Amsterdam Business School Faculty of Economics and Business
The Effects of Board of Directors and
Audit Committee Characteristics on
Real Earnings Management
MSc Accountancy & Control, specialization Accountancy
Name: Marco van Zijp
Student number: 10285040
Date of final version: June 22nd, 2015
Word count: 19,747 (excl. References and Appendix)
1
Statement of Originality
This document is written by student Marco van Zijp who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
2
Abstract
This study examines the relationships between board and audit committee characteristics and real earnings management of public U.S. firms for the period 2008-2014. I document that real earnings management increases when the board and audit committee are larger, however especially when the audit committee is larger. This implies that managers substitute accrual-based earnings management for real earnings management because larger audit committees constrain their ability to use accrual-based earnings management. I also find evidence that audit committee directors and especially board directors who serve on many other boards are associated with higher levels of real earnings management through sales manipulation. This is consistent with the hypothesis that these directors are busy and therefore less effective in constraining real earnings management. Additionally, I document evidence that boards consisting of many independent directors and with the CEO as the chairman are associated with higher levels of real earnings management through discretionary expenses. This suggests that non-independent directors seem to have the better ability to constrain real earnings management due to them being more skilled and knowledgeable of the firm’s activities, and that CEO duality has a negative influence on the monitoring effectiveness of the board. These results should be of interest to policymakers on how to structure boards and audit committees in order to minimize potentially firm-value destroying real earnings management.
Keywords: board of directors, audit committee, corporate governance, real earnings management, real activities manipulation
3
Table of Contents
Abstract ... 2
1. Introduction ... 4
2. Literature Review and Hypothesis Development... 8
2.1. Earnings Management: the Post-SOX Increase in Real Earnings Management ... 8
2.2. Agency Theory: Role of the Board and Audit Committee in Constraining REM ... 10
2.3. Board and Audit Committee Characteristics Literature Review and Hypotheses ... 12
2.3.1. Size ... 12
2.3.2. Independence ... 13
2.3.3. Serving on Other Boards ... 14
2.3.4. Financial Expertise ... 15
2.3.5. Tenure ... 16
2.3.6. CEO Duality ... 17
3. Sample and Empirical Design ... 18
3.1. Sample ... 18
3.2. Empirical Design ... 21
3.2.1. Model ... 21
3.2.2. Dependent Variable: Real Earnings Management ... 22
3.2.3. Independent Variables: Board and Audit Committee Factors ... 24
3.2.4. Control Variables ... 28
4. Results ... 30
4.1. Descriptive Statistics ... 30
4.2. Results of Hypothesis Tests ... 34
4.3. Robustness Test: Suspect Firm Analysis ... 40
4.4. Additional Analysis: Comparison of Board and Audit Committee Characteristics ... 45
5. Conclusion ... 49
4
1. Introduction
Boards of Directors and its committees are widely believed to play an important role in corporate governance, particularly in monitoring top management (Fama & Jensen, 1983). This is because top managers and shareholders can have conflicting interests, resulting in situations where management does not act in the best interests of the shareholders. For example, managers may have incentives to manipulate earnings to influence short-term stock price performance or to increase their own compensation (Healy & Wahlen, 1999). Managers have two earnings management strategies at their disposal for this purpose: either managing earnings through accrual-based earnings management or through real activities manipulation (also called real earnings management; Zang, 2012; Cohen et al., 2008).
In this thesis I examine the relationships between multiple board and audit committee characteristics and real earnings management of public U.S. firms for the period 2008-2014. Because it is the responsibility of the Board of Directors and its committees to act as the shareholders’ representative, the board and its committees should oversee managers to constrain real activities manipulation because real earnings management is likely to be costly to the firm and its shareholders as it has direct cash flow consequences (Sun et al., 2014; Cohen et al., 2008; Cohen & Zarowin, 2010). It is therefore of interest to examine whether certain characteristics regarding the composition and structure of the board and audit committee are negatively (or positively) associated with real earnings management, which would indicate if boards or audit committees with certain characteristics are more (or less) effective in constraining real earnings management. This question is the focus of this study. More specifically, I examine the association between real earnings management and the following board and audit committee characteristics: size, independence, number of other boards served on, financial expertise and tenure. In addition, I investigate the effect of having the CEO as the chairman of the board (CEO duality) on real earnings management.
This research question is motivated by the growing significance of corporate governance and real earnings management following the accounting scandals and the subsequent implementation of the Sarbanes-Oxley Act (SOX) in 2002. For example, the role of boards and audit committees has especially been the topic of active debate among policymakers, investors and academics following the accounting scandals such as at Enron or WorldCom. As a result, SOX mandated changes in corporate governance requirements of U.S. listed companies to restore confidence in financial reporting, strengthen corporate accountability and improve the quality of financial reports (Jain & Rezaee, 2006). In addition,
5 a survey by Graham et al. (2005) highlights the importance of more research on real earnings management because they find strong evidence that managers are willing to use real activities to manage earnings. They acknowledge that the accounting scandals and the subsequent implementation of SOX may have had an influence on managers’ preferences for the mix between accrual-based earnings management and real earnings management. Subsequent research by Cohen et al. (2008) actually documents this shift in managers’ preferences. They find an increase in real earnings management following the implementation of SOX, while the level of accrual-based earnings management decreased after SOX implementation. Furthermore, it can also be argued that real earnings management potentially imposes greater costs for shareholders than accrual-based earnings management because real earnings management might hurt firm value in the long run due to its negative consequences on future cash flows (Chi et al., 2011; Cohen & Zarowin, 2010). Consistent with this prediction, a study by Gunny (2005) documents significantly negative future operating performance for firms that engaged in real earnings management. Similarly, Cohen and Zarowin (2010) find evidence that shows that real earnings management is more likely to be associated with future earnings declines than accrual-based earnings management. Although these developments suggest that real earnings management has increased in importance relative to accrual-based earnings management, only limited literature exists that relates the area of corporate governance to real earnings management. Examining the association between board and audit committee characteristics and real earnings management is therefore the topic of this study.
Using a sample of 3,736 firm-year observations, I use factor analysis and multivariate regressions to examine the associations between board and audit committee characteristics and real earnings management for a sample of large U.S. firms for the period 2008-2014. In contradiction to my hypothesis, the results indicate robust evidence that real earnings management increases when the board and audit committee are larger, however especially when the audit committee is larger. This implies that managers substitute accrual-based earnings management for real earnings management because larger audit committees constrain their ability to use accrual-based earnings management. I also find evidence that audit committee directors and especially board directors who serve on many other boards are associated with higher levels of real earnings management through sales manipulation. This is consistent with the hypothesis that these directors are busy and therefore less effective in constraining real earnings management. Results also show positive associations for the tenure and financial expertise of directors with real earnings management through sales manipulation, but negative associations with real earnings management through discretionary
6 expenses. This implies that directors may view sales manipulation as less harmful or sales manipulation may be more difficult constrain. As such, there is no uniform relationship between directors’ tenure and financial expertise and the types of real earnings management. Finally, I document evidence that boards consisting of many independent directors and with the CEO as the chairman are associated with higher levels of real earnings management through discretionary expenses. This is consistent with my hypothesis that non-independent directors have the better ability to constrain real earnings management due to them being more skilled and knowledgeable of the firm’s activities (similar to Chen et al., 2012) and that CEO duality has a negative influence on the monitoring effectiveness of the board.
The results are of interest to policymakers on how to structure boards (and audit committees more specifically) in order to minimize potentially firm-value destroying real earnings management. Furthermore it indicates that the recent corporate governance reforms are not always successful in constraining real earnings management (similar to e.g. Cohen et al., 2008). For example, recent regulations such as the Sarbanes-Oxley Act of 2002 have been emphasizing directors’ independence to ensure effective monitoring. I provide further evidence that director independence by itself does not ensure effective monitoring to effectively constrain real earnings management (Chen et al., 2012). Policymakers should consider that non-independent or inside directors can prove beneficial in this regard.
This study makes several contributions to the existing literature on corporate governance and earnings management. The literature relating board and audit committee characteristics and earnings management focus extensively on accrual-based earnings management (e.g. Xie et al., 2003; Klein, 2002; Peasnell et al., 2005; Bédard et al., 2004). However, studies on the effects of board and audit committee characteristics on real earnings management are scarce. Different results could be expected for real earnings management than for accrual-based earnings management because real activities manipulation is more difficult to detect and therefore different board and audit committee characteristics are expected to have an influence. Two studies exist related to the current study. The first study by Garven (2009) examines the effect of board and audit committee characteristics on real earnings management via a reduction of discretionary expenses to avoid reporting an annual loss. Whereas Garven (2009) uses a single proxy for real earnings management (namely the reduction of discretionary expenses), I use two additional proxies as suggested by Roychowdhury (2006) because the study by Roychowdhury (2006) suggests that managers use multiple strategies for manipulating real activities interchangeably. Therefore I also examine the two other proxies of real earnings management: overproduction and sales
7 manipulation. Furthermore, whereas Garven (2009) relies on a relatively small sample of hand-collected data, I use a larger sample of firm-year observations to identify the effects of board and audit committee characteristics on real earnings management. This small sample may also have been the reason why Garven (2009) found very limited support for linkages between board and audit committee characteristics and real earnings management with this single proxy of real earnings management. The second study related to the current study is by Sun et al. (2014), which examines the association of only audit committee characteristics with real earnings management. Similar to the study of Garven (2009), Sun et al. (2014) also find limited evidence of associations between audit committee characteristics and real earnings management which also may be attributable to a limited sample size. Both studies only find support that the number of other boards served on by board and audit committee directors is positively associated with real earnings management, indicating that these directors are busy and therefore less effective in constraining real earnings management, but their studies found that none of the other characteristics examined in this study were found to be associated with the occurrence of real earnings management. Compared to the study by Sun et al. (2014), I examine the effects of both board and audit committee characteristics instead of audit committee characteristics only.
Furthermore, I contribute to both aforementioned studies by using a factor analysis methodology which identifies a latent variable from the independent variables, in order to eliminate any issues regarding correlations and multicollinearity which exist between the board and audit committee variables. This is to eliminate any bias in the results which arises as a result thereof. Finally, the authors of both studies focus on a sample of “suspect firms” which are likely to engage in real earnings management by restricting the sample to firms with small positive earnings. To build further on this prior research and to test for the robustness of the results for the full sample, I re-run the factor analysis and regressions on a second sample identified as such suspect firms. Whereas these previous authors focus on the single benchmark of zero earnings, I use two additional important benchmarks or goals that have been shown to incentivize managers to use real earnings management: previous year’s earnings and seasoned equity offerings.
This paper proceeds as follows. Section 2 provides an overview of prior literature, discusses theories and develops my hypotheses. Section 3 introduces the sample selection procedures and the empirical design. Section 4 provides and discusses the results including a robustness test and additional analysis, and section 5 concludes.
8
2. Literature Review and Hypothesis Development
2.1. Earnings Management: the Post-SOX Increase in Real Earnings Management Several definitions of earnings management exist from prior literature. Healy and Wahlen (1999) define earnings management as follows:
“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).
The earnings of a firm consist of two components: accruals and cash flows. The essence of accrual accounting is that revenues and costs are recognized at the moment they occur, even though the corresponding cash transaction occurs at a different time (Marshall et al., 2011). Because earnings consist of both accruals and cash flows, managers thus have two ways to manage earnings. Managers can influence earnings through the accruals by using accrual-based earnings management and through cash flows by using real earnings management (Schipper, 1989). More specifically, accrual-based earnings management involves accounting choices that are within Generally Accepted Accounting Principles (GAAP) in order to try to obscure the true economic performance. Intervention in the reporting process can however also occur through operational decisions by managers (Gunny, 2010; Schipper, 1989; Roychowdhury, 2006). Real earnings management can therefore be defined as a deviation from normal operational practices by managers in order to mislead stakeholders into believing that certain benchmarks regarding financial reporting have been met in the normal course of operations (Roychowdhury, 2006). As a result, an important distinction between the two forms of earnings management is that whereas real earnings management directly affects cash flows, accrual-based earnings management does not (Cohen & Zarowin, 2010). Similarly, Carcello et al. (2006) argue that real earnings management may destroy firm and shareholder value. It is however difficult to distinguish between operational practices that are undertaken to increase value for shareholders and those undertaken merely to manage earnings, therefore making real earnings management more difficult to detect than accrual-based earnings management (Schipper, 1989; Cohen et al., 2008).
A large body of academic literature is available on earnings management but the main focus of that literature has been on accrual-based earnings management. Only recently did researchers start to study real earnings management more. The large accounting scandals in
9 the early 2000s has led to further debate on whether further corporate governance and accounting regulation was needed to improve corporate governance and to enhance and restore confidence in financial reporting (Jain & Rezaee, 2006). The debate ultimately led to the introduction of the Sarbanes-Oxley Act in 2002 which required extensive changes in firms’ reporting practices.
For example, Section 302 of the SOX Act requires CEOs and CFOs to certify for each quarterly and annual report that the financial statements are a fair presentation of the financial condition and results of operations, and imposes severe penalties if deviations are found (McEnroe, 2007). Such requirements are likely to have made executives more conservative with regard to their financial reporting practices, thus ultimately leading to decreases in accrual-based earnings management following the enactment of SOX (Cohen et al., 2008). Cohen et al. (2008) provide evidence that SOX indeed led to decreases in accrual-based earnings management following the passage of SOX and that real earnings management practices have increased after SOX, implying a substitution effect from accrual-based to real earnings management since 2002. Similarly, a survey by Graham et al. (2005) shows that managers prefer using real economic actions to manage earnings. Of the participants, 80% indicated they would lower their discretionary expenses such as on R&D, advertising and maintenance in order to meet earnings benchmarks. Furthermore, more than 50% of the participants would delay starting new projects to meet earnings benchmarks, even if such projects have positive NPVs (Graham et al., 2005). An important result is that real earnings management might hurt firm value in the long run because of its negative consequences on future cash flows (Chi et al., 2011; Cohen & Zarowin, 2010). Consistent with this notion, Gunny (2005) documents that firms engaging in real earnings management exhibit significantly negative future operating performance.
Multiple methods or strategies for real earnings management have been identified in prior studies. Roychowdhury (2006) identifies and provides evidence for the existence of three “real” manipulation methods used by managers to avoid reporting an annual loss. First, managers can temporarily accelerate their sales by using increased price discounts and offering more lenient credit terms (which is essentially a price discount). This way managers are able to temporarily generate additional sales in the current period but at the expense of the next period. A second method that is used by firms for income-increasing earnings management is overproducing in order to report a lower cost of goods sold. This is accomplished because fixed costs per unit will be lower as fixed overhead costs will be spread over a larger number of units of inventory. The third method identified by
10 Roychowdhury (2006) is the reduction of discretionary expenses such as R&D, advertising and maintenance expenses. The reduction of such expenses leads to higher earnings in the current period because such expenses have to be directly expensed when they are incurred. If managers use this method of real earnings management, they will have abnormally low discretionary expenses in the current period. It is these three methods of real earnings management on which I lay focus in this study.
2.2. Agency Theory: Role of the Board and Audit Committee in Constraining REM One theory that is used to explain the existence of boards and audit committees and why such oversight bodies are expected to have an influence on real earnings management is the agency theory. According to the agency theory, an agency problem occurs when one person or party (the agent, e.g. management) makes decisions on behalf of another person or party (the principal, e.g. shareholders), thus delegating the decision making authority to the agent (Jensen & Meckling, 1976). In the agency theory it is assumed that both the principal and agent are bounded rational, risk averse and trying to maximize their own utility (Eisenhardt, 1989). An agency problem can then arise when (1) the principal and agent have conflicting desires or interests and (2) it is difficult or expensive for the principal to monitor the agent (Eisenhardt, 1989). In the latter case, the principal is not able to verify what actions the agent has undertaken on his behalf. This leads to a situation where the agent may not always act in the best interests of the principal. For example, the manager (agent) may choose to use real earnings management to maximize his own utility by maximizing short-term profits at the expense of longer-term performance (e.g. using real earnings management to meet the benchmark for a higher bonus in the current period), while this is not in the best interests of the shareholders (principal). However, the principal can limit the divergence of interests by imposing incentives for the agent or by monitoring the agent. This is where the board and its committees come in play, because one such example of a body that monitors executive behaviors is the Board of Directors. From an agency perspective, boards are used as a monitoring device to safeguard the shareholders’ interests (Fama & Jensen, 1983; Eisenhardt, 1989). More specifically, boards have the authority to validate and monitor managers’ actions, evaluate and reward them, or penalize the performance of executives (Li, 1994). As followed from the definition of earnings management by Healy and Wahlen (1999), real earnings management is undertaken with the intent to either “mislead some stakeholders about the underlying performance of the company or to influence contractual outcomes that depend on reported accounting numbers” (p. 368, emphasis added). As earnings management
11 is undertaken to mislead stakeholders (including shareholders) and it is the board’s responsibility to safeguard the shareholders’ interests, boards of directors have a responsibility to monitor and constrain earnings management practices by the executives (Fama & Jensen, 1983; Eisenhardt, 1989; Sun et al., 2014). It is therefore expected that certain characteristics of the board could influence whether executives of a firm will engage in earnings management (Xie et al., 2003). It is the focus of this study to determine whether certain characteristics of the board of directors are (negatively or positively) associated with real earnings management activities (i.e. if certain characteristics of the board constrain or facilitate real earnings management).
Similarly, the audit committee (as a subset of the board of directors) has the primary responsibility to oversee the firm’s financial reporting process (Klein, 2002). It is one of the main goals of the audit committee to detect and prevent accrual-based earnings management because it is responsible for monitoring the quality of the financial statement information and the internal audit function performance (Laux & Laux, 2009). There is however a disagreement among researchers in the prior literature about the responsibilities of the audit committee with respect to real earnings management. Carcello et al. (2006) argue that real earnings management may destroy firm value, but that it is not illegal. They argue that it is therefore beyond the scope of the audit committee’s responsibility to question the manager’s real activity choices. Conversely, Garven (2009) argues that this “GAAP-centric view” does no longer hold. Audit committees recognize that a GAAP-centric mentality is no longer acceptable and more audit committees and boards are adapting their monitoring practices as a response to higher expectations for effective governance and as a response to increased pressure on firms and their management (Garven, 2009). Similarly, Sun et al. (2014) state that SOX has required audit committees to expand their understanding of the business beyond the core competencies of accounting to also question the manager’s real activity choices. Also because real earnings management distorts financial reporting, the audit committee has a responsibility to constrain real earnings management (Sun et al., 2014). More specifically, real earnings management decreases the decision usefulness of financial reporting for investors because the firm’s earnings will not be indicative of future performance. Moreover, the audit committee is a subset of the board and thus Kieff and Paredes (2010) posit that “committee experts are directors, too, and directors are not excused from understanding, reviewing, or even controlling internal firm resource allocations – whatever their specific purpose” (p. 83). It could therefore be expected that audit committees exert (or at least should exert) influence over real earnings management choices by managers.
12 Given these contradicting signals about the role of the audit committee in constraining real earnings management, more evidence on this matter is desirable. This study will therefore also focus on the association between audit committee characteristics and real earnings management activities.
2.3. Board and Audit Committee Characteristics Literature Review and Hypotheses 2.3.1. Size
Prior research in different contexts has identified several characteristics of boards and audit committees which may have an influence on the monitoring effectiveness of these oversight bodies. One such characteristic is the size of the board or the audit committee, i.e. the number of directors serving on the board or audit committee. There is no universal agreement about the optimal size of the board or audit committee. Nonetheless, the Blue Ribbon Panel1 recommends having a minimum of three directors on the audit committee to improve the role of the audit committee in overseeing the corporate financial reporting process (SEC, 1999). The reasoning about requiring a minimum number of directors on the audit committee comes from the argument that adding more members to the board or audit committee increases the monitoring effectiveness because there will be more people to put their knowledge and experience at the board’s disposal (Vafeas, 2005; Xie et al., 2003). Consistent with this argument, Anderson et al. (2004) find that the cost of debt is decreasing in board size which implies that debtors see benefits from larger boards, for example through improvements in financial reliability and transparency. Similarly in the context of accrual-based earnings management, Peasnell et al. (2005) find that accrual-based earnings management is less prevalent when the board is larger. Xie et al. (2003) also find that the level of discretionary accruals is lower when the board is larger.
A counter to the argument that larger boards and audit committees are more effective is that larger groups may lead to poorer communication and thus poorer decision-making (John & Senbet, 1998). Consistent with this argument, Yermack (1996) finds a negative association between board size and firm value on a sample of 452 large industrial corporations between 1984 and 1991. In the real earnings management context however, the advantage of having more knowledge and experience to draw on with larger boards and audit committees probably outweighs the disadvantages of a larger board size because real earnings
1
The Blue Ribbon Panel was founded in 1998 by the New York Stock Exchange and the National Association of Securities Dealers to make recommendations to strengthen audit committees’ roles in the oversight of the corporate financial reporting process (SEC, 1999).
13 management is harder to detect than accrual-based earnings management (Cohen et al., 2008). Thus it is likely that a larger board and audit committee is more desirable to effectively constrain real earnings management. This leads to the following hypothesis:
H1: Real earnings management has a negative association with board and audit committee size.
2.3.2. Independence
A second characteristic is the independence of the board. A director can be viewed as independent when he or she has no material connection to the firm other than a board seat. One commonly held belief is that a larger proportion of independent directors on the board and audit committee leads to more effective monitoring (Xie et al., 2003) and thus less earnings management. This view is shared by the U.S. Securities and Exchange Commission (SEC) as the Sarbanes-Oxley Act of 2002 requires all members of the audit committee to be independent and prohibits the listing of any securities by issuers who do not comply with this requirement (SEC, 2003a). According to the SEC, management may be incentivized to improve short-term performance out of self-interest rather than improving long-term value for shareholders. An independent board and audit committee are believed to overcome this problem and are better able to align the corporate interests with shareholder interests (SEC, 2003a). This view can be derived from the agency theory. A study by Xie et al. (2003) in the context of accrual-based earnings management provides results that are consistent with this view. The results indicate that discretionary accruals are lower when the board is comprised of more independent outsiders (Xie et al., 2003). Similarly in the real earnings management context, Osma (2008) finds that more independent boards constrain the manipulation of R&D expenses (i.e. a form of real earnings management).
In contrast, Chen et al. (2012) suggest that non-independent boards are more effective in constraining real earnings management. They argue that director independence of management by itself is not sufficient to ensure effective monitoring. It can be argued that, particularly for constraining real earnings management, board and audit committee directors are required to be more skilled and knowledgeable of the firm’s activities (Chen et al., 2012). This is because real earnings management is difficult to detect as it is more difficult to distinguish real activities manipulation from normal business activities (Schipper, 1989; Cohen et al., 2008). In this case, inside directors (i.e. non-independent directors) might be
14 more effective in constraining real earnings management because they are exposed to day-to-day business decisions and are therefore likely to have this required firm-specific knowledge. Similarly, the amount and complexity of information received can be more difficult to process for independent directors (Lipton & Lorsch, 1992). However, the inside directors can facilitate monitoring by informing the independent directors about the firm’s activities (Armstrong et al., 2010), something which is particularly relevant for constraining real earnings management. Furthermore, Armstrong et al. (2010) argue that inside directors may have stronger incentives than independent directors to effectively monitor and to maximize shareholder value because they often have considerable holdings of options and shares in the firm.
As a result of the above discussion I hypothesize that real earnings management can be more effectively constrained by non-independent directors, in contrast to the widespread belief that independent directors are more effective monitors. Similar to Sun et al. (2014), I do not examine audit committee independence because the audit committees of all listed U.S. companies have been required to be fully independent after SOX. The hypothesis for the board is as follows:
H2: Real earnings management has a positive association with the independence of board directors.
2.3.3. Serving on Other Boards
A third characteristic is the number of other boards served on by board and audit committee directors. Research by Fich and Shivdasani (2006) posits that firms with busy boards (those with directors with three or more directorships in total) are less effective in monitoring their firms due to a lack of time. Therefore these boards and audit committees can be expected to be less effective in constraining real earnings management. Consistent with this prediction, Garven (2009) and Sun et al. (2014) find that directors with additional directorships are less effective in constraining real earnings management (yet both authors do not find any associations between any other board and audit committee characteristics and real earnings management). Many authors therefore suggest that directors do not serve on excessive amounts of other boards because this limits the time that they have to effectively monitor given the responsibilities that they have. For example, Lipton and Lorsch (1992) suggest that directors should not serve on more than three boards.
15 However, other studies do not support this “busyness hypothesis”. One example is the study by Ferris et al. (2003) who find no relation between the number of boards served on by directors and the likelihood of securities fraud litigation against the firm. They also find no evidence that a higher number of boards served on by directors harms the subsequent performance of the firm (Ferris et al., 2003). Furthermore, it can be argued that directors who hold multiple directorships may have more experience and knowledge which subsequently helps them in constraining real earnings management (Bédard et al., 2004). This suggests that serving on multiple boards by directors may actually decrease real earnings management by their firms. The sign of the association is not clear on the basis of these theories and therefore remains an empirical question. As a result, the hypothesis is as follows:
H3: Real earnings management is associated with the number of other boards served on by board and audit committee directors.
2.3.4. Financial Expertise
The fourth characteristic is the financial expertise of the board and audit committee directors. One of the rules in the U.S. that was implemented with SOX (Section 407) requires that at least one member of the audit committee has financial expertise (Bédard et al., 2004). The rules require that a company discloses whether such a financial expert is present and if not, explain why no such expert is present (SEC, 2003b). SOX Section 407 defines a financial expert as having the following attributes: (1) an understanding of GAAP and financial statements; (2) experience in applying such GAAP used in the company’s financial statements; (3) experience in preparing or auditing financial statements; (4) experience with financial reporting procedures and internal controls; and (5) an understanding of audit committee functions (SEC, 2003b). As this regulation was as a response to the corporate scandals, it was expected that the presence of financial experts would decrease accrual-based earnings management attempts. Research by Bédard et al. (2004) indeed finds that audit committees whose members have more financial expertise are more effective in constraining accrual-based earnings management. Badolato et al. (2014) find that audit committees with financial expertise are also associated with less accounting irregularities. Similarly, research by Krishnan (2005) finds that firms with audit committees with financial expertise are less likely to have material internal control weaknesses. One argument then is that it could be expected that these “financial expert” directors also have more expertise in constraining real
16 earnings management.
However, the contradicting argument is that the presence of directors with financial expertise may actually lead to increases in real earnings management attempts. If accrual-based earnings management is constrained by directors with financial expertise, managers may resort to real earnings management instead simply because it is more difficult to detect. Zang (2012) provides the evidence that managers use real earnings management and accrual-based earnings management as substitutes. Similarly, Cohen et al. (2008) document an increase in real earnings management and a decrease in accrual-based earnings management in the post-SOX period, implying that managers substituted accrual-based earnings management for real earnings management because of the SOX regulation. Furthermore, Chi et al. (2011) show that managers resort to more real earnings management in the case of higher quality auditors and Abbott et al. (2003) find that audit fees are higher if the audit committee has at least one person with financial expertise. Combining these last two studies suggests that directors with financial expertise hire higher quality auditors, meaning that accrual-based earnings management will be constrained by higher quality auditors, and managers may resort to more real earnings management as a result. In this case an increase in real earnings management attempts would occur because of directors’ financial expertise. The sign of the association is not directly clear based on the above discussion and therefore remains an empirical question. This leads to the following hypothesis:
H4: Real earnings management is associated with the financial expertise of the board and audit committee.
2.3.5. Tenure
The fifth characteristic is the tenure of board and audit committee directors. As indicated earlier, effectively constraining real earnings management requires board directors to have firm-specific knowledge and experience with the firm’s day-to-day business decisions (Chen et al., 2012). Longer tenure by board and audit committee directors makes them more familiar with the firm so that they are able to accumulate this firm-specific knowledge. Furthermore, the directors gain more experience and knowledge about, for example, the firm’s practices, operations and risks (Garven, 2009). This experience and knowledge makes them more likely to recognize and constrain real earnings management. This argument can also be denoted as the “expertise hypothesis” (Vafeas, 2003).
17 On the other hand, the effects of longer director tenure may also be explained using the “management friendliness hypothesis”. This hypothesis suggests that directors with longer tenure are likely to befriend the managers, which makes them less likely to question managers’ (real activity) choices (Vafeas, 2003). In that case directors with longer tenure are less effective monitors and any attempts in real earnings management are likely to increase. Similarly in the accrual earnings management context, Xie et al. (2003) find that directors’ tenure has a positive association with discretionary abnormal accruals. However, a study by Beasley (1996) finds that financial statement fraud becomes less likely as the tenure of outside board directors increases, which contradicts this latter management friendliness hypothesis.
On the basis of the two contradicting theories or hypotheses discussed above, the influence of directors’ tenure on real earnings management is not clear a priori. Therefore the (non-directional) hypothesis is as follows:
H5: Real earnings management is associated with board and audit committee directors’ tenure.
2.3.6. CEO Duality
In addition to the previous board and audit committee characteristics, I examine one other characteristic which relates to the board of directors as a whole only. This represents the situation where the CEO of the firm is also the chairman of the board, a situation which is also called CEO duality and which is a phenomenon commonly found in the United States (Gul & Leung, 2004; Jensen, 1993). One of the main roles of the chairman is running the board meetings and thus such a structure permits the CEO to control the information that is available to other board members, which has a negative influence on the monitoring effectiveness of the board (Jensen, 1993; Cornett et al., 2008). CEOs are not able to independently perform this function from their own personal interest (Jensen, 1993). Hence, having the CEO as the chairman makes it more difficult for the board to constrain real earnings management. I hypothesize that real earnings management is higher when the CEO is also the chairman of the board:
18
3. Sample and Empirical Design 3.1. Sample
I use the Institutional Shareholder Services (ISS) database to obtain board and audit committee data.2 The current universe of the ISS database is the S&P 1500 companies and therefore my sample consists of S&P 1500 firms. Data collection in the ISS database began in 1996, but the database has been split up into two parts: “Directors Legacy” (1996-2006) and “Directors” (2007-2013). I only focus on the latter more recent dataset due to significantly different data collection methodologies between the two datasets and the unavailability of the financial expertise data in the Directors Legacy version of the database. I exclude observations from the year 2007 because of the unavailability of the financial expertise data. Therefore my final sample consists of S&P 1500 firms with annual meetings held during the years 2008-2013.
First, I use the Compustat North America Fundamentals database to collect the data for the calculation of the real earnings management proxies. Data for all firms from the Compustat database starting from fiscal year 2006 are collected because up to two years of lagged variables are required for the real earnings management proxies. This leads to a total of 79,939 firm-year observations from the Compustat database. In line with Roychowdhury (2006) and Zang (2012), I exclude all firms in regulated industries (SIC codes between 4400-5000) and banks and financial institutions (SIC codes between 6000-6500) from the sample because of the differing practices of such firms. This reduces the sample size by 16,247 to 63,692 firm-year observations. Missing items for calculating each of the real earnings management proxies reduces the sample by 40,342 observations and missing data for the control variables reduces the sample by another 2,848 observations, leaving 20,502 firm-year observations in the sample. Subsequently merging the Compustat sample with the board and audit committee data from the ISS database and removing observations with missing board and audit committee data leaves a total of 3,736 observations in the final sample available for testing.3 The sample selection procedures are summarized in Table 1.
2 The RiskMetrics database was recently renamed to the Institutional Shareholder Services (ISS) database. 3
I also did a limited amount of hand-collecting from proxy statements (DEF-14A Forms) in case one or two board or audit committee data items were missing. Firms’ proxy statements are available for download from the SEC’s EDGAR database, available at: http://www.sec.gov/edgar.shtml
19 Table 1 – Sample Selection
# Firm-Year Observations (N)
Compustat data for fiscal years ending June 2006 to
December 2014 for calculating the REM proxies 79,939 Excluding regulated industries and banks and
financial institutions (SIC 4400-5000 and 6000-6500) (16,247) 63,692 Insufficient data for REM proxies (40,342)
23,350 Missing data for control variables (2,848) 20,502 Merging with board and audit committee data from
the ISS database for data years 2008-2013 (16,766)
Final sample 3,736
Table 2 shows the distributional properties of the final sample. First, Panel A of Table 2 shows the distribution of the sample over the fiscal years, which indicates that the sample is evenly distributed over the fiscal years (where each fiscal year represents approximately 15-18% of the total sample), but with the exception of fiscal year 2014. This exception for the fiscal year 2014 is because the ISS database reports the board and audit committee data from proxy statements at the date of the annual meeting, and the most recent version of the database only contains data for firms with annual meetings up to the end of 2013. Only a limited amount of these annual meetings held in 2013 relate to the fiscal year 2014 resulting in a lower amount of observations for the fiscal year 2014 (N = 41, i.e. 1.10% of the sample).
Panel B of Table 2 describes the distribution of the sample across industries and shows that the final sample spans a large amount of sectors of the economy, with Manufacturing (SIC codes 20 to 39) representing the largest category of industries in the sample (N = 2,169, i.e. 58,06% of the sample, which is similar to Sun et al., 2014). Business Services represents the largest individual industry in the sample with 403 observations, that is 10.79% of the final sample (see Panel B).
20 Table 2 – Final Sample Distributional Properties
Panel A – Number of Observations per Fiscal Year
Fiscal Year # Firm-Year Observations (N) % of Sample
2008 620 16.60 2009 688 18.42 2010 668 17.88 2011 595 15.93 2012 571 15.28 2013 553 14.80 2014a 41 1.10 Total 3,736 100 a
The sample consists of all firms from the ISS database with sufficient data available from the Compustat database. The ISS database contains data of firm-years with annual meetings up until the end of 2013, however some of these annual meetings held in 2013 relate to the fiscal year 2014. Therefore the sample also consists of a small portion of observations that relate to the fiscal year 2014.
Panel B – Industry Composition of the Final Sample (2-Digit SIC Codes4)
SIC Industry N %
Agriculture
01 Agricultural Production – Crops 6 0.16
Mining
10 Metal, Mining 12 0.32
12 Coal Mining 22 0.59
13 Oil & Gas Extraction 184 4.93
Construction
15 General Building Contractors 45 1.20
16 Heavy Construction, Except Building 21 0.56
Manufacturing
20 Food & Kindred Products 161 4.31
22 Textile Mill Products 8 0.21
23 Apparel & Other Textile Products 63 1.69
24 Lumber & Wood Products 20 0.54
25 Furniture & Fixtures 33 0.88
26 Paper & Allied Products 91 2.44
27 Printing & Publishing 43 1.15
28 Chemical & Allied Products 372 9.96
29 Petroleum & Coal Products 50 1.34
30 Rubber & Miscellaneous Plastics Products 31 0.83
31 Leather & Leather Products 33 0.88
32 Stone, Clay & Glass Products 22 0.59
33 Primary Metal Industries 75 2.01
34 Fabricated Metal Products 55 1.47
4 2-digit Standard Industrial Classification (SIC) codes are retrieved from the North Carolina State University,
21
SIC Industry N %
35 Industry Machinery & Equipment 314 8.40
36 Electronic & Other Electric Equipment 336 8.99
37 Transportation Equipment 146 3.91
38 Instruments & Related Products 281 7.52
39 Miscellaneous Manufacturing Industries 35 0.94
Wholesale Trade
50 Wholesale Trade – Durable Goods 113 3.02
51 Wholesale Trade – Nondurable Goods 56 1.50
Retail Trade
53 General Merchandise Stores 59 1.58
54 Food Stores 22 0.59
55 Automotive Dealers & Service Stations 34 0.91
56 Apparel & Accessory Stores 101 2.70
57 Furniture & Home furnishings Stores 23 0.62
58 Eating & Drinking Places 118 3.16
59 Miscellaneous Retail 86 2.30
Real Estate
65 Real Estate 9 0.24
67 Holding & Other Investment Offices 15 0.40
Services
72 Personal Services 2 0.05
73 Business Services 403 10.79
78 Motion Pictures 12 0.32
79 Amusement & Recreation Services 32 0.86
80 Health Services 69 1.85
82 Educational Services 32 0.86
87 Engineering & Management Services 87 2.33
Non-Classifiable Establishments
99 Non-Classifiable Establishments 4 0.11
Total 3,736 100
3.2. Empirical Design 3.2.1. Model
The model used for hypothesis testing is as follows:
REM = α0 + β1 F1_OTHERB + β2 F2_FINEXP + β3 F3_TENURE + β4 F4_SIZE +
β5 F5_INDEPDUAL + β6 MTB + β7 FIRMSIZE + β8 ROA + β9 BIG4 + β10 CRISIS +
β11-15 YEARDUMMIES + ε (1)
Where: REM equals any of the three real earnings management measures: abnormal cash flow from operations (REM_CFO), abnormal production costs (REM_PROD) and abnormal discretionary expenses (REM_DISEXP) or the combined construct for real earnings management denoted as REM_TOTAL (see the next section 3.2.2.). The independent variables used for hypothesis testing F1_OTHERB, F2_FINEXP, F3_TENURE, F4_SIZE and
22 F5_INDEPDUAL are the factors identified from the factor analysis on the board and audit committee characteristics (see Section 3.2.3.). The remaining variables MTB, FIRMSIZE, ROA, BIG4, CRISIS and YEARDUMMIES are other variables that are expected to affect the level of real earnings management and are included as control variables (see Section 3.2.4.).
3.2.2. Dependent Variable: Real Earnings Management
I rely on prior research for the proxies of real earnings management. More specifically, I use the study by Roychowdhury (2006) to examine three real earnings management proxies and thus three “real” earnings manipulation methods: (1) abnormal levels of cash flow from operations (temporarily increasing sales by using price discounts or more lenient credit terms); (2) abnormal production costs (increase production to spread fixed costs over a larger amount of units and thus lower the reported COGS); and (3) abnormal discretionary expenses (reducing SG&A, R&D and advertising expenses). Prior research provides evidence of the construct validity of these three real earnings management measures.5
The following regression is used for each year and industry in order to calculate the abnormal levels of cash flow from operations (CFO), where the industry is identified using the 2-digit SIC code:
CFOt Assetst-1= α0+ β1 1 Assetst-1+ β2 Salest Assetst-1+ β3 ∆Salest Assetst-1+ εt (2)
CFO represents the cash flows from operations (Compustat item “OANCF”), Assetst is the total assets at the end of year t (Compustat item “AT”), Salest the sales in period t (Compustat item “SALE”) and ΔSalest the change in sales where ΔSalest = Salest - Salest-1. In this case the residual (εt) from Equation (2) measures the abnormal amount of cash flow from operations for every firm-year (REM_CFO). The residual consists of the part of CFO which is not explained given the “normal” sales level and using the coefficients which were estimated from running the corresponding industry-year regression. More specifically, if firms use discounts and more lenient credit terms to manage their earnings, their cash inflows would be lower (but increasing their earnings) so that the residual term would be a larger negative number.
The second type of real earnings management identified by Roychowdhury (2006) is overproducing so that fixed costs per unit will be lower as fixed overhead costs will be spread
23 over a larger number of units of inventory. Following Roychowdhury (2006), production costs are defined as cost of goods sold (Compustat item “COGS”) plus the change in inventory (Compustat item “INVT”; and such that Prodt = COGSt + ΔInvt). The normal levels for COGS can be estimated using the following equation:
COGSt Assetst-1= α0 + β1 1 Assetst-1 + β2 Salest Assetst-1+ εt (3)
The normal levels for growth in inventory are estimated using the following model:
∆Invt Assetst-1= α0+ β1 1 Assetst-1+ β2 ∆Salest Assetst-1+ β3 ∆Salest-1 Assetst-1 + εt (4)
By combining Equations (3) and (4), the normal production costs can be estimated by running the following regression for each industry-year combination:
Prodt Assetst-1= α0+ β1 1 Assetst-1+ β2 Salest Assetst-1+ β3 ∆Salest Assetst-1+ β4 ∆Salest-1 Assetst-1 + εt (5)
The residual term in Equation (5) captures the abnormal level of production for each firm-year observation (REM_PROD) and therefore a higher positive value for the residual term indicates higher levels of real earnings management.
The third method of managing earnings using real activities identified by Roychowdhury (2006) is the reduction of selling, general and administrative expenses, R&D expenses and advertising expenses to report higher earnings in the current period. Discretionary expenses are therefore defined as the sum of SG&A expenses (Compustat item “XSGA”), R&D expenses (Compustat item “XRD”) and advertising expenses (Compustat item “XAD”; DisExpt = SG&At + R&Dt + Advt).6 The normal level of discretionary expenses, expressed as a function of lagged sales, is estimated by running the following regression for each industry-year combination7:
DisExpt Assetst-1= α0+ β1 1 Assetst-1+ β2 Salest-1 Assetst-1+ εt (6)
The residual term from Equation (6) captures the abnormal level of discretionary expenses for each firm-year observation (REM_DISEXP).
6
Consistent with Cohen et al. (2008), I set R&D and advertising expenses to zero in case they are missing.
7 Lagged sales instead of current sales are used because a mechanical problem occurs if current sales are used
24 I require that more than 10 observations for each year-industry combination are available for calculating the residuals for each of these three real earnings management regressions.8 Firms that use these three types of real earnings management would have abnormally low cash flow from operations, abnormally high production costs and abnormally low discretionary expenses given the firm’s sales levels (Cohen et al., 2008). To make for easier interpretation in further analyses, the values for abnormal cash flow from operations (REM_CFO) and abnormal discretionary expenses (REM_DISEXP) are therefore multiplied by -1 such that positive values for each proxy represent higher levels of real earnings management.
To test for the effects of board and audit committee characteristics on all of these three proxies for real earnings management simultaneously, I combine the three proxies into a single variable called REM_TOTAL.9 However, I acknowledge that the characteristics of the board and audit committee may have differing associations with the separate real earnings management methods. Therefore I also report the results for each real earnings management proxy separately (REM_CFO, REM_PROD and REM_DISEXP).
3.2.3. Independent Variables: Board and Audit Committee Factors
The empirical constructs for each of the board and audit committee characteristics are as follows and are derived from the ISS database.10 First, the size of the board (B_SIZE) and audit committee (AC_SIZE) are measured as the number of directors on the board and number of directors on the audit committee, respectively. Independence of the board is measured as the proportion of outside directors on the board, which are directors not employed by the firm and who have no material connection to the company other than a board seat (B_INDEP). To examine the effect of serving on multiple boards by directors, I empirically measure B_OTHERB and AC_OTHERB for each firm-year as the average number of other public company boards served on by their board and audit committee directors, respectively. The financial expertise of the board and audit committee are measured as the proportion of directors on the board and audit committee who have financial expertise,
8 This amount is within the range that other real earnings management researchers require in their studies. For
example, Cohen et al. (2008) require at least 8 observations, whereas Roychowdhury (2006) requires at least 15 observations.
9 REM_TOTAL = REM_CFO + REM_PROD + REM_DISEXP. Abnormal cash flow from operations
(REM_CFO) and abnormal discretionary expenses (REM_DISEXP) were multiplied by -1 such that positive values represent higher levels of real earnings management.
10 See Appendix A for more elaborate descriptions for each of the board and audit committee variables and the
25 as denoted by the firm on its proxy statement (B_FINEXP and AC_FINEXP, respectively). The tenure of a director is calculated as the current data year minus the year his or her service began. Board and audit committee directors’ tenure are then empirically measured as the average years of service by board and audit committee directors for each firm-year (B_TENURE and AC_TENURE, respectively). Finally, I measure CEO duality as an indicator variable that takes the value of one when the same person occupies both board chair and CEO positions, and zero otherwise (B_CEODUAL).
Next I employ factor analysis to extract underlying factors from the board and audit committee variables to alleviate problems of high correlation and multicollinearity among the independent variables and to further contribute to prior literature. Initial tests show that including all of the above independent variables in one regression introduces a multicollinearity problem and results would likely be biased if all independent variables were to be included into one regression model.11 Furthermore, correlation tests show that the board and audit committee variables are highly correlated (see Table 5). From the exploratory factor analysis I identify non-correlated latent constructs from the independent variables to eliminate these high correlation and multicollinearity issues and to eliminate any bias in the results which arises as a result thereof (Harman, 1976; Floyd & Widaman, 1995; Larcker et al., 2007). Thus the factor analysis summarizes the information in the original variables into multiple uncorrelated factors.
Table 3 describes the results of the factor analysis on each of the board and audit committee variables. Panel A of Table 3 shows the eigenvalues for each of the factors. I follow the Kaiser criterion and therefore retain all factors with eigenvalues larger than one which is in line with other studies (e.g. Carcello et al., 1992; Bushman et al., 2004; and Larcker et al., 2007). I also retain the fifth factor because the eigenvalue is only slightly below one and because Wood et al. (1996) found that factor overextraction leads to considerably less errors than underextraction and thus recommend factor analysts to avoid underextraction even when a risk of overextraction exists. I therefore retain a total of five factors which explain 83.74% of the total variance of the ten underlying board and audit committee variables (see Panel A).
Panel B of Table 3 subsequently shows the rotated factor loadings of each of the independent variables (that is the correlation between the factor and each of the independent
11
E.g. the (untabulated) test for multicollinearity shows VIF (Variance-Inflation-Factor) scores above 10 for both B_FINEXP and AC_FINEXP, where VIFs larger than 10 indicate serious multicollinearity problems (Street et al., 2000).
26 variables).12 The factor loadings are used for the interpretation of the factors and each factor is interpreted and assigned a name based on the underlying board and audit committee variables with the highest loading to that factor (Larcker et al., 2007).
The first factor has high positive loadings on both the number of other boards served on by board and audit committee directors (B_OTHERB and AC_OTHERB with loadings of 0.9034 and 0.9359, respectively). Therefore this factor can be interpreted as representing busy directors who serve on many other boards. As a result, I name this factor F1_OTHERB for further analyses. The second factor loads highly positively on both measures of financial expertise for the board and audit committee (B_FINEXP and AC_FINEXP with loadings of 0.9693 and 0.9646, respectively). Given that this factor measures one underlying construct, namely financial expertise, I name this second factor F2_FINEXP. Similar to the second factor, the third factor has only two high positive loadings that represent the same construct for the board and audit committee. The third factor has high positive factor loadings with both the tenure of board directors and the tenure of audit committee directors (B_TENURE and AC_TENURE, with loadings of 0.8993 and 0.9274, respectively). The third factor is therefore named F3_TENURE. The fourth factor is positively loaded on both B_SIZE (0.7771) and AC_SIZE (0.8628), and as a result is a measure for size which I name F4_SIZE. Finally, the fifth factor has significantly positive loadings with both board independence (B_INDEP, 0.6250) and CEO duality (B_CEODUAL, 0.9004). Whereas a higher value for B_INDEP would mean higher independence, a higher value for B_CEODUAL indicates lower independence, yet both variables have a positive loading with the factor. Therefore the underlying construct is not independence, but instead this factor can be interpreted as representing a board with a large proportion of independent directors but with the CEO as the chairman. As a result, I name this fifth factor F5_INDEPDUAL. These factors and how each of the board and audit committee variables load into these factors are summarized in Panel B of Table 3. These five identified factors are used as the independent variables for hypothesis testing.
12
It is general convention in the literature to subsequently rotate the reduced solution to clarify and simplify the data structure and to enhance the interpretation of the factors (Costello & Osborne, 2005; Larcker et al., 2007; Bushman et al., 2004). I choose to use an orthogonal rotation (where the rotated factors remain uncorrelated to each other) rather than an oblique rotation (where the factors can be correlated with each other). More specifically, I follow prior studies by using the Varimax rotation method as it is the most popular rotation method (Abdi, 2003; Bushman et al., 2004) and it seems suitable in the context of this study because the main reason the factor analysis is employed in this study is to alleviate any concerns regarding correlations between the independent variables.
27 Table 3 – Factor Analysis
Panel A – Eigenvalues
Factor Eigenvaluea Eigenvalue
Differenceb Proportion of Variance explainedc Cumulative Variance explained Factor 1 2.78968 0.87026 0.2790 0.2790 Factor 2 1.91943 0.44384 0.1919 0.4709 Factor 3 1.47559 0.25813 0.1476 0.6185 Factor 4 1.21746 0.24609 0.1217 0.7402 Factor 5 0.97137 0.36917 0.0971 0.8374 Factor 6 0.60219 0.05281 0.0602 0.8976 Factor 7 0.54938 0.29887 0.0549 0.9525 Factor 8 0.25052 0.06042 0.0251 0.9776 Factor 9 0.19010 0.15583 0.0190 0.9966 Factor 10 0.03427 . 0.0034 1 10 a
The sum of all eigenvalues equals the number of underlying variables (10), and can be used to calculate the proportion of the total variance explained by each factor. For example: 2.78968 / 10 = 0.2790. It is general convention to retain those factors with eigenvalues that exceed one, as suggested by the Kaiser criterion (Kaiser & Caffrey, 1965; Larcker et al., 2007; Carcello et al., 1992). Four factors are retained with eigenvalues > 1, however I also include Factor 5 given that it is only slightly below the threshold of one and Wood et al. (1996) found that factor overextraction leads to considerably less errors than underextraction. The retained factors are presented in bold font. These five factors explain 83.74% of the total variance.
b Eigenvalue Difference shows the difference between the eigenvalue of the factor with the eigenvalue of the
next factor. For example, the 0.87026 difference in eigenvalue for Factor 1 is calculated as 2.78968 – 1.91943.
c The proportion of the total variance explained by each factor.
The results of the factor analysis reported are by using the principal-component factor estimation method.
Panel B – Rotated Factor Loadings
Variable Factor 1 (F1_OTHERB) Factor 2 (F2_FINEXP) Factor 3 (F3_TENURE) Factor 4 (F4_SIZE) Factor 5 (F5_INDEPDUAL) B_SIZE 0.2864 -0.0244 -0.0256 0.7771 -0.1570 B_INDEP 0.3179 0.0448 -0.2778 0.3134 0.6250 B_OTHERB 0.9034 0.0525 -0.1392 0.2049 0.0897 B_FINEXP 0.0005 0.9693 -0.0252 -0.0126 0.0635 B_TENURE -0.1901 -0.0434 0.8993 -0.0640 -0.1118 B_CEODUAL -0.0497 -0.0133 0.0403 -0.0600 0.9004 AC_SIZE 0.0729 -0.0674 -0.0210 0.8628 0.1400 AC_OTHERB 0.9359 0.0601 -0.0785 0.0559 -0.0369 AC_FINEXP 0.0984 0.9646 -0.0373 -0.0507 -0.0619 AC_TENURE -0.0339 -0.0278 0.9274 0.0162 0.0574
High loadings (>|0.40|) are reported in bold. As a rule of thumb, any loadings >|0.40| are seen as substantial loadings used for interpretation of the factor, see e.g. Larcker et al., 2007 and Dey, 2008. The factor loadings are Varimax rotated to ease interpretation of the factors and such that the factors remain uncorrelated to each other. Variables are as defined in Appendix A.
