Amsterdam Business School
Faculty of Economics and Business, University of Amsterdam MSc Accountancy & Control, variant Accountancy and variant Control
Thesis
Predicting earnings: the information usefulness of CEO equity compensation
Name: Krishna Bhiekhemsing
Studentnumber: 10581936
Date: 21-06-2015
Supervisor: dr. B. Qin
Statement of Originality
This thesis is written by Krishna Bhiekhemsing 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
Abstract
Granting compensation is a way to align the interest of the shareholders with executives. The interest of the shareholder could be to demand earnings persistence. Earnings persistence is found to be an important variable in earnings prediction. I investigate whether current CEO equity compensation has predictive power of future earnings from a signalling perspective. My sample consists of data on North American companies between the year 2000 and 2007. Performing Fama-MacBeth regressions, I find that current earnings have predictive power of future earnings. In addition T-test results show that future earnings are generated in a sustainable way, ergo earnings persistence. I also find that equity compensation signals no predictive power of future earnings. Furthermore I encourage companies to grant their CEO more restricted shares as I find that grating solely restricted shares to a CEO signals future performance and has a positive effect on earnings persistence. Finally I find that abnormal returns around earnings announcement is low, which suggest that the market does not respond on earnings surprises as it contains low quality information, ergo earnings smoothing. In addition the market reacts positive when the CEO is granted stock options in combination with earning surprise, which signals future performance and earnings quality. By including the signalling perspective and my empirical model this thesis is unique and contributes to current literature.
Content
1. Introduction ... 5
2. Literature review and hypothesis development ... 9
2.1. Information asymmetry ... 9 2.2. Signalling theory ... 10 2.2.1. Signaler ... 11 2.2.2. Signal ... 11 2.2.3. Receiver ... 12 2.3. Hypothesis development ... 13 3. Research design ... 16 3.1. Sample selection ... 16 3.2. Methodology ... 16 3.2.1. Variable measurement ... 16 3.2.2. Control variables ... 18 3.2.3. Empirical model ... 19 4. Findings ... 24 4.1. Data collection ... 24 4.2. Descriptive statistics ... 27
4.3. Predictive power of equity compensation on future earningst+1 ... 31
4.4. Predictive power of equity compensation on earnings beyond next year’s period ... 37
4.5. Earnings smoothing ... 40
Conclusion ... 43
1. Introduction
A prosperous organization can distinguish itself from a distressing organization by sending a reasonable signal about its competence to capital markets. This signal will be convincing only if the distressing organization chooses not to mimic the good organization by sending the same signal. A signal could be passively or unintentionally presented to the market, which for example shareholders and stockholders interpret ate. This signal is very important, because the interpretation of that signal can be positively or negatively received by the receiver. In the financial market the action of the receiver could be to invest or divest. This signal is decision useful when it can indicate future performance. While there are many signals, the signal that I choose to focus on are the grants of equity compensation given to the chief executive officer (CEO). The question whether equity compensation has any predictive power of future earnings is not fully examined. With this thesis I investigate whether information regarding the equity compensation of the CEO is a useful indicator for predicting future earnings. More specific, does the current equity compensation creates earnings persistence, ergo predictive power for future earnings?
While recent studies have focused on earnings prediction, the impact of equity compensation on earnings prediction remains relatively unexplored. The large and growing literature on earnings prediction and compensation shows mixed results which lead to an empirical debate. There are external parties who predict future earnings, which are analysts. The objective of a analyst forecaster is to provide to investors and other parties forecast information. It is found that information regarding (analyst) forecast significantly influence investors’ judgments and beliefs about the value of the company, whether to invest or divest. Prior literature suggests, that accuracy is an important performance measure for analysts and investors (Mikhail, Walther, & Willis, 1997). By including more variables in the earnings prediction model, this could potentially predict more accurate future earnings. A important variable that analyst use in their earnings prediction model is prior and current earnings. In order to increase the accuracy and reduce the volatility of earnings predictions, it is easier to predict earnings when the earnings are persistent. It is the managements job to create shareholder value by generating more (persistent) earnings. According to the agency theory, one way to align the interest of the shareholder with the management in order to increase earnings persistence, is to provide equity
compensation (Jensen & Meckling, 1976). It is found that equity compensation increases productivity and efficiency, which in turn can increase shareholder wealth (Mathieu Lefebvrea, 2014, p. 28). Baber, Kang and Kumar (1998 p. 190) mention, that the persistence of accounting earnings is relevant for setting executive compensation. However they also find that the relationship with equity compensation is not positively related to earnings persistence (Baber, Kang, & Kumar, 1998, p. 171). Ashley and Yang (2004 p. 369) reveal, that the effect of high earnings persistence results in firms that focus more heavily on cash compensation (salary and bonuses) rather than on equity compensations (stock options, etc.) to compensate executives. Gong, (2011 p. 23) finds that the weight assigned to equity incentives is positively related to value creation. Also Gong and Li (2013, p.649) mention, that analysts don’t use CEO cash- or stock compensation as a variable in their model for predicting future earnings, because researchers assume that analyst might think that it will lead to forecast errors. This discussion, that executive compensation might lead to better future performance probably indicates, that equity compensation is useful for predicting future earnings. By following the stream of researcher in earnings prediction and based on the above discussion I also assume that equity compensation has predictive power of future earnings (Gong & Li, 2013; Hanlon, Rajgopal, & Shevlin, 2003; Dechow, Ge, & Schrand, 2010; Baber, Kang, & Kumar, 1998). Besides the mixed results, the mentioned literature only focuses on high rankings members (management) in general and not specifically on the CEO (compensation). Also the studies mentioned above have not taken into consideration the fact, that earnings are generated by managers and managerial behaviour such as strategy formation and investment management, which is influenced by their own incentives they face (Bertrand & Schoar, 2003, p. 1204).
The incentive of the CEO (might) differ with the incentives of the shareholders, this can result to earnings management in a particular setting. A explanation for this asymmetry is that the CEO has a strong incentives to beat or at least meet earnings expectations (Kross, Ro, & Suk, 2011). The CEO will most likely use earnings smoothing to outperform earnings expectations (Degeorge, Patel, & Zeckhauser, 1999). Kim, Li, and Zhang, (2011, p. 727) show evidence, that incentives favour the chief financial officer (CFO) in a setting where financial expertise is required and they find that CFO equity incentives are more important than CEO incentives in determining earnings management. Based on the earnings response coefficient (ERC) by explaining earnings management, zero and small positive earnings surprises are often considered
as red flags, because all information should be reflected in the stock price (Modigliani–Miller theorem, efficient market theory) and it is used by researches as a signal of low quality earnings (Keung, Lin, & Shih, 2010). The studies mentioned reveal, that the focus lays more on CFO behaviour, because of its financial expertise, direct influence on financial disclosures and decision making, which indicates the importance of CFO contribution in earning management. It is more likely that in a case of earnings management the CFO is involved. However the grant of equity compensation can motivate the CEO as much as the CFO to participate in earnings smoothing. So the question arises if equity compensation also signal the effect of earnings smoothing?
There are only two studies who have formally examined the association of executive incentives and stock options in the prediction of future earnings Hanlon, Rajgopal and Shevlin (2003) and Gong and Li (2013). However their research is mainly based on stock options (the model of Black and Scholes), while mine will be based on the relative proportion of different forms of equity incentives on total compensation. Their sample selection consist of the periods of 1992-2000, while mine will be from 2000-2007 and CEO focused. They also assume that reported earnings are an unbiased measure of firm performance. The paper of Gong and Li (2013) does have a focus on CEO incentives but their sample selection, which is one of the limitation in their study includes firms with CEO turnover. Also their model is based on current earnings as a function of next year’s earnings without the incorporation of the interaction with equity incentives. They don’t mention their choice of month they take regarding their earnings per share forecast. Furthermore they hypothesize that higher equity incentives are associated with higher future returns, while I hypothesize that (higher) equity incentive is associated with persistent earnings and therefore more useful for accurate earnings predictions. Finally their theoretical background lacks foundation, it is solely based on the agency theory while mine is also from a signalling perspective.
In my thesis I want to investigate if information regarding CEO equity compensation has any predictive power of future earnings. By disclosing specific signals to the market, whether good or bad, (potential) investors can anticipate to invest or divest. The CEO knows this and it’s the question whether they use these signals to truly indicate an increase in future shareholder value or whether this is just an act to boost short term shareholder value, ergo earnings smoothing in order to opt out. This leads to the following research question:
Does CEO equity compensation of the current year has any predictive power of future earnings?
With my thesis I want to build further on the mixed findings on earnings prediction in order to contribute to the current literature. By including the signalling perspective and my empirical model this thesis will be unique, which fills the gap in the literature. I predict that CEO equity compensation does have predictive power, because it stimulates CEO’s to create persistent earnings. By answering my research question, analyst forecasts of future earnings can potentially be improved when CEO equity compensation variable is taken into consideration. If the outcome is positive then this will indicate, that when a CEO is granted with equity compensation it is more likely that future earnings will be more persistent. If the outcome is negative, this indicates that equity compensation does not have any predictive power of future earnings and earnings quality. In other words, firms can misuse the grant of stock option in order to falsely signal the market for an expected increase in shareholder value. Falsely signalling is due to earnings manipulation. The negative outcome could also mean that the current earnings prediction model that analyst use is sufficient. By including the predictive power of equity compensation potential investors could be attracted and benefited. Investors can rely on more and accurate information, which contributes to efficient contracting, adverse selection and reducing moral hazard problems. From an academic view I will contribute to the existing literature on earning prediction, signalling and CEO compensation.
My sample consists of data on North American companies between the year 2000 and 2007. I obtain compensation data from ExecuComp, firm characteristic data from COMPUSTAT, analyst forecast data from I/B/E/S, market portfolio and securities from CRSP, internal control quality data from Audit Analytics and governance characteristics from Riskmetrics (ISS). Results show that equity- and cash compensation have no predictive power of future earnings, but restricted shares solely does have predictive power and the abnormal return around event dates is low, which means that earnings smoothing is involved.
This paper will be structured as followed, in chapter 2 I will present a literature review, where I briefly describe the agency- and signaling theory followed by a discussion of the most important theoretical and empirical papers in order to develop my hypothesis. In chapter 3 I will explain my research design. Chapter 4 will discuss descriptive statistics and the findings. Finally in chapter 5 I will present the conclusion and limitations of my thesis.
2. Literature review and hypothesis development
In this chapter I will describe key literature for my hypothesis development. I first start with explaining the theory which describes the alignment of the principal with the agent followed by the signalling theory. Furthermore I will describe in paragraph 2.3. key literature on executive compensation and literature on earnings quality in order to develop my hypothesis.
2.1. Information asymmetry
This thesis will focus on the CEO equity compensation that companies provide in order to align the interests of the top management with the shareholders for creating persistent earnings. The agency theory fits this research. Agency theory arises because of the separation of ownership from the management, which creates agency problems in corporations as agents (management) may not always act in the best interest of the principal (stockholders) (Jensen & Meckling, 1976). These agency problems arises from information asymmetry. In the macro-economic context (potential) investors use information that affects their decision-making process. The investors rely and make decisions on public available information, which is freely available. Private information is only available to a select of individuals and they are costly (Alles & Lundholm, 1993, p. 93). Information asymmetry occurs when different parties know different things. Because some information is private, this creates information asymmetries between those who hold that information and those who could potentially make better decisions if they had it (Connelly, Certo, Ireland, & Reutzel, 2011). Often are formal economic models, that assist in the decision-making process based on the assumption of perfect information, where such information asymmetries are not considered (Stiglitz J. E., 2002, p. 464). Economist believe that markets with minor information imperfections would behave the same as markets with perfect information (Stiglitz J. E., 2002, p. 464). To understand the influences of the decision-making process in the market, Stiglitz (2000, p. 1441-1445) finds that there are two types of information where asymmetry is particularly important: information about the quality and information about the intention. Information quality is about the information asymmetry, where one party is not fully aware of the characteristics of another party. The intention of information addresses information asymmetry, because the party’s intentions is not sincere (Connelly, Certo, Ireland, & Reutzel, 2011, p. 42). Literature on information asymmetry about behaviour and intentions examines the
use of compensation as a solution to reduce potential moral hazards problems (Jensen & Meckling, 1976).
The association between earnings quality and performance consequences of executive incentives is important, because next year’s performance depends upon how firms solve agency problems, which could be done through different ways (Nyberg, Fulmer, Gerhart, & Carpenter, 2010). Dalton et al. (2007) claim, that the three main principles of minimizing agency problems are: (1) board independence, viewed as improving directors’ monitoring of managers, (2) the market for corporate control, which refers to an active market that disciplines mischievous management, and (3) agent equity ownership, which leads to that managers who share ownership of their firm to embrace shareholder interests. Agency theory suggests that equity based incentives, such as common stock ownership, restricted stocks, and stock options, align the interests of executives with shareholders and promote a long-term perspective on performance (Murphy, 1999; Jensen & Meckling, 1976). If incentive alignment is strong enough, then the alignment of preferences and actions will follow. With higher equity incentives, as the firm can signal in their financial statements, I expect CEOs to work harder to generate sustainable profit and therefore earnings persistence.
2.2. Signalling theory
I focus on the activity of signalling information to public parties in order to understand how these parties anticipate on that specific information and whether information asymmetries about latent and unobservable information quality also influence their decision-making process, which constitutes to the majority of literature on signalling theory. Connelly, Certo, Ireland and Reutzel (2011) provide a clear explanation of the signaling theory. According to the signaling theory there are two primary actors: the signaler and receiver. Between these two parties there is the proces of interaction where one communicates a signal and one receives. Figure 1 presents a graphical chart of the signaling theory, where the vertical axis indicates the time (t).
0 1 2 3
SIGNALLER (services, products, projects or firm
has underlying quality)
SIGNAL is sent to receiver Receiver observes and
interprets signal
Feedback is sent to signaller
2.2.1. Signaler
The essence of the signaling theory is that signalers are insiders (e.g. executives or management), those who have access to information that is not available to outsiders. This information could be positive or negative, that outsiders would find useful in their decision-making process. This signal could for example be, specifics about the companies services, products, projects or any future performance indicator. Such information may include early stage research-and-development, results or news regarding preliminary sales. Insiders have also access to information about other aspects of the organization such as, pending lawsuits or labor union negotiations. In other words private information provides insiders with a privileged perspective regarding the underlying quality of some aspect like: the individual, product, or company (Connelly, Certo, Ireland, & Reutzel, 2011, p. 44).
2.2.2. Signal
Company’s management has both negative and positive private information and they must decide whether they want to communicate this information to the public. Companies that address signalling theory primarily focuses on disclosing positive information in an effort to conceal negative organizational characteristics. Prior literature has found that certain communication of information by the management are found to be interpreted as negative by the public. For example, issuing new shares of a company is generally considered to be a negative signal, because top-management may issue equity when they believe their company’s stock price is overvalued (Jensen, Crutchley, & Hudson, 1994, p. 314). Jensen, Crutchley and Hudson (1994) find that it even may be, because the company isn’t able to borrow funds at the private market. This may signal that the company is facing severe problems at the moment or may be in the future. So by disclosing certain information, this doesn’t necessarily mean that this information will reduce information asymmetry.
An efficient signal has two important characteristics (Connelly, Certo, Ireland, & Reutzel, 2011, p. 45). The first is signal observability, this means whether the external party is able to notice the signal. If the management discloses the signal and it is not observed by the public it is difficult to use this information to communicate with the receivers. The second characteristic is signal costs. Signal costs are central to the signalling theory, often referred to as the “theory of costly signalling” or “signalling game” (Talmor, 1981, p. 414; Bangerter, Roulin, & König, 2012, p. 720). Some signallers are in a better position than others to absorb the associated costs. The
costs associated, for example with disclosing a financial report are high, because the auditing process is time consuming and these costs make cheating or false signalling difficult. However, these financial statements are less costly for a high-quality auditing firm as compared with a low-quality auditing firm, because a low-low-quality firm would require more resources and knowledge to issue the same statement as a high-quality auditing firm. If a signaller does not have the underlying quality associated with the signal, but believes that the benefits of signalling outweigh the costs of producing the signal, the signaller may be motivated to attempt false signalling, ergo earnings smoothing (Connelly, Certo, Ireland, & Reutzel, 2011, p. 45).
2.2.3. Receiver
The final characteristic of the signalling theory is the receiver. According to the signalling theory, receivers are outsiders who lack- and are not able to gain that information about the organization but they would like to receive this information. As mentioned before signallers and receivers also have conflicting interests. If there is an interaction between them, the signallers intention is to benefit his own interest in order that the receiver will take actions that he would not otherwise had done based on that disclosed information, so a strategic effect (Connelly, Certo, Ireland, & Reutzel, 2011, p. 45). For example, the signaller signals a job advertisement and makes the choice about hiring. The receiver finds this advertisement in the paper and decides to apply for the job, an effective signal. But this also signals to other receivers like the competition, that the company is facing (not limited to) an increase in economic value, is investing or is attaining a new project.
Prior literature mentions key receivers in the financial market as shareholders (Noe & Rebello, 1996, p. 637) and debt holders (Anderson & Prezas, 2003, p. 128). Studies in marketing use customers as receivers (Mitra & Fay, 2010, p. 185). Studies in human resource management use applicants as receivers (Bangerter, Roulin, & König, 2012, p. 720). For example, customers would benefit from purchasing products and services that are associated with signals of high quality, even a brand. Similarly shareholders would benefit from buying shares of companies that signal more profitable future earnings. Or hypothetically, if companies compensate executives by granting equity, this might signal future earnings or earnings quality.
2.3. Hypothesis development
As mentioned before one way to align the interest of the shareholder with the management is to grant equity compensation. I distinguish equity compensation as common stock ownership, restricted stocks and stock options. It is found that equity compensation increases productivity and efficiency, which in turn can increase shareholder wealth (Mathieu Lefebvrea, 2014, p. 28). Performance awards can provide an incentive for employees to accomplish a variety of company targets, individual goals and objectives (Nyberg, Fulmer, Gerhart, & Carpenter, 2010). Common or restricted stock can be tailored to encourage a longer-term focus in order to increase earnings persistence than stock options, which are criticized as encouraging a short-term focus based solely to boost stock price (Mathieu Lefebvrea, 2014, p. 29). Stock options are perceived to bear only asymptotic risk (high upside returns and only the price of stock options at losses), thus leading to managerial risk seeking behaviour. As a result, CEO’s with high levels of stock options could make decisions that are riskier and therefore negatively affect earnings persistence (Nyberg, Fulmer, Gerhart, & Carpenter, 2010). To mitigate the risk seeking behavior of the CEO, companies can grant restricted shares. It is generally accepted that granting executives restricted company stock reduces risk taking behaviour of the CEO and therefore I expect a higher amount of earnings persistence (Jensen & Murphy, 1990). The amount of equity compensation that the CEO gets should be distributed in the way, where it is more likely that the proportion of stock options, restricted shares and regular stock create persistent earnings.
CEO’s are compensated based on achieving their targets. It is considered to be common, that the target is often referred as “earnings target”. There are firms who exhibit robust and sustained earnings over time (high earnings persistence) and firms which exhibit weak earnings over time (low earnings persistence). Prior literature mentions different characteristics that are related to earnings persistence such as firm size, product types, competitive powers, barriers-to-entry, market share, marketing spending, research and development intensity, derivate users (Dadalt, Gay, & Nam, 2002) and capital intensity (Baginski, Lorek, Willinger, & Branson, 1999). While other papers mentioned that earnings persistence is negatively related to growth and the risk-free rate of return (often the treasury rate) (Lev, 1983). Collins, Kothari, and Rayburn (1987) examine the effect of firm size on the performance of earnings prediction. They find that small firms have greater variance in annual earnings than large firms. One possibility is that large firms are more diversified than small firms. This indicates that larger firms have greater earnings
persistence. Watts and Zimmerman (1978) find that large firms will choose less risky investments to accompany higher returns, which has a positive effect on earning persistence. As a result of avoiding risky investments, this indicates that stock option compensations are less likely to be misused for the CEO’s own benefit when active in a large company, ergo a positive effect on earnings persistence. Firm size can also be related to a company's earnings persistence, because it indicates the strength of the company's competitive position. Based on the literature review, I hypothesize that the grants of equity compensation to the CEO is positively associated with earnings persistence, which means that it will have predictive power for future earnings. I make a distinction between CEO-stock options and restricted shares and regular shares. The hypothesises that I formulate are:
H1a: companies that grants the CEO stock options have persistent earnings.
H1b: companies that grants the CEO restricted- and regular shares have persistent earnings. Subsequent earnings persistence is hard for a company to achieve. Firms who subsequently meet or beat earnings expectations are likely to have better fundamentals, because they have been able to consistently meet or beat the market’s expectations. Investor interest in firms who consistently meet or beat earnings expectations may be heightened due to their success in the past. According to the signaling theory, firms which consistent meet or beat earnings expectations are likely to signal their future earnings performance by voluntarily disclosing more information than other firms, like management earnings forecast (Kross, Ro, & Suk, 2011). Tan, Wang and Welker (2011) find, that firms who disclose more quality information are more followed by analyst. This indicates that analyst favor the quantity of additional information. However, once a firm has met or beaten earnings expectations management feels the pressure not to break it, because of the high opportunity cost of doing so. In the pursuit of this earnings persistency, managers desire to avoid a next period failure in meeting or beating earnings expectations thus providing them with incentives to either manage earnings, ergo earnings smoothing. Earnings smoothing involves making tradeoffs between how much there has to be reported in the current period and how much to report in the next year’s period and such tradeoffs depend on managerial beliefs about future characteristics. Prior literature provides various reasons for earnings smoothing (Bouwman, 2014). The most important reason is, that earnings smoothing may reduce a firm’s perceived earnings volatility and risk, lowering the required rate of return (Trueman & Titman, 988). The volatility of reported earnings also reflects important
aspects of the accounting determination of income. One of these aspect is the quality of matching revenues to expenses (Dichev & Tang, 2009, p. 162). Dichev and Tang (2009) say that poor matching acts as noise in the economic relation between revenues and expenses, which means that the volatility of reported earnings increases in poor matching. Poor matching is associated with poor earnings predictability, because the matching noise in reported earnings obscures the underlying economic relation that governs the evolution of earnings over successive periods (Dichev & Tang, 2009, p. 162). Also earnings smoothing may lead to higher shareholder value (Thomas & Zhang, 2002).
Earnings smoothing can be detected by different ways. Recent studies have examined deferred taxes and the provision for taxes (Dhaliwal, Gleason, & Mills, 2004; Phillips, Pincus, & Rego, 2003) or assessed the variability of reported earnings relative to the variability of cash flows (Land & Lang, 2002; Lang, Smith Raedy, & Wilson, 2006; Leuz, 2003). There are a lot of studies that focus on managers’ use of discretionary accruals (Dechow, Ge, & Schrand, 2010). If analysts are rational and sufficiently competent and all information relevant to earnings management incentives is freely available, forecast errors should not be related to discretionary accrual (Wilson & Wu, 2011). However if they are irrational, or information regarding earnings management incentives is incomplete, it is possible that forecast errors will be systematically related to current period earnings management. Wilson and Wu (2011) find, that forecast errors are positively associated with discretionary accruals realized in the current period. Dechow and Dichev (2002) show that firms with larger standard deviations in matching accruals with cash flows, have less persistent earnings, longer operating cycles, larger accruals and more volatile cash flows and earnings. They also mention that these firms are smaller and are more likely to report a loss. Their findings suggest that these characteristics signal a more likelihood of estimation error in accruals, which means lower accrual quality (Dechow, Ge, & Schrand, 2010). Following Tirole and Fudenberg (1995), I assume that every manager is rational and is risk averse. In their research managers do not have biased beliefs. They also assume that every manager enjoys private benefits (salary and bonus) from keeping their job and can be fired for poor performance, bad reported earnings. To keep enjoying these benefits it is important to achieve earnings targets. As mentioned before this could shift to the negative side of granting equity compensation. I investigate the negative side of granting equity compensation to the CEO, if it might signal that the CEO is involved in earnings smoothing.
3. Research design
In this chapter I will explain my sample selection, the measurement of my dependant- and independent variables, control variables and finally my estimation of the empirical models. 3.1. Sample selection
This research focuses on North American companies specifically United States and Canadian companies between the years 2000 and 2007. I have chosen for North America, because the economy is based on the Anglo-Saxon model. This means that North America has an active economy, strong property rights, contract enforcement, low barriers to entry, large security markets, dispersed ownership and is shareholder oriented. Also the research which inspired me for my thesis and is closely related covered their data for global companies (Gong & Li, 2013). Furthermore the data availability is more accessible for these firms and it includes large and well know firms. Finally, I have chosen to investigate my research question with a sample selection between the years 2000 and 2007. During the recession it is hard for companies to have persistent earnings, so 2000 to 2007 is chosen because of avoiding the effects of the late recession and it leads to recent observations.
3.2. Methodology
In this paragraph I will describe the measurement of my dependant-, independent and control variables, which I eventually use to estimate my empirical model.
3.2.1. Variable measurement
Following prior research, I first test if there is a statistical association with current year earnings and future earnings without the use of any other independent variables with predictive power and control variables (Dichev & Tang, 2009). Prior literature on earnings prediction use baseline regression models of the following form, with future earnings as a function of current year earnings or components of earnings (Dechow, Ge, & Schrand, 2010). The first model is:
EARNINGSt+1 = α0+ α1EARNINGSt+ εT (Model 1a)
Earnings is typically scaled by assets. A higher α1 implies a more persistent earnings stream.
Intuitively, the logic behind earnings persistence being a quality indicator is as follows: If company “X” has more persistent earnings than company “Y”, then (i) in company “X”, current earnings is a useful measure of future performance; and (ii) current earnings in company “X” will give smaller valuation volatility than current earnings in company “Y”. Thus higher earnings persistence is of higher quality when the earnings is also value-relevant (Dechow, Ge, & Schrand, 2010). Persistence of earnings is often measured as the persistence of profitability (i.e. ROA). I measure EARNINGSt as income before extraordinary items scaled by the average of current year assets and previous year assets of the firm (Dechow, Ge, & Schrand, 2010).
Prior literature established correlations between stock returns and executive compensation (Jensen & Murphy, 1990; Murphy, 1999). Also between accounting earnings and executive compensation (Dechow, Ge, & Schrand, 2010). The CEO equity compensation scheme increases or decreases depending on achieving company targets. An increase in the CEO equity compensation suggests, that the CEO is achieving or has achieved his target and a decrease otherwise. Also in equity compensation the firm can signal information regarding future performance which it may not disclose, because of competitive reasons. CEO equity compensation is therefore an important measure. A straightforward way of computing CEO equity incentives, is by deflating equity incentives (EQUITYt= stock options, restricted shares and stock) with total compensation (TDC1t= salary, bonus, other annual, total value of restricted stock granted, total value of stock options granted, long-term incentive payouts, and other totals).
EQUITYt=
I distinguish equity incentives as current year, stock options (OPTAWRDSt), restricted shares (RSTKGRNTt) and stock grants (STCKWRDSt). They give the following equations:
OPTAWRDSt=
RSTKGRNTt=
STCKWRDSt=
The residual of total compensation minus the stock options, restricted shares and stock is defined as cash compensation (CASHt), which gives the following equation:
3.2.2. Control variables
In my model I use the following control variables: firm size, leverage, market value, book-to-market ratio, growth, board size, board independence, CEO-duality, CEO-compensation committee and finally if the company is big 4 audited.
I include the control variable firm size (FIRMSIZEt) in my model, because firm size has often been used in prior literature as a proxy for the quantity of information that is publicly available. Larger firms are in general more followed by analysts or investors and there is more public information available. This suggests that the market makes a better decision about the future performance of larger firms relatively to smaller firms and that there would be smaller bias for larger firms. Furthermore Bhushan (1998) finds that size and analyst following are highly correlated. Larger firms are followed by more analysts than smaller firms. So it might be that the level of analyst following influences forecast accuracy. Including firm size increases the explanatory power of the regression. I measure FIRMSIZEt by taking the logarithm of the total assets at the year end. Furthermore I include the control variable leverage (LEVERAGEt). Nissim and Penman, (2003) show that leverage is related to future earnings. They also show that an increase in leverage is more likely to depress current earnings but also lead to future improvements in earnings and the reduction of liabilities. Core and Schrand (1999) find that leverage is associated with non-linearity’s in earnings-returns relation (ERC). They predict that a firm’s debt structure will affect the magnitude of the earnings-returns relation. Leverage does not affect the informativeness of earnings but rather it affects return reactions. I measure LEVERAGEt by dividing total liabilities by total assets. The next control variable is market value. I measure market value (MVt) by the logarithm of the current year market value. In my model I also include the book to market ratio (BTMt) to control for investment opportunities and as a proxy for risk taking behavior. The book to market ratio BTMt ratio is often used in prior literature, because it indicates the value that the market places on the common equity or net assets of a firm (Ceccagnoli, 2009). BTMt reflects the ability of the management to use the assets effectively, to create value for the firm and it is linked to risk seeking behavior (Griffin & Lemmon, 2002). I measure BTMt by the percentage change in book to market, yeart-1 to yeart. The other control variable is growth, which is often used as a proxy for economic volatility. High-growth firms are more likely to reward executives with stock options than restricted stock. Also high-growth opportunities make it more difficult for shareholders or an outside board to
monitor CEO’s, which requires more CEO incentives to maximize shareholder value (Core & Guay, 1999). I measure growth (GROWTHt) as the growth in sales, (salest/ salest-1). Good corporate governance results to higher credibility and reliability of the financial statements, which results in a higher ERC. A corporate governance characteristic is board size. Jensen (1994) recommends that smaller boards are more effective in monitoring the actions of the CEO compared to a large board who places a greater emphasis on politeness and courtesy. Board size suggests that the size of the firm’s board should be inversely related to performance, because enhanced monitoring is associated with less earnings smoothing. Board size (BOARDSIZEt) is measured by the log of total number of directors that integrate the board of directors. My next governance characteristic is board independence. Klein (2002) finds that board characteristics predict lower magnitudes of discretionary accruals. Boards which contains more outside directors are in a better position to monitor and control managers. Outside directors are likely to be more independent then the company’s management, which creates better stock returns and operating performance (Rosenstein & Wyatt, 1990). Board independence (BOARDINDt) is measured by the proportion of external directors inside the board (external directors/ total directors). The next control variable that I use is CEO-duality. CEO-duality enhances the power of the CEO’s position, because the individual serves as the CEO and also has the position of the chairman of the board, which allows more room for management discretion (Jensen , 1994). But at the same time it creates 3-4% more competitive advantage compared to non-duality companies (Yang & Zhao, 2014). CEOCHAIRt is measured through a dummy variable that takes the value of 1 if there is duality of roles with the CEO and also the chairman of the board and 0 otherwise. The same counts for CEOCOMPSt, if the CEO is also the chairman of the compensation committee, which indicates CEO power on incentive contracts. I also include the control variable that indicates if the firm is audited by a big 4 accounting firm (BIG4t). Hussainey (2009, p. 341) finds that analysts are able to better anticipate future earnings when financial statements are audited by the big four accounting firms. I measure the variable BIG4t by addressing the number 1 when the auditor is a big 4 firm and 0 otherwise. Table 1 summarizes all the variable measurements.
3.2.3. Empirical model
For my different forms of compensation I estimate the following model:
EARNINGSt+1 = α0+α1EARNINGSt+α2RSTKGRNTt+α3STCKWRDSt+α4OPTAWRDSt+εt
With control variables added, I estimate the following model:
EARNINGSt+1 = α0+α1EARNINGSt+α2RSTKGRNTt+α3STCKWRDSt+α4OPTAWRDSt+
α5CASHt+α6FIRMSIZEt+α7LEVERAGEt+α8MVt+α9BTMt+
α10GROWTHt+α11BOARDSIZEt+α12BOARDINDt+α13CEOCHAIRt+
α14CEOCOMPSt+ α15BIG4t +εt (Model 1d)
For this model I use the summary measure of equity- and cash compensation, I estimate the following model:
Earningst+1(;2;3) = α0+α1EARNINGSt+α2EQUITYt+α3CASHt+α4FIRMSIZEt+α5LEVERAGEt+
α6MVt+α7BTMt+α8GROWTHt+α9BOARDSIZEt+α10BOARDINDt+
α11CEOCHAIRt+α12CEOCOMPS+α13BIG4t+εt (Model 1e)
Based on the above explanation I estimated the following empirical model where I include the interaction with total equity.
Earningst+1(;2;3) = α0+α1EARNINGSt+α2EQUITYt+α3CASHt+α4EARNINGSt*EQUITYt+
α5EARNINGSt*CASHt+α6FIRMSIZEt+α7LEVERAGEt+α8MVt+α9BTMt+
α10GROWTHt+α11BOARDSIZEt+α12BOARDINDt+α13CEOCHAIRt+
α14CEOCOMPS+α15BIG4t+εt (Model 1f)
A positive coefficient for α4 is consistent with my hypothesis that next year’s earnings is more persistent when the CEO is granted equity incentives. To make a distinction between stock options, (restricted) shares and cash compensation I estimate the following model:
Earningst+1(;2;3) = α0+α1EARNINGSt+α2RSTKGRNTt+α3STCKWRDSt+α4OPTAWRDSt+
α5CASHt+α6EARNINGSt*RSTKGRNTt+α7EARNINGSt*STCKWRDSt+
α8EARNINGSt*OPTAWRDSt+α9EARNINGSt*CASHt+α10FIRMSIZEt+
α11LEVERAGEt+α12MVt+α13BTMt+α14GROWTHt+α15BOARDSIZEt+
α16BOARDINDt+α17CEOCHAIRt+α18CEOCOMPSt+α19BIG4t +εt
(Model 1g)
The models 1f and 1g explain whether equity incentives has predictive power on future earnings. As explained in chapter 2, the negative side of granting equity incentives could motivate CEO’s to earnings smoothing. To test whether this assumption holds I am going to construct a second model (2) where I will use earnings response coefficient (ERC). Analysts are both rational forecasters and truthful reporters but they report their beliefs solely only on public information (Gu & Wu, 2003). The gap between public and private information leads to earnings surprise. ERC is a measure of investor responsiveness to earnings news. ERC needs the assumption of an efficient market, where all the prices of securities are traded at such markets that all the time fully
reflects all information that is publicly known about those securities. The price of stock is based on the sum of future dividends, if there is a reasonable signal that causes changes in the expected future dividends, the price changes. Investors change their beliefs only when there is new information. The stock price mostly changes around event days, when earnings are announced. This event results in the trading activity of investors. So the significant trading volume activity around event days could be interpreted as signals of new information, high earnings quality.
Prior literature finds that ERC is a observable proxy for earnings quality or for earnings informativeness (Ball & Shivakumar, 2008; Ball & Brown, 1968). These papers show that earnings news is related with different equity market characteristics that result investors to change their equity valuations. Based on the studies of Ball and Shivakumar (2008) and Ball and Brown (1968) I can conclude that earnings anouncements do not signal significant new information to investors. Because the most information is already anticipated by investors and the magnitude of relative importance of new information in earnings to investors is quite low. However these studies only address the decision usefulness of earnings announcements in equity valuation and not whether the results have implications for earnings quality. Liu and Thomans (2000, p.73) find evidence on the ERC as a proxy for earnings quality. They find that the ERC and the R2 of the ERC regression are high when the correlation between unexpected earnings and forecast revisions is high. They measure unexpected earnings as actual earnings for (t) minus the forecast of period (t) earnings in t-1, and earnings forecast revisions for future periods are measured using information available at time (t). So, if current unexpected earnings are informative to analysts in the way that they cause a forecast revision which suggest, that when the earnings are of higher quality the ERC is also higher. However they do not know whether earnings quality is influenecd by earnings smoothing or due to low persistence of fundamental performance. There is prior literature which show indirect evidence that ERC can be used to determin earnings smoothing. These are based on the effects of accounting methods, auditor quality, governance, firm fundamentals and leverage on earnings (Dechow, Ge, & Schrand, 2010). The market responds to this quality of information when it reflects the firms performance. When the market feels that the earnings announcements signals to much low quality information, like accruals, discretinary and earnings manipulation the market wont respond to that news. A way to measure ERC is to estimate the following model:
Where Abnormalreturnt is measured for days over a average month window centered on the announcement date. To measure abnormal returns I follow Mahajan and Singh (2011), where they measure it based on the market adjusted model. Their market model assumes that there is a stable linear relationship between the actual return and market return. Abnormal return measures the earnings response coefficient, which measures the extent of a stocks abnormal market return in response to the suprise characteristics of announced earnings. A higher β indicates that earnings provide more quality information to the market. Following Gu and Wu (2003, p. 10) earnings surprise is defined as the difference between the I/B/E/S actuals earnings per share (EPS) and the analyst forecast deflated by the standard error at the beginning of the year. Using the stock price has the advantage that both the numerator and the denominator of earnings surprise is retrieved from I/B/E/S. Unexpected earnings is measured by:
= −
For my second model I estimate the following empirical model, where I include the interaction with different forms of equity incentives and without the control variables:
Abnormalreturnt= β0+β1EARSt+β2RSTKGRNTt+β3STCKWRDSt+β4OPTAWRDSt+β5CASHt+
β6EARSt*RSTKGRNTt+β7EARSt*STCKWRDSt+β8EARSt*OPTAWRDSt+
β9EARSt*CASHt+εt (Model 2b)
A positive and significant coefficient for β6;7;8 and 9 implies that the sensitivity of abnormal returns vary as a function of the different forms of equity compensation. For my next model I estimate the following empirical model, where include the interaction with, restricted shares, stock awards, option awards and cash payments:
Abnormalreturnt= β0+β1EARSt+β2RSTKGRNTt+β3STCKWRDSt+β4OPTAWRDSt+β5CASHt+
β6EARSt*RSTKGRNTt+β7EARSt*STCKWRDSt+β8EARSt*OPTAWRDSt+
β9EARSt*CASHt+β10FIRMSIZEt+β11LEVERAGEt+β12MVt+β13BTMt+
β14GROWTHt+β15BOARDSIZEt+β16BOARDINDt+β17CEOCHAIRt+
β18CEOCOMPSt+β19BIG4t+εt (Model 2c)
A positive and significant coefficient for β6;7;8; and 9 implies that the sensitivity of abnormal returns to earnings surprises vary as a function of the amount of given compensation. Which could mean that the grant of compensation is related to higher earnings quality and less earnings smoothing, dependent on the ERC. Table 1 summarizes all the variable measurements.
Table 1. Variable measurement
Dependant variables: Measurement:
EARNINGSt+1 income before extraordinary itemst+1/average assets ((Assetst+Assetst+1)/2)
Abnormalreturnt market adjusted returns over a month window centered on the announcement
date. Actual return- S&P-market-portfolio return
Independent variables: Measurement:
EARNINGSt income before extraordinary itemst/average assets (Assetst-1+Assetst/2)
EQUITYt (stock optionst + restricted sharest + stockt) / TDC1t
RSTKGRNTt restricted sharest/ total compensationt
STCKWRDSt stock awardst/ total compensationt
OPTAWRDSt option awardst/ total compensationt
CASHt (TDC1t-(restricted shares+stock awards+option awards))t/ TDC1t
EARSt −
Control variables: Measurement:
LEVERAGEt total liabilitiest/ total assetst
FIRMSIZEt logarithm of the total assetst at the year end, log(ATt)
MVt logarithm of the market valuet
BTMt the percentage change in book to market, ((Bookvaluet+1/marketvaluet -
Bookvaluet/market valuet-1) / Bookvaluet/market valuet-1)
GROWTHt growth in sales, (salest/salest-1)
BOARDsizet the total number of directors that integrate the board of directorst
BOARDSIZEt logarithm of the total number of directors that integrate the board of directorst
CEOCOMPSt 1 if there is duality of roles between the CEO and the chairman of the
compensation committeet and 0 otherwise
CEOCHAIRt 1 if there is duality of roles between the CEO and the chairman of the boardt
and 0 otherwise
BRDINDt # external directors inside the Board (external directors/total directors)t
4. Findings
In this chapter I will discuss my data collection, my descriptive statistics and finally my results. 4.1. Data collection
I retrieve my data from Wharton Research Data Services (WRDS)1, which is a research platform and business intelligence tool, where I started collecting data from ExecuComp. ExecuComp2 is a database which contains data on the major aspects of compensations for top five executives (ranked annually by salary and bonus) at each of the firms in the S&P 500, S&P Midcap 400, and S&P SmallCap 600. Due to enhanced federal reporting requirements for fiscal years ending after December 15, 1992 I can measure incentives for the years 2000 to 2007. My initial sample contained 89759 observations with data on CEO annual compensation and CEO characteristics for the fiscal year 2000-2007, retrieved from ExecuComp. However this sample was not representative as it was influenced by CEO turnover. Clayton, Hartzell and Rosenberg (2005) mention that the turnover of top management have influence on the discretion over R&D, advertising, capital expenditures and accounting accruals. Furthermore they find that stock prices also respond more strongly to earnings announcements following turnovers. These results are informative signals of future value, which indicate possible changes in the company’s strategy and confidence about the successor CEO's ability. Lee, Matsunaga and Park (2012) investigate whether forecast accuracy provides a signal regarding CEOs’ ability to anticipate and respond to future events by examining the relation between forecast errors and CEO turnover. They find that the probability of CEO turnover is positively related to the magnitude of absolute forecast errors. Therefore, accounting earnings during CEO turnover years are less relevant for predicting future earnings. To avoid concerns related to CEO turnover, I eliminate firms in which the CEO left during the sample period, which resulted in 9764 observations. With this data I have calculated the relative proportion of different equity compensations granted to CEO’s.
Secondly I used this sample to retrieve my gvkeys to obtain data from COMPUSTAT. This database contains fundamental financial, statistical, and market data for U.S. and Canadian corporations with extensive coverage of annual income statement, balance sheet, cash flow and
supplemental data items3. My base sample consisted of 14601 observations. Which contained data on total assets, book value, market value, income, sales, liabilities and others. I have started with calculating my dependent variable EARNINGSt, where I have also created a future earnings variable which required forward looking information, EARNINGSt+1;2 and 3. Secondly I calculated my first control variable FIRMSIZEt, then LEVERAGEt, followed by GROWTHt, BTMt and MVt. I Have dropped duplicates and data with empty values. This resulted to 12346 observations. Thirdly I merged ExuComp data with COMPUSTAT data and kept only successful merges which resulted in 8160 observations. From this data I retrieved the unique keys, CUSIP.
Fourthly I used CUSIP to obtain data from Riskmetrics (ISS). RiskMetrics4 is a database which contains data of corporate governance characteristics. From the directors legacy I have retrieved data on the classification of board independence, if the CEO is in the compensation committee, is the chair of the compensation committee, is the chairman of the board, busy outside in public boards and others. I have obtained data on all directors in the board to compute the control variables BOARDSIZEt, BOARDINDt, CEOCHAIRt and CEOCOMPSt. In untabulated results I find that only a small proportion of the CEO is active in the board and also active on the compensation committee. Which suggest that a significant part of my observation has good governance regarding compensation contracts. I have merged this sample with the previous mentioned merge and obtained 7629 observations.
Fifthly I started collecting data from Audit Analytics. Audit Analytics5 provides detailed research on over 150,000 active audits and more than 10,000 companies. From this point I obtained unique keys from my last merge to retrieve data from Audit Analytics. However I couldn’t use them to find my data. I had to search the entire database. I retrieved data of the active auditor for that company in my sample for that year. To create a dummy variable if the audit is performed by a big 4 auditor. I started with a sample of 28880 observations, after computing and merging I obtained a final sample of 7621.
Finally when I obtained my complete data for the first model I dropped the years 2000 and 2007, because I had empty cells. These where empty because of creating lagged and forward data. Once I obtained my sample I winsorized all the variables (except indicator/dummy variables) to the 1st and 99th percentiles of their distribution. This eliminates extreme values
3 COMPUSTAT. (2015, April 11). Retrieved from https://wrds-web.wharton.upenn.edu/wrds/ds/comp/index.cfm
4 Riskmetrics. (2015, April 12). Retrieved from https://wrds-web.wharton.upenn.edu/wrds/ds/riskmetrics/index.cfm 5
from my data which otherwise would have led to extreme outliers. Further, similar to other studies I eliminate firms with SIC codes between 6000-6999, because the accruals of these firms are likely to be different from accruals of firms in other industries. I also eliminated empty values of (some) control variables, which leads to 3457 firm year observations from 2001 to 2006.
In addition I started collecting data on earnings surprises around announcement dates from the I/B/E/S database. I/B/E/S6 is an database which contains individual analyst forecasts of company earnings, cash flows, and other important financial items, as well as buy-sell-hold recommendations. My base sample consisted of 91734 observations, this high amount of observations was mainly due to the identification of event dates. I have calculated my earnings surprise variable for each month for the companies. This data collection was simultaneously in line with my data collection of abnormal returns, because of the available event dates. I had collected my data to calculate abnormal returns from CRSP database. CRSP7 maintains the most comprehensive collection of security price, return, and volume data for the NYSE, AMEX and NASDAQ stock markets. Additional CRSP files provide stock indices, beta- and cap-based portfolios, treasury bond and risk-free rates, mutual funds, and real estate data. For the calculation of my abnormal returns I started with a base sample of 104080 observation. I calculated the S&P-average-market-based-portfolio-return and the actual return around the event date. I have chosen to calculate the earnings surprise and abnormal returns by a month. Mainly because of avoiding merging problems with my master-file and extending the effects of the pre-and post-announcement drift, which is in line with Kaniel et. al (2012) pre-and Kothari, Lewellen, and Warner (2006). After this I merged it with my earnings surprise data and dropped duplicates and empty values, which lead to a sample of 8459 observations by year and month. The new added variables were also winsorized. Finally I merged this file with my data. Due to some restrictions of my master-file, I have chosen to keep data regarding abnormal returns and earnings surprises that contains the median of earnings surprises8, because my master-file contained annual data and the using-file monthly data. This process led to a final sample of 3133 observations where the variables are normally distrusted. Table 2 summarizes the sample derivation.
6
I/B/E/S. (2015, April 12). Retrieved from https://wrds-web.wharton.upenn.edu/wrds/ds/ibes/index.cfm
7 CRSP. (2015, April 12). Retrieved from https://wrds-web.wharton.upenn.edu/wrds/ds/crsp/index.cfm
8 Earnings surprises data was not sorted by month, so the median would not be equal to 7. So, year
t is matched with
4.2. Descriptive statistics
In table 3 you can find an overview that presents the descriptive statistics of the sample which is relevant for my models. The distributional characteristics of the variables are comparable to those in prior literature that examines similar variables over similar periods of time (Gong & Li, 2013). Earnings of the next period (mean= 0.053) is higher than current year earnings (mean=
Table 2. Sample selection
Criteria Observations
Executive Compensation base sample between 2000-2007 89759
Observations without CEO- turnover and available equity data 9764
COMPUSTAT base sample of accounting fundamentals 14601
Dropping duplicates and empty cells. 12346
Merged ExuComp- with COMPUSTAT data and kept successful merges 8160
Riskmetrics governance base sample 101409
Dropping missing values, calculating controls and merging 7629
Audit Analytics base sample 28808
Dropping missing values, calculating controls and merging 7621
Dropped the years 2000 and 2007 5789
Dropped data with SIC codes between 6000-6999 4723
Dropped empty values of governance variables 3457
Winsorized dependant- and independent variables 3457
Final sample for the first empirical model 3457
I/B/E/S base sample, calculating relevant variables 91734
CRSP base sample, calculating relevant variables 104080
Merging I/B/E/S with CRSP and dropping empty values 8459
Merging with my master-file and winsorizing control variables 3133
Final sample 3133
0.045), which indicates an average increase in profitability for the sample firms of mean 0.007. This explains that over the sample period there are a small fraction of the firms who exhibit no sustainable earnings. Of the sample firms, firms have on average 9 board members from which 69 percent is independent. On average 64 percent of the CEO’s is also the chairman of the board, which allows more room for management discretion. On average 0.1 percent of the CEO’s is also the chairman of the board and the compensation committee, which means that the CEO has no
power on its incentive contract. Interestingly CEO’s are more granted stock options (mean= 0,34) then (restricted) shares (mean= 0,09), which could indicate that they are more short term oriented then long term. Surprisingly there are firms in my sample who don’t grant equity compensation to their CEO’s and stick with cash payments (max= 1). In my sample the market is confronted by an average earnings surprise of -.00325, which suggest that analysts are conservative with their forecasts. There is also an average abnormal return of 0.037 percent in to which the market reacts during certain events. There is only a small fraction of my sample which is audited by a big 4
Table 3. Summary descriptive
Variable Obs Mean Std. Dev. Min .25 Median .75 Max
EARNINGSt+1 2964 .05250 .08417 -.44667 -0.03 0.06 0.09 .28241 EARNINGSt 3128 .04528 .09554 -.63867 -0.02 0.05 0.09 .27461 LEVERAGEt 3129 .50099 .20677 .08188 0.35 0.51 0.64 1.3074 FIRMSIZEt 3133 7.4967 1.4750 3.7524 6.42 7.36 8.40 12.073 MVt 3131 7.6262 1.4812 1.9122 6.59 7.46 8.56 12.992 BTMt 2950 .00005 .00221 -.01933 -0.00 -0.00 0.00 .01847 GROWTHt 2406 -2.359 .99349 -9.6043 -2.80 -2.19 -1.72 -.2909 BOARDSIZEt 3133 2.1731 .25987 1.3863 1.95 2.20 2.40 2.9957 BOARDsizet 3133 9.0814 2.3276 4 7 9 11 20 CEOCOMPSt 3133 .00128 .03571 0 0 0 0 1 CEOCHAIRt 3133 .64380 .47895 0 0 1 1 1 EQUITYt 3133 .44162 .32701 0 0.13 0.44 0.70 1 OPTAWRDSt 3133 .32049 .32595 0 0.00 0.26 0.58 1 RSTKGRNTt 3133 .08549 .18604 0 0.00 0.00 0.05 1 STCKWRDSt 3133 .03369 .11907 0 0.00 0.00 0.00 .75063 CASHt 3133 .52711 .33108 0 0.27 0.53 0.81 1 Abnormalreturnt 3133 .00370 .02796 -.08134 -0.00 0.00 0.03 .06494 EARSt 3133 -.0033 .17117 -.55000 -0.05 0.00 0.02 .5 BIG4t 3133 .00830 .09073 0 0.00 0.00 0.00 1
accounting firm. This indicator variable will probably add no power to my regression.
In table 4 I present a correlation matrix for my dependent- and predictor variables. The table presents the coefficient at a p-value of 0.01, 0.05, 0.10. The Pearson correlation coefficients of 0.6013 in table 4 indicates that EARNINGSt+1 and current earnings is positively and significantly correlated at a p-value of 0.01. As expected, these results suggest that current earnings are persistent and have predictive power of future earnings.
Table 4. Correlation matrix between variables
* significance level at 0.10 ** significance level at 0.05 *** significance level at 0.01 Variable measurement can be found in table 1.
EARNINGSt+1 EARNINGSt EQUITYt RSTKGRNTt STCKWRDSt OPTWRDSt CASHt Abnormalrett EARSt
EARNINGSt+1 1.0000 EARNINGSt 0.6013*** 0.0000 1.0000 EQUITYt 0.0215 0.2421 0.0224 0.2105 1.0000 RSTKGRNTt 0.0158 0.3911 0.0168 0.3484 0.3346*** 0.0000 1.0000 STCKWRDSt 0.0469** 0.0106 0.0853*** 0.0000 0.0489*** 0.0061 -0.1300*** 0.0000 1.0000 OPTAWRDSt -0.0061 0.7396 -0.0212 0.2358 0.7844*** 0.0000 -0.1848*** 0.0000 -0.2782*** 0.0000 1.0000 CASHt -0.0065 0.7223 -0.0220 0.2183 -0.8601*** 0.0000 -0.2870*** 0.0000 -0.0216 0.2267 -0.6818*** 0.0000 1.0000 Abnormalrett 0.0148 0.4213 0.0405** 0.0236 0.0325* 0.0691 0.0487*** 0.0064 -0.0242 0.1758 0.0138 0.4393 -0.0242 0.1748 1.0000 EARSt 0.0093 0.6121 0.0023 0.8991 0.0196 0.2716 0.0239 0.1804 -0.0192 0.2820 0.0148 0.4067 -0.0059 0.7420 0.0063 0.7232 1.0000
In addition, the Pearson correlation between EARNINGSt+1 and EQUITYt is positive (0.0215) but not significant. Which is not in line with my hypothesis. This suggests that current equity compensations grants to the CEO does not have predictive power for future earnings. However this includes total equity, specifically stock options and its related effect on risk seeking behavior. But when I look solely at the positive Pearson correlation of STCKWRDSt (0.0469), it suggest that stock awards might have an influence on future earnings, which is in line with my hypothesis 1b. It is found that stock awards promote a longer-term focus in order to increase earnings persistence. When I look at the positive Pearson correlation (0.0224) of current earnings with EQUITYt, this does not suggest that current year earnings influence current equity compensation. The Pearson correlation of CASHt (-0.0220) is negatively but not significantly correlated with current year earnings, which might indicate that the change in current year earnings does not have a negative effect on cash compensation. Staying optimistic, it could signal that a CEO is alternatively compensated with equity, as the Pearson correlation (0.0853) EARNINGSt with STCKWRDSt is positive and significantly correlated. Not surprisingly EQUITYt is positively and significantly correlated with RSTKGRNTt, STCKWRDSt and OPTAWRDSt, because of defining EQUITYt as the summary measure of these compensation forms. Interestingly EQUITYt has a negative and significant effect on CASHt with a Pearson correlation of (-0.8601). This result suggest that a greater proportion of equity is more likely to be related with a smaller proportion of cash compensation and a larger proportion of equity compensation. The positive Pearson correlation results (0.0063) of earnings surprise (EARSt) with Abnormalreturnt is not significantly correlated. Which indicates that the magnitude of abnormal returns has no association with the magnitude of earnings surprise revealed through public events. This is consistent with the efficient market theory, because it suggests that the market did anticipate the effect of the event on the security.
To summarize the correlation matrix, current earnings is correlated with stock awards. And current earnings is correlated with next year’s earnings, so stock awards might have predictive power of future earnings, which is in line with my hypothesis. Abnormal returns and earnings surprise is not correlated, which means that the earnings announcement signals to much low quality information and the market does not respond to that event.
