Multiple Board Positions and Earnings Management: Too busy to watch the numbers?

Master Thesis Organizational & Management Control

Multiple Board Positions and Earnings Management:

Too busy to watch the numbers?

UNIVERSITY OF GRONINGEN FACULTY OF ECONOMICS &BUSINESS

AUTHOR:MAARTEN SPIT (1274201) FIRST SUPERVISOR:T.MARRA

SECOND SUPERVISOR:R.HOOGHIEMSTRA

JEL CODES:G34,M40.

KEY WORDS: MULTIPLE DIRECTORSHIPS, BOARD MEMBER EFFECTIVENESS, AUDIT

TABLE OF CONTENTS

INTRODUCTION ... 2

1. LITERATURE REVIEW ... 4

1.1 EARNINGS MANAGEMENT ... 4

1.1.1 MOTIVATIONS FOR EARNINGS MANAGEMENT ... 6

1.2 THE SUPERVISORY BOARD AND EARNINGS MANAGEMENT ... 8

1.2.1 BOARD CHARACTERISTICS AND EARNINGS MANAGEMENT ... 9

1.3 MULTIPLE BOARD POSITIONS ... 11

2. HYPOTHESIS DEVELOPMENT ... 13

3. METHODS... 15

3.1 DATA ... 15

3.2 MULTIPLE BOARD POSITIONS ... 16

3.3 DISCRETIONARY ACCRUALS ... 18

3.4 UNIVARIATE ANALYSES ... 20

3.5 MULTIVARIATE ANALYSES ... 22

4. RESULTS ... 24

4.1 SENSITIVITY ANALYSIS ... 27

4.1.1 MODIFIED JONES MODEL ... 27

4.1.2 PANEL DATA ANALYSIS ... 31

5. CONCLUSION ... 34

6. REFERENCES ... 36

APPENDIX I.ESTIMATION RESULTS DISCRETIONARY ACCRUAL REGRESSIONS ... 41

APPENDIX II.ADDITIONAL DESCRIPTIVE STATISTICS ... 42

INTRODUCTION

Since the recent economic crisis, the functioning of boards of directors has once again received considerable attention. The functioning of boards of directors has been thoroughly studied and widely criticised for several decades. However, the worldwide prevalence of boards of directors in organizations nowadays suggests that, despite its shortcomings, it is still an important corporate governance mechanism. The non-executive board members are expected to monitor and assist the executive board members in decision-making, while serving investors’ best interests. Although the functioning of boards of directors has been widely studied, there is still little consensus on the optimal composition or structure of the board of directors. One aspect of boards of directors that has recently received considerable attention is the ‘busyness’ of boards.

The busyness hypothesis suggests that when non-executive board members engage in too many activities outside their board position, such as other non-executive or executive positions, they will be less able to adequately perform their supervisory function. Since non-executive board members have finite time to allocate to their supervisory roles, holding multiple board positions will imply that less time is available for each board position. However, the reputation hypothesis suggests that multiple board positions can signal higher board member quality (Fama and Jensen, 1983). According to the reputation hypothesis, non-executive board members holding multiple board positions will work harder to adequately fulfil their supervisory role. In addition, resource

dependency theorists argue that non-executive board members holding multiple board positions

can link the firm to valuable resources. Non-executive board members holding multiple board positions are likely to have larger social networks, more experience with board practices and superior access to valuable information, which can be used in assisting management.

Previous studies on the relation between multiple board positions and board member effectiveness have shown mixed results and seem to suggest that this relationship is more complex than has been argued by proponents of either the busyness or the reputation hypothesis. Although results are ambiguous, in the Netherlands, the Dutch Corporate Governance Code1 (2008, art.III.3.4) limits the number of non-executive board positions per supervisory board member to a maximum of five to assure “the proper performance of his duties”. In the beginning of 2010, a bill was partially passed by Dutch parliament2 that is similar to art.III.3.4 of the Dutch Corporate Governance Code (DCGC). It will legally limit non-executive board positions to a

1 _{The Dutch Corporate Governance Code or ‘Code Tabaksblat’, after its committee chairman, was} completed in December 2003 and became effective on the 1st of January 2004. The code works according to a comply-or-explain principle. In 2008 the Dutch Corporate Governance Code was updated. The articles mentioned in this text remained unchanged.

maximum of five (a maximum of two non-executive board positions for board members that also hold an executive position). In line with the DCGC, chairmanship of a supervisory board will be counted double. Dutch legislators thereby suggest that when board members hold more than five non-executive board positions (more than two non-executive board positions if they also hold a full-time position) they will not be able to adequately perform their supervisory role.This study extends research on the influence of non-executive board members holding multiple board positions on board effectiveness.

We study the influence of holding multiple board positions on board members’ ability to constrain earnings management. Numerous corporate governance reforms that have been composed in the last decades, such as the Cadbury report (1992) in the UK or the Sarbanes–Oxley Act (2002) in the US, have explicitly mentioned and focused on (among other things) financial reporting. In the Netherlands, the DCGC states that “the management board is responsible for the quality and completeness of publicly disclosed financial reports” and that “the supervisory board shall see to it that the management board fulfils this responsibility” (2008, V.1 principle). When a separate audit committee (AC) is appointed the AC will be responsible for these duties (2008, III.5 principle). It states that the AC shall supervise the provision of financial information by (the management board of) the company, which includes (among other things) the choice of accounting policies, application and assessment of the effects of new rules, information about the handling of estimated items in the annual accounts, forecasts and work of internal and external auditors (2008, III.5.4.b).

tests of the relationships between AC members holding multiple board positions and the degree of earnings management. We regress DAs against several measures of AC members holding multiple board positions and control variables.

This study contributes to literature on the desirability of board members holding multiple board positions. In addition, it provides insights into the determinants of earnings management and the measurement of discretionary accruals. The first section of this paper will review previous literature on the practice of earnings management and the role of the AC in constraining earnings management. The second section contains the theoretical framework and hypotheses concerning the relationship between board members holding multiple board positions and the degree of earnings management. The third section describes the data and methodology used in this study. The fourth section presents the results from our main multivariate analysis and sensitivity analyses. The final section contains the conclusions following from these results and discusses limitations of this study in the form of recommendations for future research on this topic.

1. LITERATURE REVIEW

1.1 EARNINGS MANAGEMENT

The financial reporting process has received considerable attention in the last decade since several high-profile accounting scandals, such as the demise of Enron in 2001, revived the debate about the importance of (control on) financial statements. The purpose of financial statements is to effectively present a firm's economic position and performance in a timely and credible manner (Healy and Wahlen, 1999). Reported earnings, as the most visible and influential accounting number, should thus reflect a firm’s performance as closely as possible. However, managers, having the ultimate responsibility for the financial reporting process, may face various incentives to present financial statements and reported earnings that do not accurately reflect a firm’s current position. In the extreme, such deviations may lead to fines for financial statement fraud or even corporate failure.

management in this article. For a review of ‘real’ earnings management literature we refer to Xu et al. (2007).

The nature of accrual accounting gives managers a great deal of influence over reported earnings in any given period (Xie, Davidson and DaDalt, 2003). One of the main principles of accrual accounting is that revenues and costs should match, which is called the matching principle. This may lead to income received or expenditures spent in a certain period to be accrued and divided as revenues or costs, respectively, over subsequent periods. For example, a machine purchased in one period will not lead to drastic increases in costs in that period but will lead to depreciation costs over all periods in which the machine is expected to create products and thus revenues. The difference between accrued costs and accrued revenues in a specific period equals the difference between cash flow from operations and net income in that period. These differences are called ‘net accruals’ and include depreciation and amortization expenses and changes in working capital items such as receivables, inventories and payables. The height of net accruals is partially dependent on historical expenses and economic circumstances which are often similar for companies in the same industry. These accruals are referred to as normal or non-discretionary accruals. However, the height of net accruals is also dependent on managerial discretion. Managers may increase or decrease provisions for doubtful accounts, change inventory valuations or decide on the timing and amount of write-offs on goodwill or provisions. These accruals are called discretionary or abnormal accruals, through which managers may exert influence on reported earnings.

reported earnings resulting from actual cash flows show higher power in predicting future earnings than do earnings resulting from accruals, investors do not differentiate between these two. This inability of users to correctly value accruals may make earnings management desirable to managers.

1.1.1 MOTIVATIONS FOR EARNINGS MANAGEMENT

Managers can be faced with various different incentives to present reported earnings in a way that does not optimally reflect a firm’s current situation. We will discuss some of the motivations to engage in earnings management.

Managers may engage in earnings management to avoid the violation of debt covenants. When firms attract debt financing, the lenders may include debt covenants in the loan agreement to stay informed about firms’ financial performance and decrease the possibility the firm will default on its payments. These debt covenants are often tied to accounting figures such as reported earnings or comparable accounting figures. For example, a bank may demand the right to renegotiate the terms of a loan when a firm’s interest coverage ratio (ebit / interest expense) falls below a certain threshold, to make sure that the firm will be able to keep making its interest payments. In general, debt covenant violations will lead to higher financing costs. Since debt contracts are usually rigid, managers have a strong incentive to prevent debt covenants from being violated. Therefore, managers may want to manage earnings upward to prevent a debt covenant violation. Sweeney (1994), studying 130 US firms that report debt covenants violations in their annual reports, finds that managers of firms approaching debt covenant violations make income-increasing accounting choices to (temporarily) defer default costs. DeFond and Jiambalvo (1994) and Jaggi and Lee (2002) studying debt covenant violations and discretionary accruals find similar results.

related to initial firm value but negatively related to subsequent firm performance, indicating that managing earnings upward prior to the IPO increases IPO proceeds.

In addition to reducing financing costs, managers might use earnings management to reduce political costs or increase political gains. Jones (1991) investigates earnings management in 23 US firms in 5 different industries during import relief investigations. She hypothesizes and finds evidence that firms in industries facing these investigations manage earnings downwards to increase the probability of receiving import relief. Han & Wang (1998) hypothesize that firms will manage earnings downwards if reporting (excessively) high earnings will be politically sensitive. They study earnings management at 76 oil companies during the 1990 Persian Gulf Crisis. They find that oil companies that expected to profit from the crisis, due to higher oil prices, managed quarterly earnings downwards to reduce political sensitivity.

Another motivation for managers to engage in earnings management is to meet earnings expectations. According to Skinner and Sloan (2002) firms are punished by the market for not meeting earnings expectations. They find that the magnitude of negative stock returns for not meeting earnings expectations is greater than the magnitude of positive stock returns for exceeding earnings expectations. Managers therefore have an incentive to manage earnings upwards to meet earnings expectations or to manage earnings downwards if earnings expectations are already met to increase the probability of meeting earnings expectations in subsequent periods. Peasnell, Pope and Young (2000) study the influence of the Cadbury governance reform on the role of board of directors in constraining earnings management for 360 listed UK firms. They find that, both pre- and post-Cadbury, firms use income-increasing accruals to meet earnings expectations.

lowering his compensation, thereby increasing his expected compensation in subsequent periods. Managers may also want to realize low-volatile compensation. Since managers are usually assumed to be risk-averse, they will prefer a constant compensation over a volatile compensation. They might therefore want to create a constant flow of earnings, by managing relatively high earnings downward and relatively low earnings upward. In addition, this strategy, called ‘income smoothing’, might reduce the probability of dismissal since it will lower the probability of reporting low earnings (Weisbach, 1988).

1.2 THE SUPERVISORY BOARD AND EARNINGS MANAGEMENT

The inability or unwillingness (too costly) of shareholders to check whether earnings were managed can make earnings management desirable to managers. Hermalin and Weisbach (2003) state, that in most large corporations the diffusion of shareholders leads to a free-rider problem concerning the collection of information. However, dispersed shareholders have an additional instrument at their disposal to constrain such corporate governance issues; the board of directors. Boards of directors are required by law in most countries nowadays. However, they predate regulation and are thus part of the market solution to contracting problems inside organizations (Hermalin and Weisbach, 2003). In most Anglo-Saxon countries firms have a one-tier structure. In a one-tier structure there is one board of directors which consists of executive and non-executive or supervisory directors. In countries with primarily two-tier structure firms, such as the Netherlands, the board of directors is split up into two different bodies; the managing board, consisting solely of executive directors, and the supervisory board, consisting solely of non- executive directors.

The supervisory board has the duty to monitor and assist management in decision making. In doing so, the supervisory board has the responsibility to weigh the interests of different stakeholders (DCGC, 2008, p.6). As stated in the introduction, one of the duties of the supervisory board, according to the DCGC, is to ensure the quality and completeness of publicly disclosed financial reports. This encompasses making sure financial reports best reflect a firm's economic position and performance and that earnings therefore are not manipulated. The supervisory board is likely to obtain information about the manager and the firm that is superior to shareholders’ information. With respect to earnings management, the supervisory board will therefore be better at judging whether financial reports and reported earnings best reflect information available to the manager or that earnings were manipulated.

it that the management board fulfils this responsibility” (2008, V.1 principle). In addition, it recommends installing a separate audit committee (AC) when the supervisory board exceeds four members (2008, III.5 principle). This AC shall supervise the provision of financial information by (the management board of) the company, which includes (among other things) the choice of accounting policies, application and assessment of the effects of new rules, information about the handling of estimated items in the annual accounts, forecasts and work of internal and external auditors (2008, III.5.4.b). When a separate AC is not appointed the AC’s duties will be the responsibility of the entire supervisory board (2008, III.5 principle). The extent to which supervisory boards and their ACs are able (or willing) to constrain earnings management depends on the quality and quantity of the control exercised.

1.2.1 BOARD CHARACTERISTICS AND EARNINGS MANAGEMENT

The quality of supervision by the board has often been linked to general characteristics of the supervisory board such as board independence, board size and board activity. With respect to constraining earnings management, characteristics of the supervisory board that are associated specifically with being able to detect earnings management, such as the presence of an AC and the presence of financial (accounting) expertise on the board, are relevant as well. Several studies have examined the association between different board characteristics and the degree of earnings management, accounting fraud, financial restatements, etc. In the presence of a separate AC, some studies have also examined AC independence, AC size, AC activity and the presence of financial expertise on the AC. We discuss some of the previous literature on board and AC characteristics and focus on literature that studied some form of earnings management.

Chairman of the Board, and (iii) less likely to have an audit committee. Abbot et al. (2004) examine the likelihood of financial restatements and find a negative association with audit committee independence. In line with these results, Klein (2002), studying a large sample of publicly traded U.S. firms, finds negative relationships between board & AC independence and abnormal or discretionary accruals, implying that more independent boards are better at constraining earnings management. Other studies that examine the relationship between board/AC independence and abnormal accruals find similar results (Xie et al., 2003, Bedard, Chtourou and Courteau, 2004, Peasnell, Pope and Young, 2005, Carcello, Holingsworth, Klein and Neal, 2006, Sarkar, Sarkar and Sen, 2008). Hooghiemstra, Lammerink and Marra (2008), studying 99 Dutch listed firms, find no significant association between supervisory board/AC independence and abnormal accruals.

With respect to board size, several authors have found that larger boards are less effective. According to Jensen (1993), boards that exceed seven or eight members are less likely to “…function effectively and are easier for the CEO to control” (Jensen, 1993, p. 865). Both Yermack (1996), Eisenberg (1998) and Mak and Kusnadi (2005) find, for varying board sizes, that firms with larger boards perform worse. Results on the influence of board/AC size on earnings management are not that straightforward. Xie et al. (2003), studying 110 US firms, find a negative association between board size and abnormal accruals. This may suggest that larger boards are better at constraining earnings management. However, Chin, Firth and Rui (2006), studying 313 Hong Kong-based firms find a positive association between board size and abnormal accruals. Hooghiemstra et al. (2008) firms, find no significant association between supervisory board/AC size and abnormal accruals for 99 Dutch listed firms.

Board activity usually refers to the amount or the nature of the board’s meetings. When boards meet more often or, for example, meet frequently without management present, the quality of board supervision is argued to be higher. In line with this rationale, Xie et al. (2003) and Abbott, Parker and Peters (2004) find that higher board and AC meeting frequency are associated with, respectively, reduced levels of discretionary accruals and reduced likelihood of financial restatements. Hooghiemstra et al. (2008), using a more comprehensive measure of board and AC activity, also find that board and AC activity are negatively related to the amount of abnormal accruals.

Peasnell et al. (2005) find no evidence that the presence of an audit committee directly affects the extent of earnings manipulations. However, Hooghiemstra et al. (2008) find that the presence of a separate AC is negatively related to the amount of abnormal accruals, suggesting that ACs are more effective in constraining earnings management.

Board members with financial expertise are argued to be better at constraining earnings management. Xie et al. (2003), Bedard et al. (2004) and Carcello et al. (2006), all find that the presence of board and AC members with financial backgrounds and/or financial expertise is associated with smaller discretionary accruals. Abbott et al. (2004) examine the likelihood of financial restatements and find a significant negative association with the presence of an AC that includes at least one financial expert. Defond, Hann and Hu (2005), find that the market reacts more favorably to the appointment of accounting financial experts on the AC than the appointment of non-accounting financial experts.

1.3 MULTIPLE BOARD POSITIONS

As stated in the introduction, the influence of multiple board positions3 on the quality of board control is not that straightforward. The busyness hypothesis suggests that board members holding multiple board positions will not be able to produce the required effort in monitoring the firm and its management. Board members have finite time to allocate to their supervisory roles and holding multiple board positions will imply that less time is available for each board position. However, Fama and Jensen (1983) suggest that multiple board positions can signal higher board member quality. The appointment to multiple board positions might be the result of superior performance of the firms for which board members previously served. They suggest that board members holding multiple board positions will work harder to fulfil their supervisory roles. In addition, resource dependency theorists argue that holding multiple board positions allows board members to link firms to valuable resources more easily.

Li (2000) argues that, with respect to multiple board positions, a trade-off exists between the quantity and the quality of board members’ time. Non-executive board members holding multiple board positions are likely to have less time available to fulfil one particular board position (busyness hypothesis). However, if holding multiple board positions is a consequence of board member quality and/or leads to access to valuable resources (reputation hypothesis), the quality of the time of non-executive board members holding multiple board positions is likely to be higher. Chen (2009) investigates the trade-off between the costs and the benefits of non-executive board members holding multiple board positions. He argues that non-non-executive board

members holding multiple board positions will be less effective in monitoring management but more effective in providing management with valuable advice. He argues that firms with high growth opportunities (Tobin’s Q) will have higher advising needs and that firms with high agency conflicts (low governance index) will have higher monitoring needs. Chen finds a positive association between non-executive board members holding multiple board positions and firm performance for firms with low monitoring needs and high advising needs and a negative association between non-executive board members holding multiple board positions and firm performance for firms with high monitoring needs and low advising needs. These findings indicate that whether holding multiple board positions is beneficial or costly may depend on which actions are required from non-executive board members, which in turn depends on firm circumstances.

Several studies have found evidence that supports the busyness hypothesis. Core, Holthausen and Larcker (1999) find that non-executive board members holding three or more (six or more for retired board members) board positions pay higher levels of CEO compensation, which in turn leads to poor firm performance. Fich and Shivdasani (2006) find that firms with boards in which the majority of non-executive board members hold three or more board positions, exhibit lower market-to-book ratios, weaker profitability, and a lower sensitivity of CEO turnover to firm performance. Jiraporn, Kim and Davidson (2008) investigate the diversification discount for a large sample of US firms. Using the same measures as used in Fich and Shivdasani (2006), they find that firms with more board members holding multiple board positions have larger diversification discounts and thus lower firm value. Ahn, Jiraporn and Kim (2010) find that boards with more than half of its board members holding three or more board positions, are associated with significantly lower acquirer returns.Jiraporn, Davidson, Dadalt and Ning (2009) find that non-executive board members holding a higher number of board positions exhibit a higher tendency to be absent from board meetings. Sharma, Naiker and Lee (2009), studying the determinants of audit committee (AC) meeting frequency, find that non-executive board members holding multiple board positions exhibit a higher tendency to be absent from AC meetings.

positions exhibit higher acquisition performance. Di Pietra, Grambovas, Raonic and Riccaboni (2008) study board busyness of Italian firms and find that firms with more board members holding multiple board positions have a higher market valuation. Sarkar and Sarkar (2009) find similar results for Indian firms. They find a positive association between non-executive board members holding multiple board positions and firm valuation. In addition, they find that non-executive board members holding multiple board positions attend more board meetings and are more likely to attend the annual shareholders meeting. Spit (2010, WP), investigating the busyness of two-tier supervisory boards, finds no evidence that board members holding multiple board positions pay their CEOs higher levels of compensation. He does find that firms with more board members holding multiple board positions exhibit higher market valuations, higher subsequent operating performance and higher subsequent stock performance.

Several regulators and legislators have recommended limiting the number of board positions a board member is allowed to have. In the US, the National Association of Corporate Directors (2001) recommends to limit the number of non-executive board positions to a maximum of six per board member and a maximum of three non-executive board positions for board members that also hold an executive position. In the Netherlands the Dutch Corporate Governance Code (2008) recommends limiting the number of supervisory board positions to a maximum of five per board member, to assure “the proper performance of his duties” (2008, III.3.4). Although the DCGC works according a comply-or-explain principle, in the beginning of 2010 a bill that will legally restrict the number of supervisory board positions to a maximum of five (two for non-retired board members) per board member was partially approved by Dutch parliament. By limiting the number of supervisory board memberships to assure board members proper performance, legislators implicitly assume the busyness hypothesis to be correct, although previous literature has provided us with ambiguous results. This study extends research on the influence of holding multiple board positions on the functioning of board members in their supervisory role.

2.HYPOTHESIS DEVELOPMENT

more board positions, firms have significantly higher discretionary accruals, as measured by the modified Jones’ model. This suggests that board members holding more than two board positions are worse at constraining earnings management. However, the reputation hypothesis suggests that supervisory board members holding multiple board positions will work harder to fulfil their duties. In addition, these supervisory board members may have gained experience or have access to resources that may help in monitoring the financial reporting process and constraining earnings management. Xie et al. (2003) find that non-executive board members with corporate experience are associated with lower levels of discretionary accruals. With respect to holding multiple board positions, as suggested by Li (2000) and found by Chen (2009), a trade-off between the quantity and quality of time is likely to exist.

We argue that, up to a certain threshold, holding more board positions will be positively related to supervisory board members’ ability to constrain earnings management. Up to that threshold, additional board positions may signal higher board member quality or link board members to valuable resources, without causing board members’ time to be too limited to adequately perform their supervisory function. When this threshold has been surpassed, holding more board positions will put too much strain on supervisory board members’ time and decrease their ability to constrain earnings management. This suggests an inverted U-relationship between the number of board positions board members hold and their ability to constrain earnings management. An inverted U-relationship implies that we can divide board members into three different categories: (i) board members that hold (too) few board positions, (ii) board members that hold (close to) the ‘optimal’ number of board positions and (iii) board members that hold (too) many board positions. Consequentially, we hypothesize that (H1) firms with high percentages of board members (on their ACs) holding (too) few board positions will exhibit higher degrees of earnings management, that (H2) firms with high percentages of board members (on their ACs) holding (close to) the ‘optimal’ amount of board positions will exhibit lower degrees of earnings management and that (H3) firms with high percentages of board members (on their ACs) holding (too) many board positions will also exhibit higher degrees of earnings management.

H1: There is a positive relationship between the percentage of AC/board members holding (too) few board positions and the degree of earnings management.

H3: There is a positive relationship between the percentage of AC/board members holding (too) many board positions and the degree of earnings management.

3. METHODS

This study extends research on the advantages and disadvantages of non-executive board members holding multiple board positions. We study whether, with respect to holding multiple board positions, a trade-off between the quality and the quantity of board members’ time exists. We examine the influence of holding multiple board positions on board member’s ability to constrain the degree of earnings management. We hypothesize that, up to a certain threshold, holding multiple board positions can increase board member’s ability to constrain the degree of earnings management. We use discretionary accruals to proxies for the degree of earnings management. Since ACs are responsible for supervising the provision of financial information by the company (DCGC, 2008, art. III.5.4.b.), we examine the association between AC members holding multiple board positions and discretionary accruals. We perform both univariate and multivariate analyses in order to study the hypothesized threshold.

3.1 DATA

3.2 MULTIPLE BOARD POSITIONS

Most previous studies4 that examine multiple board positions have examined whether holding (too) many board positions is associated with board member effectiveness. Two measures often used by these studies are: (i) the percentage of total non-executive board members that are ‘busy’ (Core et al, 1999) and (ii) a dummy variable assuming the value of 1 if 50% or more of total non-executive board members are ‘busy’ (Fich and Shivdasani, 2006). Previous studies have used different definitions of what is considered to be a ‘busy’ board member. Ferris et al. (2003) consider board members (both executive and non-executive) to be busy when they hold more than two directorships. Fich and Shivdasani (2006) also consider non-executive board members to be busy when they hold more than two directorships, which, according to them, is in line with recommendations of the Council for Institutional Investors. Core et al. (1999) make a distinction between retired and non-retired board members. When board members are retired they are considered to be busy when they hold more than five directorships. Board members that are not retired are considered to be busy when they hold more than two directorships. Their consideration is in line with NACD guidelines (2001) and with the DCGC and the proposed Dutch legislation. Core et al. (1999) only consider board positions in listed firms in their study, whereas Dutch legislation also counts board positions in private firms and non-profit organizations. Overall, choices for when a board member is considered to have too many board positions in previous literature are somewhat arbitrary, but are mostly (loosely) based on guidelines or recommendations by institutions concerned with corporate governance. Since current considerations for when a board member holds too many board positions are rather ambiguous we have no ex ante prediction for where our hypothesized threshold will be. Univariate analyses will have to indicate where the hypothesized threshold lies and how many board positions will be considered (too) few and (too) many.

Most of previously mentioned studies solely consider board positions at listed firms for these measures. In our sample the average number of supervisory board positions at listed firms held by supervisory board members is approximately 1.9. The DCGC (2008, III.4, Principle) states that the chairman of the supervisory board, among other things, (i) is responsible for the functioning of the entire supervisory board and its committees, (ii) is the main contact for the management board and shareholders and (iii) is responsible for the efficient and orderly conduct of board meetings. Because of these and other additional duties of the chairman a chairmanship

4

position is counted double for the DCGC regulations (2008, art. III.3.4.). Counting chairmanships of supervisory boards double, the average number of positions per board members increases to 2.3. Core et al. (1999) also differentiates between retired and non-retired board members. They consider retired non-executive board members ‘busy’ when they hold more than five board positions and non-retired non-executive board members ‘busy’ when they hold more than two board positions, which is also in line with the Dutch bill that is currently under review by the Senate. In our sample 36% of our board members hold executive positions at other firms or other full-time positions. We weigh the full-time positions as three supervisory board positions, as implicitly assumed by Core et al. (1999) and Dutch legislators. Including full-time positions, the average number of positions held by board members is 3.4.

Table 1. Board Positions held by Supervisory Board Members

(1) (2) (3)

Supervisory board positions at listed firms

Counting chairmanship double

Including executive (full-time) positions # of positions

held per board member # of board members % of board members # of board members % of board members # of board members % of board members 1 1131 55% 930 45% 494 24% 2 378 18% 466 23% 289 14% 3 301 15% 213 10% 153 7% 4 129 6% 191 9% 582 28% 5 85 4% 129 6% 297 14% 6 15 <1% 49 2% 101 5% 7 16 <1% 48 2% 87 4% 8 0 0% 23 1% 32 2% 9 0 0% 6 <1% 14 <1% 10 or more 0 0% 0 0% 6 <1% Total 2055 100% 2055 100% 2055 100% Average # of positions per board member 1.9 2.3 3.4

For our univariate analysis, in order to broadly determine the location of our hypothesized threshold, we divide the board members in our sample into small intervals based on the number of positions they hold as reported in column (3) of table 1 (counting chairmanships double and including full-time positions). Based on the number of positions we divide all board members in 7 different categories (1-2, 3-4, 5-6, >3, >4, >5, >6). For each AC we then calculate the percentage of total AC members that is divided in each of 7 categories. %AC1-2 is the percentage of AC members that hold 1 or 2 positions. %AC3-4 is the percentage of AC members that hold 3 or 4 positions. %AC5-6 is the percentage of AC members that hold 5 or 6 positions. %AC>3 is the percentage of AC members that hold more than 3 positions. %AC>4 is the percentage of AC members that hold more than 4 positions. %AC>5 is the percentage of AC members that hold more than 5 positions. %AC>6 is the percentage of AC members that hold more than 6 positions. Table 2 shows descriptives of variables %AC1-2, %AC3-4, %AC5-6, %AC>3, %AC>4, %AC>5 and %AC>6.

3.3 DISCRETIONARY ACCRUALS

The most common proxy for the degree of earnings management is discretionary accruals (Klein, 2002, Xie et al., 2003, Bedard et al., 2004, Peasnell et al, 2002, 2005, Carcello et al., 2006, Larcker et al., 2007, Hooghiemstra et al. 2008, Sarkar et al., 2008). The most common approach to measuring discretionary accruals is based on a model by Jones (1991). This model calculates discretionary accruals by taking the difference between actual total accruals and an estimation of total discretionary accruals. The assumption underlying this model is that the height of non-discretionary accruals is dependent on firm-specific economic circumstances. Controlling for changes in firms’ economic circumstances, and thus expected changes in non-discretionary accruals, additional changes in total accruals are than considered discretionary. Although the methods for estimating non-discretionary accruals have differed and changed over time, the rationale behind different models (calculating discretionary accruals from actual accruals and estimated non-discretionary accruals) hasn’t changed.

regression of total accruals against several firm characteristics that are hypothesized to determine the height of non-discretionary accruals. The regression model can be specified as follows;

(1) TAs_i,t =

α

β

β

β

β

ε

In line with Klein (2002) and Larcker et al. (2007), total accruals (TAs) are calculated as the difference between income before extraordinary items and operating cash flow (income before extraordinary items - operating cash flow). In line with Jones (1991), changes in revenues (∆Sales) are used to control for changes in total accruals due to economic circumstances that are (largely) out of the managers’ control. As suggested by Dechow et al. (1995, p.199), the changes in revenues are adjusted for changes in trade receivables (∆REC), since “… it is easier to manage earnings by exercising discretion over the recognition of revenue on credit sales”. Gross property, plant and equipment (PPE) is used to control for the portion of total accruals related to non-discretionary depreciation costs, which is in line with Jones’ model. Since PPE acquired in the year under consideration can contribute substantially to the height of depreciation costs, we take the average of gross PPE at the beginning and end of period t. In line with Larcker et al. (2007) we control for firm growth and current operating performance. Firms with large consistent growth can show steady increases in total accruals, which may lead to the overestimation of discretionary accruals and incorrect conclusions about earnings management behaviour (McNichols, 2000). We use a firm’s book-to-market ratio of equity (BM) to proxy for its growth opportunities. Since a firm’s BM is a somewhat course proxy for growth opportunities and may change substantially within short periods of time due to price fluctuations, we take the average of a firm’s BM over the past three years. Dechow et al. (1995) find that the probability of misspecification is higher for firms with high levels of performance. They suggest including a measure of firm performance in the earnings management regression. In line with Larcker et al. (2007) and Hooghiemstra et al. (2008), we include current operating cash flow (CFO) in our model. All variables, except for BM, are scaled by total assets at the beginning of year t.

we estimate regression (1) over four years (2003 – 2006)5. We excluded 8 of 269 firm year observations, because of outlier values6 for TACC scaled by total assets. This leaves a final sample for regression (1) of 8 industries, 69 firms and 261 firm year observations. In line with previous studies using cross-sectional models (DeFond & Jiambalvo, 1994, Defond & Subramanyam, 1998 and Larcker et al., 2007), the portion of total accruals (TAs) that is unexplained by the independent variables (i.e. the residual values; ε) are labelled discretionary accruals (DAs). This implies that DAs are calculated as the difference between the total accruals for each firm less the expected industry-specific non-discretionary accruals for that firm based on the estimated coefficients from regression (1). The specification for DAs is:

) ) ( ( _{1 , 2 , 3 , 4 ,} , ,t it it it it it

i TAs Sales REC PPE BM CFO

DAs = −

α

β

β

β

β

In line with previous studies and as expected we find negative coefficients on PPE. Our coefficients for (∆Sales - ∆REV) are inconsistent. Only, in 4 out of 8 industry regressions do these coefficients show the expected negative sign. Our coefficients for BM show expected positive coefficients in 7 out of 8 industry regressions. Coefficients for CFO show consistent expected negative signs in 7 out of 8 industry regressions. Explanatory power of our industry models (adjusted R2) vary from 19 to 88%, with an average of 53%. The descriptives of TAs and DAs can be found in table 2. The estimation results of regression (1) can be found in Appendix I.

3.4 UNIVARIATE ANALYSES

We examine the Pearson correlations between our measures of multiple board positions (%AC1-2, %AC3-4, %AC5-6, %AC>3, %AC>4, %AC>5, %AC>6) and discretionary accruals. In this study we are not interested in the sign of earnings management but in the degree of earnings management. Both highly negative and highly positive discretionary accruals are indications of high levels of earnings management. In line with Klein (2002) we therefore take the absolute value of discretionary accruals as derived from regression (1) (ABS_DA). The univariate analyses provides insight into the whereabouts of the threshold of holding multiple board positions. Pearson correlations can be found in table 2.

5

The regressions in our main multivariate analyses in section 3.5 are performed over three years (2004 – 2006) because available supervisory board data is restricted to that period.

Table 2. Descriptives of discretionary accruals and our measures for multiple board positions. The percentage of positive observations for TAs and DAs are 26% and 50% respectively. *, **, *** reflect significance at the 10%, 5% and 1% level, respectively. P-values for Pearson correlations are in parentheses.

Our univariate analyses seem to confirm the existence of our hypothesized threshold. The Pearson correlation between %AC1-2 and ABS_DA of 0.13 indicates that firms with higher percentages of AC members holding 1 or 2 board positions are associated with higher levels of discretionary accruals. The Pearson correlation between %AC3-4 and ABS_DA of -0.17 indicates that firms with higher percentages of AC members holding 3 or 4 positions are associated with lower levels of discretionary accruals. This may indicate that, as hypothesized, up to a certain threshold, holding more positions increases board member’s ability to constrain earnings management. The Pearson correlation between %AC5-6 and ABS_DA of -0.03 is still slightly negative, but not significant at the 10%-level. This may indicate that beyond 4 or 5 board positions the positive effect of holding multiple board positions strongly diminishes. The positive coefficients for %AC>4, %AC>5 and %AC>6 and increasing from %AC>4 to %AC>6, seem to confirm that holding more than approximately 4 or 5 board positions will hurt board members’ ability to constrain earnings management.

Table 2. Descriptives Discretionary Accruals & Multiple Board Positions

3.5 MULTIVARIATE ANALYSES

For our multivariate analyses we perform pooled data estimations using Ordinary Least Squares (OLS) analysis. Based on our univariate analysis, we regress the absolute value of discretionary accruals (ABS_DA) against four of our measures of multiple board positions: (i) the percentage of AC members that hold 1 or 2 positions (%AC1-2), (ii) the percentage of AC members that hold 3 or 4 positions (%AC3-4), (iii) the percentage of AC members that hold more than 4 positions (%AC>4) and, (iv) the percentage of AC members that hold more than 5 positions (%AC>5). In line with results from our univariate analysis and with hypotheses H1, H2 and H3, our expectations are: (i) that ACs with high percentages of AC members holding 1 or 2 board positions (%AC1-2) will have larger discretionary accruals, (ii) that ACs with high percentages of AC members holding 3 or 4 (%AC3-4) board positions will have smaller discretionary accruals and, (iii) that ACs with high percentages of AC members holding more than 4 (%AC>4) and 5 (%AC>5) board positions will have larger discretionary accruals. To avoid multicollinearity7 we regress each of these variables separately against ABS_DA. We use heteroscedasticity consistent standard error estimates (White, 1980).

In our regressions, we control for several AC and firm characteristics. In line with previous studies (mentioned in section 1.2.1) we control for AC independence, AC size, AC meeting frequency and financial expertise on the AC. We measure AC independence (AC_INDEP) using a dummy variable that assumes the value of 1 if all AC members are independent. To determine whether supervisory board members are independent or dependent we use the criteria of the DCGC (2008, art III.2.2). We expect more independent ACs to be associated with lower levels of discretionary accruals. We measure AC size (AC_SIZE) by the number of AC members, which is in line with Xie et al. (2003), Chin et al. (2006) and Hooghiemstra et al. (2008). Because of ambiguous results from previous studies concerning the relation between AC size and discretionary accruals we have no ex ante prediction for AC_SIZE. In line with Xie et al. (2003) and Abbott et al. (2004), AC meeting frequency (AC_MEET) is calculated as the total number of meetings of the AC. We expect ACs that meet more frequently to be associated with lower levels of discretionary accruals. We control for financial expertise on the AC, using a dummy variable (AC_FINEXP) assuming the value of 1 if one or more of the AC members can be labeled as a financial expert. In line with Xie et al. (2003), Bedard et al. (2004) and Carcello et al. (2006), we expect the presence of a financial expert of the AC to be associated with lower levels of discretionary accruals. We control for a firm’s debt ratio (LEVERAGE). LEVERAGE is calculated as long-term debt divided by total assets. Carcello et al. (2006) find

that firms with higher debt ratios, which are likely to be subject to increased investor monitoring, have lower discretionary accruals. In line with Klein (2002), we control for the change in net income (∆NETINC), negative net income in prior years (–NETINC) and firm size (ASSET).

∆_{NETINC is calculated as the absolute value of the change in net income scaled by total assets.}

–NETINC is a dummy variable assuming the value of 1 if at least two consecutive periods directly preceding period t had negative net income. Firms that perform poorly may face higher incentives to engage in earnings management. We control for firm size using the natural logarithm of total assets (ASSET). In general, larger firms are more visible and are more likely to be subject to high political costs. We expect larger firms to be associated with lower levels of discretionary accruals. Finally, since the firms in our sample were obliged to change their accounting standards to IFRS during our sample period we control for the accounting standards used. We use a dummy variable (IFRS) assuming the value of 1 if IFRS was applied in the year over which DAs were calculated. The general specification of our multivariate analyses is as follows.

(2) ABS_DAi,t =

α

β

β

ε

Where ABS_DA is the absolute value of the discretionary accruals as derived from regression 1.

AC are the variables for AC members holding multiple board positions (%AC1-2,

%AC3-4, %AC>4 and %AC>5). CONTROL are the control variables for AC characteristics (AC_INDEP, AC_SIZE, AC_MEET and AC_FINEXP) and firm characteristics (DEBT, ∆NETINC, –NETINC, ASSET and IFRS). Descriptives of all variables can be found in table 3.

Since the dependent variable, ABS_DA, does not meet the requirement, imposed by the use of OLS analysis, of a normal distribution, we also perform a separate regression (3), with the following specification.

(3) LnABS_DAi,t =

α

β

β

ε

Where LnABS_DA is the natural logarithm of ABS_DA. Since, taking the natural logarithm of values close to zero returns large negative values, we winsorize the left tail of LnABS_DA at the 95th percentile. This implies that absolute discretionary accruals (ABS_DA) of approx. 0.0025 or closer to zero8 (mean and median ABS_DA are 0.035 and 0.026, respectively) will return the

value of approx. -6. Taking the natural logarithm of ABS_DA lowers the Jarque-Bera statistic9, from 214 to 6.2 (after winsorization). Descriptives of LnABS_DA can be found in table 3. AC and CONTROL are the same variables as incorporated in regression (2).

Table 3. Descriptives of dependent and independent variables incorporated in regressions (2) and (3). The Jarque-Bera statistic of ABS_DA and the natural logarithm of ABS_DA (LnABS_DA) are 214 and 6.4, respectively. *, **, *** reflect significance at the 10%, 5% and 1% level, respectively. P-values in parentheses.

4.RESULTS

Table 4 presents the estimation results for regressions (2) and (3). In line with results from our univariate analysis we find positive coefficients for the percentage of AC members

9 _{The Jarque-Bera statistic is used to calculate the probability of a distribution being a ‘normal’ distribution} from a distribution’s skewness and kurtosis. A Jarque-Bera statistic of <6, indicates normality (5%-level).

Table 3. Descriptive Statistics Dependent and Independent Variables

holding 1 or 2 board positions (%AC1-2), more than 4 board positions (%AC>4) and more than 5 board positions (%AC>5) and negative coefficients for the percentage of AC members holding 3 or 4 board positions (%AC3-4). The positive coefficients for %AC1-2 are not significant at the 10%-level. There is little indication that the percentage of AC members holding only 1 or 2 board positions is positively related to the absolute value of discretionary accruals. The negative coefficients for %AC3-4 are both significant at the 1%-level. The percentage of AC members holding 3 of 4 board positions is strongly negatively related to the absolute value of discretionary accruals, suggesting that AC members holding 3 or 4 board positions are better at constraining earnings management. The positive coefficients for %AC>4 are significant at the 5%- and 10%-level. This indicates that ACs with higher percentages of AC members holding more than 4 board positions are associated with higher absolute discretionary accruals. The positive coefficients for %AC>5 are significant at the 1%-level, indicating that the percentage of AC members holding more than 5 board positions is strongly positively related to the absolute value of discretionary accruals. In line with our univariate analysis the coefficients for %AC>5 are higher than the coefficients for %AC>4. These findings suggest that holding more than 4 board positions decreases AC members’ ability to constrain earnings management and that holding more than 5 board positions even further decreases this ability.

In line with Xie et al. (2003), Bedard et al. (2004) and Carcello et al. (2006), we find evidence that financial expertise on the AC reduces earnings management. Coefficients for the presence of a financial expert on the AC (AC_FINEXP) are negative for all regressions and significant at the 5%- or 10%-level for 5 out of 8 regressions. In line with Klein (2002), we find that firms that reported negative net income in two or more years preceding the observation year (-NETINC) have higher absolute discretionary accruals. Coefficients for –NETINC are all positive and significant at the 5%-level or higher. Finally, we find some evidence that firm size (ASSET) is negatively associated with absolute discretionary accruals. Coefficients for ASSET are negative for all regressions, but only significant at the 10%-level for 2 out of 8 regressions.

Table 4. Estimation results of regression (2) on the absolute value of discretionary accruals (ABS_DA), as calculated by Larcker et al. (2007), and of regression (3) on the natural logarithm of ABS_DA (LnABS_DA). The number of observations in all reported regressions is 104. *, **, ***, reflect significance at the 10%, 5% and 1% level, respectively. T-values are in parentheses.

Table 4. Estimation Results Regression (2) & (3) on the Absolute Value of Discretionary Accruals

Variable Predicted

sign Coefficient estimates of regression (2) on ABS_DA Coefficient estimates of regression (3) on LnABS_DA

4.1 SENSITIVITY ANALYSIS

4.1.1 MODIFIED JONES MODEL

The results from our main multivariate analysis are strongly dependent on our ability to successfully measure discretionary accruals. Klein (2002) argues that any study on earnings management is simultaneously a test of the ability of the adopted model to decompose total accruals into discretionary and non-discretionary accruals. To test the robustness of our results we use another model for the estimation of discretionary accruals; the Modified Jones model from Dechow et al. (1995). This model is most widely used to estimate discretionary accruals and, according to Dechow et al. (1995), is superior to the original Jones model. The Modified Jones model can be specified as follows.

(4) TAs_i,t =

β

β

β

ε

Where total accruals (TAs) are calculated as the difference between income before extraordinary items and operating cash flow (income before extraordinary items - operating cash flow). (1/total_assets) is incorporated to account for firm size. ∆Sales represents the absolute change in sales over period t, to control for changes in total accruals due to economic circumstances. The absolute change in sales is adjusted for the absolute change in trade receivables (∆REC), to account for changes in revenue that may have been realized through exercising discretion over the recognition of revenue on credit sales. Gross property, plant and equipment (PPE) is used to control for the portion of total accruals related to non-discretionary depreciation costs. We take the average of gross PPE at the beginning and end of period t. All variables, except for (1/total_assets), which is already scaled, are scaled by total assets at the beginning of period t. Unlike the model used by Larcker et al. (2007), the modified Jones model does not incorporate a constant (α). In line with our model from regression (1) we estimate regression (4) over the period 2003-2006 for the same 8 two-digit industry BIK-codes separately. The specification for the discretionary accruals calculated with this model is as follows.

) ) ( ) _ / 1 ( ( 1 2 , 3 , , ,t it it it

i TAs total assets Sales REC PPE

DAs = −

β

β

β

coefficients on (∆Sales - ∆REV) for 7 out of 8 industry regressions. Only one of these coefficients is significant at the 5%-level. Coefficients for PPE show the expected negative sign in 7 out of 8 industry regressions, all significant at the 5%-level or higher. Explanatory power of the Modified Jones model of regression (4) is substantially lower than the explanatory power of the model of regression (1), with an average adjusted-R2 of 15%10. We compare the estimated discretionary accruals from the Modified Jones model with the model used in regression (1), based on Larcker et al. (2007). Descriptives from DAs from both models and Pearson correlations between TAs and DAs can be found in table 5.

Table 5. Descriptives of and Pearson correlations between total accruals, estimated discretionary accruals from regression (1) (Larcker et al., 2007) and estimated discretionary accruals from regression (4) (Dechow et al., 1995). The percentages of positive observations for TAs, DAs (1) and DAs (4) are 26%, 50% and 51%, respectively. *, **, *** reflect significance at the 10%, 5% and 1% level, respectively.

The Pearson correlation between discretionary accruals from the two different models is 0.63. The correlation between the estimated discretionary accruals from the Modified Jones model and total accruals is 0.90. This high correlation, that has also been documented by Dechow, Richardson and Tuna (2003), may indicate that the cross-sectional Modified Jones model is not sophisticated enough to decompose total accruals into non-discretionary and discretionary accruals, but merely reflects the differences in the height of total accruals between firms in the same industry.

10 _{The average adjusted-R}2 _{from the industry models of regression (1) was 53%.}

Table 5. Descriptives and Correlations: DAs of Regressions (1) and (4)

Variable Mean Median Min. Max. Standard

Deviation TAs -0.037 -0.034 -0.29 0.16 0.069 DAs (1) -0.000 0.000 -0.16 0.06 0.052 DAs (4) -0.002 0.003 -0.23 0.19 0.065 ABS_DA (1) 0.029 0.025 0.000 0.16 0.026 ABS_DA (4) 0.048 0.036 0.000 0.23 0.045 Pearson

correlations TAs DAs (1) DAs (4) ABS_DA (1) ABS_DA (4)

TAs 1

DAs (1) 0.57*** 1

DAs (4) 0.90*** 0.63*** 1

ABS_DA (1) -0.32*** -0.32*** -0.32*** 1

For robustness we regress the absolute value of discretionary accruals as estimated by the Modified Jones’ model (Dechow et al., 1995) against our measures of AC members holding multiple board positions (%AC1-2, %AC3-4, %AC>4 and %AC>5). As in regression (3), we take the natural logarithm of absolute discretionary accruals to increase normality. We control for the same AC and firm characteristics as in regression (2) and (3). However, since the Modified Jones model does not account for operating performance (CFO) or indications of growth (BM), we include CFO and BM in the regression. Regression (5) can be specified as follows.

(5) LnABS_DAi,t =

α

β

β

ε

Where LnABS_DA is the natural logarithm of the absolute value of discretionary accruals, as measured by the Modified Jones model (Dechow et al., 1995). AC are the variables for AC members holding multiple board positions (%AC1-2, %AC3-4, %AC>4 and %AC>5).

CONTROL are the control variables for AC characteristics (AC_INDEP, AC_SIZE, AC_MEET

and AC_FINEXP) and firm characteristics (DEBT, ∆NETINC, –NETINC, ASSET, IFRS, CFO and BM). Descriptives of all variables can be found in Appendix II.Results of regression (5) are reported in table 6.

In line with results from regression (2) and (3), the signs of the coefficients for %AC1-2, %AC3-4, and %AC>5 are as predicted. However, the coefficient estimate for the percentage of AC members holding more than 4 board positions (%AC>4) is not. Furthermore, the significance of the coefficient estimates is substantially reduced. The percentage of AC members that hold 1 or 2 board positions, the percentage of AC members holding 3 or 4 board positions and the percentage of AC members holding more than 4 board positions are not significantly related to the absolute value of discretionary accruals at the 10%-level. However, the percentage of AC members holding more than 5 board positions is significantly positively related to the absolute discretionary accruals at the 5%-level, suggesting that holding more than 5 board positions reduces the ability of AC members to constrain earnings management.

Table 6. Estimation results of regression (5) on the natural logarithm of the absolute value of discretionary accruals, as measured by the Modified Jones model (Dechow et al., 1995). The number of observations in all reported regressions is 104. *, **, ***, reflect significance at the 10%, 5% and 1% level, respectively. T-values are in parentheses.

Overall, our estimation model from regression (5) proves poorly in predicting absolute discretionary accruals as calculated by the modified Jones model. The average adjusted-R2 is approx. 0.9, which is substantially lower than the average of 6.2 from regression (3). The lower explanatory power may indicate that the discretionary accruals as calculated by the modified Jones model are a poorer proxy for the degree of earnings management. The absence of a

Table 6. Estimation Results Regression (5) on LnABS_DA (Modified Jones Model)

Variable Predicted

sign Coefficient estimates

significant association between financial expertise on the AC and the absolute value of

discretionary accruals as found in regression (2) and (3), which is intuitively appealing, seems to confirm this.

4.1.2 PANEL DATA ANALYSIS

Klein (2002) argues that in the estimation of and regressions on discretionary accruals the probability of omitting relevant variables is high. If the dependent variable is correlated to potentially omitted variables, the estimation results may be distorted. To reduce the effect of and account for potentially omitted variables we perform a panel data estimation of regression (3).

Thus far we have only considered firms with a separate AC in our main multivariate analyses (regressions (2), (3) and (5)). However the estimation models for discretionary accruals (regression (1) and (4)) were performed over the entire sample of firms, including firms that had not appointed a separate AC. In our sample the mean and median absolute discretionary accruals (for both estimation models) for firms with a separate AC (0.029 and 0.025) are slightly lower than the mean and median absolute discretionary accruals of the entire sample (0.034 and 0.026). Hooghiemstra et al. (2008) also find that for Dutch listed firms the presence of an AC is associated with lower discretionary accruals. To verify that the results from regressions (2) and (3) were not influenced by this difference we estimate regression (6) over our entire sample of 69 firms and 195 firm-year observations.

(6) LnABS_DAi,t =

α

β

β

ε

auditing issues, we use the total number of board meetings to proxy for AC activity11. We control for the same firm characteristics as in regression (2) and (3) (DEBT, ∆NETINC, –NETINC, ASSET and IFRS). Results from regression (6) are reported in table 7.

In line with results from regression (2) and (3), the signs of the coefficients for %AC1-2, %AC3-4, %AC>4 and %AC>5 are all as predicted, confirming the hypothesized inverted U-relation between the number of board positions AC members hold and their ability to constrain earnings management. The percentage of AC members holding only 1 or 2 board positions (%AC1-2) is significantly positively related to the absolute value of discretionary accruals at the 5%-level. This suggests that AC members holding only 1 or 2 board positions are worse at constraining earnings management. The percentage of AC members holding 3 or 4 board positions (%AC3-4) is significantly negatively related to the absolute discretionary accruals at the 1%-level, indicating the AC members holding 3 or 4 board positions are better at constraining earnings management. The percentage of AC members holding more than 4 board positions (%AC>4) is significantly positively related to the absolute discretionary accruals at the 10%-level, indicating that AC members holding more than 4 board positions are worse at constraining earnings management. Unlike in regressions (2), (3) and (5) the coefficient for the percentage of AC members holding more than 5 board positions (%AC>5) is not larger than the coefficient for %AC>4 and is not significant at the 10%-level. A possible explanation is that the low amount of AC members holding more than 5 board positions (12%, see table 1.) and the consequentially skewed distribution of %AC>5 (mean and median value of 0.08 and 0) distort the coefficient estimate.

In line with Klein, 2002, Xie et al., 2003, Bedard et al., 2004, Peasnell et al., 2005, Carcello et al., 2006 and Sarkar et al., 2008, we find evidence that AC independence (AC_INDEP) is significantly negatively related to the absolute value of discretionary accruals, indicating that more independent ACs are better at constraining earning management. In line with Carcello et al. (2006) we find that firms with higher debt ratios (DEBT) have lower absolute values of discretionary accruals. Coefficients for both AC_INDEP and DEBT are significant at the 5%-level. In line with, Xie et al. (2003) and Abbott et al. (2004) we find that higher board meeting frequency (BOARD_MEET) is associated with lower absolute discretionary accruals. Finally, we find some, but not very convincing, evidence that the presence of a financial expert on the AC (AC_FINEXP) is associated with lower absolute discretionary accruals. Coefficients for AC_FINEXP are all negative, but only one is significant at the 10%-level.