The effect of CEO’s social media reputation on earnings management

The effect of CEO’s social media

reputation on earnings management

University of Groningen

Faculteit Economie en Bedrijfskunde

MSc. Accountancy & Controlling

Master Thesis

Abstract: This study focuses on CEO’s online media reputation and its effect on earnings management (EM). Prior literature examines two (contradicting) economic perspectives. The efficient contracting perspective predicts a negative relationship between CEO reputation and EM. In contrast, the rent extraction perspective predicts a positive relationship between CEO reputation and EM. In this study a proxy of Twitter data and social media awards has been used for measuring CEO’s online media reputation. Furthermore two channels of EM, accruals-based management (AM) and real activities management (RAM) are taken into consideration. The sample consists of CEOs of the S&P 500 over the 10-year period 2005-2014, including 3,614 (AM) and 2,294 (RAM) firm-year observations from 719 CEOs and 388 firms for AM (RAM: 416 CEOs and 256 firms). Concerning RAM, this study finds after addressing the selection bias problem, a significant negative relationship between CEO reputation and RAM. This result holds after a robustness check. The found relationship is supportive of the efficient contracting hypothesis. It indicates that more reputable CEOs are less likely to indulge themselves in RAM. Thus more reputable CEOs execute more often activities that are in the best interests of the company they work for.

Student name: Martin Douwes

Student number: s1903896

Student e-mail: m.douwes@student.rug.nl

Supervisor: Dr. Y. Karaibrahimoglu

Second Assessor: Dr. N. Hussein

Date: 23-01-2017

2

Table of contents

1 INTRODUCTION ... 3 2 THEORETICAL FRAMEWORK ... 7 2.1 AGENCY THEORY ... 7 3 LITERATURE REVIEW ... 8

3.1 SOCIAL MEDIA AND ITS IMPORTANCE ... 8

3.2 EARNINGS MANAGEMENT ... 9 4 HYPOTHESIS DEVELOPMENT ... 12 5 RESEARCH METHOD ... 15 5.1 SAMPLE ... 15 5.2 RESEARCH MODEL ... 15 5.3 MEASUREMENT OF VARIABLES ... 16 6 RESULTS ... 20 6.1 SUMMARY STATISTICS ... 20 6.2 PEARSON CORRELATION MATRIX ... 22 6.3 REGRESSION ANALYSIS ... 23 6.4 ADDITIONAL ANALYSES ... 27 7 CONCLUSION AND DISCUSSION ... 39 8 REFERENCES ... 42

3 1

Introduction

On August 24, 2011, Steve Jobs announced his resignation as Chief Executive Officer (CEO) of Apple. Apple’s stock prices immediately dropped by 3 percent after this announcement, equal to about $10 billion of company value. This decline of Apple’s stock resulted from a lack of investors’ confidence that the company would continue to deliver the positive results as it did when Steve Jobs was CEO (Fetscherin, 2015). This example illustrates the importance of CEO’s reputation for a company. The impact of CEO’s reputation can be significant according to David Larcker, a professor at Stanford University. He states that a 10% increase in a CEO’s reputation leads to a 24% improvement of a firm’s market value (Gaines-Ross, 2000). CEOs are becoming more and more the public image of their organisations and therefore treated as a brand on their own (Bendisch et al., 2013). In the United States CEOs can even receive awards, like movie stars, such as a ‘CEO of the year’ award. Examples of other famous CEOs are Michael O’Leary (Ryanair), Richard Branson (Virgin) and Bill Gates (Microsoft). The reputation of such CEOs can take up to roughly 50 percent of the public’s opinion of a company’s reputation (Burson-Marsteller, 2006; Kitchen & Laurence, 2003). These studies indicate that CEO reputation constitutes an important aspect of the organisation. Currently CEOs become so popular, that they built a personal reputation – voluntarily or not – that exceeds the organisational boundaries (Rosenberger, 2015). This raises the question of whether CEO reputation is a paragon or a parasite for organisational practices?

The CEO’s position is at the top of an organisation and is the highest ranked executive. Because of this position the CEO has relatively higher power and influence in comparison with any other individual organisation member. Therefore a CEO is regarded as the key decision maker over corporate decisions, such as in investments, financing and operating decisions (Boivie, Lange, McDonald & Westphal, 2011; Finkelstein et al., 2009). The traditional economic theory has suggested a limited role of managerial influence on organisational outcomes (Koh, 2012; Francis et al., 2008). A commonly cited study supportive of this view is Lieberson and O’connor (1972), which shows that a firm’s profitability is mostly affected by forces beyond a CEO’s immediate control (Finkelstein et al., 2009). However, there is an emerging stream of studies documenting that CEOs do affect organisational outcomes, like corporate performance (Bertrand & Schoar, 2003). Several studies have shown that observable characteristics of CEOs have significant effects on a company’s strategical decisions and performance, such as age, tenure and education (Carpenter et al., 2004; Finkelstein & Hambrick, 1996). Moreover, recent studies also relate various managerial personal traits to organisational outcomes. For example, narcissism (Olsen et al., 2014; Chatterjee & Hambrick, 2007), overconfidence (Malmendier & Tate, 2008), risk aversion (Graham et al, 2013) and reputation (Francis et al., 2008; Malmendier & Tate, 2009; Jian & Lee, 2011). Given the growing evidence of associations between managerial traits and firm outcomes, I think it’s reasonable to assume that CEOs have an impact on organisational outcomes.

4

In this study I’ll examine a specific managerial trait, namely the reputation of a CEO. In particular, I’ll focus on CEO’s online media reputation and its effect on financial reporting practices. There is some literature on CEO’s reputation (e.g. Francis et al., 2008; Malmendier & Tate, 2009; Jian & Lee, 2011). However, research aimed at specific CEO’s online media reputation is virtually absent. This may be explained because the emergence of social media platforms is very recent (e.g. Twitter’s starting date was in 2006). Nevertheless, the lack of research on this topic is surprising. Academics have started to conduct research on corporate’s use of social media and therefore is highly related to this study. Twitter has been regarded as an important tool for the communication between an organisation and their stakeholders (Alexander & Gentry, 2014). Social media enables a company to have an active dialogue with its stakeholders, which enhances their engagement with the company and therefore corporate reputation (Castriotta et al., 2013; Floreddu et al., 2014). Thus, the recent available internet technologies have a significant impact on the interaction between companies and other related parties (Kietzmann et al., 2011; Hanna et al., 2011). This applies to CEOs and their relationships with other stakeholders as well. In the past CEOs could execute their activities without paying attention to their social media presence (Holmes, 2016; Weber Shandwick, 2015). Nowadays, however, this has changed. As the public image of their firm, CEOs need to get active and be social (Gaines-Ross, 2013a). Currently the (social) media is more present than ever and CEOs become aware of this fact. A survey of Weber Shandwick (2013) shows that ‘CEO sociability’ increased from 36% to 66% between 2010 and 2012. ‘CEO sociability’ means the presence on social media, writing a blog or posting messages on the company’s website. Gaines-Ross (2013a, p.5) states that “social media is no longer the wave of the future.” In today’s digital connected world, social media has ‘become modern PR for executives’ (BRANDfog, 2014, p 3). Being active on social media allows CEOs “to be accessible, establish relationships and listen to and communicate with individuals” (Gaines-Ross, 2013b, p.1). Through (online) social media channels, CEOs can build on their personal image and reputation (Gaines-Ross, 2013b). If stakeholders identify themselves with your brand, “they are inspired by you, want to be around you, work with you and do business with you” (Bates, 2012, p. 13).

Public listed companies need to convey information about their financial performance to external capital providers and other stakeholder. Financial statements play an important role in this communication with external stakeholders. The Financial Accounting Standards Board (FASB) state that financial statements are ‘a principal means of communicating financial information to those outside an entity’ (Statement of Financial Accounting Concepts No. 5, 1984). Financial statement users consider CEO’s reputation in assessing the quality of financial reporting (AICPA, 1994). Also, results of a survey held under Chief Financial Officers (CFOs) indicate that manager’s concern regarding his external reputation influences financial reporting decisions (Graham et al., 2005). However, it’s unclear whether the reputation of a CEO has a positive or negative effect on financial reporting practices. The reason for this uncertainty has its origin in the classical ‘agency problem’

5

(Fetscherin, 2015). This problem exists in publicly traded companies where control and ownership are separated (Fama, 1980). In such companies, CEOs have substantial influence and power (Jensen & Meckling, 1976) and incentive problems in CEO’s decision-making may be an issue (Fama, 1980). In other words, it is not necessary the case that the interests of the CEO aligns with the interests of the shareholders. In their financial reporting practices, CEOs have a considerable amount of discretion. Standard setters provide accounting policies (for example GAAP) from which a CEO has to choose. They permit executives to use their judgement in financial reporting in order to enable them to convey (inside) information to external stakeholders. This way executives can make use of their knowledge and are able to select the accounting policies that reflect the organisation’s true underlying economic performance. However, executives can also make use of their judgement to engage in earnings management (EM). When this is the case, they do not necessarily select accounting policies that correctly reflect their organisation’s true underlying economic performance (Healy & Wahlen, 1999). Furthermore, most studies use accruals management (AM) to estimate EM (Francis et al., 2008). However CEOs can also indulge in EM through real activities management (RAM). Cohen et al. (2008) suggest that there has been a shift from AM to RAM after the introduction of SOX. Therefore it’s important to consider these two channels of EM, since the sample used in this study is from a post-SOX period (2005-2014). Also, Fields et al. (2001) argue that one channel cannot explain the overall effect of EM. Therefore, in this study, I investigate how CEO’s reputation affects financial reporting practices. In particular, I examine the effect of CEO’s social media presence on earnings management. A substantial amount of research on firm’s reporting decisions primarily focuses on the effects of firm-specific characteristics (e.g. size, growth, leverage) and ignore the possible effects of manager-specific traits. However, recent studies show that those firm-manager-specific characteristics don’t explain the heterogeneity found in firms’ reporting decisions (Larcker, Richardson, & Tuna, 2005; Bowen, Rajpogal, & Venkatachalam, 2008). The contribution of this paper to the existing (accounting) literature is providing empirical evidence to explain the variation in financial reporting practices from a managerial perspective, namely CEO’s social media reputation. Up to my knowledge, I’m the first one to research this form of reputation and its impact on EM. Furthermore I take both forms of EM (AM and RAM) in consideration, where most previous studies only have the single focus on AM. Currently, it is not well understood what CEO reputation is (Bendisch et al., 2013) and there is no ‘widely-agreed’ framework to measure CEO reputation. This is because CEO reputation is ‘multidimensional and unobservable’ (Francis et al., 2008, p. 141). Recent studies investigating CEO reputation have mainly used CEO’s press coverage as a proxy to measure CEO reputation (e.g. Francis et al., 2008; Jian & Lee, 2011; Milbourn, 2003). However, this proxy may be biased which leads to measurement error (Core et al., 2008; Miller, 2006; La fond, 2008; Koh, 2012). Other researchers have tried to proxy CEO reputation by using awards that CEOs won (e.g. Malmendier and Tate, 2009) and

6

by experiments (Cianci and Kaplan, 2010). These examples illustrate the difficulties concerned with measuring the reputation of a CEO. Because research up to now uses different proxies, their findings are also different and can even be contradicting. These contradicting results ask for more research to construct a more valid proxy for CEO reputation (LaFond, 2008). In this study, I use CEO’s social media reputation to proxy for overall CEO reputation. This proxy consists of Twitter data and social media awards. I contribute to the existing literature by providing empirical testing of this dimension of CEO reputation.

The remainder of this thesis is as follows. Chapter 2 explains the theory behind the ‘agency problem’. In chapter 3 a literature review will be provided on the importance of social media and earnings management. Subsequently, the hypothesis development will be explained in chapter 4. Explanation of the used methodology will follow in chapter 5. In chapter 6 an overview of the results will be given and chapter 7 provides the discussion and conclusion of this thesis.

7 2

Theoretical Framework

2.1

Agency Theory

The agency theory is defined as the contractual relationship between the owner (the principal) and the CEO (the agent). The CEO works on behalf of the owner and therefore (parts of) the decision making authority is delegated to the CEO (Jensen & Meckling, 1976). This separation of control and ownership is mainly to be found in publicly traded companies (Fama, 1980). In such companies, CEOs have substantial influence and power (Jensen & Meckling, 1976) and incentive problems in CEO’s decision making may be an issue (Fama, 1980). In other words, it is not necessary the case that the interests of the CEO aligns with the interests of the shareholders. Shareholders (principle) have the possibility to buy shares of different companies and thereby diversifying their risk. Therefore they’re risk neutral. However, a CEO (agent) is not able to diversify his risk and as a result is risk averse. Thus, the principle and agent have different risk preferences (Eisenhardt, 1989). To align these interests a principal-agent contract can be established. To be able to establish such a contract, the principle needs to have the ability to know what the agent is doing. This is unlikely, because shareholders (i.e. principle) often don’t have the resources and time to assess what the executive is doing. This leads to information asymmetry. Information asymmetry might be the result of adverse selection and moral hazard. When a principle is hiring an executive to carry out work for him, he might exaggerate his competences and skills to get the job. The adverse selection problem arises here when the principle is not able to verify the executive’s true capabilities before hiring. The second problem, moral hazard, arises after the hiring. The information asymmetry exists here because the executive possesses inside information, which the shareholders don’t have access to. This is especially the case in listed public companies, where ownership is dispersed and as a result the shareholders are too small and don’t have the resources to assess the activities undertaken by the executive. Since the executive has a different risk preference and interest, he might shirk or may not execute his duties as agreed with the shareholders.

Agency costs are incurred by the principle in trying to minimize the unobservable behaviour arising from information asymmetry. The principle has two options in doing so. The first option is to monitor the actions of the agent. The other option is to assign a part of the agent’s compensation to a performance measure. This way the principle can reduce the divergence of interests between him and the agent (Eisenhardt, 1989; Jensen & Meckling, 1976). In addition, the agent can put an effort in ensuring that he won’t undertake any actions that are not aligned with the interests of the principle. These are called bonding costs. However, in no situation there will be assurance that the interests of the agent are fully aligned with the interests of the principle. Therefore there will always be a residual loss. Thus, agency costs consists of the sum of monitoring expenditures incurred by the principle, bonding expenditures by the agent and the residual loss (Jensen & Meckling, 1976).

8 3

Literature Review

3.1

Social Media and its Importance

In today’s digital world internet constitutes an important aspect of people’s life. 87% of the American adults use the internet and 73% of the American adults who are active online are using social networking platforms (Pew Research Internet Project, 2014). In the last decade, social media platforms such as Twitter have experienced an exponential growth. With starting date March 2006, Twitter has grown to 313 million active users a month who sent 500 million tweets on a daily basis. Social media changed the way on how people interact with each other and with companies (Kietzmann et al., 2011). Unlike the traditional information channels (for example the business press), companies are able to directly communicate with investors and other stakeholders. Blankespoor et al. (2014) label these new platforms (e.g. Twitter) as direct-access information technologies (DAITs), since they allow organisations to directly communicate with stakeholders. It provides a two-way interaction tool (Aula ,2011) and reduces the resources needed by stakeholders to absorb the information (Blankespoor et al., 2014). Since the emergence of social media is very recent, research on this topic is scarce. Up to my knowledge, I’m the first to research the relationship between CEO’s social media reputation and EM. I only found one study that explores CEOs usage of Twitter and its impact on corporate performance. This study, Oh and Bunkanwanicha (2016), finds that CEO’s age and compensation level are positively linked with a CEO having a Twitter account. Regarding Twitter and its impact on corporate performance they don’t find consistent results. However, more studies are available that examine the corporate use of social media and therefore are closely related to this study. These studies will be considered next.

Social media provides new ways for a company to communicate directly with all parties who are interested. Social media allows firms to have an active relationship through interactive dialog with their stakeholders (Floreddu & Cabiddu, 2016). Not only companies use social media to interact with their stakeholders, but they also use social media management tools (SMMT) “to gain insights into the brand perception among users” (Risius & Beck, 2015, p. 824). A recent report published by Burson-Marsteller (2012) shows that 87% of the worldwide companies use at least one social media platform, with Twitter being the most commonly used (82%). Nowadays DAITs “have become an integral component of firm communications and investor relations” (Blankespoor et al., 2014, p. 84). Results of a study by Weber Shandwick (2013) show that social media can be a useful tool that can improve a CEO’s communications strategy. Kitchen and Laurence (2003) point out that the CEO is the main spokesman of all the communication activities of the organisation. The reason why a CEO is the main communicator of an organisation is because the position is inherently seen as the most visible leadership role in an organisation and establishes relationships with all the stakeholders. Survey results of Weber Shandwick (2013) show that 78% of the executives say that social media has a positive

9

impact on the company’s reputation. Therefore the CEO plays a major role in a company’s public relations. Public relations executives consider social network platforms as very important, because it provides an organisation with new opportunities to interact with a wide range of stakeholders (DiStaso, McCorkindale & Wright, 2011; Aula, 2011) and thus starting to adopt social media tools (Eyrich, Padman, & Sweetser, 2008). Corporate communications plays an important role in building a corporate reputation, because “a firm, through its chosen messages, enables stakeholders to understand the firm's operations, and it positively loads the perception of the firm’s activities, which can lead to an overall positive evaluation of the company” (Floreddu & Cabiddu, 2016, p. 490). Thus, social networks are considered to be valuable tools in gaining a corporate reputation (Kietzmann et al., 2011; Bunting & Lipski, 2000) and Floreddu and Cabiddu (2016) provide evidence in their case study that high reputable companies use social media to manage their corporate reputation.

3.2

Earnings Management

There is ample evidence that executives engage in EM (Roychowdhury, 2006). Burgstahler and Dichev (1997) provide evidence that 30% to 44% of the companies indulge in EM to avoid reporting a loss. Executives have the possibility to engage in EM, because they have a certain amount of discretion in the use of accounting numbers. Standard setters provide accounting policies (for example, GAAP) from which an executive has to choose. They permit executives to use their judgement in financial reporting, to enable them to convey (inside) information to external stakeholders. This way executives can make use of their knowledge and are able to select the accounting policies that reflect the firms’ true underlying economic performance. However, executives can also make use of their judgement to engage in earnings management. When this is the case, they do not necessarily select accounting policies that correctly reflect their firm’ true underlying economic performance (Healy & Wahlen, 1999). EM is defined by Healy and Wahlen (1999) as: “Earnings management occurs when managers use judgment in financial reporting and in structuring transactions to alter financial reports to either mislead some stakeholders about the underlying economic performance of the company or to influence contractual outcomes that depend on reported accounting numbers” (p. 368). Executives can engage in earnings management through two channels: accruals management (AM) and real activities management (RAM) (Kothari et al., 2016).

Accruals management. Research conducted on the topic of earnings management is mainly

focused on accrual-based earnings (Zang, 2012; Dechow et al., 1995; Roychowdhury, 2006). When executing accruals-based earnings management, executives make use of their judgement regarding accounting estimates and policies to bias reported earnings in a certain direction (Zang, 2012). The earnings of a company consists of two components, namely cash flow from operations and (net) accruals. Subsequently, accruals is decomposed into non-discretionary and discretionary accruals (Scott, 2015). Non-discretionary accruals are mandated by accounting policies (i.e. GAAP). For

10

instance, the depreciation method of fixed assets must be applied in a systematic manner. However, accounting policies also provide room for judgement. Discretionary accruals are adjustments to the cash flows where an executive can use his judgement. For instance, an executive can choose to accelerate or delay the depreciation method of fixed assets. Thus, the difference between these types of accruals is, is that over discretionary accruals an executive has considerable influence. Since accruals affect the timing of reported net income, an executive can shift earnings between periods and so enables an executive to engage in earnings management (Healy, 1985). Besides this opportunity for EM, it’s important to realize that accruals reverse. Meaning, an executive that adjust accruals upwards, will be confronted in subsequent periods with the reversal of these accruals to adjust future earnings downwards (Scott, 2015).

Healy and Wahlen (1999) give four examples where an executive’s judgement is involved. The first example where an executive has considerable discretion to use his judgement is in estimating live expectancy and residual values of fixed assets. Also judgement is involved in estimating obligations for pension benefits, impairments and provisions for bad debts. Secondly, executives need to decide which accounting method they use for fixed assets. For example the straight-line and accelerated depreciation methods. For inventory valuation methods they can choose between FIFO, LIFO or weighted average. Also the timing of costs and revenues is subject to judgement (for example, R&D and maintenance). Finally, there is judgement involved in deciding how to show certain transactions in the financial statement, like lease contracts on- or off balance.

Real activities management. The second channel to engage in earnings management is through

real activities management (RAM). RAM is defined by Roychowdhury (2006) as: “departures from normal operational practices, motivated by managers’ desire to mislead at least some stakeholders into believing certain financial reporting goals have been met in the normal course of operations” (p. 337). When managing accruals, there is no direct cash flow consequence. This is different when executives engage in real activities management. When conducting RAM, operational activities change. As a result, current and future cash flows are affected and therefore company value (Roychowdhury, 2006). The survey of Graham et al. (2005) find evidence that executives are willing to engage in RAM, in order to meet earnings targets (for example, analyst expectations), even though this manipulation potentially reduces company value. The survey of Graham et al. (2005) also shows that executives are more willing to manipulate earnings through RAM rather than AM. Cohen et al. (2008) give as a possible explanation for this shift the implementation of the Sarbanes-Oxley Act (SOX) in 2002. The SOX act ensured that questionable accounting practices (i.e. accrual choices) came under greater scrutiny and RAM is harder to detect than AM. As a result, executives switched from accrual-based manipulation to real activities management after the implementation of SOX. This is particularly relevant for this study, because the data used is over the years 2005-2014 (post-SOX).

11

Roychowdhury (2006) explains three opportunities to engage in RAM: sales manipulation, reduction of discretionary expenditures and overproduction. The first opportunity, sales manipulation, is realised by selling products at a reduced selling price or making the credit terms more flexible. This way sales can be (temporarily) increased. Another way to engage in RAM, executives are able to adjust discretionary expenses (for example, R&D, maintenance and advertising). When cutting or delaying these expenses, current outgoing cash flows are reduced. While this may result in meeting the current earning targets, it has the possible risk of lower future cash flows. The third opportunity is overproduction. When producing more products, fixed overhead costs can be spread over a larger number of products. As a result, the cost of goods sold (COGS) is lowered and therefore the executive can show a higher profit.

Incentives for earnings management. An executive can have several incentives to indulge in

earnings management. These motives may have an internal or external origin. Healy and Wahlen (1999) states three different incentives of earnings management: expectations and valuation of the capital market, contracts based on accounting information and government or anti-trust regulation. First of all, accounting numbers are an important source of information for investors to come to a valuation of a company’s stock price. Executives can have the incentive to manage the earnings in order to influence this valuation. For instance, an executive might want to boost the earnings before a stock offering. As a result, the company can maximize the income gained by a stock offering. Also shareholders don’t like much variation in profits and prefer a steady increase. This is because variation in profits could mean more risks for an investor. This is called ‘income smoothing’ and means a steady increase in profits and thereby an executive prevents a decline in share price. Furthermore expectations of the market (i.e. financial analysts) are important for a company. Failure to meet these expectations might have negative consequences. For these reasons executives might have the incentive to indulge in EM. Another frequent method, besides ‘income’ smoothing, is taking a ‘big bath’. The manipulation here is that a company reports a bigger loss than the actual loss. Future costs can already be allocated to the current year. This way a bigger loss will be reported, but in the subsequent years it will be easier to show a profit. Especially newly appointed executives might indulge in such behaviour; they can blame the previous executive for the incurred extra costs. A second incentive Healy and Wahlen (1999) state is because of contract reasons. A form of such a contract is executive remuneration. A CEO’s bonus may be linked with accounting variables, like a firm’s profit. When this is the case, a CEO wants to maximize the profits in order to receive higher compensation. Also lending contracts often include covenants. These covenants are designed so the money issuer (e.g. bank) possesses greater security that the loan will be repaid. A company needs to comply with these covenants, because violation means paying a penalty. So in order to avoid a violation, an executive may want to manage the accounting numbers to avoid high costs. The last incentive Healy and Wahlen (1999) state

12

for EM is because of government or anti-trust regulation. Some industries, like banking and insurance, face a rigid set of regulation in order for the government to monitor such industries. Accounting numbers (e.g. minimum capital requirements) are used for monitoring, so the government is able to assess the financial health of these companies. When a company fails to meet the requirements, it can face a number of penalties or higher monitoring intensity. An executive therefore has an incentive to prevent these problems and might manage the earnings upwards so the firm doesn’t drop below the requirements. On the contrary, there are reasons to minimize earnings. Reasons for this form of earnings management are not wanting to attract attention from competitors and regulators and payment of less taxes.

Because of the reasons explained above, one might assume EM is mainly used for negative purposes. There is, however, another (positive) side of EM. Executives can use their (inside) knowledge of the company to make financial reports more informative for the various stakeholders (Dechow et al., 2010). In other words, earnings management is not inherently bad and executives can use their discretion provided by the accounting standards to close the information gap between them and the external stakeholders.

4

Hypothesis development

Research in many different disciplines have discussed the effect of CEOs on companies, but is rather scant in the accounting literature. For instance in the finance and economics literature, Bertrand and Schoar (2003) show that differences in managerial style affect corporate decisions. The CEO also

13

influences corporate capital investments (Jian & Lee, 2011), stock returns and earnings performance (Johnson et al., 1993) and Graham et al. (2009) find that several CEO traits affect corporate financial decisions. Besides financial aspects, the CEO can also have an influence on non-financial aspects of companies. Examples are CSR investments (Borghesi, Houston & naranjo; 2014), fraudulent behaviour (Rijsenbilt & Commandeur, 2013) and the trust of analysts (Gaines-Ross, 2000). Regarding earnings management, survey results show that CEO’s reputation helps financial statements users to assess the quality of financial reports (AICPA, 1994). Given the evidence provided by the aforementioned studies, I predict an association between reputable CEOs and financial reporting practices.

In this study, I build on prior literature to examine two contradicting economic perspectives on the association between CEO reputation and earnings management. These two perspectives are the

efficient contracting and the rent extraction perspective.

The first, efficient contracting, is the ‘good’ side of CEO reputation and argues that reputable CEOs execute activities that are in the best interests of their companies. This perspective builds on the model of Fama (1980), in which market observers use a CEO’s history to derive some personal characteristics like credibility (Jian & Lee, 2011). For instance, CEOs build up a financial reporting reputation based on the accuracy of prior earnings forecasts (Williams, 1996). This acquired reputation by CEOs for credible financial reporting is considered to be an “enduring trait about a firm’s managers, referring to investors’ perceptions of managers’ competence and trustworthiness” (Mercer, 2004, p. 186). The reputation effect can help to mitigate the agency problems arising from information asymmetry between the CEO and investors (Cao et al., 2012). The SEC acknowledges the importance of reputation gained on the internet and therefore started to promote the use of internet technologies to disseminate information as a mechanism to minimize information asymmetry (SEC, 2008). Blankespoor et al. (2014) found that making use of Twitter to disseminate corporate’s news mitigates the information asymmetry gap. In the USA it’s common that CEOs can receive high-profile awards. These awards are based upon certain performance criteria that are acknowledged by key-stakeholders as credible and legitimate (Wade et al., 2006). As a result, receiving such an award boosts a CEO’s reputation. In order to retain this reputation, reputable CEOs have career-related incentives to pursue activities that align with stakeholder’s interests. Francis et al. (2008) state that “loss of reputation serves as a deterrent to reporting poor quality earnings when the capital value of the consequences of such an action is greater than the benefit of reporting low quality earnings” (p. 115). Furthermore, reputable CEOs are able to withstand short-term variance in their firm’s accounting performance without significant changes in stakeholder’s trust (Koh, 2012). Therefore, reputable CEO’s have less incentives to indulge themselves in accruals and/or real activities management. So according to the

14

trustworthy a CEO is in his financial reporting. This leads to a negative relationship between CEO reputation and earnings management.

H1a. The CEO’s social media reputation has a negative effect on accruals management. H1b. The CEO’s social media reputation has a negative effect on real activities management.

In contrast, the rent extraction perspective is the ‘bad’ side of CEO reputation. High reputable CEOs can acquire this status by winning prestigious awards. Malmendier and Tate (2009) find that such ‘celebrity CEOs’ create market and analyst increased expectations of future company performance. Reputable CEOs have incentives to meet these market expectations, because unable to meet is a potential indication of a CEO’s failure that might limit an executive’s future career employment possibilities (Graham et al., 2005). High reputable CEOs might also exploit their status by extracting rents from the firm and indulge themselves into perks, like private use of a corporate plane. Due to extensive use of self-indulgent behaviour, they might find it increasingly difficult to meet (or beat) the market expectations (Francis et al., 2008). The failure to meet (or beat) the expectations will lead to degrading the CEOs reputation. To prevent this, high reputable CEOs who experience difficulties in meeting or beating the expectations might submerge themselves into earnings management. Thus according to rent extraction perspective, I posit a positive relationship between CEO reputation and earnings management.

H2a. The CEO’s social media reputation has a positive effect on accruals management H2b. The CEO’s social media reputation has a positive effect on real activities management.

15 5

Research method

In the first paragraph the sample selection will be discussed. Paragraph 5.2 introduces the research model. Subsequently in paragraph 5.3, the data collection and the measurement of the variables used in the research model will be explained.

5.1

Sample

I begin my sample selection process by selecting all CEOs of the S&P 500 companies available on ExecuComp database. Since the aim of this research is on CEO’s reputation, I specifically take CEOs from the United States. The reason for this focus is that CEOs in the United States receive a higher level of attention from their stakeholders compared to other parts of the world (Graffin et al., 2012). My initial sample consists of 4,755 CEO-years observations from 920 CEOs and 500 firms over the 2005-2014 period. I exclude 867 CEO-years observations from financial firms (SIC 6000-6799) and 72 CEO-years observations from firms with ambiguity regarding CEO information. During this process, I eliminate for my AAM and ARAM sample 202 and 1,522 respectively observations with missing specific variables. My final sample consists of 3,614 (AAM) and 2,294 (ARAM) firm-year observations from 719 CEOs and 388 firms (AAM) (ARAM: 416 CEOs and 256 firms) over the 2005-2014 period.

5.2

Research model

In the following regression models, I include all the variables that I expect to have an effect on earnings management (both AAM and ARAMRD).

AAM = β0 + β1(CEO_social) + β2(SIZEit) + β3(LEVit) + β4(ROAit) + β5(AGE) + β6(TEN) +

β7(GEN) + β8(DUA) + year dummies +

ε

ARAMRD = β0 + β1(CEO_social) + β2(SIZEit) + β3(LEVit) + β4(ROAit) + β5(AGE) + β6(TEN) +

β7(GEN) + β8(DUA) + year dummies +

ε

Explanatory variables

CEO_social = CEO_social is measured by using two different measures: (i) Twacc

and (ii) factor analysis of Twacc & SocAw.

Twacc = Twitter account. 1 if CEO has Twitter account, 0 otherwise

SocAw = Social media award. 1 if CEO is on the list, 0 otherwise

Firm control variables

SIZE = Firm size

LEV = Leverage

ROA = Return on Assets

CEO control variables

AGE = CEO’s age

TEN = CEO’s tenure

GEN = CEO’s gender. 1 if male, 0 if female

16

The outcomes of the dependent variable (EM) can be positive and negative. This depends on the direction of managing the earnings upwards or downwards. To control for this (contrary) effect in my analysis, I’ll draw the analysis on the absolute amount of EM. These are called Absolute Accruals Management (AAM) and Absolute Real Activities Management Research & Development (ARAMRD). The higher the value of these variables, the higher EM is conducted. The closer the results are to 0, the less EM is conducted. When conducting the analyses on the absolute values, the direction of EM can’t be measured. In the additional analyses, I conduct the analysis on only the negative and the positive values to determine the direction of EM.

5.3

Measurement of variables

CEO’s social media reputation. I use the Twitter search function to verify which CEOs

possess a Twitter account (dummy variable). Furthermore I collect the following data regarding the Twitter account: CEO links to the company website (1 if yes/0 otherwise), number of tweets, number of followers, number of following, picture (1 if suit/0 otherwise) and picture (1 if smiling/0 not smiling). I also collect data on if a CEO has received a social media award. The awards I used in this research are from the digital platforms Richtopia and Hootsuite. Richtopia compiled a list from top CEOs. This list is based on an algorithm based on CEO’s social media influence (e.g. on Twitter/Facebook/Linkedin) and Klout scores. Also Hootsuite compiled an extensive list of influential CEOs on the social media. Hootsuite based their list in terms of how much a CEO is creating added value content for their followers using social media. This variable takes 1 if the CEO is on either list, 0 otherwise.

Earnings management. In order to measure earnings management, following Kothari et al.

(2016), I use two different measures; (i) real activities management and (ii) accruals-based earnings managements. All the data regarding both channels of earnings management (RAM and AM) is gathered from Compustat (wrds). The codes of the variables in compustat are provided in parentheses after each variable.

Measuring real activities management. Kothari et al. (2016) follow partly the systematic for

measuring RAM of Roychowdhury (2006). Roychowdhury (2006) explains three opportunities to engage in RAM: sales manipulation, reduction of discretionary expenditures and overproduction. However, Kothari et al. (2016) solely focus on the reduction of discretionary expenditures. In particular the R&D discretionary expenses. The reason for this single aim is, is that a reduction in a discretionary expense not only leads to increased earnings, but also improves a firm’s profit margins and cash flow from operations (CFO). Whereas sales manipulation and overproduction negatively affect profit margins and CFO (Kothari et al., 2016; Roychowdhury, 2006). Therefore stakeholders will have a better perception of a company’s performance when a reduction of discretionary

17

expenditure has occurred. Furthermore, as discussed before, the results of a survey conducted by Graham et al. (2005) show that executives prefer to manage the earnings through a reduction of discretionary expenses. The model for measuring RAM is:

𝑅&𝐷!" 𝐴𝑠𝑠𝑒𝑡𝑠_!"!!=

∝

!" !+ Δ!" !+ 𝛷!"∗ 𝑅&𝐷!"!! 𝐴𝑠𝑠𝑒𝑡𝑠_!"!! + 𝛾!"∗ 𝑆𝑎𝑙𝑒𝑠!"!! 𝐴𝑠𝑠𝑒𝑡𝑠_!"!! +

ε

!" !"

𝑅&𝐷!" = R&D expense of firm i at time period t (XRD).

𝑅&𝐷!"!! = R&D expense of firm i at time period t-1.

𝐴𝑠𝑠𝑒𝑡𝑠!"!! = Total assets of firm i at time period t-1 (TA).

Δ!" !

=

Difference between a firm’s annual R&D expense and the cross-sectional

mean of R&D expense in that year.

𝛷!" = Annual deviation between R&D expense and the cross-sectional mean differenced

from the deviation in the previous year.

𝛾!" = Annual deviation between sales and the cross-sectional mean differenced

from the g deviation in the previous year.

𝑆𝑎𝑙𝑒𝑠!"!! = Total sales of firm i at time period t-1 (SALE).

ε

!" !" = Error term of firm i at time period t.

The abnormal R&D expenditure (error term) is estimated with a fixed-effect first-order autoregressive model. This model controls for heteroscedastic. Heteroscedastic means that there is an uneven distribution in the variance. It’s reasonable to assume that larger companies have larger discretionary expenses (i.c. R&D) compared with smaller companies. To control for this effect the model adjusts these expenses by size. Size is measured by (lagged) total assets. Secondly this model controls for firm-specific and year-specific effects. To control for firm-specific effects every firm’s annual R&D expense is differenced from the cross-sectional mean in that year. To control for year-specific effects, the annual deviation between R&D expense and the cross-sectional mean is differenced from the deviation in the previous year. These two steps are also performed on sales. Then the model is estimated using panel data of the years 2004-2014. For every firm the model calculates the error term. To obtain a firm’s abnormal R&D expense in every year the error term is subtracted by the mean of all the error terms across all the years for this specific firm. This firm’s abnormal R&D expense is the amount of RAM conducted.

Measuring accruals-based earnings management. Total accruals are defined as “the change

in non-cash current assets (ACT-CHE) minus the change in current liabilities (LCT) net of the current portion of long-term debt (DLC), minus depreciation and amortization (DP), divided by lagged total assets (AT)” (Kothari et al., 2016, p. 563). Total accruals exists of two parts, namely discretionary

18

accruals (DA) and non-discretionary accruals (NDA). The difference between these two parts is that over DA executives can use their discretion. This is the part where AM is potentially conducted and therefore the focal point. The modified-jones model (1991) is a frequently used model to measure accruals management (Dechow et al., 1995). In this study, as previous stated, I use the model provided by Kothari et al. (2016). This model is based on the modified-jones model, however it is augmented for net income. The model for measuring AM is:

𝑇𝐴_!" 𝐴𝑠𝑠𝑒𝑡𝑠!"!!= 𝛽!+ 𝛥!" !+ 𝛷!"∗ 𝑇𝐴!"!!+ 𝛽!∗ 1 𝐴𝑠𝑠𝑒𝑡𝑠!"!! + 𝛽! 𝛥𝑆𝑎𝑙𝑒𝑠_!" 𝐴𝑠𝑠𝑒𝑡𝑠!"!!− 𝛥𝐴𝑅!" + 𝛽! 𝑃𝑃𝐸_!" 𝐴𝑠𝑠𝑒𝑡𝑠!"!! + 𝛽! 𝑁𝑒𝑡𝐼𝑛𝑐𝑜𝑚𝑒_!" 𝐴𝑠𝑠𝑒𝑡𝑠!"!! + 𝜐!"

𝑇𝐴!" = Total accruals of firm i at time period t .

𝑇𝐴!"!! = Total accruals of firm i at time period t-1.

𝐴𝑠𝑠𝑒𝑡𝑠_!"!! = Total assets of firm i at time period t-1 (TA).

Δ!" !

=

Difference between a firm’s annual total accruals and the cross-sectional

mean of total accruals in that year.

𝛷_!" = Annual deviation between total accruals and the cross-sectional mean differenced from the deviation in the previous year.

𝛥𝑆𝑎𝑙𝑒𝑠!" = Change in sales net of accounts receivable of firm i at time period t (SALE).

𝛥𝐴𝑅_!" = Change in accounts of receivable of firm i at time period t (RECT). 𝑃𝑃𝐸!" = Net property, plant and equipment of firm i at time period t (PPENT).

𝑁𝑒𝑡𝐼𝑛𝑐𝑜𝑚𝑒_!" = Net income of firm i at time period t (NI).

𝜐_!" = Error term (amount of AM) of firm i at time period t.

As stated earlier, the focal point is the part of where an executive can use his discretion (DA). This

correspondents in the formula with 𝜐!". The remainder of this formula constitutes the NDA part. The

same calculation, as provided by the RAM, is used to measure 𝜐_!" (abnormal accruals).

Control variables. To control for influences on EM that do not have the focus of this study, I

include a number of control variables. Prior research have shown the influence of several variables on accruals (Kothari et al., 2005; Dechow et al., 1995). These variables are firm size, leverage and return on assets (ROA). Thus I include these as control variables in this study.

I include firm size as a control variable, because larger companies have normally higher accruals than smaller companies. Also larger companies tend to have a more rigorous governance board and as a result are expected to have lower information asymmetry (Meek et al., 2007). Therefore I expect a

19

negative relationship. I use the logarithm of total assets (TA) as a measure for firm size. CEOs have incentives to avoid debt covenants violations and as a result indulge in EM. Therefore I include leverage as a proxy for this incentive and expect a positive relationship. Leverage is computed as total liabilities (LT) / total assets (TA). Also Dechow et al. (1995) argue that firms with higher earnings are more likely to indulge in earnings management to ‘smooth’ their earnings. Because of this I include ROA as a control variable as well and expect a positive relationship. ROA is calculated as Operating Income Before Depreciation (OIBDP) / total assets (AT).

Besides firm specific factors, I also include CEO-specific factors. The upper echelon theory suggests that CEO’s demographic characteristics, such as age and gender, influence CEO’s values and explain differences in their disclosure behaviour (Hambrick & Mason, 1984; Ge et al., 2011). I include age and gender as control variables. However the effect of gender and age on accounting outcomes are mixed (Ge et al, 2011; Dyreng et al., 2010). Thus I don’t make an expectation of these variables in a certain direction. Furthermore Aier, Comprix, Gunlock and Lee (2005) show that a manager’s tenure (year worked in a company) negatively influences the amount of restatements. As a result I expect a negative relationship between tenure and earnings management. CEOs who are also chairman of the board exhibit stronger control over the firm and have therefore greater ability to pursue their own private interests. Such powerful CEOs are less likely to choose a big4 (high quality) audit firm (Lennox, 2005) and may compromise a board’s ability to monitor CEO’s actions (Finkelstein & D’aveni, 1994). Therefore I expect a positive relation between a CEO who is also the chairman of the board and earnings management.

20 6

Results

Paragraph 6.1 starts with the interpretation of the summary statistics. Subsequently paragraph 6.2 introduces the Pearson Correlation Matrix. The results of the regression analysis will be provided in paragraph 6.3. The additional analyses will follow in paragraph 6.4.

6.1

Summary statistics

Table 1 provides the summary statistics between the non-Twitter and the Twitter sample. First of all significant differences (P<0.05) are found in both the earnings management measures (AAM and ARAMRD). The mean for AAM in the non-Twitter sample is 0.57 (Twitter sample: 0.68). Thus compared with their counterparts there is a higher use of AAM in the Twitter sample. Regarding ARAMRD, the mean value of ARAMRD is higher in the non-Twitter sample (0.31) compared with the Twitter sample (0.22). This indicates that the CEOs in the Twitter sample use EM to manage the earnings upwards by managing the AM upwards and ARAMRD downwards. Since I divide both samples by CEOs having a (non)-Twitter account the mean of TwAcc is logically 0 in the non-Twitter sample and 1 in the Twitter sample. Also it’s logical that CEOs who have won a social media award (SocAw) are more likely to possess a Twitter account. The mean for SocAw is 0.00 in the Non-Twitter sample and 0.35 in the Twitter sample. This difference is significant on the 1% level and thus corresponds with the expectation. Furthermore CEO_rep is a factor variable between TwAcc and SocAw. As was the case with TwAcc and SocAw, a significant difference (1% level) is found between the Non-Twitter (mean: -0,25) and Twitter sample (mean:1.73). Concerning LnTA, ROA and LEV, the mean in the Non-Twitter sample is respectively 9.30, 0.16 and 0.59. In the Twitter sample these values are 9.67, 0.16 and 0.56. LnTA is significantly different (1% level) and indicates that CEOs with Twitter are working in bigger companies. The reason for this might be that bigger companies have more external stakeholders in comparison with smaller companies. As a result a CEO might use a Twitter account to convey information to the external stakeholders. LEV is also significantly different (5% level) and indicates that CEOs with a Twitter account work in less leveraged companies. ROA is not significantly different. The youngest CEO in the non-Twitter sample is 34 (Twitter sample: 27) and the oldest 89 (Twitter sample: 84). With mean values of 56.23 and 54.31 respectively. Thus CEOs in the Twitter account are in general younger than their non-Twitter counterparts. The explanation for this significant difference (1% level) is simple. Younger people are more brought up with the internet and are using social media channels more often than older people. In the Twitter sample more CEOs are female (mean: 0.93) than in the non-Twitter sample (mean: 0.97; P<0.01). This might be due to a bigger interest of the public towards female CEOs. Regarding duality it is less common that CEOs with a Twitter account are also chairman of the board (P<0.01). CEOs who are also the chairman have a bigger influence over the company and might therefore have less incentives to convey information to external parties. A weak significant level (5% level) exists in tenure. CEOs that are longer on the job have more often a Twitter account.