It is not bad management if you do not get caught: Evidence
on the relationship between CEO power and the information
environment.
July 2015
by
Sebastiaan Thuis
University of Amsterdam
MSc Business Economics, Finance track Master Thesis
Thesis supervisor: Florian Peters Student No.: 6055818
Statement of Originality
This document is written by Sebastiaan Thuis who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgements
This thesis would not have been possible without the support and help from others. First and foremost I would like to thank dr. F.S. Peters, my thesis supervisor, for his advice, guidance and insights. I am grateful for his contributions, which allowed me to further enhance the quality of this study and inspired me to think out of the box. Additionally I would like to thank my parents and brothers, for supporting me during my studies and this thesis in particular.
Abstract
One of the most salient agency-principal relationships is the one between the CEO, board and investors. Although CEO power most likely plays a role with regard to this relationship, there has been relatively little research on this topic. This paper examines the effects of CEO power on the quality and quantity of information that finds its way to investors. Thereby examining whether CEOs entrench themselves by hindering monitoring and how CEO power affects the information environment of investors in general. So far there have been only two studies on this subject. Both report conflicting results. By adopting a broader set of proxies for the information environment, different instrumental variables and a more expansive model, we are able to shed more light on this topic. More powerful CEOs will only engage in the manipulation of the information environment if the market is able to discipline these managers for unwanted behavior. Additionally evidence if found that too little CEO power can lead to more difficulty in assessing future company performance. This is hypothesized to be caused by subordinate opposition or dissent, which can affect the CEOs’ ability to direct the company. As a consequence analysts and market makers are argued to have more difficulty in assessing company strategy and performance. However these conclusions should be interpreted with prudence, the findings of causality are not robust to different methods and the instruments are not completely exogenous. This is the first study that, to our knowledge, finds robust evidence on the effects of CEO power with regard to the information environment. Additionally, evidence is found that CEOs will indeed empower themselves through the structural decrease in monitoring of investors. Furthermore, it nuances previous empirical studies that argue that a better corporate governance will improve the information available to investors. Lastly, it indicates that CEO power in some cases should be examined with non-linear models.
1
Table of Contents
2 Introduction ... 6 3 Literature Review ... 8 3.1 Definitions... 8 3.2 Literature Review ... 9 3.3 Relevancy ... 14 4 Methodology ... 16 5 Data ... 23 6 Results ... 26 6.1 Endogenous Regressions ... 266.2 IV-regressions & Delta of Dependent variables ... 33
6.3 Interpretation ... 35 7 Robustness checks ... 36 7.1 CEO power ... 36 7.2 Disclosure Quality ... 39 7.3 Company Forecasts ... 41 8 Conslusion ... 42 8.1 Discussion ... 42
8.2 Summary & Findings ... 43
9 Bibliography... 47
2
Introduction
Information is arguably one of the most important commodities in the financial markets. With many investors trying to determine the true value of companies and the stocks or bonds that go with it, information effectively determines almost all aspects of the capital markets. This information is not only used in trading decisions, but is also an important tool in monitoring the behavior of the managers. The chief executive officer, as the most senior manager, holds the ultimate responsibility for maximizing shareholder value. Being able to evaluate whether the CEO indeed pursues this goal is not in the least dependent on the information that is available to investors. However, being transparent might not be in the interests of the CEO. A large amount of transparency can inhibit his or her ability to extract personal rents or engage in activities that are self-serving at the expense of shareholders. This study examines whether more powerful CEOs will manipulate the information flow to investors more than their weak counterparts. Moreover, inspecting this relationship will also shed light on whether CEOs entrench themselves by maintaining a more opaque information environment and thus inhibit monitoring from investors.
Previous empirical studies have shown that CEOs in some instances will indeed manipulate the information that finds its way to investors. Examples of this are the casting of conference calls with analysts (Cohen, Lou & Malloy, 2012), earnings management (Bergstresser & Philippon, 2006) and the timing of company forecasts (Aboody & Kasnik, 2000). However these studies only show that the CEO will manipulate the information environment if this leads to a direct financial gain. This study extends on these findings by studying whether CEOs will use their power to obscure the information environment structurally. Illustrating whether the information environment of investors is dependent on the degree of CEO power and whether CEOs will purposely try to increase information asymmetry to inhibit monitoring and market discipline. To our knowledge there have been only two studies on this subject and both found contradictory evidence (Liu & Jiraporn, 2011; Jiraporn, Liu & Kim, 2014). One study argues that more powerful CEOs will maintain a more opaque information environment, to prevent market discipline (Liu & Jiraporn, 2011). While the other argues that more powerful CEOs do not need to obscure the information environment because their power prevents them from being disciplined (Jiraporn, Liu & Kim, 2014). This contradiction alongside the lack of a broad investigation of multiple information environment proxies warranted a further examination of this topic.
Understanding the effects of CEO power is relevant for investors and policy makers. Identifying both the upsides and downsides of power, could help in mitigating the general agency costs that arise from separating ownership from control. Not only could downsides be mitigated by resolving the negative symptoms of CEO power, but it could also lead to
insights on what constitutes an optimal amount of power to be given to the CEO. Aside from its relevance with regard to corporate governance, it could also contribute to academics. Currently there have been no theories of mathematical models incorporating the effects of CEO power. Empirical evidence can help in identifying aspects that need to be incorporated in future models to better reflect reality.
The relationship between CEO power and the quality and quantity of information that is available to investors (the information environment) is studied by examining panel data, which is gathered from multiple datasets supplied by the University of Amsterdam. This results in a sample that covers 11823 observations, 3738 CEOs and 2342 companies in the United States from 1993 to 2013. CEO power is measured by a novel proxy developed by Bebchuk, Cremers and Peyer (2011), who validate that the proportion of CEO compensation compared to the payment to the top five executives will capture the influence of the CEO. A robustness study is done on the CEO power index developed by Adams, Almeida & Ferreira (2005). The information environment is measured through multiple proxies, which have been identified by previous literature as valid measures. These include analyst dispersion, the bid-ask spread, discretionary accruals, disclosure quality and whether or not a CEO will start voluntary management forecasts.
This study finds evidence that CEOs will engage in more manipulation of the information environment if there are incentives to do so. In companies with poor corporate governance, powerful CEOs have no incentives to manipulate the information environment, because the CEO faces no negative consequences from performing poorly or engaging in unwanted behavior. In companies with strong corporate governance the manipulation of the information environment does take place, possibly to prevent investors from being alarmed and trying to discipline the CEO. This corresponds with the idea that CEOs will obscure the information environment in order to entrench themselves. Therefore evidence is found in favor of the ‘quiet life hypothesis’ as proposed by Jiraporn, Liu and Kim (2014) and Armstrong, Balakrishnan and Cohen (2011).
Additionally, evidence is found that indicates that CEO power can also be beneficial
if the CEO is weak. An U-shaped curve is found between CEO power with regard to analyst
dispersion and the bid-ask spread. Together with the exponential relationship between CEO power and discretionary accruals leads us to believe that a certain degree of CEO power might be beneficial in estimating future company performance. A hypothesis is formulated that argues that when CEOs are relatively weak, the opposition of subordinates increases. This increased dissent within the top executive team, might make it harder to predict future company policies and strategies, thus making it harder to assess company performance. Although evidence aligns with this hypotheses, the time frame did not allow to exclude other possibilities. These findings should be interpreted with prudence. Although causal evidence is
found for the general relationship between CEO power and the information environment, the causal results were not robust to different methods that test for causality. This could be caused by the non-linearity of the relationship, but more research is needed before strong conclusions can be made. Furthermore, the interaction terms were not tested for causality. Theory does not raise any cause for concern with regard to reverse causality, but a better examination is warranted.
This thesis is organized as follows. Section 2, examines the definitions of CEO power and discusses the relevant literature and theory on this topic, while also further deepening our contribution to scientific literature. Section 3, presents the methodology on how this topic is studied. This includes the exact specification of the variable construction and how to final model is specified. Section 4, offers the specifics of the data and the choices that were made with regard to the calculation the information proxies. Section 5, reviews the results of the initial regressions, their economic significance and relation to other scientific findings. Section 6, discusses the results of the robustness studies. Section 7, concludes this study by arguing possible drawbacks and the implications that can be derived from the previous results.
3
Literature Review
3.1 Definitions
Both CEO power and the information environment for investors are broad and ambiguous terms. To prevent any confusion or misinterpretation through the use of these terms, we shall shortly discuss their definitions. These definitions might deviate slightly from their official meaning, with the goal of being able to explain concepts more concisely and precisely. For
CEO power, the definition by Finkelstein (1992) is adopted. He argues that CEO power is the
degree to which the CEO is able to exert his or her will over subordinates and an organization as a whole. This effectively means that power is a sum of many different forms of influence rather than a specific concept such as social capital, leadership, charisma, power from legal rights and authority.
It is substantially harder to define information environment, because information itself is substantially more ambiguous than power. Edmunds et al. (2000) argued that
information can be interpreted in multiple ways and that this could lead to general confusion
and seemingly irreconcilable theories and findings. We define the information environment surrounding a stock as the quality, quantity and comprehensibility of the information that is easily available to investors. Concretely, this means that all of the following contribute to a
poorer information environment: wrong information, manipulated information, too little information, information that is too complex, and too much irrelevant information. Although this seems all encompassing, the concept of information uncertainty is excluded. Information
uncertainty is defined as the degree to which things are uncertain to both insiders and
outsiders of a company. Examples include risk or uncertainty about immeasurable aspects, such as the happiness of the employees.
The information environment overlaps with information asymmetry between the shareholders, the board and the CEO. Information asymmetry is the gap of information between the agent and principal. An increase in this gap could be achieved through an increase in quantity, comprehensibility, usability or quality of the information that is available to the CEO or through a decrease of these aspects with regard to the information of investors. Many previous studies focus on information asymmetry but use the information environment as a proxy for information asymmetry (e.g., Moeller, Schlingemann and Stulz, 2007; Thomas, 2002; Krishnaswami and Subramaniam, 1999). Our results might complement empirical research that has been done on information asymmetry; but we consider these terms to be different.
3.2 Literature Review
Literature on the effects of CEO power is intrinsically tied to corporate governance. Although general corporate governance literature does not provide any theories or models of the effects of CEO power, predictions can be inferred from the concepts that lie at the heart of this field. One of the most relevant theories is ‘Agency Theory’, which was first introduced simultaneously and independently by Stephen Ross (1973) and Barry Mitnick (1973). The theory claims that, when work is delegated to another party (the agent), costs will be incurred for the party that delegates the work (the principal). These costs can arise because of information asymmetry, incomplete contracts, the price of monitoring and different risk attitudes (Eisenhardt, 1989). An important factor in the agent-principal relationship is the degree of information asymmetry between both parties. This theory shows the importance of information for aligning the goals of the agent. When there is no information asymmetry or when there are no costs involved with monitoring the agent, the agency costs dissipate.
A typical principal-agent relationship is the one between the owner and manager, which matches closely with the relationship between the CEO, board and shareholders. In this theoretical relationship, the owner signs a contract with the manager. This contract specifies what the manager should do with the funds that are attributed to him and how he should distribute the returns that come from his activities (Shleifer and Vishny, 1997; Eisenhardt,
1989). Traditionally, this was assumed to be a complete contract, which is a contract that perfectly describes the right course of action in all future states of the world. Grossman and Hart (1986) contend this claim and propose a theory of incomplete contracts. By replacing complete contracts with incomplete contracts, a situation is created in which there are residual control rights (Grossman and Hart, 1986; Hart and Moore, 1990). These residual control
rights are the control rights in situations that are not covered by the contract. Ownership
implicitly means the ownership of these rights; however, as Shleifer and Vishny (1997) argue, the opposite will most likely be in the interest of the owners. If the owners own all the residual rights, they need to make a decision every time something unexpected happens. Since they do not have the expertise and information necessary to make all of these decisions, however, the residual rights will most likely be given to the manager. Owners could impose certain limits on the manager’s discretion; however, this would only decrease manager discretion partially. A large strand in corporate governance literature deals with these restrictions in behavior (Shleifer and Vishny, 1997). However, restrictions still allow a certain degree of freedom for self-interested managers to expropriate shareholder wealth and pursue other opportunistic behavior (Shleifer and Vishny, 1997).
Another addition to agency theory is contract theory, and more specifically the concept of moral hazard. The term originates from literature on insurance, where it describes the increase in risky behavior that occurs when people are insured (Pauly, 1968; Arrow, 1971; Spence and Zeckhauser, 1971). The concept was quickly applied to other contractual relationships such as the principal-agent relationship (Harris and Raviv, 1979; Holmström, 1979). Holmström (1979) modelled this relationship and predicted that imperfect proxies of measuring action plus outcome-based contracts could alleviate some of agency costs. The general conclusion is that any additional information that reveals the decision quality or effort by the agent is valuable for the principal and could help him in aligning the goals of the agent. This notion demonstrates the value of information in agent-principal relationships. More information about the actions of the agent leads to a decrease in agency costs.
These theories alone do not carry any specific predictions for the effects of powerful CEOs on the information environment. However, the intuition behind these ideas do provide a direction. If increased information would increase the ability of the shareholder to monitor the CEO, it might impair the CEO and his or her ability to expropriate wealth or act opportunistically. Therefore, self-interested CEOs have an incentive to limit the (relevant) information that finds its way to shareholders. We argue that relatively powerful CEOs will have more influence over the information that finds its way to shareholders and would therefore opt to decrease quality and quantity of (relevant) information flow more. It is important to note that the incentives to decrease the information flow can take many forms. For instance, a CEO might opt to release less information or manipulated information about a
company’s financial performance so as to hide his or her poor performance. Or the CEO might engage in earnings management to meet or beat the analyst forecasts and secure bonuses.
Empirical evidence on this specific relationship is slim: only two studies examine this topic indirectly. The findings of these studies oppose each other, even though the methodology is very similar. Liu and Jiraporn (2010) examine the effects of a CEO power proxy called the CEO pay slice on bond ratings and yields. The CEO pay slice measures the proportion of CEO salary in relation to the salaries of the top five executives (Bebchuk et al, 2011). They found that more powerful CEOs result in a higher cost of debt. The paper further studies this effect by examining possible causes. Since firms with more powerful CEOs have higher bid-ask spreads, the authors argue that these CEOs maintain a more opaque information environment to extract more personal rent (Liu et al, 2010). However, this finding could be debunked by contending that it is due to correlation rather than to causation. Theory holds that CEO power can both cause and be caused by a poor information environment. Too little information impairs the ability of shareholders to monitor, and more powerful CEOs can weaken shareholders’ monitoring ability.
The paper by Jiraporn, Kim and Liu (2014) builds on the findings of the previous study. The authors consider whether more powerful CEOs will result in less analyst coverage, a higher bid-ask spread and a higher probability of insider trading. Contrary to the previous study, an inverse relationship is discovered between CEO power and these variables. The authors argue that this can be attributed to the fact that more powerful CEOs will also be more entrenched and are less easily punished for wrongdoing. Although an IV-regression is used in this study, neither instrument is necessarily exogenous in this context. The first instrument is adopted by the study of Bebchuk et al (2011), which considers the average power of CEOs in the industry. When an industry is characterized by high information asymmetry, more CEOs will be powerful. In this case, the information is not entirely exogenous. The other variable is CEO age might capture some aspects of experience, rather than measuring the probability of the CEO wanting to retire and decrease his or her importance for the organization.
Moreover the latter evidence is opposed by circumstantial evidence. Shivdasani et al. (1999) find that the appointment of an independent director with CEO involvement garners less positive stock market reactions than independent directors that are appointed without involvement. This indicates that CEOs are possibly prone to manipulate shareholders’ monitoring ability by appointing independent directors that might collude with him or her. Moreover, CEOs have been shown to manipulate the information supply for direct personal gain. CEOs have been shown to manipulate analysts before insider trading (Cohen et al, 2012), use earnings management before insider trading (Beneish et al., 2002) and use
earnings management to meet analyst forecasts in order to reap higher bonuses and avoid getting fired (Mande et al., 2012). Although these studies do not show that relatively powerful CEOs engage in these activities more, it does show that CEOs in general have incentives to engage in the manipulation of the information environment. The contradictory evidence on the effects of CEO power and the general evidence on the CEOs that manipulate the information environment leads us to believe that relatively powerful CEOs will decrease the information environment of investors.
H1: Firms with relatively powerful CEOs maintain a more opaque information environment in the financial markets.
Information on companies is not used only to align the goals of management and shareholders. Many market participants use the same information to evaluate trade decisions or act as information intermediaries to lure clients. Although there are many theories about how the markets are affected by information (e.g., the efficient market hypothesis), models about how the information environment can be measured are not as widespread.
The paper by Barron et al. (1998) offers a model, which shows that the behavior by financial analysts can be used to study the information environment. Financial analysts, or sell-side equity analysts, publicly announce their forecasts on the future performance of a company, to attract clients for their brokerage. Barron et al. (1998) theorize that analysts process public information to form a private signal (or opinion) about the future earnings per share. Errors within this private signal could be caused by a mistake in the public information (common error) or by the processing of the information (idiosyncratic error). The authors theorize that analyst dispersion (the degree to which analyst forecasts differ) is a function of idiosyncratic errors, whereas forecasts errors are a function of both. Both errors are argued to be a product of a poor information environment. A common error is the result of faulty information; whereas the idiosyncratic error is a function of the comprehensibility, complexity and vagueness of the information environment. For example, the CEO might manipulate public signals, which will result in a larger common error. Or the CEO might pursue more opaque company financials, which makes predicting company performance harder, thus leading to more processing errors.
The model of Barron et al. (1998) does not account for two aspects that might bias this measure of information uncertainty. The first is herding. Scharfstein and Stein (1990) model the behavior of managers and show that managers might choose to forego their personal inclinations in favour of mimicking the investment decisions of other managers to rationally enhance their reputations. A similar argument is made by Trueman (1994), which models herding behavior under analysts. Inexperienced analysts are motivated to herd in
forecasts because of career concerns (Hong et al., 2000). Lee (Lee, 2007) argues that uncertainty could aggravate herding behavior and therefore bias the effects of analyst dispersion and forecast errors. Although an opposing effect might be present, we argue that analyst dispersion might still function as a decent proxy for information asymmetry. The reason for this is that analyst dispersion is widely used in the literature as a proxy for the information environment, and that this method has been validated through several studies (Barron and Stuerke, 1998; Diether, Malloy, and Scherbina, 2002; Imhoff and Lobo, 1992; Lang and Lundholm, 1996; Moeller, Schlingemann and Stulz, 2007; Thomas, 2002). Moreover, we argue that uncertainty is something different than the quality of the information environment. A poor information environment might make it harder to assess the future company performance, which could lead to more analyst herding. Conversely, the difficulty in assessing future performance might actually increase the anchoring of their private signal and thus decrease herding. The latter has been shown to be the case among investors (Daniel, Hirshleifer and Subrahmanyam, 1998).
Secondly, the model by Barron et al. (1998) might be biased in the context of researching CEO power. CEOs have been shown to have substantial incentives to meet the analyst consensus: e.g., they wish to acquire larger bonuses and avoid being fired (Matsunga and Park, 2001; Farrell and Whidbee, 2003). This is further strengthened by the fact that firms with more powerful CEOs have been shown to meet or just beat analyst consensus more often (Mande et al. 2012). The direct empirical evidence against the validity of analyst forecast errors inhibits the use of the model proposed by Barron et al. (1998). Therefore, we opt to measure analyst dispersion rather than a variable that incorporates both aspects. Additionally, both the analyst dispersion before the earnings announcement and after the earnings announcement are researched, because each variable might capture different aspects of the information environment. Analyst dispersion after the earnings announcement might be a stronger function of disclosure quality than analyst dispersion before the earnings announcement, while the latter might be a stronger function of the general information environment (i.e., news, internet, etcetera).
H1.1: Analyst forecast dispersion will be larger for firms with more powerful CEOs.
Another market participant that is dependent on information is comprised of market makers. Kyle (1985) models the market liquidity supplied by market makers. This model demonstrates that the amount and the precision of public information are positively related to market liquidity and thus result in a lower bid-ask spread. Market makers post prices for stocks on the exchange in order to provide investors with the option to sell and buy shares with greater ease. The bid-ask spread is the profit they generate from actively trading; it
compensates the market makers for the risks they are taking. When a company becomes harder to appreciate, selling and buying shares will naturally become more risky. This then results in a larger bid-ask spread and smaller share debt or less market liquidity. There are several studies that validate this measure of the information environment. Roulstone (2003) finds that more analyst coverage will lead to more market liquidity (smaller bid-ask spread), while more analyst dispersion leads to a lower market liquidity. The measure is further validated by other studies (Coller and Yohn, 1997; Amiran, Owens and Rosenbaum, 2013).
H1.2: The bid-ask spread will be larger for firms with more powerful CEOs.
Aside from studying the information environment indirectly through the behavior of market agents, we will also study the information environment directly. One of the best-researched ways in which management can mislead investors and other stakeholders is through earning management. In general, this constitutes the manipulation of earnings to extract personal gain. Although we believe that there are many other ways in which management and CEOs can influence the information environment, we feel that investigating the information environment directly could serve as a robustness check for our study.
Previous research has identified strong incentives for CEOs to manipulate earnings and the financial statement. Earnings management is often exercised right before trades by management (Sawicki et al, 2008). It is used to attain certain performance measures (Bergstresser et al., 2006), to maintain their social status (Malmendier et al., 2008) and to achieve higher bonuses (Matsunaga et al., 2001). These findings seem to indicate that more powerful CEOs will engage in more earnings management. The study by Mande et al. (2012) finds evidence for this hypothesis. The study finds that firms with more powerful CEOs are more likely to just meet the analyst consensus or to outperform it by one cent. This leads to our last hypothesis.
H1.3: Companies with more powerful CEOs engage in more earnings management.
3.3 Relevancy
A substantial portion of the corporate governance literature deals with the restrictions of manager discretion and the mitigation of agency costs in general (Shleifer and Vishny, 1997). CEO power has received relatively little attention within this field, possibly because of the
complexity of measuring and quantifying power (Pfeffer, 1981). Finkelstein (1992) was one of the first researchers to propose a number of measures that quantify the degree of influence of the chief executive; however, this influence remained hard to measure and the methods contained considerable flaws. The topic gained new attention with the introduction of the proxy developed by Bebchuk et al (2011). Rather than seeing salary as a way to mitigate agency costs, it was argued and shown to reflect agency costs (Bebchuk, 2003; Bebchuk et al., 2011). By measuring the proportion of salary that is paid to the CEO relative to the top five executives, the authors argued for measuring both the CEOs ability to extract personal rents and the importance of the CEO within the organization. This CEO pay slice (CPS) was shown to be linked with many indicators of agency costs, such as lower firm value, lower acquisition returns, lucky options grants, greater tendency to reward the CEO for luck, lower CEO turnover and lower variability of stock returns. Later research identified higher costs of debt (Liu et al., 2010), cost of equity (Chen et al., 2013) and meeting analyst consensus (Mande et al., 2012).
The evidence that more powerful CEOs will result in more agency costs is relatively robust. This study researches the effects of CEO power on the information environment, which has not been researched extensively before. Jiraporn, Liu and Kim (2014) focused on measuring the effects of powerful CEOs on analyst coverage, the bid-ask spread and adoption of the ask spread called the probability of insider trading (derived from bid-ask spread). Similarly the paper by Liu and Jiraporn (2010) examine the effects of CEO power on the bid-ask spread as a robustness check. The results from both papers contract each other, which warrants further research. With this study, we hope to resolve this contradiction by studying the information environment more extensively. Aside from examining the bid-ask spread, we also explore analyst dispersion and earnings management, disclosure quality and voluntary company forecasts. No measures except the bid-ask spread have been researched in this context before.
The results of this study could contribute to our understanding of how CEOs manipulate investors and how CEO power influences this manipulation; they could also offer insights into how to mitigate these agency costs through policy changes. Additionally, this is the first study to research CEO power in a non-linear fashion. The different aspects of CEO power might interact so as to create a situation in which additional power increases exponentially with regard to agency costs. Moreover, previous research relied on two instruments proposed by Bebchuk et al. (2011) to prove causality. This paper introduces another possibly exogenous instrument. Lastly, we add to the scientific literature by exploring another possible cause of a poorer information environment.
4
Methodology
Analyst dispersion has been widely adopted as a proxy for the quality of the information environment. A number of different approaches exist for measuring analyst dispersion. This study adopts the approach of Zhang (2006). Zhang measures analyst dispersion as the standard deviation of forecasts scaled by a company’s end-year stock price. This approach is supplemented by that of Hirshleifers et al. (2009), which excludes forecasts that have not been reviewed in the last 60 days. This means that all forecasts that have been reviewed 60 days prior to the announcement date will be included to calculate the dispersion. When multiple forecasts of one analyst are reviewed in the last 60 days, the most recent one is adopted. The basic formula for calculating analyst dispersion after the earnings announcement remains the same. The criterion for the inclusion of forecasts is that the forecasts must be posted at least 60 days after the earnings announcement. If there are multiple active forecasts, the last one is used. The ‘i’ is a company identifier and the ‘y’ stands for year.
𝐴𝑛𝑎𝑙𝑦𝑠𝑡 𝑑𝑖𝑠𝑝𝑒𝑟𝑠𝑖𝑜𝑛 𝑖,𝑦=
𝑆𝑡. 𝐷𝑒𝑣. (𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡𝑠𝑖,𝑦) 𝑠𝑡𝑜𝑐𝑘𝑝𝑟𝑖𝑐𝑒 𝑦𝑒𝑎𝑟 𝑒𝑛𝑑𝑦,𝑖
The bid-ask spread is another important indicator of a poor information environment. CRSP reports the highest and lowest prices for each day, which are reported as the bid and ask prices for stocks. Corwin and Schultz (2012) argue that this is incorrect because this bid-ask spread also includes stock variance as opposed to just the bid-ask spread. The paper presents a novel high-frequency proxy for the bid-ask spread, which is empirically shown to be a better estimator of the real bid-ask spread than other measures such as the covariance estimator by Roll (1984) and the measure proposed by Lesmond, Ogden and Trzcinka (1999). The intuition behind the formulas below is this: the variance of stocks increases steadily across time, while the bid-ask spread will maintain similar values across short intervals. The authors devise a model that separates the variance of the high and low prices of two days from its bid-ask spread. This model does result in a substantial number of negative spreads. These observations are not included in the year average. Then the available two-day bid-ask spreads are averaged across the year, as suggested by Corwin and Schultz (2012). The ‘i’ is a company identifier, the ‘y’ stands for year and the ‘d’ stands for day.
𝛽𝑖,𝑑 = ln (ℎ𝑖𝑔ℎ 𝑝𝑟𝑖𝑐𝑒𝑙𝑜𝑤 𝑝𝑟𝑖𝑐𝑒𝑖,𝑑 𝑖,𝑑) + ln ( ℎ𝑖𝑔ℎ 𝑝𝑟𝑖𝑐𝑒𝑖,𝑑−1 𝑙𝑜𝑤 𝑝𝑟𝑖𝑐𝑒𝑖,𝑑−1) 2
𝜆𝑖,𝑑= ln (ℎ𝑖𝑔ℎ 𝑝𝑟𝑖𝑐𝑒𝑙𝑜𝑤 𝑝𝑟𝑖𝑐𝑒𝑖,𝑑 𝑎𝑛𝑑 𝑑−1 𝑖,𝑑 𝑎𝑛𝑑 𝑑−1) 𝛼𝑖,𝑑= √2 ∗ 𝛽 − √𝛽 3 − 2 ∗ √2 − √ 𝜆 3 − 2 ∗ √2 𝑆𝑝𝑟𝑒𝑎𝑑𝑖,𝑑= 2 ∗ (𝑒𝛼− 1) 1 + 𝑒𝛼 𝐵𝑖𝑑 − 𝑎𝑠𝑘 𝑠𝑝𝑟𝑒𝑎𝑑𝑖,𝑦= ∑ 𝐷𝑎𝑖𝑙𝑦 𝑠𝑝𝑟𝑒𝑎𝑑𝑠 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑑𝑎𝑖𝑙𝑦 𝑠𝑝𝑟𝑒𝑎𝑑𝑠
The last way in which we will measure the information environment is by measuring discretionary accruals, which reflect the degree of earnings management. The paper by Dechow et al. (1995) inspects which methods have the most power to detect earnings management. The adjusted Jones model (1991) performs the best, and is therefore adopted here. The general intuition is that accruals of companies can be divided into two categories: non-discretionary accruals and discretionary accruals. Non-discretionary accruals are caused by normal company operations, while discretionary accruals are caused by conscious decisions of management. The adjusted Jones Model starts by calculating the ‘normal’ number of accruals for a company. This is done through a regression specified below. The predicted ‘normal’ accruals are then calculated for each company year and subtracted from the actual accruals. This results in the discretionary accruals. Each of the former terms is scaled by lagged total assets. The ‘i’ is a company identifier and the ‘y’ stands for year.
𝐴𝑐𝑐𝑟𝑢𝑎𝑙𝑠𝑖,𝑦= (𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝑎𝑠𝑠𝑒𝑡𝑠𝑖,𝑦−𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝑎𝑠𝑠𝑒𝑡𝑠𝑖,𝑦−1)−(𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠𝑖,𝑦−𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝑎𝑠𝑠𝑒𝑡𝑠𝑖,𝑦−1)−𝐷𝑒𝑝𝑟𝑒𝑐𝑖𝑎𝑡𝑖𝑜𝑛𝑖,𝑦 𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠𝑦−1 OLS: 𝐴𝑐𝑐𝑟𝑢𝑎𝑙𝑠𝑖,𝑦= 𝛼 + 𝛽0 1 𝐴𝑠𝑠𝑒𝑡𝑠𝑦−1 + 𝛽1 ∆𝑆𝑎𝑙𝑒𝑠 − ∆𝑅𝑒𝑐𝑒𝑖𝑣𝑎𝑏𝑙𝑒𝑠 𝐴𝑠𝑠𝑒𝑡𝑠𝑦−1 + 𝛽2 𝑃𝑟𝑜𝑝𝑒𝑟𝑡𝑦, 𝑃𝑙𝑎𝑛𝑡, 𝐸𝑞𝑢𝑖𝑝𝑚𝑒𝑛𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑦−1 %𝐷𝑖𝑠𝑐𝑟𝑒𝑡𝑖𝑜𝑛𝑎𝑟𝑦 𝑎𝑐𝑐𝑟𝑢𝑎𝑙𝑠𝑖,𝑦 = %𝑇𝑜𝑡𝑎𝑙 𝐴𝑐𝑐𝑟𝑢𝑎𝑙𝑠 − %𝑁𝑜𝑛 𝐷𝑖𝑠𝑐𝑟𝑒𝑡𝑖𝑜𝑛𝑎𝑟𝑦 𝑎𝑐𝑐𝑟𝑢𝑎𝑙𝑠𝑖,𝑦
In order to conduct this study, the degree of CEO power needs to measured and quantified. Bebchuk et al. (2011) offer a novel proxy called the CEO pay slice (CPS). The CPS is the proportion of CEO salary in comparison to the top five executives (including the CEO). The authors theorize that the CPS measures the relative importance of the CEO and its ability to extract rents from the company. Both aspects are closely related to our definition of CEO
power, which is the ability a CEO has to force his or her will on the company and
subordinates. Moreover, the CPS carries significant benefits over other proxies that have been proposed in scientific literature. First and foremost, the CPS captures the product of both the observable and the unobservable aspects of CEO power. Previous studies rely strongly on measures that are based on observable characteristics of CEO power, such as CEO duality, number of titles, and being the only insider on the board (Finkelstein, 1992; Harrison et al. 1998; Davidson et al. 2004; Adams et al., 2005). The second benefit of the CPS is the fact that it is continuous. Other proxies of CEO power are indexes of dichotomous variables (Adams et al., 2011) or rely on ordinal variables (Ashbaugh-Skaife et al. 2006). The continuous nature allows for a more insightful examination of the relationship between CEO power and the information environment. The CPS achieves these benefits by looking at the consequence of CEO power rather than by measuring the causes. This method does warrant some caution, since it might also measure other aspects that influence CEO pay and the general compensation of company executives. For instance, the CPS might be correlated with certain company policies on salary (see tournament theory) and/or CEO ability (e.g., more capable CEO are awarded more salary/bonuses). These problems can be resolved partially by including control variables; however, the CPS might still absorb some of these aspects. Therefore, a robustness study is done with another variable of CEO power.
The CPS is calculated on the basis of total compensation given to the executives. This includes salary, bonuses, restricted stock grants, stock options, long-term incentive pay-outs and all other compensation (as measured by the Execucomp item TDC1). Additionally, we examine whether the relationship is non-linear. Non-linearity could be present, since the CPS is a ratio. An increase of CPS from 0.7 to 0.8 arguably requires substantially more power than an increase from 0.4 to 0.5. This non-linear representation of power could lead to an exponential curve if the costs of obscuring the information environment do not increase in the same proportion. The ‘i’ is a company identifier, whereas the ‘y’ stands for the year to which the observation applies.
𝐶𝐸𝑂 𝑝𝑎𝑦 𝑠𝑙𝑖𝑐𝑒𝑖,𝑦= 𝐶𝑃𝑆𝑖,𝑦 = 𝐶𝐸𝑂 𝑐𝑜𝑚𝑝𝑒𝑛𝑠𝑎𝑡𝑖𝑜𝑛𝑖,𝑦
Another problem with examining the relationship between CEO power and the information environment is reverse causality. Theory states that the degree of information asymmetry might influence the degree of manager discretion (freedom) and its ability to extract personal rents. This study examines whether the opposite is also true. Therefore, each model is also examined with an IV-regression. Literature provides two instruments with regard to the CPS, which are industry mean CPS and CEO age (Bebchuk et al., 2011; Liu et al., 2010; Jiraporn et al., 2014; Mande et al., 2012). The former is not exogenous in this context, since certain industries might be characterized by a certain degree of information availability, asymmetry or transparency. CEO age is arguably endogenous as well, as it could reflect a certain ability or experience. However, we have chosen to adopt this measure, because the ability or experience of the CEO does not necessarily influence the information environment. Moreover, two instruments are needed for testing the causality of a non-linear regression. The second instrument has not been used for CPS before this study. It was adopted from Adams, Almeida and Ferreira (2005), in which a dummy was created that is equal to one of the founders are dead. The argument behind the use of this proxy is that, if the founder is dead, then the CEO has more freedom to lead the company. Fahlenbrach his (2009) approach is used to calculate a proxy for this variable. Rather than collecting data on founder deaths, companies which are incorporated before 1950 are argued to have dead founders. Although imperfect, previous studies have identified this variable as exogenous (Fahlenbrach 2009; Nagarajan, Schlingemann, Poel and Yalin, 2013). The shortage of true exogenous instruments motivate another method for illustrating the causal nature of the CEO power and information environment relationship. Therefore, the change in the dependent variables is examined. We argue that the change in the information environment in future years, does not affect the current CPS. The deltas of the information environment proxies are examined, with periods of one and three years. For the delta to be included in the regression, the CEO must have served during the entire period.
Control variables are kept roughly the same across regressions to ensure that the results can be compared. These variables are adopted from a number of studies that have identified certain control variables to be correlated with the information environment, specific proxies of the information environment, or CEO power, or that are included to ensure the validity of the CPS variable. Zhang (2006) identifies a number of variables to be controlled with the information environment: company size, company age, analyst coverage, analyst dispersion, cash flow volatility and return volatility. Company size is argued to affect the information environment because of the high fixed costs of attaining and disclosing information. Veldkamp her (2006) model shows that, because of this high fixed cost, information will gather around larger companies, for which the costs can be spread among many investors and then used for other companies. Moreover, Zhang (2006) argues that larger
companies can afford to disclose better information because of the marginal effects of these costs. Although the author opts for the use of market value, we adopt the approach by Bebchuk et al. (2011) and measure the log value of total assets. Company age is also significantly related to the information environment (Barry and Brown, 1985; Zhang, 2006). The reason is that when a company has been on the public market longer, the investors have more information about the company. The age is measured as the number of years since the first observation was recorded in the CRSP dataset (Bebchuk et al., 2011). Number of analysts is a relatively obvious aspect of the information environment. When more analysts cover a company, investors have access to more opinions about future company performance, which thereby improves the information environment (Zhang, 2006). Return volatility and cash flow volatily are measures of the uncertainty about company future performance. Additionally, revenue volatility is added in the equation, which might result in collinearity between the terms. All aspects are expected to result in a poorer information environment. Research and development has more to do with information asymmetry. Aboody and Lev (2000) validate this theory by showing that, when the R&D increases, insider-trading results in more insider gains. We therefore expect that more R&D expenses will deteriorate the information environment of investors. Unfortunately, a substantial number of values are missing in the Compustat database. These missing values are replaced with a zero and have a dummy that is equal to one if missing in order to retain a sizable number of observations. This approach is adopted by Bebchuk et al. (2011).
A number of control variables that control for company-specific aspects are also included. First, the average trading volume in millions of dollars is included. Second, the average proportion of outstanding shares that are traded is added. These measures are included because they are correlated with the bid-ask spread, but they could also show the demand for information. A greater demand in information will most likely increase the general information environment, as discussed by Veldkamp (2006). The Eindex is a measure of corporate governance. This variable is proposed by Bebchuk, Cohen and Ferrell (2009); it indexes a number of dichotomous variables that indicate poor corporate governance. The Eindex is preffered over the Gindex by Gompers, Ishii and Metrick (2001) because of data availability. The data needed to construct the Gindex is available from 1990 to 2006, whereas the Eindex can be calculated from 1990 until 2014. Li, Li and Wu (2010) find evidence to support the view that corporate governance could indeed lead to a better information environment. However, the opposite is also found. Armstrong, Balakrishan and Cohen (2012) find that when managers are more entrenched (i.e., poor corporate governance), firms actually have a better information environment. However it is expected that a better corporate governance will not allow managers to engage in more manipulation of the earnings environment.
The bargaining power of a CEO can influence the CPS considerably. Therefore, two variables are included to control for these aspects. First, the abnormal salary that is awarded to the top executive team is calculated. Second, the number of vice-presidents in the top executive team is considered. As argued by Bebchuk et al. (2011), these methods indirectly quantify the number of available candidates for the CEO position, which decreases the bargaining power. The abnormal salary is the residual of the regression in which the log of total compensation of the top executive team is regressed on the log of the company total assets with industry and year-fixed effects. A high residual indicates that executives get paid substantially more than a similar company in the same industry and year, which in turn suggests that the company has difficulty in attracting top talent. A high abnormal salary indicates that the company has most likely had difficulty attracting top talent. The sign of this coefficient is unclear, but its inclusion is necessary to adjust for bargaining power rather than CEO power. The number of vice-presidents, is the proportion of the top five executives that also carry the title vice-president. The number of vice-presidents are argued to increase the pool of suitable candidates for an CEO succession, thus decreases the bargaining power of a new CEO.
The return on assets, leverage and capital expenditures are also company-specific variables that are added into the regression. The return on assets is added because Liu et al. (2010) argue that it could reflect some degree of company risk, which inadvertently affects the information proxies. Leverage is widely considered in the literature as a way to discipline managers (Jensen, 1986; Stulz, 1990; Zwiebel, 1996). Capital expenditures, similar to research and development expenses, could indicate a degree of information asymmetry. These aspects are all expected to be inversely related to the information proxies.
Aside from variables that have to do with the information environment, the company and the industry, a number of variables are added that are tied to the CEO. First is the degree of CEO incentives, originally proposed by Bergstresser et al. (2006). They found that CEOs with more stock incentives are more likely to engage in earnings management. Although this variable is especially relevant for the accruals, it is also included in regressions on other dependent variables. The reason is that it might be easier to achieve certain stock performance if the information environment is poorer. The variable measures the increase in stock-based salary from a 1% increase in the stock price divided by the increase in total compensation that results from a 1% increase in the stock price. Ownership, founder status and CEO Duality have all been linked to CEO power in previous studies (Bebchuk et al. 2011; Finkelstein, 1992; Adams et al., 2005); they are therefore added into the regression to ensure the robustness of the CPS variable. All coefficients are expected to be positive.
Tenure is added for a number of reasons. The variable could capture some aspects of CEO power, since a longer tenure indicates greater power. But it could also indicate that
the CEO was able to fortify his or her power. Similar to other power variables an inverse relationship is expected with the information proxies. The male dummy is added because the variable might absorb some effects. The sign of the coefficient is unclear. The last CEO-specific variable is overconfidence. Malmendier et al. (2008) found that overconfidence leads to greater expropriation of shareholder wealth. Moreover, overconfidence can lead to excessive manipulation of the information environment, as the CEO might underestimate his or her chances of getting caught or disciplined for such actions. Thus, a positive sign is expected for this coefficient. Unfortunately, the data that was used in the study by Malmendier et al. (2008) was not available to us. Hirshleifer, Low and Teoh (2012) offer a valid substitute. For each year, a dummy is created that is equal to one if the exercisable options are at least 67% in the money. Compustat does not allow for an exact calculation of the moneyness of the options; but as proposed by Hirshleifer, the method of Campbell et al. (2009) is sufficient for calculating the average moneyness. The ‘i’ is a company identifier and the ‘y’ stands for year.
𝑀𝑜𝑛𝑒𝑦𝑛𝑒𝑠𝑠𝑖,𝑦 = 𝑆𝑡𝑟𝑜𝑐𝑘𝑝𝑟𝑖𝑐𝑒 𝑓𝑖𝑠𝑐𝑎𝑙 𝑦𝑒𝑎𝑟 𝑒𝑛𝑑𝑖,𝑦 𝐸𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑 𝑒𝑥𝑒𝑟𝑐𝑖𝑠𝑒 𝑝𝑟𝑖𝑐𝑒𝑖,𝑦− 1 𝐸𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑 𝑒𝑥𝑒𝑟𝑐𝑖𝑠𝑒 𝑝𝑟𝑖𝑐𝑒𝑖,𝑦
𝑆𝑡𝑟𝑜𝑐𝑘 𝑟𝑖𝑐𝑒 𝑓𝑖𝑠𝑐𝑎𝑙 𝑦𝑒𝑎𝑟 𝑒𝑛𝑑 − 𝑉𝑎𝑙𝑢𝑒 𝑢𝑛𝑒𝑥𝑒𝑟𝑐𝑖𝑠𝑒𝑑 𝑜𝑝𝑡𝑖𝑜𝑛𝑠 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑢𝑛𝑒𝑥𝑒𝑟𝑐𝑖𝑠𝑒𝑑 𝑜𝑝𝑡𝑖𝑜𝑛𝑠
The Gini coefficient is the last ‘traditional’ control variable and was introduced by Bebchuk et al. (2011). The Gini coefficient was originally used to measure the degree of inequality in incomes (Gini, 1912). Bebchuk et al. (2011) use this measure to control for the general inequality of the incomes of top executives. Like abnormal salary and the number of vice-presidents, this variable attempts to adjust for aspects that might bias the use of the CPS.
Lastly, four interaction terms are added to the regression. The interaction term with corporate governance is added, as one might expect that more power is more useful to CEOs under poor corporate governance. The interaction between CEOs and CEO incentives is added, as stronger incentives could be enacted more strongly among more powerful CEOs. Moreover, Bebchuk et al. (2011) find that there is an interaction between CEO power and abnormal salary awarded to top executives with regard to many agency costs. This will also be adopted. Lastly, the interaction term between CEO power and overconfidence is added. CEO overconfidence might interact with CEO power, as a more powerful CEO is better able to act on his or her overconfidence than a weak CEO. The general expectation is that these
effects all exacerbate the positive relationship between CEO power and the information environment.
The regression between discretionary accruals and the CPS requires some additional controls and one adjustment. Based on the study by Bergstresser et al. (2006), the lagged leverage, the change in earnings of the last year and the book-to-market ratio need to be added. In some instances, the book-to-market ratio is missing, which is handled in the same way as the missing values of research and development.
𝐴𝑛𝑎𝑙𝑦𝑠𝑡 𝑑𝑖𝑠𝑝𝑒𝑟𝑠𝑖𝑜𝑛 (𝑎𝑓𝑡𝑒𝑟 𝑡ℎ𝑒 𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑠 𝑎𝑛𝑛𝑜𝑢𝑛𝑐𝑒𝑚𝑒𝑛𝑡)𝑖,𝑦= 𝛼 + 𝛽1𝐶𝑃𝑆 + 𝛽2𝐶𝑃𝑆2+ 𝛽3𝐺𝑖𝑛𝑑𝑒𝑥 + 𝛽4𝐶𝑃𝑆 ∗ 𝐺𝑖𝑛𝑑𝑒𝑥 + 𝛽5𝑂𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 + 𝛽6𝐶𝑃𝑆 ∗ 𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 + 𝛽7𝐶𝐸𝑂 𝑠𝑡𝑜𝑐𝑘 𝑖𝑛𝑐𝑒𝑛𝑡𝑖𝑣𝑒𝑠 + 𝛽8𝐶𝑃𝑆 ∗ 𝐶𝐸𝑂 𝑠𝑡𝑜𝑐𝑘𝑖𝑛𝑐𝑒𝑛𝑡𝑖𝑣𝑒𝑠 + 𝛽9𝐴𝑏𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑎𝑙𝑎𝑟𝑦 + 𝛽10𝐶𝑃𝑆 ∗ 𝐴𝑏𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑎𝑙𝑎𝑟𝑦 + 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 𝐴𝑛𝑎𝑙𝑦𝑠𝑡 𝑑𝑖𝑠𝑝𝑒𝑟𝑠𝑖𝑜𝑛 (𝑏𝑒𝑓𝑜𝑟𝑒 𝑡ℎ𝑒 𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑠 𝑎𝑛𝑛𝑜𝑢𝑛𝑐𝑒𝑚𝑒𝑛𝑡)𝑖,𝑦= 𝛼 + 𝛽1𝐶𝑃𝑆 + 𝛽2𝐶𝑃𝑆2+ 𝛽3𝐺𝑖𝑛𝑑𝑒𝑥 + 𝛽4𝐶𝑃𝑆 ∗ 𝐺𝑖𝑛𝑑𝑒𝑥 + 𝛽5𝑂𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 + 𝛽6𝐶𝑃𝑆 ∗ 𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 + 𝛽7𝐶𝐸𝑂 𝑠𝑡𝑜𝑐𝑘 𝑖𝑛𝑐𝑒𝑛𝑡𝑖𝑣𝑒𝑠 + 𝛽8𝐶𝑃𝑆 ∗ 𝐶𝐸𝑂 𝑠𝑡𝑜𝑐𝑘𝑖𝑛𝑐𝑒𝑛𝑡𝑖𝑣𝑒𝑠 + 𝛽9𝐴𝑏𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑎𝑙𝑎𝑟𝑦 + 𝛽10𝐶𝑃𝑆 ∗ 𝐴𝑏𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑎𝑙𝑎𝑟𝑦 + 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 𝐵𝑖𝑑 − 𝐴𝑠𝑘 𝑆𝑝𝑟𝑒𝑎𝑑𝑖,𝑦= 𝛼 + 𝛽1𝐶𝑃𝑆 + 𝛽2𝐶𝑃𝑆2+ 𝛽3𝐺𝑖𝑛𝑑𝑒𝑥 + 𝛽4𝐶𝑃𝑆 ∗ 𝐺𝑖𝑛𝑑𝑒𝑥 + 𝛽5𝑂𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 + 𝛽6𝐶𝑃𝑆 ∗ 𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 + 𝛽7𝐶𝐸𝑂 𝑠𝑡𝑜𝑐𝑘 𝑖𝑛𝑐𝑒𝑛𝑡𝑖𝑣𝑒𝑠 + 𝛽8𝐶𝑃𝑆 ∗ 𝐶𝐸𝑂 𝑠𝑡𝑜𝑐𝑘𝑖𝑛𝑐𝑒𝑛𝑡𝑖𝑣𝑒𝑠 + 𝛽9𝐴𝑏𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑎𝑙𝑎𝑟𝑦 + 𝛽10𝐶𝑃𝑆 ∗ 𝐴𝑏𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑎𝑙𝑎𝑟𝑦 + 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 %𝐷𝑖𝑠𝑐𝑟𝑒𝑡𝑖𝑜𝑛𝑎𝑟𝑦 𝑎𝑐𝑐𝑟𝑢𝑎𝑙𝑠𝑖,𝑦 = 𝛼 + 𝛽1𝐶𝑃𝑆 + 𝛽2𝐶𝑃𝑆2+ 𝛽 3𝐺𝑖𝑛𝑑𝑒𝑥 + 𝛽4𝐶𝑃𝑆 ∗ 𝐺𝑖𝑛𝑑𝑒𝑥 + 𝛽5𝑂𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 + 𝛽6𝐶𝑃𝑆 ∗ 𝑣𝑒𝑟𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 + 𝛽7𝐶𝐸𝑂 𝑠𝑡𝑜𝑐𝑘 𝑖𝑛𝑐𝑒𝑛𝑡𝑖𝑣𝑒𝑠 + 𝛽8𝐶𝑃𝑆 ∗ 𝐶𝐸𝑂 𝑠𝑡𝑜𝑐𝑘𝑖𝑛𝑐𝑒𝑛𝑡𝑖𝑣𝑒𝑠 + 𝛽9𝐴𝑏𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑎𝑙𝑎𝑟𝑦 + 𝛽10𝐶𝑃𝑆 ∗ 𝐴𝑏𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑎𝑙𝑎𝑟𝑦 + 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠
5
Data
The diversity and quantity of the dependent and independent variables requires us to combine several databases. We gather data from Compustat Execucomp, Compustat Annual Fundamentals, I/B/E/S Details, I/B/E/S Guidance, I/B/E/S Summary, CRSP and ISS
(formerly known as riskmetrics). In order to match these datasets, we use two linking tables and the CUSIP identifier. The linking tables were supplied by Wharton WRDS and the University of Amsterdam1. After merging all data, we are left with 11823 observations on
2342 companies and 3738 CEOs in the United States from 1992 to 2013. The dataset is unbalanced and contains gaps. These gaps result from missing data or the exclusion of certain observations.
- insert Table 3.1 here -
Table 3.1 contains the summary statistics for all the variables that are used in the
initial regressions. A number of variables include a substantial number of missing values. We were able to retrieve some of these values by cross-referencing the data with other variables, which is discussed below. We could not prevent the substantial drop in observations that arise from using lagged components as seen in discretionary accruals (3713 observations missing). A number of variables contained outliers. A number of variables contained outliers. We resolved these outliers by winsorizing a number of specific variables on the 99th percentile.
The choice for winsorization over trimming data was made, because our study is especially interested in the tails of the distributions. Furthermore a drop in the observations is prevented by winsorizing over trimming. The variables that are winsorized can be identified by the (w) behind the variable name in Table 3.1. Lastly, we replaced the negative values from the variable discretionary accruals with zero.
The Execucomp database is central to our study. On the basis of the TDC1 variable that is provided in Execucomp, we were able to calculate the CEO pay slice. Company years that contain less than five non-missing values on executive compensation were dropped from the dataset. Some companies report more executives, but these observations are not used for the calculation of the CEO pay slice. In addition, we keep only company years in which the CEO served the entire year. Calculating the CPS from these observations would result in abnormally low values. This could not be resolved by extrapolating the salary in relation to the time served as a CEO, as the salary might include compensation that was gained before the promotion.
The variables CEO age, date at which the executive joined the company and the date at which the executive became a CEO contain many missing values. We mitigate this problem by cross-referencing these values with other variables. CEO age variable is supplemented by calculating the age in that fiscal year based on the current age of the CEO
1 Florian Peters, assistant professor at the University of Amsterdam was kind enough to
supply us with a linking table for CRSP and IBES, which is an adjusted version of the linking table on Wharton WRDS.
(variable PAGE in Execucomp). Although this is not a perfect proxy, the ages never deviate more than a year. The joined company date is supplemented by the ending date of the fiscal year in which the CEO was first observed working at the company in Execucomp. Although this is a very imprecise supplement, we feel that it is warranted because we use this date only to determine whether the CEO can be considered an insider. The cut-off point for being considered an insider is when the CEO worked longer than a year at the firm before he became CEO. The dates on which the CEO started working were augmented by looking at the CEO flag. If the value of the CEO was missing, and if it was not the first value of the executive for this company, this fiscal year end would be as date. Furthermore, we drop company years that are covered by less than three analysts. This is done because the spread measure would not make sense without at least three analysts. These observations are also dropped in the samples for the other regressions in order to ensure that all samples are comparable.
- insert Graph 2.1 here -
The CEO pay slice shows that, on average, 38% of the total salary of the top five executives is paid to the CEO. This is higher than the 35% reported by Bebchuk et al. (2011). This can be explained by the rising trend in the CPS (Graph 2.1). There is no substantial difference between the lagged CPS and the CPS that is predicted by the instrumental variable. The average dispersion of analysts is 0.33 % of the stock price before the announcement date. After the earnings announcement, this jumps to 0.52%. The average dispersion is relatively small, but can be substantially bigger, as indicated by the standard deviation of 1.099. Lastly, the average accruals are around 2.3% of all accruals.
- insert Graph 2.2, 2.3 and 2.4 here -
If we look at the CEO pay slice over the course of a couple years, it seems to be increasing at a steady rate. There is no clear reason to why this should be so. CEOs could have become more powerful, corporate governance could have decreased, risks for the CEO could have increased or some other reason. To prevent this trend from affecting the regression estimates, we will include time-fixed effects. The proxies for the information environment are stable across the years. There are a number of deviations from its base line. Analyst dispersion has a strong spike in the years after 2007 (Graph 2.2 and 2.3). It is very likely that this is caused by the 2008 recession. A similar spike is seen in the bid-ask spread (Graph
2.4). The bid-ask spread also has a spike around the 2000-2001 recession, which seems to
discretionary accruals have spikes in 1995, 2000 and 2005. It is unclear what causes these spikes; however, to determine this is beyond the scope of this study. The data illustrates that regulations on corporate disclosures cannot be used as exogenous shocks to prove causality. The regulation Fair Disclosure, introduced in 2001, limited the freedom companies have in disclosing information, which theoretically should have improved the information environment. Although analyst dispersion clearly drops after 2001, the bid-ask spread does not show this effect.
- insert Graph 2.6, 2.7, 2.8 and 2.9 here -
Additionally the firms with the weakest CEOs (i.e. lowest quintile) per year are compared with the firms with the most powerful CEOs (i.e. highest quintile). The comparison results with indicative evidence against our hypotheses. Firms with a higher CPS seem to have a generally better information environment than firms with a lower CPS (Graph 2.6,
2.7, 2.8 and 2.9). One could argue that this might be because of the correlation with another
important information aspect, such as size (i.e., the groups are structurally different, which makes it hard to compare). There are differences between quintiles (Table 2.2) if these affect the graphs. The firms with more powerful CEOs are shown to be smaller, revenue volatility and research development expenses being lower.
6
Results
In this paragraph, the results of the first regressions are discussed. The quantity and size of the regressions make it hard to discuss the results collectively while maintaining clarity. Therefore, the following structure is adopted to discuss the findings. As discussed earlier, four information proxies are examined and the regressions for each of these dependent variables are grouped. The results of these regressions will be discussed sequentially and per group in the following order: analyst dispersion after the earnings announcement, analyst dispersion before the earnings announcement, the bid-ask spread and discretionary accruals. After these regressions, the findings of the IV and the diff-in-diff regressions are reviewed. We will end with the conclusions that are derived from the general results. The graphs and tables are included in the appendix.
6.1 Endogenous Regressions
The results of the regression on analyst dispersion in the first 60 days after the earnings announcement are reported in Table 5.1. The initial regressions do not show a non-linear relationship. The quadratic term only becomes significant after the CEO-specific control variables are added. However, there are strong reasons to believe that the quadratic specification is correct. First, when the coefficients are replaced with linear coefficients, they start out significant but become insignificant after the company variables are added (Table
5.2). Moreover, regressions on the other information proxies also show a similar non-linear
relationship in which the quadratic component maintains significance across regressions.
- insert Graph 5.1 here -
Although a degree of non-linearity was expected, the coefficients deviated from our expectation. Rather than an exponential relationship between the CPS and analyst dispersion, the coefficients show a U-shaped curve (Graph 5.1 in the appendix). The coefficients seem to indicate that both CEOs with relatively little and CEOs with relatively large amounts of CEO power are correlated with a poorer information environment. Research directly related to the CPS offers no explanation for why this may be, possibly because no studies have examined the CPS in a non-linear context. We propose two possible explanations. First, the CPS does not measure CEO power directly and could therefore deviate in some specific instances from the parameter we are trying to study. In this particular case, when the CPS becomes abnormally low (i.e., lower than 20%), it results in a situation in which the CEO gets paid less than his or her direct subordinates. One could argue that the CPS fails to measure CEO power in these specific instances. For instance, it could be that these CEOs are rewarded in ways that are not measured in Compustat (e.g., perks), that they are compensated in ways that are not easily measured in general (e.g., long-term options), or that they choose not to get paid a substantial amount (e.g., Steve Jobs, Elon Musk and Eric Schmidt). Second, analyst dispersion could deviate from the information environment in specific instances. Landier, Sraaer and Thesmar (2005) argue that independently minded executives can exercise strong discipline on their CEO even though they are subordinates. They find evidence that independently minded executives will oppose and disagree with some of the strategies and decisions of the CEO. This consequently increases the quality of company policies and strategies. We build on this theory and hypothesize that when CEO power is truly low, the CEO might face strong opposition from other executives in the implementation of his plans and in making important decisions. This opposition could increase the uncertainty of company strategy, which might make it harder to predict future performance. In this case, the relationship between analyst dispersion and CEO power might absorb some aspects of information uncertainty, even though the model attempts to control for these aspects.
