Barber, B.M., Monitoring the Monitor: Evaluating CalPER's Activism The Journal of Investing 16, Baum, C.F., Instrumental Variables Estimation in Stata. Baum, C.F., Schaffer, M.E., ivreg2: Stata module for extended instrumental variables/2sls, Gmm and AC/HAC, LIML and k-class regressoin. URL Bebchuk, L.A., Cohen, A., Ferrel, A., What Matters in Corporate Governance? Review of Financial Studies 22, Bebchuk, L.A., Weisbach, M.S., The State of Corporate Governance Research. Review of Financial Studies 23, Berle, A.A., Means, G.C., The Modern Corporation and Private Property. Bhagat, S., Bolton, B., Corporate Governance and Financial Performance Journal of Corporate Finance 14, Bhagat, S., Bolton, B., Romano, R., The Promise and Peril of Corporate Governance Indices. Columbia Law Review 108, Bhagat, S., Jefferis, J., Richard H., The Econometrics of Corporate Governance Studies. The MIT Press, Cambridge, Massachusettes. Boone, A.L., Field, L.C., Karpoff, M.J., Raheja, C.G., The determinants of corporate board size and composition: An empirical analysis. Journal of Financial Economics 85, Brav, A., Jiang, W., Partnoy, F., Thomas, R., Hedge Fund Activism, Corporate Governance, and Firm Performance. The Journal of Finance 63, Brickley, J.A., Coles, J.L., Jarrell, G., Leadership structure: Separating the CEO and the Chairman of the Board. Journal of Corporate Finance 3, Brickley, J.A., Linck, J.S., Coles, J.L., What happens to CEOs after they retire? New evidence on career concerns, horizon problems, and CEO incentives. Journal of Financial Economics 52, Bushee, B.J., Core, J.E., Guay, W., Hamm, S.J.W., The Role of the Business Press as an Informational Intermediary. Journal of Accounting Research 48, Çelik, S., Isaksson, M., Institutional Investors as Owners: Who Are They and What Do They Do? OECD Corporate Governance Working Papers 11. Çelik, S., Isaksson, M., Institutional Investors as Owners - Who are They and What Do They Do?. Document Slide. Chang, X., Dasgupta, S., Hilary, G., Analyst Coverage and Financing Decisions. The Journal of Finance 61, Cho, M.-H., Ownership structure, investment, and the corporate value: An empirical analysis. Journal of Financial Economics 47, Coffee, J.C.J., Palia, D., The Wolf at the Door: The Impact of Hedge Fund Activism on Corporate Governance Annals of Corporate Governance 1, 1-94 Coles, J.L., Daniel, D.N., Naveen, L., Co-opted Boards. The Review of Financial Studies 27,

104 Cornett, M.M., Marcus, A.J., Saunders, A., Tehranian, H., The impact of institutional ownership on corporate operating performance. Journal of Banking and Finance 31, Del Guercio, D., Seery, L., Woidtke, T., Do boards pay attention when institutional activists "just vote no"? Journal of Financial Economics 90, Djankov, S., McLiesh, C., Nenova, T., Shleifer, A., Who Owns the Media? Journal of Law and Economics 46, Dyck, A., Volchkova, N., Zingales, L., The Corporate Governance Role of the Media: Evidence from Russia. The Journal of Finance 46, Dyck, A., Zingales, L., The Corporate Governance Role of the Media. The World Bank, Washington DC. Dyck, A., Zingales, L., The Media and Asset Prices. Working Paper. Edmans, A., Blockholder Trading, Market Efficiency, and Managerial Myopia. The Journal of Finance 64, Edmans, A., Holderness, C.G., Blockholders: A Survey of Theory and Evidence. Working Paper. Eisenhardt, K.M., Agency Theory: An Assessment and Review. The Academy of Management Review 14, Elyasiani, E., Jia, J., Distribution of institutional ownership and corporate firm performance. Journal of Banking and Finance 34, Ember, S., The New York Times Co. Reports Loss as Digital Subscriptions Grow. In: The New York Times. The New York Times. Fama, E.F., Agency Problems and the Theory of the Firm. The Journal of Political Economy 88, Fama, E.F., Jensen, M.C., Separation of Ownership and Control. Journal of Law and Economics 26, Fang, L., Peress, J., Media Coverage and the Cross-section of Stock Returns. The Journal of Finance 64, Farrell, K.A., Whidbee, D.A., Monitoring by the Financial Press and Forced CEO Turnover. Journal of Banking and Finance 12, Garcia-Castro, R., Arino, M.A., Canela, M.A., Does Social Performance Really lead to Financial Performance? Accounting for Endogeneity. Journal of Business Ethics 92, Gompers, P., Ishii, J., Metrick, A., Corporate Governance and Equity Prices. The Quarterly Journal of Economics 118, Gormley, T.A., Matsa, D.A., Common Errors: How to (and Not to) Control for Unobserved Heterogeneity. Review of Financial Studies 27, Gottfried, J., Shearer, E., News Use Across Social Media Platforms Greenwood, S., Perrin, A., Duggan, M., Social Media Update In: Pew Research Center. Grossman, S.J., Hart, O.D., Take-Over Bids, the Free-Rider Problem, and the Theory of the Corporation. The Bell Journal of Economics 11, Hart, O., Corporate Governance: Some Theory and Implications. The Economic Journal 105, He, J.J., Tian, X., The dark side of analysts coverage: The case of innovation. Journal of Financial Economics 109,

105 Hermalin, B.E., Weisbach, M.S., Boards of Directors as an Endogenously Determined Institution: A Survey of the Economic Literature. Economic Policy Review 9, Hirschman, A., Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States. Harvard University Press, Cambridge, MA. Holderness, C.G., The Myth of Diffuse Ownership in the United States. Review of Financial Studies 22, Jensen, M.C., Meckling, W.H., Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure. Journal of Financial Economics 3, Jensen, M.C., Murphy, K.J., Performance Pay and Top-Management Incentives. The University of Chicago Press 98, Joe, J.R., Louis, H., Robinson, D., Managers' and Investors' Responses to Media Exposure on Board Ineffectiveness. Journal of Financial and Quantitative Analysis 44, Karpoff, M.J., Malatesta, P.H., Walkling, R.A., Corporate governance and shareholder initiatives: Empirical evidence. Journal of Financial Economics 42, Klein, A., Zur, E., Entrepreneurial Shareholder Activism: Hedge Funds and Other Private Investors. The Journal of Finance 64, Knyazeva, A., Knyazena, D., Masulis, R.W., The Supply of Corporate Directors and Board Independence. The Review of Financial Studies 26, Loughran, T., McDonald, B., When is a Liability Not a Liability? Textual Analysis, Dictionaries, and 10-Ks. The Journal of Finance 66, Lowenstein, L., Financial Transparency and Corporate Governance: You Manage What You Measure. Chicago Law Review 96, Lowenstein, L., Corporate Governance and the Voice of the Paparazzi. The Center for Law and Economics Studies Working Paper No Lui, B., McConnell, J.J., The Role of the Media in Corporate Governance: Do the Media Influence Managers' Capital Allocation Decisions. The Journal of Financial Economics 110, Manne, H.G., Mergers and the Market for Corporate Control The Journal of Political Economy 73, McCahery, J.A., Sautner, Z., Starks, L.T., Behind the Scenes: The Corporate Governance Preferences of Institutional Investors. The Journal of Finance 71, Miller, G.S., The Press as a Watchdog for Accounting Fraud. Journal of Accounting Research 44, Mitchell, A., Gottfried, J., Kiley, J., Matsa, K.E., Political Polarization & Media Habits. In: Pew Research Center. Morck, R., Shleifer, A., Vishny, R.W., Management Ownership and Market Valuation: An Empirical Analysis. Journal of Financial Economics 20, Mueller, D.C., A Life Cycle Theory of the Firm Journal of Industrial Economics 20, Murphy, K.J., Zábojník, J., Managerial Capital and the Market for CEOs. Working Paper. Ross, S.A., Westerfield, R.W., Jaffe, J., Corporate Finance. In:McGraw-Hill Irwin, New York, NY. 101

106 Rubin, A., Smith, D.R., Institutional ownership, volatility and dividends. Journal of Banking and Finance 33, Shleifer, A., Vishny, R.W., Large Shareholders and Corporate Control. Journal of Political Economy 94, Smith, M.P., Shareholder Activism by Institutional Investors: Evidence from CalPERS. 51 1, Snyder, J.M.J., Stromberg, D., Press Coverage and Political Accountability. The Journal of Political Economy 118, Stynes, T., Circulation Up at Journal, Times - Rise Reflects Gains in Digital Subscribres; Readers for Top U.S. Newspapers Decline Overall. In: The Wall Street Jounral. Tetlock, P.C., Tsechansky, M.S., MacKassy, S., More Than Words: Quantifying Language to Measure Firms Fundamentals. The Journal of Finance 63, Tricker, B., The Role of the Institutional Investor in Corporate Governance. Corporate Governance 6, Vega, C., Stock Price Reaction to Public and Private Information. Journal of Financial Economics 82, Weisbach, M.S., Outside Directors and CEO Turnover. Journal of Financial Economics 20, Williamson, O.E., Markets and Hierarchies: Some Elementary Considerations. The American Economic Review 63, Wooldridge, J.M., Econometric Analysis of Cross Section and Panel Data (2). MIT Press, Cambridge, US. Wooldridge, J.M., Introductory Econometrics: A Modern Approach. South- Western Cengage Learning, Mason, OH. Yu, F.F., Journal of Financial Economics. Journal of Financial Economics 88,

107 Appendix A: Summary and Regression Tables 46 Exploratory and Descriptive Tables Table 1: Media following on Facebook and Twitter Table 1: Political media bias Table 3: Twitter versus Facebook users Table 4: News category Table 5: Variable definitions Table 6: Industry distribution Table 7: Summary statistics from Compustat firm universe vs. sample Table 8: Financial vs. non-financial news sources Table 9: Mainstream vs. non-mainstream news sources Table 10: Paywall vs. non-paywall news sources Table 11: Political leanings of news sources Table 12: Financial vs. non-financial summary statistics Table 13: Mainstream vs. non-mainstream summary statistics Table 14: Paywall vs. non-paywall summary statistics Table 15: Political leaning summary statistics Table 16: T-tests and ANOVA Table 17: Chi-square tests Table 18: Instrumental variable descriptions Table 19: Instrumental variables summary statistics Table 20: Instrumental variable regressed on endogenous dependent variable Table 21: IV tests for media models Table 22: Hausman test for media models Complete Sample OLS Regressions Table 23: Tobin s Q as dependent variable Table 24: ROA as dependent variable Table 25: Percent of independents as dependent variable Table 26: Hedge fund ownership as dependent variable Table 27: Mutual fund ownership as dependent variable Table 28: Leverage as dependent variable Table 29: Log(total coverage) as dependent variable Table 30: Log(Facebook shares) as dependent variable Complete Sample OLS with Interaction Regressions Table 31: Tobin s Q as dependent variable Table 32: ROA as dependent variable Table 33: Financial vs. non-financial media Table 34: Mainstream vs. non-mainstream media Table 35: Paywall vs. non-paywall Simultaneous with Instrumental Variable Regressions Table 36: IV models In all tables reporting regression results, the dependent variable is identified in the table title; robust standard errors are presented in parentheses. Also, * denotes p-values < 0.10; **, 0.05; and ***,

108 Exploratory and Descriptive Tables Social Media Exploratory Tables Table 1: Media following on Facebook and Twitter The following table was constructed by looking up the number of followers of each page from each social media platform. Numbers are subject to change and were collected in February Media Source Facebook Followers (mn) Twitter Followers (mn) Barrons Chicago Tribune CNBC Economist Forbes Fox News LA Times Market Watch New York Times NPR Reuters USA Today Wall Street Journal Washington Post Yahoo Finance Average

109 Table 2: Political media bias Table 2 compares research conducted on media bias. Allsides is a website that presents news from multiple perspectives to their readers, with the goal of exposing media bias. Pew Research Center is a highly regarded nonpartisan fact tank which provides much of the current research about media. 47 Media Source Allsides Bias Rating 48 Pew Research Center 49 L = Left LL = Lean Left Barrons C = Center Chicago Tribune C LR = Lean Right CNBC C R = Right Economist LL LL Forbes C Fox News R R Los Angeles Times LL Market Watch New York Times LL L NPR C L Reuters C USA Today C Wall Street Journal C C Washington Post LL LL Yahoo Finance C (Mitchell et al. 2014) 105

110 Table 3: Twitter versus Facebook users This table, replicated from Pew Research Center 50, compares a recent panel of Facebook users to Twitter users. The table indicates that demographically, the average Twitter user is quite similar to the average Facebook user. The largest difference between Twitter and Facebook users is the age of users. There are also small discrepancies in education and income. The proportion of Twitter users rises as education increases. Facebook users are more uniform in income than Twitter, but there is no clear trend income trend of Twitter users. % of online adults who use Facebook Twitter Relative Proportion Facebook Relative Proportion Twitter All Gender Men Women Age Education High school Some college College Income > 30, ,000-49, ,000-75, , Location Urban Suburban Rural Table 4: News category Table 4 lists the different news sources included in the analysis and why the specific sources are included. Media Category Previously studied Financial/Business Regional Political Leaning Mainstream News Sources New York Times, Wall Street Journal, Business Week Barrons, CNBC, Economist, Forbes, Market Watch, Reuters, Yahoo Finance Los Angeles Times, Chicago Tribune Fox News NPR, USA Today, Washington Post 50 (Gottfried & Shearer 2016) 106

111 Complete Sample OLS Exploratory and Summary Statistics Tables Table 5: Variable definitions Variable Description Data Source Financial Performance Tobin Tobin s Q: Total market value/total assets Compustat ROA Return on assets: Net income/total assets Compustat Governance Pct. Independent Percent of independents on the board ISS/Orbis Log(board size) Log of board size ISS/Orbis Duality 1 if CEO is also chairman of the board, 0 otherwise ISS/Orbis Ownership Hedge Fund Count of known hedge fund owners Osiris Mutual Fund 51 Count of known mutual fund owners Osiris Capital Structure Leverage (long-term debt + current debt)/total assets Compustat Media log(positive Score) log(negative Score) log(total Coverage) log(facebook Shares) Controls RDA log(property, Plant, & Equipment) The log of average positive score (classified using the Loughran/McDonald dictionary) aggregated by firm by year The log of average negative score (classified using the Loughran/McDonald dictionary) aggregated by firm by year The log of total number of articles about a firm aggregated by year, extensive margin The log of average Facebook shares about a firm aggregated by year, intensive margin Research, development, and advertising expenses, with missing values coded as 0 Log of property, plant and equipment See Data Section Compustat Compustat Volatility Standard deviation of the monthly stock return for the five preceding years CRSP Industry Leverage Average leverage per industry per year Compustat 51 The variable mutual funds is actually a combined count of the number of mutual funds and pension funds. For simplicity, throughout this research, the variable is referred to as just mutual funds. 107

112 Table 6: Industry distribution Table 6 shows industry distributions 52 for the media sample used in this research compared to the Compustat database. Compustat is one of the most comprehensive databases of firm data, so the comparison gives a good idea of how representative the sample is compared to the universe of firms in the United States. Both samples exclude FIRE firms, which are treated differently in corporate finance research. The comparison shows that the sample used in this research is fairly similar to the overall Compustat universe, however is over-representative of firms from the following section: Eating and Drinking (SIC: 58), Apparel and Accessory (SIC: 56), Food Products (SIC: 20) and General Merchandise (SIC: 53). The sample under-represents firms from: Chemical Products (SIC: 28), Metal Mining (SIC: 10), Electronic and Electrical Equipment (SIC: 36), and Water Transportation (SIC: 44). SIC Industry 01 Agriculture crops Agriculture & livestock Forestry Metal Mining Coal Mining Oil and Gas Extraction Mining of Non-metallic General Construction Heavy Construction Special Trade Construction Food Products Tobacco Products Textile Products Apparel and Fabrics Lumber and Wood Products Furniture and Fixtures Paper and Allied Products Printing and Publishing Chemicals and Allied Products Petroleum Refining Rubber and Plastics Leather Stone, Clay, Glass, Concrete Products Primary Metal Industries Fabricated Metal Products Industrial, Commercial Machinery, Computer Electronic Equipment w/o Computers Transportation Equipment Measuring, Analyzing, Controlling Instruments Miscellaneous Manufacturing Railroad Transportation Motor Freight Water Transportation Air Transportation Pct. Media Pct. Compustat 52 Industry descriptions are from 108

113 46 Pipelines, w/o Natural Gas Transportation Services Communications Electric, Gas, Sanitary Services Wholesale Trade: Durable Wholesale Trade: Non-durable Building Materials, Hardware, Garden Supply General Merchandise Food Stores Automotive Dealers Apparel and Accessory Home Furniture, Furnishings Eating and Drinking Miscellaneous Retail Hotels, Camps Personal Services Business Services Automotive Repair, parking Motion Pictures Amusement and Recreation Health Services Legal Services Educational Services Engineering, Accounting, Research, Management Non-classified

114 Table 7: Summary statistics from Compustat firm universe vs. sample Summary statistics from the Compustat firm sample (excluding FIRE firms) are provided to be transparent about differences between the sample used in this research and the sample used regularly in other corporate finance research. Research, development, and advertising (RDA) is the only variable that stands out as being very different across the samples. RDA is much lower in the sample used in this research, because industries such as Chemical Products and Metal Mining are under-represented in the sample. N Mean Min p50 Max SD Comp. Sample Comp. Sample Comp. Sample Comp. Sample Comp. Sample Comp. Sample Tobin ROA Pct. Independent Duality Board Size Hedge Funds Mutual Funds Leverage RDA Log(ppe) Volatility Negative Score na 3357 na na 0 na na 379 na Positive Score na 3357 na na 0 na na 198 na Facebook Shares na 3357 na na 0 na na na Total Coverage na 3357 na na 1 na 3 na 7150 na

115 Segmentation Exploratory Tables and Summary Statistics Table 8: Financial vs. non-financial news sources Financial sources were chosen based off financial or business prominence in the United States. News organizations like the New York Times, which is considered a reputable source of business and financial news, was categorized as non-financial, because the New York Times does not strictly focus on business and financial news. Financial Non-Financial CNBC, Barrons, Forbes, Market Watch, Reuters, The Economist, The Wall Street Journal, Yahoo Finance The Chicago Tribune, Fox News, The Los Angeles Times, National Public Radio, The New York Times, USA Today, The Washington Post Table 9: Mainstream vs. non-mainstream news sources Mainstream sources were chosen based on their Facebook and Twitter followings Table 1 and prominence as a media source in the United States. All news sources labeled as mainstream are the top 5 most followed Facebook pages from the relative sample. Mainstream Non-Mainstream Fox News, The New York Times, USA Today, The Washington Post, CNBC, Barrons, Chicago Tribune, Forbes, Los Angeles Times, Reuters, The Economist, Yahoo Finance Table 10: Paywall vs. non-paywall news sources The different sources were looked-up to see if they had a paywall or not. Paywall is defined as providing some free content, for example 10 free articles a month, but then requiring the users to be a subscriber. Paywall No Paywall Barrons, The New York Times, The Economist, The Wall Street Journal, The Washington Post, The Chicago Tribune, The Los Angeles Times CNBC, Forbes, Fox News, Market Watch, NPR, Reuters, USA Today, Yahoo Finance Table 11: Political leanings of news sources Political leanings were categorized using Table 2. NPR was left un-categorized because the Pew bias ratings conflicted with the Allsides bias ratings. Left Right Center Not-Categorized The Los Angeles Times, The New York Times, The Economist, The Washington Post Fox News The Chicago Tribune, Forbes, Reuters, USA Today, The Wall Street Journal CNBC, Barrons, Market Watch, NPR, Yahoo Finance 111

116 Table 12: Financial vs. non-financial summary statistics Financial Non-Financial N Mean Min p50 Max N Mean Min p50 Max Tobin ROA Pct. Independent Duality Board Size Hedge Mutual Leverage RDA Log(ppe) Volatility Log(Positive Score) Log(Negative Score) Log(Facebook Shares) Log(Total Coverage) Table 13: Mainstream vs. non-mainstream summary statistics Not Mainstream Mainstream N Mean Min p50 Max N Mean Min p50 Max Tobin ROA Pct. Independent Duality Board Size Hedge Mutual Leverage RDA Log(ppe) Volatility Log(Positive Score) Log(Negative Score) Log(Facebook Shares) Log(Total Coverage)

117 Table 14: Paywall vs. non-paywall summary statistics No Paywall Paywall N Mean Min p50 Max N Mean Min p50 Max Tobin ROA Pct. Independent Duality Board Size Hedge Mutual Leverage RDA Log(ppe) Volatility Log(Positive Score) Log(Negative Score) Log(Facebook Shares) Log(Total Coverage)

118 Table 15: Political leaning summary statistics Center Left Right N Mean Min p50 Max N Mean Min p50 Max N Mean Min p50 Max Tobin ROA Pct. Independent Duality Board Size Hedge Mutual Leverage RDA Log(ppe) Volatility Log(Positive Score) Log(Negative Score) Log(Facebook Shares) Log(Total Coverage)

119 Table 16: T-tests and ANOVA T-tests of the mean and ANOVA tests to see if groups are statistically different than each other. The null hypothesis is β Y = β N, or in the cases of political β L = β C = β R. P-values are reported. In all but two cases the results indicate that the null hypothesis can be rejected, suggesting that there is a difference in the means across sentiment and coverage for the different segments. This indicates that different types of media report differently in terms of coverage and sentiment. Positive Score Negative Score Facebook Shares Financial Mainstream Political Paywall Total Coverage Table 17: Chi-square tests The Chi-square test is used to test if there is a significant difference in frequencies between two (or more) groups. The dependent variable is the segment, and the independent variable is industry. The Chi-square tests if there is a statistical difference in the distribution of industries across segments. P-values are reported. The high p-values indicate that across all segments except for mainstream, the media reports on similar industries. Industry Financial Mainstream Political Paywall

120 Simultaneous Modeling Exploratory and Descriptive Tables Table 18: Instrumental variable descriptions Instrument Variable Definition Financial Performance Tobin s Q IA 53 Tobin s Q industry average Tobin s Q ROA IA ROA industry average ROA Governance Multiple Directorships Average number of boards that a firm s directors serve on Percent Board Shareholder Percent of the board that is a shareholder of the firm Ownership Stability Log(share volume/shares outstanding) Dividend Yield Average yearly dividend as a percentage of share price Positive Earnings 1 if net income is greater than 0, 0 otherwise SP 500 Constituent 1 if firm is in the S&P 500, 0 otherwise 54 Capital Structure 1.2*(working capital/total assets) + 1.4*(retained earnings/total Altman s Z-score assets) + 3.3*(EBIT/total assets) +0.6*(market value/total liabilities) +1.0*(sales/total assets) Media Facebook Shares IA Log(Facebook shares) Log(industry average Facebook shares) Total Coverage IA Log(Total coverage) Log(industry average total coverage) Table 19: Instrumental variables summary statistics N Mean Min p50 Max Tobin s Q IA ROA IA Multiple Directorships Percent Board Shareholder Stability Dividend Yield Positive Earnings SP 500 Constituent Altman s Z-score Facebook Shares IA Total Coverage IA IA stands for industry adjusted 54 Data on S&P 500 composition came from the website 500/, which contains historical data about the companies that compose the S&P

121 Table 20: Instrumental variable regressed on endogenous dependent variable A simple test to see if instruments meet the first condition of strong instruments, highly correlated with the variable they are instrumenting for, is to regress the endogenous variable on the instrument. In every case, there is a statistically significant relationship between the endogenous variable and the instrumental variable Variable Instrument Coefficient P-value Financial Performance Tobin Industry-Adjusted Tobin s Q (0.000) ROA Industry-Adjusted ROA (0.000) Governance Percent Independent Multiple Directorships (0.000) Percent Independent Percent Board Shareholder (0.000) Ownership Number of Hedge Funds Log(Share Volume/Shares Outstanding) (0.000) Number of Hedge Funds Dividend Yield (0.000) Number of Hedge Funds Positive Earnings (0.000) Number of Mutual Funds Log(Share Volume/Shares Outstanding) (0.000) Number of Mutual Funds SP 500 Constituent (0.000) Number of Mutual Fund s Dividend Yield (0.052) Number of Mutual Funds Positive Earnings (0.000) Capital Structure Leverage Altman s Z-score (0.000) Media Facebook Shares Industry-Adjusted Facebook Shares (0.000) Total Coverage Industry-Adjusted Total Coverage (0.000) 55 Altman s Z-score is the standard formula used to predict bankruptcy (Ross et al. 2010) 117

122 Table 21: IV tests for media models The following tests are represented below: F-Test, Sanderson-Windmeijer (SW) Chi-square, Sanderson-Windmeijer (SW) F-test, and Hausman test. The F-test tests the relevance of excluded instruments in the first-stage regression by testing for joint significance (Rubin & Smith 2009). The low F-test p-values indicate the instruments are jointly significant. The SW Chi-square test tests for under-identification. The low SW Chi-square p-values indicate the models are identified. The SW F-test tests for weak instruments (Baum & Schaffer 2007). Using a 10-percent significance level (Baum 2007), the SW F-test indicates some of the instruments, with values less than 11.46, are weak. The Hausman tests for consistency. and the low Hausman p-values indicate the 2SLS is most consistent than the LSDV model. Test F-test SW Chi-square SW F-test Hausman test F-statistic p-value statistic p-value statistic H-statistic p-value Hedge Hedge (0.000) Tobin s Q Mutual (0.000) ROA Tobin (0.000) Leverage ROA (0.000) Pct. Indep Pct. Independent (0.000) Facebook shares Leverage (0.000) Total coverage Total Coverage (0.000) Mutual Facebook Shares (0.000) Tobin s Q ROA SW F-test Leverage Critical Sig. Value Pct. Indep % Facebook shares % Total coverage % 6.65 Tobin s Q 30% 4.92 Mutual Funds Hedge Funds Leverage Pct. Indep Facebook shares Total coverage ROA Mutual Funds Hedge Funds Leverage Pct. Indep Facebook shares Total coverage

123 Table 21: IV tests for media models (continued) Test F-test SW Chi-square SW F-test F-statistic p-value statistic p-value statistic SW F-test Pct. Independent Sig. Critical Value Mutual Funds % Hedge Funds % Leverage % 6.65 Tobin s Q % 4.92 ROA Facebook shares Total coverage Leverage Mutual Funds Hedge Funds Tobin s Q ROA Pct. Independent Facebook Shares Total Coverage Total Coverage Mutual Funds Hedge Funds Leverage ROA Tobin s Q Pct. Indep Facebook Shares Mutual Funds Hedge Funds Leverage ROA Tobin s Q Pct. Indep Table 22: Hausman test for media models The Hausman test suggests the 3SLS model is more consistent than the LSDV model. However, the failure to reject the null hypothesis when comparing the 2SLS model with the 3SLS model. indicates that the 3SLS isn t any more consistent than the 2SLS model. Hausman Test p-value 3SLS vs. OLS SLS vs. 2SLS

124 Complete Sample OLS Regression Tables Financial Performance Table 23: Tobin s Q as dependent variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLES OLS LSDV OLS LSDV OLS LSDV OLS LSDV OLS LSDV Leverage t *** *** *** *** *** *** *** *** *** *** (0.193) (0.201) (0.193) (0.201) (0.194) (0.201) (0.194) (0.201) (0.202) (0.210) Pct. Independent t *** * *** * *** * *** * * (0.258) (0.255) (0.258) (0.254) (0.259) (0.255) (0.258) (0.254) (0.274) (0.267) Duality t (0.054) (0.055) (0.054) (0.055) (0.055) (0.055) (0.055) (0.055) (0.058) (0.060) Log(Board Size) t * * * * * ** * * (0.135) (0.133) (0.135) (0.133) (0.135) (0.133) (0.135) (0.134) (0.142) (0.142) RDA t *** 3.602*** 4.109*** 3.623*** 4.132*** 3.596*** 4.119*** 3.615*** 3.772*** 3.345*** (0.515) (0.617) (0.513) (0.617) (0.518) (0.621) (0.519) (0.621) (0.602) (0.722) Hedge Funds t (0.015) (0.015) (0.015) (0.015) (0.015) (0.015) (0.016) (0.016) (0.016) (0.016) Mutual Funds t (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.003) Log(Property, Plant, Equipment) t *** *** *** *** *** *** *** *** *** *** (0.016) (0.021) (0.016) (0.021) (0.016) (0.021) (0.016) (0.021) (0.019) (0.027) Monthly Volatility t *** *** *** *** *** *** *** *** *** *** (0.532) (0.543) (0.532) (0.543) (0.535) (0.544) (0.536) (0.544) (0.548) (0.555) Log(Positive Score) t (0.038) (0.037) (0.051) (0.050) (0.059) (0.058) Log(Negative Score) t (0.037) (0.037) (0.049) (0.050) (0.055) (0.058) Log(Facebook Shares) t (0.017) (0.015) Log(Total Coverage) t *** 0.087*** (0.026) (0.028) Constant 3.683*** 4.104*** 3.802*** 4.144*** 3.697*** 4.091*** 3.774*** 4.120*** 3.606*** 3.999*** (0.369) (0.419) (0.391) (0.438) (0.387) (0.437) (0.395) (0.442) (0.432) (0.482) Observations 1,738 1,738 1,736 1,736 1,731 1,731 1,729 1,729 1,524 1,524 R-squared Industry FE YES YES YES YES YES Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<

125 Table 24: ROA as dependent variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLES OLS LSDV OLS LSDV OLS LSDV OLS LSDV OLS LSDV Leverage t *** *** *** *** *** *** *** *** *** *** (0.014) (0.015) (0.014) (0.015) (0.014) (0.015) (0.014) (0.015) (0.015) (0.016) Pct. Independent t (0.026) (0.026) (0.026) (0.026) (0.026) (0.026) (0.026) (0.026) (0.028) (0.029) Duality t * * * * ** (0.005) (0.006) (0.005) (0.006) (0.005) (0.006) (0.005) (0.006) (0.005) (0.007) Log(Board Size) t *** 0.028** 0.047*** 0.027** 0.046*** 0.027** 0.047*** 0.028** 0.049*** 0.030** (0.012) (0.011) (0.012) (0.011) (0.012) (0.011) (0.012) (0.011) (0.012) (0.012) RDA t ** ** ** ** (0.033) (0.038) (0.033) (0.038) (0.033) (0.038) (0.033) (0.038) (0.036) (0.044) Hedge Funds t (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Mutual Funds t * * ** ** *** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Log(Property, Plant, Equipment) t *** *** *** *** *** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.002) Monthly Volatility t *** *** *** *** *** *** *** *** ** ** (0.048) (0.042) (0.048) (0.042) (0.049) (0.042) (0.049) (0.042) (0.051) (0.047) Log(Positive Score) t * (0.003) (0.003) (0.004) (0.005) (0.006) (0.005) Log(Negative Score) t (0.003) (0.003) (0.005) (0.005) (0.006) (0.005) Log(Facebook Shares) t * (0.002) (0.001) Log(Total Coverage) t *** 0.005** (0.002) (0.002) Constant (0.033) (0.034) (0.035) (0.036) (0.032) (0.035) (0.034) (0.036) (0.036) (0.039) Observations 1,738 1,738 1,736 1,736 1,731 1,731 1,729 1,729 1,524 1,524 R-squared Industry FE YES YES YES YES YES 121

126 Governance Table 25: Percent of independents as dependent variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLES OLS LSDV OLS LSDV OLS LSDV OLS LSDV OLS LSDV Leverage t *** *** *** *** *** *** *** *** * *** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Pct. Independent t *** 0.222*** 0.141** 0.214*** 0.149*** 0.225*** 0.147** 0.221*** 0.122** 0.209*** (0.057) (0.061) (0.057) (0.060) (0.058) (0.061) (0.057) (0.060) (0.060) (0.064) Duality t (0.023) (0.025) (0.023) (0.025) (0.023) (0.025) (0.023) (0.025) (0.025) (0.027) Log(Board Size) t *** 0.037*** 0.036*** 0.037*** 0.036*** 0.037*** 0.036*** 0.036*** 0.039*** 0.038*** (0.006) (0.007) (0.006) (0.007) (0.006) (0.007) (0.006) (0.007) (0.007) (0.007) RDA t *** 0.083*** 0.082*** 0.082*** 0.083*** 0.083*** 0.080*** 0.080*** 0.086*** 0.084*** (0.015) (0.015) (0.014) (0.015) (0.015) (0.015) (0.014) (0.015) (0.015) (0.016) Hedge Funds t ** ** ** ** ** (0.043) (0.054) (0.044) (0.055) (0.044) (0.055) (0.044) (0.055) (0.049) (0.064) Mutual Funds t (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Log(Property, Plant, Equipment) t *** *** *** *** *** *** *** *** *** *** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Monthly Volatility t (0.038) (0.036) (0.037) (0.036) (0.038) (0.037) (0.038) (0.037) (0.040) (0.039) Log(Positive Score) t * *** *** (0.004) (0.004) (0.006) (0.006) (0.006) (0.007) Log(Negative Score) t * ** 0.016*** 0.020*** 0.022*** (0.004) (0.005) (0.006) (0.006) (0.007) (0.007) Log(Facebook Shares) t *** *** (0.002) (0.002) Log(Total Coverage) t (0.002) (0.002) Constant 0.618*** 0.677*** 0.595*** 0.656*** 0.619*** 0.683*** 0.597*** 0.659*** 0.613*** 0.669*** (0.036) (0.076) (0.039) (0.077) (0.038) (0.076) (0.039) (0.076) (0.043) (0.077) Observations 1,715 1,715 1,713 1,713 1,708 1,708 1,706 1,706 1,503 1,503 R-squared Industry FE YES YES YES YES YES 122

127 Ownership Table 26: Hedge fund ownership as dependent variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLES OLS LSDV OLS LSDV OLS LSDV OLS LSDV OLS LSDV Leverage t (0.057) (0.058) (0.057) (0.058) (0.057) (0.058) (0.057) (0.058) (0.059) (0.060) Pct. Independent t * * (0.987) (1.034) (0.985) (1.030) (0.988) (1.033) (0.988) (1.034) (1.068) (1.118) Duality t *** 1.043*** 0.934*** 1.043*** 0.929*** 1.043*** 0.935*** 1.050*** 0.881*** 0.882** (0.313) (0.369) (0.313) (0.370) (0.314) (0.370) (0.314) (0.370) (0.330) (0.387) Log(Board Size) t (0.523) (0.511) (0.525) (0.513) (0.525) (0.514) (0.527) (0.516) (0.573) (0.565) RDA t (0.097) (0.103) (0.097) (0.103) (0.097) (0.103) (0.097) (0.104) (0.107) (0.114) Hedge Funds t *** *** *** *** *** *** *** *** *** *** (0.244) (0.256) (0.245) (0.257) (0.244) (0.256) (0.245) (0.257) (0.263) (0.281) Mutual Funds t *** 2.457*** 2.151*** 2.497*** 2.193*** 2.566*** 2.199*** 2.541*** 2.033** 2.778*** (0.723) (0.874) (0.722) (0.873) (0.724) (0.877) (0.724) (0.878) (0.790) (0.970) Log(Property, Plant, Equipment) t ** 0.080** 0.060** 0.083** 0.061** 0.088** 0.060** 0.087** * (0.028) (0.036) (0.028) (0.036) (0.028) (0.036) (0.028) (0.036) (0.033) (0.045) Monthly Volatility t *** 2.785*** 2.898*** 2.792*** 2.957*** 2.862*** 2.965*** 2.867*** 3.168*** 3.038*** (1.063) (1.029) (1.055) (1.018) (1.059) (1.018) (1.060) (1.017) (1.179) (1.122) Log(Positive Score) t ** ** * ** * (0.062) (0.064) (0.080) (0.082) (0.096) (0.097) Log(Negative Score) t (0.065) (0.067) (0.086) (0.088) (0.097) (0.101) Log(Facebook Shares) t * (0.026) (0.026) Log(Total Coverage) t (0.037) (0.044) Constant 3.128*** 5.889*** 3.297*** 6.108*** 3.387*** 6.176*** 3.356*** 6.148*** 3.314*** 5.815*** (0.662) (1.222) (0.672) (1.221) (0.668) (1.216) (0.673) (1.221) (0.778) (1.287) Observations 1,736 1,736 1,734 1,734 1,729 1,729 1,727 1,727 1,522 1,522 R-squared Industry FE YES YES YES YES YES 123

128 Table 27: Mutual fund ownership as dependent variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLES OLS LSDV OLS LSDV OLS LSDV OLS LSDV OLS LSDV Leverage t *** *** *** *** *** *** *** *** *** *** (0.550) (0.579) (0.550) (0.579) (0.551) (0.581) (0.551) (0.581) (0.578) (0.616) Pct. Independent t ** *** ** *** *** *** *** *** *** *** (9.032) (9.091) (9.017) (9.085) (9.084) (9.141) (9.080) (9.136) (9.640) (9.835) Duality t *** *** *** *** *** (3.125) (3.431) (3.120) (3.417) (3.130) (3.424) (3.129) (3.423) (3.298) (3.634) Log(Board Size) t *** *** *** *** *** *** *** *** *** ** (4.802) (4.765) (4.808) (4.766) (4.819) (4.772) (4.830) (4.779) (5.214) (5.097) RDA t *** * *** * *** * *** * ** (0.993) (1.016) (0.994) (1.016) (0.994) (1.017) (0.996) (1.018) (1.072) (1.102) Hedge Funds t *** *** *** *** *** *** *** *** *** *** (2.281) (2.291) (2.283) (2.285) (2.285) (2.294) (2.288) (2.294) (2.424) (2.423) Mutual Funds t *** *** *** *** ** (6.880) (8.390) (6.868) (8.347) (6.885) (8.390) (6.892) (8.382) (7.614) (9.183) Log(Property, Plant, Equipment) t *** *** *** *** *** *** *** *** *** * (0.297) (0.381) (0.297) (0.382) (0.296) (0.382) (0.298) (0.382) (0.336) (0.464) Monthly Volatility t * * * * (11.023) (10.480) (10.887) (10.316) (10.936) (10.304) (10.909) (10.275) (11.885) (10.848) Log(Positive Score) t ** ** * (0.655) (0.679) (0.870) (0.895) (1.030) (1.042) Log(Negative Score) t * ** (0.707) (0.722) (0.944) (0.958) (1.066) (1.080) Log(Facebook Shares) t *** 0.999*** (0.296) (0.289) Log(Total Coverage) t *** *** (0.358) (0.390) Constant *** *** *** *** *** *** *** *** *** *** (6.284) (7.734) (6.482) (7.789) (6.438) (7.782) (6.505) (7.828) (7.544) (8.403) Observations 1,736 1,736 1,734 1,734 1,729 1,729 1,727 1,727 1,522 1,522 R-squared Industry FE YES YES YES YES YES 124

129 Capital Structure Table 28: Leverage as dependent variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLES OLS LSDV OLS LSDV OLS LSDV OLS LSDV OLS LSDV Tobin's Q t *** *** *** *** *** *** *** *** *** *** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) ROA t * * * * (0.077) (0.074) (0.077) (0.074) (0.078) (0.074) (0.078) (0.074) (0.084) (0.079) Pct. Independent t * * * * (0.041) (0.042) (0.041) (0.042) (0.041) (0.042) (0.041) (0.042) (0.044) (0.044) Duality t (0.008) (0.008) (0.008) (0.008) (0.008) (0.008) (0.008) (0.008) (0.009) (0.009) Log(Board Size) t *** 0.038* 0.066*** 0.038* 0.067*** 0.039* 0.067*** 0.039* 0.076*** 0.046** (0.020) (0.021) (0.020) (0.021) (0.020) (0.021) (0.020) (0.021) (0.021) (0.022) RDA t ** ** 0.103** ** 0.103** ** 0.102** ** 0.149** (0.052) (0.059) (0.052) (0.059) (0.052) (0.059) (0.052) (0.059) (0.058) (0.067) Hedge Funds t *** 0.009*** 0.009*** 0.009*** 0.009*** 0.009*** 0.009*** 0.009*** 0.009*** 0.008*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) Mutual Funds t ** 0.001*** 0.001** 0.001*** 0.001** 0.001*** 0.001** 0.001*** ** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Log(Property, Plant, Equipment) t *** 0.013*** 0.007*** 0.013*** 0.007*** 0.013*** 0.007*** 0.013*** 0.008** 0.016*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.004) Monthly Volatility t (0.058) (0.049) (0.058) (0.049) (0.058) (0.049) (0.058) (0.049) (0.064) (0.054) Industry Leverage t *** 0.468** 0.647*** 0.472** 0.645*** 0.470** 0.647*** 0.471** 0.645*** (0.059) (0.203) (0.059) (0.204) (0.060) (0.205) (0.060) (0.205) (0.064) (0.220) Log(Positive Score) t (0.005) (0.005) (0.007) (0.007) (0.008) (0.008) Log(Negative Score) t (0.006) (0.005) (0.007) (0.007) (0.009) (0.008) Log(Facebook Shares) t ** (0.002) (0.002) Log(Total Coverage) t *** *** (0.003) (0.003) Constant *** ** *** ** *** (0.055) (0.101) (0.058) (0.103) (0.056) (0.102) (0.058) (0.103) (0.062) (0.104) Observations 1,739 1,739 1,737 1,737 1,732 1,732 1,730 1,730 1,525 1,525 R-squared

130 Media Table 29: Log(total coverage) as dependent variable (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES OLS LSDV OLS LSDV OLS LSDV OLS LSDV Leverage t *** 0.229*** 0.257*** 0.234*** 0.253*** 0.229*** 0.256*** 0.233*** (0.049) (0.046) (0.050) (0.046) (0.050) (0.046) (0.050) (0.046) Pct. Independent t * ** * * (0.721) (0.698) (0.722) (0.698) (0.726) (0.700) (0.724) (0.700) Duality t *** ** *** ** *** ** *** ** (0.336) (0.309) (0.336) (0.310) (0.337) (0.311) (0.337) (0.311) Log(Board Size) t * * * * (0.070) (0.068) (0.070) (0.068) (0.071) (0.068) (0.071) (0.068) RDA t ** ** ** ** (0.180) (0.163) (0.180) (0.163) (0.180) (0.163) (0.180) (0.163) Hedge Funds t * * ** * (0.202) (0.210) (0.201) (0.210) (0.203) (0.211) (0.202) (0.210) Mutual Funds t *** 3.426*** 2.279*** 3.373*** 2.250*** 3.428*** 2.276*** 3.386*** (0.548) (0.619) (0.549) (0.621) (0.551) (0.622) (0.552) (0.625) Log(Property, Plant, Equipment) t * * * * (0.023) (0.021) (0.022) (0.021) (0.023) (0.021) (0.022) (0.021) Monthly Volatility t *** 0.019*** 0.017*** 0.020*** 0.016*** 0.019*** 0.017*** 0.020*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) Log(Positive Score) t *** 0.553*** 0.345*** 0.549*** 0.348*** 0.553*** 0.345*** 0.552*** (0.023) (0.027) (0.023) (0.027) (0.023) (0.027) (0.023) (0.027) Log(Negative Score) t ** ** ** ** (0.354) (0.367) (0.353) (0.359) (0.355) (0.367) (0.350) (0.358) Log(Facebook Shares) t ** 0.119*** 0.154*** 0.211*** (0.044) (0.044) (0.055) (0.057) Log(Total Coverage) t ** (0.044) (0.043) (0.055) (0.056) Constant ** *** *** *** *** *** *** *** (0.409) (0.486) (0.425) (0.506) (0.422) (0.496) (0.428) (0.509) Observations 1,739 1,739 1,737 1,737 1,732 1,732 1,730 1,730 R-squared Industry FE YES YES YES YES 126

131 Table 30: Log(Facebook shares) as dependent variable (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES OLS LSDV OLS LSDV OLS LSDV OLS LSDV Leverage t ** ** ** * (0.047) (0.049) (0.047) (0.049) (0.047) (0.049) (0.046) (0.049) Pct. Independent t ** * * * * (0.839) (0.892) (0.836) (0.888) (0.839) (0.888) (0.839) (0.887) Duality t *** *** *** *** *** *** *** *** (0.347) (0.346) (0.350) (0.347) (0.347) (0.346) (0.350) (0.348) Log(Board Size) t (0.081) (0.085) (0.081) (0.085) (0.081) (0.085) (0.081) (0.085) RDA t (0.190) (0.195) (0.189) (0.195) (0.190) (0.196) (0.190) (0.196) Hedge Funds t (0.249) (0.275) (0.250) (0.275) (0.249) (0.274) (0.249) (0.274) Mutual Funds t (0.567) (0.712) (0.566) (0.712) (0.573) (0.718) (0.573) (0.719) Log(Property, Plant, Equipment) t ** ** ** ** ** ** ** ** (0.023) (0.024) (0.023) (0.025) (0.023) (0.025) (0.023) (0.024) Monthly Volatility t *** 0.030*** 0.028*** 0.030*** 0.028*** 0.030*** 0.028*** 0.029*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) Log(Positive Score) t *** 0.145*** 0.083*** 0.147*** 0.085*** 0.155*** 0.083*** 0.152*** (0.025) (0.032) (0.025) (0.032) (0.025) (0.032) (0.025) (0.032) Log(Negative Score) t (0.668) (0.686) (0.671) (0.691) (0.650) (0.675) (0.651) (0.676) Log(Facebook Shares) t ** ** (0.064) (0.067) (0.084) (0.088) Log(Total Coverage) t *** *** *** ** (0.068) (0.069) (0.086) (0.088) Constant 4.265*** 5.122*** 4.604*** 5.464*** 4.725*** 5.543*** 4.724*** 5.554*** (0.496) (0.841) (0.519) (0.842) (0.519) (0.837) (0.523) (0.842) Observations 1,720 1,720 1,718 1,718 1,713 1,713 1,711 1,711 R-squared Industry FE YES YES YES YES 127

132 Complete Sample OLS with Interactions Regression Tables Table 31: Tobin s Q as dependent variable VARIABLES (1) (2) (3) (4) (5) (6) (7) Tobin s Tobin s Tobin s Tobin s Tobin s Q Q Q Q Q Tobin s Q Tobin s Q Leverage t *** 1.369*** 1.375*** 1.379*** 1.262*** 1.245*** 1.240*** (0.201) (0.201) (0.201) (0.201) (0.210) (0.209) (0.212) Pct. Independent t * * * *** (0.254) (1.333) (0.255) (1.209) (0.267) (0.679) (0.390) Duality t (0.055) (0.055) (0.055) (0.056) (0.060) (0.060) (0.061) Log(Board Size) t * * ** * (0.133) (0.134) (0.133) (0.135) (0.142) (0.142) (0.144) RDA t *** 3.614*** 3.596*** 3.621*** 3.345*** 3.304*** 3.377*** (0.617) (0.618) (0.621) (0.616) (0.722) (0.721) (0.718) Hedge Funds t (0.015) (0.071) (0.015) (0.060) (0.016) (0.034) (0.027) Mutual Funds t (0.002) (0.012) (0.002) (0.010) (0.003) (0.006) (0.004) Log(Property, Plant, Equipment) 0.076*** 0.075*** 0.076*** 0.075*** 0.131*** 0.134*** 0.131*** t-1 (0.021) (0.021) (0.021) (0.021) (0.027) (0.027) (0.027) Monthly Volatility t *** 1.831*** 1.876*** 1.867*** 1.723*** 1.683*** 1.731*** (0.543) (0.544) (0.544) (0.545) (0.555) (0.559) (0.555) Log(Negative Score) t (0.037) (0.382) (0.058) (0.058) (0.058) Log(Neg)*Hedge Funds t (0.021) Log(Neg)*Mutual Funds t (0.004) Log(Neg)*Pct. Ind. t (0.406) Log(Pos) t * (0.037) (0.431) (0.058) (0.058) (0.058) Log(Pos)*Hedge Funds t (0.022) Log(Pos)*Mutual Funds t (0.004) Log(Pos)*Pct. Ind. t (0.456) Log(Facebook Shares) t ** (0.015) (0.131) (0.015) Log(Total Coverage) t *** 0.102*** (0.028) (0.028) (0.162) Log(FB shares)*hedge Funds t (0.008) Log(FB shares)*mutual Funds t (0.001) Log(FB shares)*pct. Ind. t ** (0.144) Log(Total Cover)*Hedge Funds t (0.012) Log(Total Cove)*Mutual Funds t (0.002) Log(Total Cover)*Pct. Ind. t (0.166) Constant 4.144*** 5.507*** 4.091*** 5.888*** 3.999*** 5.145*** 3.888*** (0.438) (1.317) (0.437) (1.189) (0.482) (0.751) (0.506) Observations 1,736 1,736 1,731 1,731 1,524 1,524 1,524 R-squared

133 Table 32: ROA as dependent variable (1) (2) (3) (4) (5) (6) (7) VARIABLES ROA ROA ROA ROA ROA ROA ROA Leverage t *** *** *** *** *** *** *** (0.015) (0.015) (0.015) (0.015) (0.016) (0.016) (0.017) Pct. Independent t (0.026) (0.099) (0.026) (0.092) (0.029) (0.074) (0.052) Duality t (0.006) (0.006) (0.006) (0.006) (0.007) (0.007) (0.008) Log(Board Size) t ** 0.027** 0.027** 0.027** 0.030** 0.029** 0.029** (0.011) (0.011) (0.011) (0.011) (0.012) (0.012) (0.012) RDA t (0.038) (0.038) (0.038) (0.038) (0.044) (0.044) (0.045) Hedge Funds t (0.001) (0.006) (0.001) (0.004) (0.001) (0.004) (0.002) Mutual Funds t * ** *** (0.000) (0.001) (0.000) (0.001) (0.000) (0.000) (0.000) Log(Property, Plant, Equipment) t (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) Monthly Volatility t *** *** *** *** ** ** ** (0.042) (0.042) (0.042) (0.042) (0.047) (0.047) (0.047) Log(Negative Score) t (0.003) (0.025) (0.005) (0.005) (0.005) Log(Neg)*Hedge Funds t (0.002) Log(Neg)*Mutual Funds t (0.000) Log(Neg)*Pct. Ind. t (0.029) Log(Pos) t (0.003) (0.032) (0.005) (0.005) (0.005) Log(Pos)*Hedge Funds t (0.002) Log(Pos)*Mutual Funds t (0.000) Log(Pos)*Pct. Ind. t (0.037) Log(Facebook Shares) t * * (0.001) (0.011) (0.001) Log(Total Coverage) t ** 0.006** (0.002) (0.002) (0.015) Log(FB shares)*hedge Funds t * (0.001) Log(FB shares)*mutual Funds t (0.000) Log(FB shares)*pct. Ind. t (0.013) Log(Total Cover)*Hedge Funds t (0.001) Log(Total Cove)*Mutual Funds t (0.000) Log(Total Cover)*Pct. Ind. t (0.017) Constant (0.036) (0.087) (0.035) (0.084) (0.039) (0.068) (0.050) Observations 1,736 1,736 1,731 1,731 1,524 1,524 1,524 R-squared

134 Segmented Sample OLS Regression Tables Table 33: Financial vs. non-financial media VARIABLES (1) (2) (3) (4) (5) (6) (7) (8) Tobin s Q ROA ROA Pct. Indep. Pct. Indep. Mutual Funds Fin. = 0 Fin. = 1 Fin. = 0 Fin. = 1 Fin. = 0 Fin. = 1 Tobin s Q Fin. = 1 Mutual Funds Fin. = 0 Tobin's Q t * * *** *** (0.005) (0.007) (0.806) (0.978) ROA t ** 0.207* *** *** (0.090) (0.109) (14.209) (14.525) Leverage t *** ** *** *** (0.332) (0.308) (0.028) (0.022) (0.037) (0.046) (5.384) (5.620) Duality t *** 0.030** * (0.095) (0.086) (0.013) (0.007) (0.010) (0.012) (1.718) (1.528) Log(Board Size) t ** 0.041** *** 0.086*** *** *** (0.203) (0.206) (0.018) (0.016) (0.025) (0.030) (3.415) (3.732) Hedge Funds t ** (0.023) (0.023) (0.002) (0.002) (0.003) (0.003) Mutual Funds t *** *** (0.004) (0.004) (0.000) (0.000) (0.001) (0.001) Log(Property, Plant, Equipment) t *** *** * ** (0.038) (0.035) (0.003) (0.004) (0.004) (0.005) (0.594) (0.721) Monthly Volatility t * ** * 0.163*** (1.247) (0.576) (0.114) (0.039) (0.057) (0.050) (12.922) (13.889) RDA t *** 3.829*** ** ** (1.002) (0.986) (0.061) (0.062) (0.081) (0.114) (12.444) (15.255) Log(Positive Score) t *** * (0.079) (0.069) (0.008) (0.005) (0.008) (0.009) (1.209) (1.272) Log(Negative Score) t ** 0.025*** (0.078) (0.067) (0.007) (0.005) (0.008) (0.008) (1.356) (1.262) Log(Total Coverage) t * 0.186*** ** *** ** (0.069) (0.065) (0.005) (0.005) (0.005) (0.007) (0.787) (1.042) Log(Facebook Shares) t * (0.020) (0.020) (0.002) (0.002) (0.003) (0.003) (0.380) (0.407) Pct. Independent t * ** (0.390) (0.381) (0.041) (0.040) (7.499) (6.696) Constant 3.962*** 3.224*** *** 0.249*** *** *** (0.679) (0.473) (0.061) (0.038) (0.106) (0.064) (10.538) (9.977) Observations R-squared

135 Table 33:Financial vs. non-financial media (continued) (9) (10) (11) (12) (13) (14) (15) (16) VARIABLES Hedge Funds Fin. =1 Hedge Funds Fin. = 0 Leverage Fin. = 1 Leverage Fin. = 0 Total Coverage Fin. = 1 Total Coverage Fin. = 0 FB Shares Fin. = 1 FB Shares Fin. = 0 Tobin's Q t *** *** 0.169*** 0.113** 0.171* (0.082) (0.094) (0.006) (0.007) (0.042) (0.051) (0.051) (0.090) ROA t * * * ** (1.387) (1.743) (0.118) (0.114) (0.563) (0.648) (0.933) (1.401) Pct. Independent t *** (0.843) (0.740) (0.065) (0.061) (0.266) (0.326) (0.462) (0.623) Duality t (0.185) (0.154) (0.013) (0.013) (0.058) (0.061) (0.098) (0.140) Log(Board Size) t *** 1.003*** *** (0.423) (0.367) (0.030) (0.030) (0.135) (0.177) (0.217) (0.351) Hedge Funds t *** *** (0.004) (0.004) (0.017) (0.015) (0.029) (0.038) Mutual Funds t ** 0.011*** 0.010*** 0.026*** 0.031*** (0.001) (0.001) (0.002) (0.002) (0.004) (0.006) Monthly Volatility t ** (1.446) (1.533) (0.128) (0.062) (0.452) (0.276) (0.668) (1.180) RDA t * 3.030** ** *** (1.224) (1.418) (0.088) (0.114) (0.460) (0.544) (0.752) (1.303) Industry Leverage t * (0.289) (0.341) Log(Positive Score) t ** *** (0.122) (0.120) (0.009) (0.011) (0.047) (0.036) (0.076) (0.113) Log(Negative Score) t * *** 0.112*** 0.122* (0.122) (0.128) (0.010) (0.011) (0.044) (0.037) (0.074) (0.111) Log(Total Coverage) t * (0.095) (0.103) (0.007) (0.008) Log(Facebook Shares) t (0.038) (0.037) (0.003) (0.003) Leverage t * ** (0.617) (0.505) (0.201) (0.207) (0.359) (0.507) Log(Property, Plant, Equip.) t * *** 0.244*** 0.300*** 0.234*** (0.059) (0.066) (0.024) (0.027) (0.039) (0.053) Constant 7.156*** 3.878*** ** *** *** 1.913** 3.246*** (1.720) (0.910) (0.138) (0.145) (0.496) (0.379) (0.785) (0.878) Observations R-squared

136 Table 34: Mainstream vs. non-mainstream media (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES Tobin s Q Main. = 1 Tobin s Q Main. = 0 ROA Main. = 1 ROA Main. = 0 Pct. Indep. Main. = 1 Pct. Indep. Main. = 0 Mutual Funds Main. = 1 Mutual Funds Main. = 0 Tobin's Q t ** *** *** (0.009) (0.004) (1.372) (0.732) ROA t *** *** *** (0.155) (0.073) (23.983) (11.185) Leverage t *** *** ** *** ** ** (0.363) (0.270) (0.031) (0.020) (0.052) (0.035) (6.294) (4.679) Duality t ** *** 0.032*** * (0.107) (0.080) (0.010) (0.010) (0.013) (0.009) (1.861) (1.473) Log(Board Size) t ** 0.101*** 0.085*** *** *** (0.237) (0.177) (0.024) (0.014) (0.036) (0.022) (4.687) (3.038) Hedge Funds t * (0.024) (0.024) (0.002) (0.002) (0.004) (0.003) Mutual Funds t *** * *** (0.005) (0.003) (0.000) (0.000) (0.001) (0.000) Log(Property, Plant, Equipment) t *** *** *** (0.046) (0.031) (0.004) (0.003) (0.006) (0.004) (0.812) (0.557) Monthly Volatility t ** ** ** *** (0.991) (0.758) (0.086) (0.065) (0.081) (0.054) (14.747) (9.548) RDA t *** 3.692*** (0.962) (0.939) (0.080) (0.053) (0.124) (0.078) (18.301) (11.166) Log(Positive Score) t ** 0.011** * (0.087) (0.063) (0.007) (0.005) (0.009) (0.007) (1.520) (1.042) Log(Negative Score) t * * 0.017** * (0.085) (0.062) (0.008) (0.005) (0.009) (0.008) (1.639) (1.174) Log(Total Coverage) t ** ** * (0.078) (0.058) (0.006) (0.004) (0.008) (0.005) (1.187) (0.758) Log(Facebook Shares) t ** *** (0.026) (0.019) (0.002) (0.002) (0.004) (0.002) (0.521) (0.335) Pct. Independent t ** (0.446) (0.349) (0.036) (0.040) (8.335) (6.761) Constant 4.490*** 3.665*** 0.106* *** 0.666*** *** *** (0.689) (0.611) (0.063) (0.049) (0.090) (0.094) (13.113) (9.419) Observations R-squared

137 Table 34:Mainstream vs. non-mainstream (continued) (9) (10) (11) (12) (13) (14) (15) (16) VARIABLES Hedge Funds Main. = 1 Hedge Funds Main. = 0 Leverage Main. = 1 Leverage Main. = 0 Total Coverage Main. = 1 Total Coverage Main. = 0 FB Shares Main. = 1 FB Shares Main. = 0 Tobin's Q t * *** 0.151** 0.106*** * (0.111) (0.073) (0.008) (0.006) (0.063) (0.037) (0.080) (0.067) ROA t * * ** ** (1.875) (1.373) (0.146) (0.101) (0.976) (0.434) (1.248) (1.226) Pct. Independent t * * (1.187) (0.645) (0.091) (0.056) (0.375) (0.245) (0.706) (0.526) Duality t (0.225) (0.141) (0.016) (0.012) (0.079) (0.050) (0.160) (0.115) Log(Board Size) t *** *** 0.096** 0.070** (0.607) (0.326) (0.040) (0.027) (0.204) (0.136) (0.387) (0.284) Hedge Funds t *** * * (0.004) (0.004) (0.022) (0.016) (0.038) (0.039) Mutual Funds t ** 0.011*** 0.008*** 0.032*** 0.034*** (0.001) (0.000) (0.003) (0.002) (0.006) (0.005) Monthly Volatility t *** 3.990** ** (2.399) (1.643) (0.118) (0.126) (0.596) (0.417) (1.472) (1.683) RDA t ** *** (1.846) (1.153) (0.127) (0.080) (0.614) (0.429) (1.300) (0.957) Industry Leverage t (0.439) (0.286) Log(Positive Score) t (0.144) (0.095) (0.012) (0.008) (0.053) (0.036) (0.110) (0.089) Log(Negative Score) t *** 0.138*** 0.214* 0.259*** (0.148) (0.097) (0.012) (0.009) (0.056) (0.035) (0.115) (0.088) Log(Total Coverage) t ** (0.113) (0.093) (0.009) (0.007) Log(Facebook Shares) t (0.053) (0.030) (0.005) (0.003) Leverage t ** *** (0.762) (0.520) (0.279) (0.153) (0.516) (0.438) Log(Property, Plant, Equip.) t * 0.299*** 0.221*** 0.288*** 0.263*** (0.083) (0.053) (0.036) (0.022) (0.062) (0.050) Constant *** 2.906*** ** *** 4.036*** 3.747*** (2.045) (0.975) (0.197) (0.134) (0.669) (0.521) (1.185) (1.149) Observations , ,150 R-squared

138 Table 35: Paywall vs. non-paywall VARIABLES (1) (2) (3) (4) (5) (6) (7) (8) Tobin ROA ROA Pct. Indep. Pct. Indep. Mutual Paywall =0 Paywall = 1 Paywall = 0 Paywall = 1 Paywall = 0 Paywall = 1 Tobin Paywall= 1 Mutual Paywall = 0 Tobin's Q t ** *** *** (0.006) (0.006) (1.058) (0.797) ROA t ** *** *** (0.095) (0.089) (16.163) (13.967) Leverage t *** *** *** ** *** (0.282) (0.330) (0.021) (0.025) (0.040) (0.043) (5.137) (5.884) Duality t *** 0.051*** ** (0.084) (0.093) (0.007) (0.012) (0.010) (0.012) (1.557) (1.725) Log(Board Size) t * ** 0.076*** 0.112*** *** *** (0.219) (0.194) (0.019) (0.019) (0.028) (0.027) (3.600) (3.607) Hedge Funds t (0.021) (0.025) (0.002) (0.002) (0.003) (0.003) Mutual Funds t ** *** *** (0.004) (0.004) (0.000) (0.000) (0.001) (0.001) Log(Property, Plant, Equipment) t *** *** ** (0.034) (0.040) (0.003) (0.004) (0.005) (0.005) (0.694) (0.646) Monthly Volatility t ** ** * * (1.094) (1.011) (0.064) (0.083) (0.063) (0.065) (20.161) (15.186) RDA t *** 3.920*** (0.919) (1.150) (0.049) (0.088) (0.081) (0.108) (13.396) (13.618) Log(Positive Score) t ** * (0.065) (0.075) (0.004) (0.009) (0.007) (0.010) (1.111) (1.452) Log(Negative Score) t *** (0.070) (0.073) (0.005) (0.008) (0.008) (0.010) (1.164) (1.520) Log(Total Coverage) t *** *** *** (0.049) (0.059) (0.004) (0.005) (0.006) (0.006) (0.874) (0.920) Log(Facebook Shares) t *** * *** ** ** (0.020) (0.024) (0.001) (0.002) (0.002) (0.003) (0.335) (0.430) Pct. Independent t *** ** (0.353) (0.420) (0.038) (0.046) (7.304) (7.670) Constant 4.339*** 3.512*** *** 0.775*** *** *** (0.712) (0.672) (0.061) (0.061) (0.112) (0.082) (12.529) (10.770) Observations R-squared

139 Table 35: Paywall vs. non-paywall (continued) (9) (10) (11) (12) (13) (14) (15) (16) VARIABLES Hedge Funds Paywall = 1 Hedge Funds Paywall = 0 Leverage Paywall = 1 Leverage Paywall = 0 Coverage Paywall = 1 Coverage Paywall = 0 FB Shares Paywall = 1 FB Shares Paywall = 0 Tobin's Q t *** *** 0.158*** 0.085** 0.155** (0.081) (0.100) (0.007) (0.006) (0.045) (0.034) (0.063) (0.098) ROA t ** *** (1.333) (1.969) (0.111) (0.119) (0.577) (0.502) (1.091) (1.827) Pct. Independent t *** (0.883) (0.759) (0.069) (0.068) (0.251) (0.310) (0.560) (0.773) Duality t (0.149) (0.198) (0.012) (0.014) (0.057) (0.059) (0.120) (0.181) Log(Board Size) t *** *** 0.085*** 0.070** (0.352) (0.515) (0.029) (0.034) (0.158) (0.136) (0.309) (0.420) Hedge Funds t ** 0.010** (0.004) (0.005) (0.018) (0.019) (0.039) (0.055) Mutual Funds t ** *** 0.010*** 0.033*** 0.042*** (0.001) (0.001) (0.002) (0.002) (0.005) (0.007) Monthly Volatility t *** 3.081* ** (2.253) (1.684) (0.123) (0.135) (0.498) (0.408) (1.199) (2.123) RDA t ** * 1.573*** (1.239) (1.554) (0.089) (0.102) (0.517) (0.493) (0.926) (1.395) Industry Leverage t (0.322) (0.369) Log(Positive Score) t (0.106) (0.138) (0.009) (0.011) (0.045) (0.038) (0.098) (0.147) Log(Negative Score) t *** 0.195*** 0.204** 0.274* (0.116) (0.137) (0.009) (0.011) (0.045) (0.034) (0.097) (0.143) Log(Total Coverage) t ** (0.101) (0.113) (0.007) (0.008) Log(Facebook Shares) t (0.036) (0.035) (0.003) (0.003) Leverage t * 1.354* * *** (0.513) (0.714) (0.172) (0.191) (0.406) (0.634) Log(Property, Plant, Equip.) t *** 0.198*** 0.327*** 0.292*** (0.062) (0.069) (0.027) (0.022) (0.049) (0.068) Constant 5.899*** 5.657*** *** ** 2.586*** 4.168*** (1.834) (1.808) (0.152) (0.158) (0.796) (0.476) (0.953) (1.525) Observations R-squared

140 Simultaneous with Instrumental Variable Regression Tables Table 36: IV models VARIABLES (1) (2) (3) (4) (5) (6) (7) (8) (9) Tobin s Q Tobin s Q ROA ROA ROA Pct. Indep. Pct. Indep. 2SLS 3SLS LSDV 2SLS 3SLS LSDV 2SLS Tobin s Q LSDV Pct. Indep 3SLS Tobin's Q t *** ** *** (0.004) (0.029) (0.019) ROA t *** (0.068) (0.297) (0.187) Leverage t *** * *** *** (0.194) (3.922) (3.019) (0.019) (0.321) (0.270) (0.024) (0.308) (0.221) Duality t ** 1.398** ** 0.125** 0.039*** 0.121*** 0.146*** (0.058) (0.913) (0.545) (0.006) (0.075) (0.052) (0.007) (0.027) (0.023) Log(Board Size) t ** ** 0.028** * ** 0.089*** (0.135) (1.354) (0.921) (0.013) (0.111) (0.084) (0.017) (0.051) (0.046) RDA t *** *** *** ** 3.212*** ** 3.810*** (0.490) (16.456) (9.825) (0.048) (1.349) (0.937) (0.061) (1.106) (0.852) Hedge Funds t * (0.016) (0.412) (0.314) (0.002) (0.034) (0.028) (0.002) (0.023) (0.019) Mutual Funds t ** * * *** *** *** (0.003) (0.084) (0.054) (0.000) (0.007) (0.005) (0.000) (0.002) (0.002) Log(Property, Plant, Equipment) t *** (0.024) (0.187) (0.127) (0.002) (0.015) (0.012) (0.003) (0.012) (0.009) Monthly Volatility t *** *** * (0.400) (1.915) (1.754) (0.039) (0.157) (0.146) (0.049) (0.121) (0.110) Log(Positive Score) t *** (0.056) (0.379) (0.305) (0.006) (0.031) (0.026) (0.007) (0.020) (0.018) Log(Negative Score) t *** (0.057) (0.301) (0.279) (0.006) (0.025) (0.023) (0.007) (0.018) (0.017) Log(Facebook Shares) t *** *** (0.017) (0.114) (0.036) (0.002) (0.009) (0.005) (0.002) (0.007) (0.005) Log(Total Coverage) t *** ** *** 0.006** * *** ** *** (0.023) (0.222) (0.094) (0.002) (0.018) (0.011) (0.003) (0.015) (0.010) Pct. Independent t * ** ** (0.268) (11.151) (6.639) (0.026) (0.914) (0.638) Constant 3.986*** * ** ** 0.671*** 0.938*** 0.894*** (0.657) (12.572) (8.037) (0.065) (1.031) (0.753) (0.078) (0.211) (0.182) Observations 1, , ,

141 Table 36: IV models (continued) VARIABLES (10) (11) (12) (13) (14) (15) (16) (17) (18) Hedge Funds Hedge Funds Hedge Funds Mutual Funds Mutual Funds Mutual Funds Lev. Lev. Lev. LSDV 2SLS 3SLS LSDV 2SLS 3SLS LSDV 2SLS 3SLS Tobin's Q t *** * *** *** *** (0.063) (0.291) (0.253) (0.616) (3.274) (2.998) (0.005) (0.021) (0.017) ROA t *** ** ** (1.050) (2.727) (2.413) (10.282) (30.701) (28.960) (0.082) (0.217) (0.177) Leverage t ** *** ** (0.370) (2.998) (2.672) (3.624) (33.760) (30.407) Duality t (0.110) (0.346) (0.312) (1.076) (3.892) (3.604) (0.009) (0.064) (0.048) Log(Board Size) t *** * ** *** * *** (0.252) (0.528) (0.488) (2.467) (5.950) (5.567) (0.018) (0.094) (0.072) RDA t *** * ** (0.941) (12.124) (10.685) (9.209) ( ) ( ) (0.073) (1.184) (0.925) Hedge Funds t *** (0.002) (0.027) (0.022) Mutual Funds t ** (0.000) (0.006) (0.005) Log(Property, Plant, Equipment) t ** *** *** (0.047) (0.115) (0.105) (0.456) (1.299) (1.183) Monthly Volatility t *** ** (0.755) (1.220) (1.144) (7.388) (13.732) (12.981) (0.059) (0.095) (0.089) Log(Positive Score) t * * (0.106) (0.169) (0.160) (1.038) (1.900) (1.811) (0.008) (0.021) (0.018) Log(Negative Score) t (0.108) (0.171) (0.164) (1.053) (1.923) (1.845) (0.008) (0.017) (0.015) Log(Facebook Shares) t * ** *** ** ** 0.005** 0.010* (0.030) (0.065) (0.055) (0.291) (0.730) (0.672) (0.003) (0.005) (0.004) Log(Total Coverage) t *** * * (0.043) (0.157) (0.136) (0.425) (1.770) (1.600) (0.003) (0.014) (0.011) Pct. Independent t ** ** ** *** *** (0.502) (2.433) (2.229) (4.911) (27.395) (25.632) (0.040) (0.722) (0.540) Industry Leverage t * (0.241) (1.318) (0.992) Constant 5.770*** *** *** *** *** *** (1.225) (2.900) (2.657) (11.991) (32.651) (30.588) (0.107) (0.794) (0.599) Observations 1, , ,

142 Table 36: IV models (continued) VARIABLES (19) (20) (21) (22) (23) (24) FB Shares FB Shares Total Coverage Total Coverage 2SLS 3SLS LSDV 2SLS FB Shares LSDV Total Coverage 3SLS Tobin's Q t *** *** (0.050) (0.281) (0.235) (0.040) (0.501) (0.188) ROA t * (0.847) (3.185) (2.695) (0.673) (5.678) (1.807) Leverage t ** ** (0.298) (3.061) (2.696) (0.237) (5.456) (3.808) Duality t ** 2.269*** 1.809*** (0.088) (0.442) (0.399) (0.070) (0.789) (0.612) Log(Board Size) t ** ** (0.205) (0.699) (0.642) (0.163) (1.246) (1.048) RDA t ** ** 2.996*** *** *** (0.748) (5.946) (5.611) (0.595) (10.600) (9.431) Hedge Funds t ** ** *** ** (0.024) (0.262) (0.227) (0.019) (0.468) (0.359) Mutual Funds t *** *** (0.004) (0.049) (0.043) (0.003) (0.087) (0.064) Log(Property, Plant, Equipment) t *** 0.230** *** 0.525*** 0.422*** (0.032) (0.107) (0.096) (0.025) (0.190) (0.141) Monthly Volatility t (0.609) (1.180) (1.129) (0.484) (2.103) (2.006) Log(Positive Score) t ** ** (0.084) (0.238) (0.216) (0.066) (0.424) (0.353) Log(Negative Score) t *** (0.085) (0.192) (0.183) (0.068) (0.343) (0.322) Log(Facebook Shares) t-1 Log(Total Coverage) t-1 Pct. Independent t *** * ** *** ** ** (0.402) (6.044) (5.401) (0.320) (10.775) (7.978) Constant 5.125*** * *** *** * (0.999) (7.445) (6.658) (0.794) (13.274) (9.878) Observations 1, ,

143 Appendix B: Figures 56 Twitter Media Market Analysis Figure 1: New York Times and Wall Street Journal media market overlap with other sources from this analysis (in millions) 56 All visualizations presented in this analysis are in color. Readers should refer to the online copy for Figures

144 Figure 2: Media market overlap (in millions) 140

145 Media Market Figures Sentiment Distributions Figure 3: Sentiment score distribution across media organization 141

146 Figure 4: Sentiment score segmentations 142

147 Daily Posts Figure 5: Daily posts by media organization 143

148 Figure 6: Daily posts segmentation 144

149 Daily Share Per Post Figure 7: Daily shares per post by media organization 145

150 Figure 8: Daily shares per post segmentation 146