Good Monitor or Good Networker?
A Study on Social Networks of Board of Directors, Earnings
Management, and Corporate Governance Codes
Jeroen Jurgens
1University of Groningen
Abstract
This paper studies the relationship between social networks of boards of directors and monitoring by boards of directors. Furthermore, this paper focuses on the moderating effect of the installment of the British Combined Code in 2003 on this relationship. Using a sample of 213 and 204 firms for 2002 and 2010, respectively, I find that social networks have a negative effect on monitoring. In addition, I find that the British Combined Code does not have a positive effect on the relationship between social networks and monitoring. These findings indicate that social networks of directors play an important role in monitoring and that social networks are an important topic for further research. Furthermore, my results suggest that constraining the number of directorships is ill-advised as long as the social network is not taking into account.
JEL-Classification: G34; G38
Key-words: Social networks; Earnings management; Corporate governance codes; Board of directors
Date: 10-09-2012
1
2
1 Introduction“Old Boys need to count supervising jobs” - Quote from an article in De Volkskrant, 26th of June 2012.
In recent years there has been an increase of interest, both in popular press and in literature, on the number of directorships of members of board of directors. This increase of interest is also reflected in politics. The quote mentioned above is derived from a newspaper article on the plans of several political parties in the Netherlands on constraining the number of positions supervisors are allowed to have. The main care of these political parties is that supervisors do not have sufficient time for each of their supervisory jobs, resulting in ineffective monitoring. Besides the time aspect, the political parties criticize the “old-boys network”. According to the political parties, “it is time to restrict this cronyism”.
Not only the Netherlands, but also the United States, Germany, and the United Kingdom have focused on restricting the number of jobs of supervisors2. In this paper I argue that simply counting the number of positions of supervisors is not an appropriate way of measuring time spent on directorships. I propose that social networks better proxy time spent on directorships. Besides a better measure of busyness, there is another reason why it is important to consider social networks when examining the effectiveness of monitoring. Upper-class cohesion (also called: old-boys network) points to a negative effect of social networks on monitoring. Following methodology of Barnea and Guedj (2007), Andres and Lehmann (2010), and Renneboog and Zhao (2011), I calculate three measures that capture the weight and intensity of networks. I use these measures to examine how monitoring is affected by directors’ networks. Since countries focus on constraining the number of positions, this paper is focused on the moderating effect of the installment of the British Combined Code in 2003 on the relationship between social networks and monitoring.
Boards of directors play an important role in monitoring top management (Fama and Jensen, 1983). It is often investigated by scholars whether the number of positions of members of boards of directors has an influence on monitoring by these directors. As no unambiguous evidence on the influence of the number of positions exists (see, among others, Fama and Jensen, 1983; Shivdasani and Yermack, 1999; Core, Holthausen, and Larcker, 1999; Ferris, Jagannathan, and Pritchard, 2003; Fich and Shivdasani, 2006; and Andres and Lehmann, 2010), and because
2
3
the results on the busyness of directors seem to be sensitive to the definition of busy directors (Fich and Shivdasani, 2006), there possibly is a better method to measure busyness. Current literature implicitly assumes all directorships equally weighted, however, the time-consumption of directorships can vary notably. It is not only the number of directorships that makes a director overcommitted, but it is also the weight and intensity of these positions that have an influence.An example of a corporate governance code that focuses on the number of positions is the British Combined Code. According to this code, “The board should not agree to a full time executive director taking on more than one non-executive directorship in a FTSE 100 company nor the chairmanship of such a company” (Combined Code, 2003, p.9). However, it might be that this restriction does not achieve its intended goal (i.e. better monitoring). There is a possibility that, due to regulation, a director needs to abdicate directorships but that a director abdicates directorships that are not important and do not come with a large network since their network gives directors status and other benefits. When the quality of monitoring depends on time spent on monitoring and social networks better proxy time spent on directorships, a limit on the number of directorships does not sort the effect it was installed for: the network does not seriously decline, time spent on directorships does not increase, and consequently, monitoring is not improved.
In this study I examine the possible moderating effect of the installment of the British3 Combined Code in 2003 on the relationship between social networks of directors and monitoring by directors. By comparing data from 2002 and 2010 I am able to examine this effect. The primary research question of this study is:
“What is the effect of the British Combined Code on the relationship between social networks4 of boards of directors and on monitoring by boards of directors?”
3
In this study the UK definition of a director is used. This means that a director can be either an executive or a non-executive director. Executive directors are members of the board and exert a senior management position in the company (in the US, they would usually be called officers). The non-executive directors (in the US often called directors) are board members who are not involved in the daily management; they often are managers in other firms (Renneboog and Zhao, 2011).
4
4
To eventually draw conclusions, I have to address the following two research questions first. Firstly, the influence of social networks on monitoring has to be examined. I use earnings management as a measure of the quality of monitoring. The influence on monitoring is studied by answering the following research question:“What is the influence of social networks of members of board of directors on earnings management?”
Secondly, it is necessary to analyze the influence of the installment of the British Combined Code on the number of positions and on the social network of members of boards of directors. This is examined with the following research question:
“What is the influence of the installment of the British Combined Code in 2003 on the number of positions and on the social network of members of boards of directors?”
The analysis reported in this study is based on a dataset comprising directors of companies listed on the FTSE 350 in 2002 and / or 2010.
5
performance, will be a good indicator of future performance, and will accurately annuitize the intrinsic value of the company (Dechow and Schrand, 2004). Good monitoring can prevent earnings management and induce high quality earnings.Lastly, the focus of my paper is the United Kingdom, in contrast to prior research that focused on the United States. The fact that directors in the UK do not have dense networks (Veen, van and Kratzer, 2011) makes the UK an interesting country to examine. If a relationship between social networks and earnings management is found in the UK, it is plausible that this relationship will also hold for other countries with denser networks.
From my findings it appears that the focus of corporate governance codes on solely the number of positions might be inappropriate. By only limiting the number of positions, problems with busyness and upper-class cohesion are not solved. This study shows that it is not only the number of positions that has an influence on monitoring, but also the weight, intensity, and time consumption of these positions. This is confirmed by the finding that all three measures of social networks have a negative influence on monitoring. It has to be noted that two measures have an insignificant negative effect. Turning to the primary research question of this study, I do not find a moderating effect of the British Combined Code on the relationship between social networks and earnings management. Therefore, more research on this subject is necessary to confirm that the focus of corporate governance codes on solely the number of positions might be inappropriate.
The remainder of this paper is organized as follows. Section 2 contains the literature review and hypotheses development. I describe the dataset, the construction of different network measures, and the methodology in section 3. In section 4 I present the main results. Lastly, in section 5 I explain my conclusion.
2 Literature review and hypotheses development
6
2.1 Corporate GovernanceCorporate governance refers to the set of mechanisms and processes that help ensure that companies are directed and managed to create value for their owners while concurrently fulfilling responsibilities to other stakeholders (Merchant & van der Stede, 2007, p.577). Corporate governance’ focus is on controlling the behaviors of top management and, through their direction, those of all the other employees in the firm. Corporate governance is therefore important in each firm; it does not only assure that top managers act in the correct way, but changes in corporate governance mechanisms and practices will usually also have immediate and direct effect on the effectiveness of mechanisms that ensure proper behavior of employees in the organization (i.e. management control systems; Merchant & van der Stede, 2007, p.577). Examples of mechanisms of corporate governance are the board of directors, institutional shareholders, and the operation of the market for corporate control (Larcker, Richardson, and Tuna, 2007). In this study I focus on the board of directors.
7
important stakeholders or other important entities (Burt, 1980; and Hillman, Keim, and Luce, 2001), and facilitating access to resources such as capital (Mizruchi and Stearns, 1988).There is extensive research on the influence of several board characteristics on firm performance and / or on the quality of monitoring. Literature on the influence of several board characteristics often present mixed evidence, for example on the relationship between board independency and firm performance. Some studies find that outside directors help to protect shareholders’ interests by being independent of management’s influence (Byrd and Hickman, 1992; Brickley, Coles and Terry, 1994; and Cotter, Shivdasani and Zenner, 1997). Other scholars, however, find an insignificant or negative impact of outside directors (MacAvoy, Cantor, Dana, and Peck, 1983; Hermalin and Weisbach, 1991; and Bhagat and Black, 2000).
The example of board independence present evidence that board characteristics can have a different effect on the monitoring function than on the provision of resources function. On the one hand, board independence will positively affect the relationship between a board's ability to monitor and actual monitoring. When it comes to the provision of resources, however, a board's independence may be undesirable. While connections to the current CEO/organization may be a disincentive to monitor, they may also be an incentive to provide resources to the firm (Hillman and Dalziel, 2003). These contradictory effects of board dependence on both functions also may explain why so many previous empirical studies of board dependence and firm performance have yielded mixed or insignificant results (Dalton, Daily, Johnson, and Ellstrand, 1999). Although the resource dependence theory might propose a negative relation between social networks and earnings management (i.e. connections of directors are used to obtain information about monitoring and about methods to constrain earnings management), prior literature on social networks and monitoring points to a negative relation. This will be further emphasized in the hypotheses development section.
2.2 Earnings Management
8
the external financial reporting process, with the intent of obtaining some private gain”. Healy and Wahlen (1999) state that “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”. A consequence of earnings management might be the creation of certain agency costs: increased compensation for managers (Holthausen, Larcker, and Sloan, 1995) and reduced likelihood of dismissal when performance is low (Weisbach, 1988). It is clear that earnings management should be constrained. An example of an institution that can do this, and thereby can reduce potential agency costs, is an effective board of directors. Monitoring by boards of directors can reduce agency costs that are inherent in the separation of ownership and control and that are created by earnings management. In this way, firm performance is improved (Fama, 1980; Mizruchi, 1983; and Zahra and Pearce, 1989).It should be noted that earnings management is not by definition costly to shareholders. Subramanyam (1996) concludes that earnings management can be beneficial for shareholders when managers use it as signaling device. He shows that earnings management is used to signal future profitability and dividend changes, which benefits shareholders. Evidence shows that at least some firms appear to manage earnings for these stock market reasons. However, whether the frequency of this behavior is widespread or infrequent is still an open question (Heealy and Wahlen, 1999). Further, there is conflicting evidence on whether it actually has an effect on stock prices. Several recent studies indicate that there are situations in which investors do not see through earnings management. In the banking and property casualty industries investors do see through earnings management because of disclosure regulations (Heealy and Wahlen, 1999). Because in this study I focus on non-financial companies and since the evidence on the signaling explanation is inconclusive, I assume earnings management is costly to shareholders and, therefore, should be constrained.
2.3 Social Networks
9
economic outcomes for three main reasons. Firstly, social networks affect the flow and quality of information. For example, Uzzi (1996) find that manufacturers share certain information only to close ties (e.g. certain buyers), giving these close ties an informational advantage. Secondly, social networks are an important source of reward and punishment. It is proven that business connections affect hiring and compensation (Kuhnen, 2009). In the mutual funds industry, advisors receive higher pay when they are more connected to the fund directors. The pay is not compensated by higher performance. Besides this, fund boards award contracts preferentially to advisory firms which have had more business relationships with the funds’ directors (Kuhnen, 2009). Thirdly, social networks build trust that others in the social network will do the “right” thing despite a clear balance of incentives to the contrary. This is shown by Uzzi (1999), who finds that firms with embedded relations and high network complementarity are more likely to be deemed credit eligible and to receive lower cost financing.Within the corporate setting an important social network is created by members of boards of directors. Relationships between directors may affect independence and behavior, and therefore these relationships are an important corporate governance issue (Brown et al., 2009). Several studies focus on the social structure of members of board of directors. Dooley (1969) performed one of the first academic studies on board interlocks5. He finds that less solvent firms are heavily interlocked with banks and that large corporations in the U.S. are interlocked with other large corporations. He states that the widely accepted view of Berle and Means (1932), the view that the modern corporation is an independent and self-sufficient organ ruled by its own self-perpetuating management, maybe needs to be modified. Levine (1972) investigates the network of board members of major banks in the US and finds that boards of these banks are interlocked with the boards of major industrials.
The studies mentioned here have not focused on the relation between social networks and firm performance and/or quality of monitoring. The studies that do have a focus on this relation will be discussed below.
2.4 Hypotheses development
As stated in the introduction, before I answer the primary research question of this study, two other research questions need to be addressed first. Therefore, I start the hypotheses development section with developing a
5
10
hypothesis for the second research question: “Do Social Networks of Members of Board of Directors Have an Influence on Earnings Management?”Commonly, two different views on the influence of multiple directorships are used by scholars: the busyness view and the reputational view. The busyness view suggests that firms with busy directors are associated with weak corporate governance and monitoring (Ferris et al. 2003; Andres and Lehmann, 2010). Lipton and Lorsch (1992) state that: ‘’Based on our experience, the most widely shared problem directors have is lack of time to carry out their duties’’. Time constraints are often related to in the busyness view. This scarcity of time of directors can be explained by two factors (Hooghiemstra, Marra, and Hermes, 2011). Firstly, the majority of firms adopt the calendar year as their fiscal year. Consequently, this leads to a concentration of board meetings in specific periods of the year. Secondly, as a consequence of regulation, listed firms are required to eastablish board committees6. This leads to additional meetings for directors. According to Kahneman (1973), humans have limited information processing capacity. Competing demands (caused by membership of several boards) can compromise decision making quality (Harris and Shimizu, 2004). As the number of additional directorships on other firms’ boards increases, demands on the individual board member's time decrease the amount of time available for the director to fulfill monitoring responsibilities at a single firm. Therefore, time constraints limit directors’ ability to monitor effectively. This is supported by Core et al. (1999). They find that CEO compensation is higher when outside directors serve on more than three other boards. Investigating financial statement fraud, Beasley (1996) shows that when outside directors of fraud firms (i.e. firms with an occurrence of financial statement fraud publicly reported) hold more directorships in other firms the likelihood of financial statement fraud increases. This result is consistent with the view that additional directorships held by outside directors of fraud firms distract those outside directors from their monitoring responsibilities, thereby increasing the likelihood of financial statement fraud (Beasley, 1996). According to Fich and Shivdasani (2006), who define busy boards as boards in which the majority of outside directors hold three or more directorships, busy boards are not effective monitors. Firms with busy boards are associated with weaker corporate governance: these firms exhibit lower market-to-book ratios, weaker profitability, and lower sensitivity of CEO turnover to firm performance. Sarkar, Sarkar, and Lee (2006) and
11
Berberich and Niu (2011) provide evidence on the relationship between busyness and earnings management. They conclude that the number of positions has a positive influence on earnings management, and thus a negative influence on monitoring. To conclude, an implication of the busyness hypothesis is that the presence of directors with multiple positions on a firm’s board reduces oversight of management and, as a result, reduces market value and increases agency costs (Ferris et al., 2003).Contrastingly, the reputational view promotes holding more directorships. This view is based on Fama (1980) and Fama and Jensen (1983). They state that the number of board memberships signal the quality (reputation) of the monitor. The market for corporate control creates incentives that help ensure directors act in shareholders’ best interests. Directors who perform their monitoring task correctly receive more offers to serve on other boards. Better performing directors are therefore expected to hold more and better quality directorships, implying that firms with board members with a large number of directorships are better monitored. This also implies that directors serving on multiple boards are motivated to perform their monitoring duties vigilantly because they do not want to destroy their reputation (Hooghiemstra et al., 2011). The reputational view is confirmed by, for example, Gilson (1990). He finds that directors who resign from financially distressed firms subsequently serve on fewer boards of other companies. This would imply that directors’ reputational capital is important and that a market for corporate directors exists. Kaplan and Reishus (1990) also observe that there is a market for outside directors that values managerial performance. They find that top managers of poorer performing companies are less likely to obtain directorships in other companies than are managers of better performing counterparts. Keys and Li (2005) present evidence that the market for professional directors is working effectively. They find that there is a limited supply of high quality outside directors and as a result these high quality directors are offered several directorships. These multiple positions reflect the reputation as director and, according to Keys and Li (2005), “Professional directors generally have valuable general human capital that more than offsets the costs of multiple directorships”. To conclude, the reputational view points out that incentives to monitor management also depend on the value of reputational capital proxied by the number of board seats held by a director.
12
misses two important aspects: firstly, how important are these positions? And secondly, how time-consuming are these positions? The measure can be improved by looking at the social networks of directors. In a recent discussion, Andres and Lehmann (2010) focus on the influence of social networks on firm performance. They cite Mayhew and Levinger (1976) who state that the time that can be allocated to any relation is limited and that the number of contacts an agent can sustain will decline as the size of the network increases. It is time-consuming to establish and maintain social ties and this implies that important players in the network will have less time for other obligations (e.g. monitoring) since other members will use the central position of important players in the network as a communication channel. The size and importance (e.g. weight) of social networks can be an improved indication of busyness (e.g. time spending) of directors. Not only does it include the commonly used counting measure (e.g. more positions result by definition in a larger social network), it is also a better proxy for estimating the time directors spent on their directorships.13
number of directors belong to the same social networks (e.g. upper-class cohesion), the CEO is less likely to be dismissed for poor performance.Three explanations can be found why upper-class cohesion negatively influences monitoring. Firstly, according to Janis (1982) and Mullen, Anthony, Salas, and Driskell (1994), cohesive groups perform worse in decision making and often suffer from a reduction in independent critical thinking and are thus worse monitors.
Secondly, Gupta, Otley, and Young (2008) present evidence that upper-class cohesion and the old-boys network of directors can be important in acquiring more directorships. They find that directors of superior performing firms are awarded with better quality directorships7. This confirms the aforementioned reputational view. However, some appointments do not correlate with performance, suggesting that other factors may drive the selection process. The old-boys network can be one of these other factors. This would imply that directors are selected on the basis of informal networks and friendships rather than merit (Gupta et al., 2008). Consequently, monitoring could be inadequate when directors have intense and important networks.
Lastly, Barnea and Guedj (2007) point to the provision of security by upper-class cohesion. They find strong evidence that in firms whose directors are more central in the network CEO pay is higher, CEO pay is less sensitive to firm performance, CEO turnover is less sensitive to firm performance, and forced CEO turnover is less likely to occur. They partly agree with the aforementioned reputational view. When directors are not well connected (i.e. they do not have an intense and important network) they build their reputation by providing superior monitoring, as is in line with the reputational view. However, when they are highly connected they provide softer monitoring of the CEO as they feel that their status in the network is secure. According to Barnea and Guedj (2007), “Connected directors – being central in their social and professional circles – do not need to exert effort to perform monitoring as their centrality in the network serves them well enough”, implying that the market for directors does not function perfectly. This might be explained by the fact that the market for directors is not a single market, but a collection of partially overlapping markets (Horton et al., 2012). Some of the companies, typically in the same industries, compete for the same candidates, but, as whole, the pool of directors
7
14
is not universal. Besides this, since the network is composed of partially-overlapping markets, no single firm is fully aware of the entire pool of directors (Horton et al., 2012). Better-connected directors are likely to be more visible to the firm and this visibility might be a reason to select a director.Although the reputational view provides explanations for a positive relationship between social networks of boards of directors and monitoring (Fama 1980; Fama and Jensen, 1983; Gilson, 1990; Kaplan and Reishus, 1990; and Keys and Li, 2005), most of the evidence is indicating a negative relationship. The busyness hypothesis (Core et al., 1999; Fich and Shivdasani, 2006; Barnea and Guedj, 2007; Andres and Lehmann, 2010; and Renneboog and Zhao, 2011) and upper-class cohesion (Barnea and Guedj, 2007; Non and Franses, 2007; Gupta et al., 2008; Andres and Lehmann, 2010; Renneboog and Zhao, 2011; and Horton et. al, 2012), provide explanations for a negative relationship between social networks and monitoring, and thus a positive realtionship between social networks and earnings management. Therefore, I hypothesize that:
H1: Social networks of members of board of directors have a positivee influence on earnings management.
When a negative influence on monitoring exists, social networks should be constrained. Currently, however, several corporate governance codes restrict the number of positions of members of boards of directors although evidence on the influence of multiple positions is mixed. For example, the Dutch Corporate Governance Code states that “A management board member may not be a member of the supervisory board of more than two listed companies. Nor may a management board member be the chairman of the supervisory board of a listed company” (Dutch Corporate Governance Code, 2009, p. 13). In the US, the corporate governance policy developed by the Council of Institutional Investors states that: “Absent unusual, specified circumstances, directors with full-time jobs should not serve on more than two other boards. Currently serving CEOs should not serve as a director of more than one other company, and then only if the CEO’s own company is in the top half of its peer group. No other director should serve on more than five for-profit company boards” (Corporate Governance Policies, 2011, p.6).
non-15
executive directors. Together with the contributions of a group led by Sir Robert Smith, who wrote a review on audit committees, the contributions of Higgs (2003) led to the publication of the Code by the Financial Reporting Council Ltd. This code applied for reporting years beginning on or after 1 November 2003. One recommendation focuses on the time directors spend on their role. According to the Code: “The board should not agree to a full time executive director taking on more than one non-executive directorship in a FTSE 100 company nor the chairmanship of such a company” (Combined Code, 2003, p.9). Besides this, the code prescribes that “By accepting this appointment, you have confirmed that you are able to allocate sufficient time to meet the expectations of your role. The agreement of the chairman should be sought before accepting additional commitments that might impact on the time you are able to devote to your role as a non-executive director of the company” (Combined Code, 2003, p. 73). About the appointment of board members the code concludes “Care should be taken to ensure that appointees have enough time available to devote to the job. This is particularly important in the case of chairmanships” (Combined Code, 2003, p. 8). It can be concluded that, according to the code, directors need to spend the time required on their directorships.16
extent comply with corporate governance codes. Therefore, I can expect that the installment of the Code in 2003 negatively influenced the number of directorships of director:H2a: The installment of the British Combined Code in 2003 has a negative influence on the number of positions of a director.
A negative influence of the Code on the number of positions of a director might have a negative effect on the social network of a director. Since the director possesses fewer directorships, his social network declines. Although the Code does not have a provision on the social networks of directors, I hypothesize that:
H2b: The installment of the British Combined Code in 2003 has a negative influence on the social networks of a director.
17
the installment of the Code. Therefore, I hypothesize that the negative influence of social networks on monitoring is moderated by the installment of the Code:H3: The installment of the British Combined Code in 2003 moderates the negative influence of social networks on monitoring.
3 Research design
In this section I describe the research design of this study. Firstly, I describe the sample selection of companies and directors. Secondly, the network measures are discussed. The last section contains an explanation of the analyses used to test the three hypotheses.
3.1 Sample Selection companies and directors
The sample of companies for this study is based on all non-financial8 companies that were part of the FTSE 350 in 2002 and on all companies that were part of the FTSE 350 in 2010. Companies that ended their fiscal year (irrespective of exact date) in 2002 and / or 2010 and had a listing during their entire fiscal year were selected. This results in an unbalanced sample of 213 companies for 2002 and 204 companies for 20109. Accounting data is selected in accordance with the company’s fiscal year and is obtained from Compustat and Thomson’s Datastream10. Missing accounting data is hand-collected from annual reports. All figures are presented in UK pounds and hand-collected data is converted into pounds using the same exchange rate as is used in Thomson’s Datastream. TabIe 1 presents descriptive statistics of several firm characteristics.
8 Companies with a SIC code 6,000-6,999 (Division H - Finance, Insurance, And Real Estate) are excluded from all regressions, since financial companies have fundamentally different accruals processes that are not captured by the different models to estimate earnings management. 9
This sample is unbalanced since the overlap between the years is 119 firms. Firms delisted from the FTSE 350 for several reasons, for example, bankruptcy or mergers.
10
18
Table 1Firm Data Description
This table contains descriptive statistics for 213 UK firms for the period 2002 and 204 UK firms for 2010. Leverage is defined as book value of total debt divided by book value of total assets. Return on assets is defined as profit before taxes divided by total assets.
Variable Mean Median SD
2002 2010 2002 2010 2002 2010
Market value of equity (bn UK£) 4,242 7,267 0,939 1,573 12,715 17,655
Book value of assets (bn UK£) 4,756 8,281 1,315 1,744 14,846 23,757
Leverage 0.26 0.22 0.24 0.20 0.18 0.16
Return on assets 4.51% 9.04% 6.16% 7.03% 12.36% 11.10%
As Table 1 shows, market capitalization of firms was higher in 2010 than in 2002. Besides this, the average value of total assets increased between 2002 and 2010. Furthermore, the financial performance, measured by return on assets, was better in 2010 compared to 2002.
19
is included. This can be explained with the help of the network of Oliver Stocken. He is director at two financial companies which are not included in the regression analyses, Standard Chartered plc and 3i Group, and he is director at Home Retail Group plc, which is included in the regression analyses. However, the directorships of Oliver Stocken at the two financial companies do influence his busyness and, presumably, his monitoring at Home Retail Group plc. Therefore, the directors of financial companies are included when calculating the network measures of directors of non-financial companies, but financial companies are not included in the regression analyses.The information collected results in a database with about 2,800 (2,200) directors and approximately 35,000 (26,000) connections for 2002 (2010). Panel A of Table 2 presents descriptive statistics regarding the directorships of selected directors and regarding the boards of selected companies. Panel B of Table 2 shows data regarding the “busyness” of these directors.
Table 2
Director Data Description
This table provides data on 3,230 (2,629) directorships and 2,755 (2,225) directors for 2002 (2010). Panel A presents descriptive statistics regarding the directorships of selected directors and regarding the boards of selected companies. Panel B presents descriptive statistics regarding the “busyness” of directors.
Panel A
Variable Mean Median SD
2002 2010 2002 2010 2002 2010
Board size 10.66 8.73 10 8.00 3.47 2.71
Number of directorships per director 1.18 1.18 1.00 1.00 0.51 0.46
Percentage of executives per board 40.64 34.64 42.68 33.33 18.07 11.74
Percentage of independent directors per board11 41.10 51.64 40.00 54.20 15.38 14.78
Percentage of dependent directors (excluding executives) per board 18.26 13.72 15.38 11.81 16.36 13.85
Panel B
Variable Mean
2002 2010
Percentage of busy directors12 3.19 2.47
Percentage of busy boards13 0.00 0.00
Percentage of directors with more than 1 position 14.16 15.33
As can be seen in Panel A of Table 2, the average board size declined from 10.66 directors in 2002 to 8.73 directors in 2010 and the median board size declined from 8.73 directors in 2002 to 8.00 directors in 2010. The average number of directorships per director remained stable: 1.18 directorships per director. The median number
11
Independent according to the standards of the British Combined Code 2003. 12 A director who holds three or more positions (Core et al., 1999 and Ferris et al., 2003). 13
20
of directorships is in both years 1.00. Panel B of Table 2 shows that the percentage of busy directors, defined as a director who holds three or more positions (Core et al., 1999 and Ferris et al., 2003), declined from 3.19 in 2002 to 2.47 in 2010, and that in both years no busy boards, defined as boards of which the majority of the directors holds three or more positions (Fich and Shivdasani, 2006), existed. This is very low compared to, for example, Germany. Andres and Lehmann (2010) find that in Germany in the period 2003-2006 the average board size is 13.58, the number of directorships per director is 3.49, the percentage of busy directors is 33.07, and the percentage of busy boards is 19.36. It can be concluded that directors in the UK do not hold much directorships and that busyness, according to the definitions of Ferris et al. (2003) and Fich and Shivdasani (2006), is not common. The percentage of executives decreased from 40.64 to 34.64. Another interesting development is the increase of independent14 directors from 41.10 per cent in 2002 to 51.64 per cent in 2010. Furthermore, the percentage of dependent directors decreased from 18.26 per cent in 2002 to 13.72 per cent in 201015. This might be attributed to the installment of the Code in 200316.3.2 Network measures
To investigate the social network of members of board of directors, I follow the method of Andres and Lehmann (2010) and Renneboog and Zhao (2011). The members of a network and their connections can be viewed as a structure of nodes (individual directors) and ties (relationships between directors). Two directors are connected (i.e. there is a tie) when they serve on the same board. The structure is best explained on the basis of an example. Figure 1 depicts the network of Oliver Stocken in 2010. As stated before, Oliver Stocken is the deputy chairman at 3i Group plc, an international investment company. Oliver Stocken was also a non-executive director at the retail company Home Retail Group plc and at the international bank Standard Chartered plc. The three directorships of Oliver Stocken create connections with 31 directors. Standard Chartered plc and 3i Group plc share besides Oliver Stocken a second director: Richard Meddings. Richard Meddings was a non-executive director at 3i Group plc and a finance director at Standard Chartered plc. A sequence between two directors, visiting no directors more than
14 Independent according to the standards of the British Combined Code 2003. 15
These numbers do not add up to 100% since I do not classify executives as dependent of independent. 16
21
once, is a called a path. In Figure 1, multiple paths exist between, for example, Jaspal Bindra, director of Standard Chartered plc, and John Allan, non-executive director at 3i Group plc. For example, Bindra – Stocken – Allan, Bindra – Meddings – Allan, and Bindra – Stocken – Cox – Allan. The length of a path is the number of links it comprises and the shortest path between two directors is called a geodesic path (Scott, 2000). In the case of Bindra and Allan the geodesic path is Bindra – Stocken – Allen or Bindra – Meddings – Allen.Figure 1: Example of a director’s professional network
This figure depicts the network surrounding Oliver Stocken (white circle in the middle of the graph). Directors in the three companies served by Oliver Stocken are represented as different forms. In this figure, each shape stands for a director (node). The lines between the shapes are the links (ties) between directors.
After information on all directors in the sample is collected, the next step is to build a matrix in which each director is represented by a column and a row. Whenever two directors i and j know each other, the value of the intersection point (cell (i, j)) is 1, otherwise it is 0. Since relationships are always bilateral (director i knows director j, which implies that j must know i), this procedure results in a symmetric matrix, with the diagonal (the relation between i and i) being 0 by definition.
22
are used. Based on a method by Renneboog and Zhao (2010), I illustrate the concept of centrality and the calculation of the centrality measures with the creation of a hypothetical network of five companies and nine directors. In Appendix B, the numbers refer to firms and the letters represent directors. Figure 1 and Panel A of Appendix B present an overview of the director network. Panel B of Appendix B contains the aforementioned matrix.The first measure, degree centrality (variable: DEGREE, ), is based on the number of direct ties of a director. According to Andres and Lehmann (2010), degree centrality comes theoretically closest to the “traditional” measure of busyness because it only captures the number of connections. Two directors who have the same number of direct connections will get assigned the same value irrespective of how well connected their contacts are. In the hypothetical director network of Appendix B, the number of links – the degree centrality – for director a is 7. Panel A shows that director a is connected to 5 directors in company 4 and 2 directors in company 5 (directors d and i are “new” connections).
23
g is the total number of directors or nodes and is the number of geodesic paths between j and k that run through i:, (1)
Compared to degree, the betweenness measure is better capable of measuring the importance, and thus the commitment, of directors.
The third measure, eigenvalue centrality (variable: EIGENVALUE, ), captures the extent to which a director is connected to other well connected directors. It is developed by Bonacich (1972, 1987), who states that the quality of the connections should be taken into account when assessing the centrality of a director. Whenever a director gets connected to another well connected agent, this will not only boost his own centrality, but also the centrality of other directors who are connected to him. Eigenvalue centrality can be defined as follows:
, (2)
where stands for the intersection of row i and column j in the matrix described above. Bonacich (1972) shows that there exists a positive eigenvalue λ for every Matrix W that results in a corresponding eigenvector CC that only consists of positive values or 0. This condition is met for the largest positive eigenvalue. Therefore, λ equals the largest positive eigenvalue.
I follow Horton et al. (2012) who normalize the centrality measures by dividing the centrality score by the maximum centrality score in the sample. Therefore, the normalized centrality scores range from zero to one. A higher degree, betweenness, and / or eigenvalue centrality indicates a better-networked director.
24
The measures of centrality can be computed with the help of the UCINET software package v6.171 (Borgatti, Everett, and Freeman, 2002). According to Huisman and van Duijn (2005) the UCINET software package is one of the most useful software packages when analyzing network data.Panels A and B in table 3 present the ten most central directors for each of the three measures in 2002 and 2010, respectively.
Table 3 Most Central Directors
This table contains a list of the ten most central directors in the network of UK firms, subdivided into degree ( ), betweenness ( ), and eigenvalue ( ) centrality measures for 2002 (Panel A) and 2010 (Panel B). represents the score on centrality measure i and n represents the normalized score
on centrality measure i. n is calculated by dividing the centrality score by the highest centrality score in the sample. Numbers presented in the table are used to indicate the level of centrality and to rank directors on the basis of their centrality. No other interpretations can be made on the value of the measures.
Panel A: Top 10 Degree, Betweenness, and Eigenvalue centrality for 2002
No. Name n # Name n # Name n
1 Colin Marshall 67 1.00 1 Colin Marshall 353,573 1.00 1 Colin Marshall 0.267 1.00
2 Charles Sinclair 60 0.90 2 Oliver Stocken 291,834 0.83 2 Philip Hampton 0.160 0.60
3 Ralph Robins 60 0.90 3 John Banham 231,753 0.66 3 Mark Moody-Stuart 0.156 0.58
4 Robert Wilson 54 0.81 4 Raymond Seitz 216,906 0.61 4 Iain Vallance 0.152 0.57
5 Angus Grossart 53 0.79 5 John Walls 197,067 0.56 5 Maarten van den
Bergh
0.151 0.57
6 John Banham 53 0.79 6 George Russell 195,191 0.55 6 James Anderson 0.148 0.55
7 Mark Moody-Stuart 52 0.78 7 Pierre Danon 187,968 0.53 7 Edward Isdell 0.148 0.55
8 Philip Hampton 50 0.75 8 Ralph Robins 187,332 0.53 8 John Weston 0.141 0.53
9 Oliver Stocken 49 0.73 9 Philip Hampton 184,357 0.52 9 Carl Symon 0.141 0.53
10 Robin Renwick of Clifton
49 0.73 10 Norman Broadhurst
173,636 0.49 10 Brian Moffat 0.140 0.52
Panel B: Top 10 Degree, Betweenness, and Eigenvalue centrality for 2010
No. Name n # Name n # Name n
1 John Buchanan 49 1.00 1 Timothy Ingram 237,027 1.00 1 Douglas Flint 0.245 1.00
2 Paul Anderson 45 0.92 2 John McAdam 197,320 0.83 2 Simon Robertson 0.219 0.89
3 Charles Gregson 42 0.86 3 Anita Frew 194,301 0.82 3 Narayana Murthy 0.214 0.87
4 Anita Frew 42 0.86 4 David Tyler 184,080 0.78 4 Rona Fairhead 0.207 0.84
5 Nick Land 42 0.86 5 Philip Rogerson 173,703 0.73 5 Iain Conn 0.089 0.36
6 Anthony Watson 42 0.86 6 Robert Rowley 165,638 0.70 6 Paul Anderson 0.081 0.33
7 Rudolph Markham 41 0.84 7 John Allan 153,941 0.65 7 Byron Grote 0.079 0.32
8 John McAdam 41 0.84 8 Iain Ferguson 149,450 0.63 8 Andy Inglis 0.073 0.30
9 Thomas Parker 39 0.80 9 Simon Fraser 142,471 0.60 9 Cynthia Carroll 0.065 0.27
10 Michael Rake 38 0.78 10 Stephen King 141,785 0.60 10 Brendan Nelson 0.061 0.25
25
and chairman at Rentokil Initial plc and United Utilities Group plc. These four positions resulted in 41 direct connections (degree centrality); only six directors have more direct connections. However, looking at betweenness centrality, only one director has a higher score. This means that the connections of John McAdam are relatively unique. When looking at betweenness centrality, Timothy Ingram is an even better example. Timothy Ingram is not in the top 10 of directors with the highest degree measure; however, he is number one director when it comes to betweenness. His three positions, CEO at Caledonia Investments plc and lead independent director at The Sage Group plc and Savills plc give him a unique position: other directors cannot use other paths to connect to each other. This has the effect that Timothy Ingram can control the flow of information between these companies. As stated before, eigenvalue centrality emphasizes the quality of connections. Douglas Flint holds positions at BP plc and HSBC Holdings plc. Both are very well connected firms, resulting in a high eigenvalue centrality score for Douglas Flint. Table 3 also shows that the centrality of the top ten connected directors in 2010 is much lower compared to the centrality of the top ten connected directors in 2002. Furthermore, Table 3 shows that the network centrality distribution in both years is highly concentrated. In 2002, the 10th best connected director on degree, betweenness, and eigenvalue centrality has a score of 0.73, 0.49, and 0.52, respectively. In 2010, the 10th best connected director on degree, betweenness, and eigenvalue centrality has a score of 0.78, 0.60, and 0.25, respectively. This is very low given sample sizes of directors of 2,755 in 2002 and 2,225 in 2010, respectively. When the network centrality distribution would not be highly concentrated, one would expect the 10th best connected director to have normalized centrality scores close to 1.00 (e.g. 0.98).In Table 4, the descriptive statistics of the three network centrality measures are presented.
Table 4
Descriptive statistics network measures
This table contains descriptive statistics of the network centrality measures degree, betweenness, and, eigenvalue for a sample of 2,755 (2,225) directors for 2002 (2010)
Descriptive statistics centrality measures individual directors
Variable Mean Median SD
2002 2010 2002 2010 2002 2010
DEGREE 12.77 11.71 11.00 10.00 7.03 6.18
BETWEENNESS 8,066.93 8,601.71 0.00 0.00 25,684.81 23,887.9
26
Table 4 shows that the mean degree centrality and eigenvalue centrality in 2010 are lower than in 2002. The average number of connections of a director (degree centrality) declined from 12.77 in 2002 to 11.71 in 2010 and the median of degree centrality declines from 11 to 10. The median betweenness score is in both years 0.00. This can be explained by the fact that the betweenness score by definition is zero when a director does not have any connections outside of “his” board. This holds true for about 85% of the directors in 2002 and 2010, respectively. Eigenvalue centrality declined from 0.0065 in 2002 to 0.0035 in 2010. Whether the values between 2002 and 2010 significantly differ will be emphasized in Section 4 when I test hypothesis 2.The last step is to aggregate the data of individual directors to firm level. In this way, I can draw conclusions concerning the extent to which directors of a firm’s board, and thus the board as a whole, might be busy or overcommitted. This last step will be emphasized in the next section.
3.3 Design of analyses
In this study, I test three hypotheses developed in Section 2. In this section, I describe the design of the analyses of each of these hypotheses.
Hypothesis 1
To test the first hypothesis, I use an OLS regression model based on data of 2002. The dependent, explanatory, and control variables are described below.
Dependent variable
27
be classified as legitimate (nondiscretionary) accruals or managed (discretionary) accruals. The greater the degree of discretion in an accrual, the greater the opportunity for earnings management (Dechow and Schrand, 2004).I use discretionary or abnormal accruals as the dependent variable. Discretionary accruals are a common measure of earnings quality (Hooghiemstra et al., 2011). Management can impound bias into financial statements and the degree of this bias can be measured by the unexpected portion of the reported discretionary accruals (Hunton, Hoitash, and Thibodeau, 2011).
To calculate discretionary accruals, I first estimate total accruals. As stated before, total accruals (TA) can be decomposed into non-discretionary accruals (NA) and discretionary accruals (DA), with NA representing the expected (normal) accruals that would be reported without managerial manipulation such that:
(3) where i and t are firm and period indicators, respectively.
Following prior studies (Jones, 1991; and Dechow, Sloan, and Sweeney, 1995), I compute TA17 as follows:
, (4)
where, for sample firm i in year t,
= the change in current assets ,
= the change in cash and cash equivalents ,
= the change in current liabilities ,
= the change in portion of debt included in current liabilities , and
= the change in depreciation and amortization expense .
Secondly, total accruals have to be seperated into discretionary and non-discretionary accruals. Several methods have been proposed in literature for separating operating accruals into discretionary and
28
discretionary components (Dechow et al., 1995; and Peasnell et al., 2005). In this study, I use the cross-sectional18 modified Jones model to estimate discretionary accruals. According to Dechow et al. (1995), a modified version of the model developed by Jones (1991) is most powerful in detecting earnings management. The modified Jones model estimates discretionary accruals as follows:, (5)
where
= discretionary accruals scaled by ,
= total accruals scaled by ,
= firm-specific parameters, = total assets at t-1,
= the change in revenues of firm i scaled by , = the change in receivables of firm i scaled by
= gross property, plant, and equipment of firm i scaled by ,
= the error term,
j = the industry indicator,
i = the firm indicator, and
t = the year indicator.
Estimates of the firm-specific parameters are generated using the following model in the estimation period:
, (6)
29
where denote the OLS estimates of . To control for differences in non-discretionary accruals across industries, I estimate the model (6) for different industry groups based on work by French (2012)19 who distinguishes 5 different industry groups20. The industry-specific parameters of the modified Jones model are estimated using 2002 and 2010 data21.Signed (income increasing or income decreasing) or unsigned (absolute) measures of earnings management can be used to test for overall differences in earnings quality or the general propensity to manage earnings (Hribar and Nichols, 2007). Signed measures of earnings management are used when hypotheses are one-directionally. Since I do not have developed a hypothesis regarding managers using their discretion to manage earnings up- or downwards, the absolute value of is used.
All continuous data for earnings management is winsorized at 1 per cent and 99 per cent to mitigate the influence of outliers. I investigate whether the association between the level of earnings management and social network measures remains similar when I change the measurement of my earnings management variable. To test this, I use the classical Jones (1991) model and McNichols’ (2002) modification of the Dechow and Dichev (2002) model.
Explanatory variables
The last step is to aggregate the data of individual directors to a firm level. As stated before, I use three normalized network measures (DEGREE, BETWEENNESS, and EIGENVALUE). A higher degree, betweenness, and / or eigenvalue centrality indicates a better-networked director. I follow Horton et al. (2012) by defining the firm centrality measures as the sum of the firm’s directors’ individual centrality measures. DEGREEBOARD, BETWEENNESSBOARD, and EIGENVALUEBOARD represent normalized centrality scores at firm level.
19 Available at http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ftp/Industry_Definitions.zip. 20
See Appendix C for an extensive description of the different industry groups. 21
30
Control variablesTo test the pure influence of social networks on earnings management, I include both board and firm characteristics as control variables. Carcello, Hollingsworth, Klein, and Neal (2008) and Hoitash, Hoitash, and Bedard (2009) developed a composite governance score measuring board strength. The score identifies four characteristics that are associated with board quality and that have an influence on earnings management: board independence, board size, the separation of the chair of the board from the CEO, and whether the CEO is the only executive on the board. Because in the sample of 2002 only 0.65 (0.0007) per cent of the CEOs is also chairman of the board (is the only executive on the board), I exclude the last two characteristics. Evidence on the use of board size is confirmed by Yermack (1996), Core et al. (1999) and, Cheng (2008). Evidence on board independence is also found by Peasnell et al. (2005) and Beasly (1996). Because I use only two measures of the original composite governance score, I include them separately in the regression analysis: BOARDIND and BOARDSIZE. Consistent with prior studies (Jensen, 1993; Yermack, 1996; and Larcker et al., 2007) I view smaller, more independent boards as stronger boards. As Hermalin and Weisbach (1991) show, the relation between economic outcomes and board independence may not be linear. Therefore, for BOARDIND, I calculate the median and assign the value of one for companies above the sample median for percentage of outside directors on the board. For BOARDSIZE, I follow Yermack (1996) who uses a log specification for the board-size variable (based upon the convex association between board size and market value). Piot and Janin (2007) show that the presence of an audit committee is associated with less earnings management. However, in 98 per cent of the firms in the UK an audit committee is present (Renneboog and Zhao, 2011). This indicates too little variation in my dataset and therefore I do not include the presence of an audit committee as a control variable.
31
book value divided by equity book value + liabilities book value) as control variable. Klein (2002) shows that firms having two or more consecutive years of negative income are positively associated with the level of earnings management. Therefore, I include NEGINC, a dummy variable assuming the value of one for firms having two or more consecutive years of negative income, and zero otherwise. Evidence (Healy and Wahlen, 1999; and Hooghiemstra et al., 2011) shows that the issuance of equity capital in the following fiscal year is positively related to earnings management. Therefore, I include ISSUE, a dummy variable assuming the value of one if the firm issues capital in the following fiscal year, and zero otherwise22. To control for errors in the measurement of abnormal accruals (Peasnell et al., 2005), I include CFO, measured as operating cash flow over beginning-of-year total assets. Lastly, I control for industry effects by including industry dummies based on work by French (2012). Appendix D defines all variables used in my analysis.To conclude, to test the first hypothesis I regress the absolute value of abnormal accruals using the following OLS regression model based on data of 2002:
,
(7)
where represents absolute discretionary accruals estimated in (5), represents the constant, the Network measure included is based on the network centrality measure I intend to test (DEGREEBOARD, BETWEENNESSBOARD, or EIGENVALUEBOARD), the represent the coefficients calculated in the regression analysis, and represents the error term.
Hypothesis 2
Hypothesis 2a and 2b are tested using an independent-samples T-test. This T-test is used to compare the mean score on continuous variables for two different groups. In this study, the continuous variables are the average number of positions, the average score on degree centrality, the average score on betweenness centrality, and the average score on eigenvalue centrality. The two different groups are based on date: 2002 or 2010.
22
32
Hypothesis 3I test hypothesis 3 by regressing the absolute value of abnormal accruals using the OLS regression model specified in (7). However, I modify (7) on several aspects. Firstly, I use data of 2002 and 2010. Secondly, I insert a dummy variable to control for the difference between 2002 and 2010. This dummy variable, YEAR, assumes the value of zero for firms of 2002 and the value of one for firms of 2010. Lastly, I include an interaction term to test the possible moderating influence of the installment of the Code. Interactions terms are estimated to infer how the effect of one independent variable on the dependent variable depends on the magnitude of another independent variable (Ai and Norton, 2003). The interaction term is calculated as Network measure * YEAR. The network measures I use are DEGREEBOARD, BETWEENNESSBOARD, and, EIGENVALUEBOARD. This leads to the interaction variables DEGREEMODERATING, BETWEENNESSMODERATING, and EIGENVALUEMODERATING. The coefficient of the interaction term indicates the moderating influence of the installment of the Code on the realtionship between the network measure and earnings management. To conclude, to test the third hypothesis, I regress the absolute value of abnormal accruals using the following OLS regression model based on data of 2002 and 2010:
,
(8)
where represents absolute discretionary accruals, calculated using the modified Jones approach, represents the constant, the Network measure included is based on the network centrality measure I intend to test (DEGREEBOARD, BETWEENNESSBOARD, or, EIGENVALUEBOARD), the represent the coefficients calculated in the regression analysis, and represents the error term.
4 Results
33
4.1 Results hypothesis 1Descriptive statistics
Table 5 presents the descriptive statistics for the variables used in the empirical analysis. Panel A shows information regarding the dependent and control variables. Panel B shows the descriptive statistics of the explanatory variables.
Table 5
Descriptive statistics dependent, control, and explanatory variables
Panel A of this table contains statistics of the dependent and control variables used in the regression analysis. Panel B of this table presents descriptive statistics of the explanatory variables. All variables are described in the main text and in Appendix D. Data in this table is based on a sample of 213 firms in 2002.
Panel A: Descriptive statistics dependent and control variables
Variable Mean Std. Dev. Min Max
MJACCRUALS 0.0894 0.1247 0.0004 0.7722
BOARDIND 0.3900 0.4890 0.0000 1.0000
(dummy for percentage independent directors above median)
BOARDSIZE 2.3453 0.2849 1.6094 3.0445
(log of board size)
SIZE 14.2194 1.3605 11.5640 18.9031
(log of total assets)
LEVERAGE -1.6193 1.1202 -8.1117 0.0000
(log of debt / total assets)
GROWTH 0.4111 0.4355 -0.4158 2.7236
(log of Tobin’s Q)
NEGINC 0.0600 0.2400 0.0000 1.0000
(dummy for two or more consecutive years of negative income)
ISSUE 0.0900 0.2860 0.0000 1.0000
(dummy for issuance of capital in following year)
CFO 0.1473 0.0903 0.0012 0.5049
Panel B: Descriptive statistics explanatory variables
Variable Mean Std. Dev. Min Max
DEGREEBOARD 2.4525 1.4974 0.2985 9.8060
(Sum of directors’ DEGREE)
BETWEENNESSBOARD 0.5504 0.5738 0.0000 3.9778
(Sum of directors’ BETWEENNESS)
EIGENVALUEBOARD 0.2755 0.8556 0.0000 10.7079
(Sum of directors’ EIGENVALUE)
34
negative income and only nine percent of the firms issued shares in the year following the sample year. Almost 40 per cent of the firms have more independent directors than the median. Besides this, almost 65 per cent of the firms issued shares in the year following the sample year.The results shown in panel B of Table 5 demonstrate the firm centrality scores. A higher score on DEGREEBOARD, BETWEENNESSBOARD, and / or EIGENVALUEBOARD indicates a firm with better-networked directors. These directors are busier, are more central in the organizational elites and old-boys network. The statistics in Table 5 show that differences between the centrality measures are fairly large. The firms’ score on betweenness centrality and eigenvalue centrality are much lower than the firms’ score on degree centrality. This is reflected by the fact that in 2002 85 per cent of the directors had a betweenness centrality of 0.00 and 44 per cent of the directors had an eigenvalue centrality of 0.00.
Multivariate test results
Table 6 presents the results of testing hypothesis 1, i.e. whether social networks of members of board of directors have a negative influence on monitoring. Column (1) shows the results of testing the pure model. Columns (2), (3), and (4) show the results of testing the influence of the firm’s degree centrality, betweenness centrality, and eigenvalue centrality, respectively.
Table 6
Director networks and monitoring
This table contains the results of OLS regressions (7) of earnings management, measured using the modified Jones model, on several firm characteristics for a sample of 213 firms for 2002. All variables are described in the main text and in Appendix D. P-values are presented in parentheses below each coefficient estimate. Asterisks denote statistical significance at the 0.01 (**) and 0.10 (*)-level.
Variable Predicted sign Estimate
(1) (2) (3) (4)
DEGREEBOARD + 0.015
(Sum of directors’ DEGREE) (0.034)*
BETWEENNESSBOARD + 0.011
(Sum of directors’ BETWEENNESS) (0.495)
EIGENVALUEBOARD + 0.014
(Sum of directors’ EIGENVALUE) (0.150)
BOARDIND - 0.001 -0.011 -0.002 -0.001
(dummy for percentage independent directors above median) (0.976) (0.531) (0.929) (0.940)
BOARDSIZE + 0.016 0.017 0.017 0.017
(log of board size) (0.545) (0.534) (0.521) (0.524)
SIZE - -0.037 -0.046 -0.040 -0.040
(log of total assets) (0.000)** (0.000)** (0.000)** (0.000)**
LEVERAGE + 0.003 0.002 0.003 0.003
35
Table 6 (continued)Variable Predicted sign Estimate
(1) (2) (3) (4)
GROWTH + 0.016 0.010 0.014 0.013
(log of Tobin’s Q) (0.465) (0.657) (0.534) (0.563)
NEGINC + -0.014 -0.022 -0.012 -0.012
(dummy for two or more consecutive years of negative income)
(0.671) (0.494) (0.704) (0.714)
ISSUE + -0.018 -0.022 -0.017 -0.019
(dummy for issuance of capital in following year) (0.501) (0.412) (0.539) (0.478)
CFO 0.042 0.038 0.049 0.047
(0.689) (0.712) (0.641) (0.653)
Industry dummies Yes Yes Yes Yes
Constant 0.628 0.730 0.657 0.675
(0.000)** (0.000)** (0.000)** (0.000)**
Observations 213 213 213 213
Adjusted R-squared 0.214 0.228 0.215 0.218
The results in Table 6 show mixed findings. The adjusted R-squared of a regression with only control variables (column (1)) is 0.214. The adjusted R-squared of the regression using firms’ directors’ degree centrality is 0.228, so that firms’ directors’ degree centrality increase the adjusted R-squared by 0.014. BETWEENNESSBOARD and EIGENVALUEBOARD do not add much explanatory power. The coefficient of DEGREEBOARD is positive and significant (at the 0.10-level) as predicted. This suggests that firms with boards consisting of directors with a large number of connections are associated with more earnings management and hence, worse monitoring. However, the coefficients of BETWEENNESSBOARD and EIGENVALUEBOARD are, although positive as predicted, insignificant. This might be related to the aforementioned fact that in 2002 85 per cent of the directors had a betweenness centrality of 0.00 and 44 per cent of the directors had an eigenvalue centrality of 0.00.
