The mediating role of CEO compensation in the effect of director compensation on firm performance

Amsterdam Business School

THE

M

EDIATING

R

OLE OF

CEO

C

OMPENSATION IN THE

E

FFECT OF

D

IRECTOR

C

OMPENSATION ON

F

IRM

P

ERFORMANCE

Name: Yunfeng Hu

Student No.: 11086238

Date: June 18, 2016

Thesis Supervisor: Nan, Jiang

MSc Accountancy & Control

C

ONTENTS

1.

I

... 4

2. T

F

H

D

... 6

2.1 Literature Review ... 6

2.1.1 Researches on supervising view ... 6

2.1.2 Researches on collusion view ... 7

2.1.3 Researches on CEO compensation and firm performance .... 9

2.2 Hypothesis Development ... 11

3.

R

M

... 14

3.1 Data Source and Sample Selection ... 14

3.2 Decomposition of Director Compensation ... 15

3.3 Mediating Variable Model ... 20

3.3.1 Mediation test of CEO compensation between director

supervising compensation and firm performance ... 21

3.3.2 Mediation test of CEO compensation between director excess

compensation and firm performance ... 27

3.3.3 Comparison between two different mediations ... 32

3.4 Additional Analyses ... 33

4. D

C

... 38

A

Using the data set of 1089 US listed firms for the year 2014 mainly from Compustat and Execucomp, this paper examines the estimated different effects of different compositions of director compensation on subsequent firm performance, and the expected mediation effect of CEO compensation intervening in the total effects. Regressions of director compensation on explanatory factors are processed to decompose the director compensation into supervising compensation and excess compensation. Mediation tests are then performed based on the classification, followed by the further analyses. In light of the empirical evidence, the hypotheses are both confirmed that director supervising compensation positively affects firm performance whilst director excess compensation sheds negative influence, and both links are mediated by CEO compensation.

1.

I

From the dimension of compensation, this paper is to investigate the relation between the director and the CEO and to what extent this relation affects firm’s future performance. Specifically, there exists the influence of director compensation on firm performance, and the influence is expected to be shed through the mediation effect of CEO compensation. The study examines the influence and corresponding mediation.

Theoretically, the principal-agent relation contributes to the main framework of researching the compensation of senior management. According to the definition given by Jensen and Meckling (1976), an agency relation is “a contract under which one or more persons (the principal(s)) engage another person (the agent) to perform some service on their behalf which involves delegating some decision-making authorities to the agent.” (Jensen and Meckling, 1976, p.4). If both parties to the relation are utility maximizers there is the reason to believe that the agent will not always act in the best interests of the principal, which brings the agency cost, “a high cost as the sum of (1) the monitoring expenditures by the principal, (2) the bonding expenditures by the agent, and (3) the residual loss.” (Jensen and Meckling, 1976, p.4)

In practice field, in terms of the monitoring expenditures, it is a heavy burden of cost to monitor and incentivize agents or CEOs directly by principals or shareholders mainly because of the information asymmetry. Thus, the director system is employed, where directors with specialist knowledge and experience monitor and incentivize agents on behalf of the shareholders, which can optimize the cost of agents’ misbehaviors and the cost of monitoring. Hence, distinguishing the function difference of managers and directors contributes to the development of classic agency theory, and the original “principal-agent relation” gradually evolves into “principal-supervisor-agent relation” where directors mainly play the role of supervisor.

However, on the other hand, this three-tier hierarchy is not always efficient especially when some parts of the chain are contaminated. Specifically, the presence of the collusion between supervisor and agent may manipulate the information received by the principal to benefit one or several participants of the collusion, namely, the supervisor and the agent themselves, at the expense of the principal’s benefits (Tirole, 1986). Hence, is the director-manager relation a supervising relation or a collusion? Based on the question, from the dimension of compensation arrangement, this paper examine the effect of director’s pay on CEO’s pay and afterward on firm performance.

Consistent with the method of Brick et al. (2006), this study distinguishes different compositions of director compensation by regressing the actual compensation against factors that reflect the need and difficulty of directors’ supervising works on CEOs. The proportion explained by these factors can be regarded as director supervising compensation, with the residuals treated as the collusion-related excess compensation.

Mediation tests are then performed using the mediation test model of MacKinnon et al. (2007) to examine the linkages between the different compositions of director compensation and firm performance, and the mediation effect of CEO compensation. This is followed by the further analyses emphasizing on the possible direct influences of different proportions of CEO compensation, namely, the part derived from director due work and the one derived from the collusion with directors, on the subsequent firm performance. Expectations are finally arrived at and enhanced that there exists the relation between director compensation and firm performance, and the direction (positive or negative) of the influence is opposite for different compositions of director compensation. These indirect influences are both mediated by the role of CEO compensation.

As for the main contribution of the research, (1) compared to prior researches which did not pay much attention to the difference between director compensation and CEO compensation, this paper explicitly distinguishes the two compensations, examines the relation between them, and especially divides the director compensation into supervising compensation and excess compensation, a classification that provides empirical support for “supervising view” and “collusion view” separately. (2) Compared to the prior researches which did not provide empirical evidence for the mechanism of how board of directors can affect firm’s performance, this paper establishes a correlated chain “director – CEO – firm performance” by using the mediation effect model and the mediation test regarding different compositions of director compensation is the emphasis of the research. Thus, this paper deepens the research on the relation between the board of directors and firm’s subsequent overall performance. (3) Compared to the prior researches which relied mainly on the two-tier hierarchy, namely, principal-agent hierarchy, this paper adopts three-tier hierarchy developed by Tirole, thus providing a relatively new research.

The remainder of the paper is organized as follows. Section 2 reviews the prior researches with respect to different relationships between directors and CEOs and the respective impacts, followed by the literature regarding the link between CEO

compensation and firm performance. Accordingly, hypotheses are then proposed. In Section 3 several models are indicated to decompose director compensation and examine the mediation effect to response to the research objectives. Discussion and conclusion are stated in Section 4.

2. T

F

H

D

2.1 Literature Review

In terms of the relation between board of directors and CEO, two contradictory views emerges, one of which is that directors play a role of supervisor who can improve the firm’s overall performance by regulating and incentivizing CEO’s behaviors through reporting on the agent's effort or on the productivity parameter, a view which is summarized as “supervising view”; on the other hand, another view comes to assert that the relation between board of directors and CEO is far more intimate than it should be and this intimacy forms a kind of conspiracy which benefits conspirators their own at the expense of principal’s interests, a view summarized as “ collusion view”.

2.1.1 Researches on supervising view

Singh and Harianto (1989) stresses that creating different committees such as compensation committee and nominating committee is one of the most considerable means to improve the boards’ decision-making process. The importance of the existence of compensation committee is clear. According to Conyon and Peck (1998), executives may try to award themselves by raising pays that are not consistent with shareholder interests and compensation committee decreases this opportunity for executives so the committee potentially plays an important role in exercising boardroom control. But compensation committee not just performs the work of reducing senior management self-serving compensation, but more generally determining appropriate level and structure of management compensation and aligning the interest between management and shareholders (Conyon et al., 1995; Ezzamel & Watson, 1997). With respect to earnings management, according to the research of Klein (2002), firms’ abnormal accruals are negatively affected by the audit committee independence. The same relation also applies to board independence and abnormal accruals. That earnings management reflecting CEOs’ intention is decreased can be

regarded as the benefit from directors’ supervision.

Prior research has emphasized the role of the board of directors in supervising and providing consulting for the CEO and the importance of board structure (e.g., Adams and Ferreira 2007; Fama and Jensen 1983; Weisbach 1988). Tirole (2001) uses the term “active monitoring” to reflect director’s task of ensuring that firms are run by competent managers who act in their shareholders’ interest. “An active monitor collects information about the firm’s operations or the manager’s ability and—in contrast to a passive monitor such as a financial analyst, rating agency, or the media—is in a position to fire the manager if he or she believes firm value is not being maximized.”(Cornelli, et al, 2013, p.1)

Furthermore, Cornelli et al. (2013) find that when directors fire CEOs, information about firms' operation or CEOs' competence which is relatively difficult to be verified exerts more influence than verifiable information such as the completion of firms' targets does. And this decision of dismissal is always fair and responsible to the extent that the board fires the CEO only based on the concerns about the CEO’s ability and in turn about the company’s future performance “rather than for poor performance that is the result of bad luck or a decision that was wrong ex-post but reasonable ex-ante.” Cornelli et al, 2013, p.6)

As for whether forced CEO turnover leads to improved performance, Denis and Denis (1995) and Huson, Malatesta, and Parrino (2004) find such improvements after forced CEO replacements, then Cornelli et al. (2013) advance this line of research by showing that the effect is causal. Also, Knyazeva et al. (2013) find partly due to a better consistence between CEOs' and shareholders' interests, increased board independence enhances firm performance and profitability. More specifically, more independent boards lead to a higher proportion of incentive-based compensation for CEOs, and all else equal, greater CEO turnover-performance sensitivity.

2.1.2 Researches on collusion view

“There are strong theoretical reasons for expecting a Board sub-committee such as the remuneration committee to exert an influence on top executive pay. And that influence should be in the interests of the owners, i.e. the shareholder” (Main and Johnston, 1993, p.4). But the due work of compensation committee can be swayed by many factors including the strong influence from senior management. Based on the

Ezzamel & Watson (1997) “bid-ding-up” hypothesis, executives underpaid relative to the market level have their pay raised, but the counterpart paid more than market average has no corresponding downwards adjustment from the compensation committee. They explain it as the existence of a “cozy collusion” between executives and non-executive directors, where the executives earnings are bid up. For the empirical evidence, Main and Johnston (1993) find that where firms set compensation committees, executive compensations are relatively much higher.

Jensen (1993) argues that the board of directors often fails to effectively monitor the firm’s management because board culture may inhibit constructive criticism. Brick et al. (2006) found that director and CEO compensation are strongly positively related partly because of an internal environment of cronyism where board members and management are colluded and do not stand for shareholder’s interests. Another study illustrates that “compensation arrangements have often deviated from arm's-length contracting because directors have been influenced by management, sympathetic to executives, insufficiently motivated to insist on shareholder-serving compensation, or simply ineffectual”(Bebchuk and Fried, 2006, p.14). CEOs' influence over directors enables them to grab "rents-benefits” greater than the due compensation under true arm's-length contracting and they thus in return reward compliant directors using their power over corporate resources.

This kind of cronyism is more notable when the CEO has a stronger power on board decision. Lorsch and Young (1990) report that CEOs are essentially the formal leader in the vast majority of boardrooms and that CEOs can often influence the perception of their performance assessed by directors. They can influence the board decision by nominating board candidates who may be sympathetic to them (Shivdasani and Yermack, 1999). Or they incline to place management-friendly directors in key committee such as nominating committee and compensation committee (Vafeas, 2003). Also, they can achieve the same goal just by colluding with board members (Beetsma et al., 2000). According to another study, CEO compensation is higher when the CEO is also the board chair or the outside directors are appointed by the CEO or are considered “grey” directors”(Core et al, 1998, p.12), indicating that CEO’s power compromises director’s original role, which is also consistent with the view of O' Reilly et al. (1988). When CEOs dominate the board of director, they are more likely to influence the board’s perception through the sources at their disposal thus the CEO-dominated boards will be more likely than outsider-CEO-dominated boards to have higher

levels of assessed CEO efficacy (Haleblian and Rajagopalan, 2006).

From the perspective of earnings management, top management usually has an influential and critical opinion on the final accounting numbers on the financial statement regardless of the emphasis of director’s especially the audit committee’s role. Antle and Nalebuff (1991) indicate that management and external auditor sometimes have divergence in how to best apply accounting standards, and this divergence results either in the external auditor being dismissed or in negotiated and compromised released financial statements. Similarly, Nelson et al. (2000) reported that many earnings numbers on the statements are negotiated due to the intervention of the management. Thus, the role of the director is somehow more like a broker, reconciling the difference between the management and external auditor to produce a negotiated and harmonized, but maybe not fairly accurate financial statement.

2.1.3 Researches on CEO compensation and firm performance

A number of the literature studying the relation between CEO compensation and firm performance focus on how CEO compensation changes with the fluctuation of firm performance. Kato and Kubo (2006), Firth et al. (2006), and Ozkan (2011) confirm the CEO pay--firm performance elasticity using Japan, China, and the UK panel data respectively and the findings are relatively similar and consistent. For the influence of CEO compensation on firm performance, Brick et al. (2006) find that CEOs’ excess compensation is associated with firm underperformance. With respect to incentive compensation, Carpenter and Sanders (2002) conclude that incentive-related CEO compensation is an important predictor of variation in firm performance. In the context of financial restatements, Cheng and Farber (2008) report that in the two years following the financial restatement, a decrease in option-based compensation results in improved profitability because it reduces CEOs’ incentives to excessively risky investments. When treating CEO compensation and top management compensation differently, Zajac (1990) presents that CEOs’ compensation structure exhibits a positive relation with firm performance, and this relation is mediated or increased by the stronger positive effect of top management team compensation. However, the research examining the effects of both different compositions of CEO compensation on firm performance is relatively limited.

structure becomes important to indicate the influence of different director compensation. It is clear that cash compensation and performance-based compensation such as equity and option are the main part of executive compensation. If the collusion does exist, as I expected, CEO may prefer cash compensation than performance-based compensation because direct cash grant can allow more room for discretion than compensation based on well-measured performance. Jensen and Murphy (1990) suggest that equity-based rather than cash compensation for CEOs can incentivize them to maximize firm value. This is consistent with the researches of Jensen and Meckling (1976) and Amihud and Lev (1981) although they interpret it as the risk-averse appetite. From their perspectives, managers prefer structured compensation (cash compensation) than fluctuated one (equity-based compensation) to bear less personal risks because the latter is tied to the firm’s stock return which is, of course, volatile and to some extent, beyond their control. In order to avoid the potential loss related to performance-tied compensation, managers may engage in activities which could negatively affect firm value. In this article, the process is interpreted into the outcome of collusion between directors and CEOs where executives prefer cash compensation to seek benefit with less restriction and choose activities that may ultimately negatively affect firm performance.

However, the executive compensation is not determined by themselves or by shareholders. The level and structure of top executive compensation are usually set by directors. In terms of board structure, there is growing evidence showing that compared to inside directors, outside directors are more independent of top management thus can make a more fair decision on executive compensation. Rosenstein and Wyatt (1990) suggest that on average appointing outside directors can result in positive stock returns. Similarly, Weisbach (1988) finds that CEOs of underperforming firms are more likely to be fired with outsider-dominated boards than insider-dominated ones. Also, Fama (1980) states that inside directors are not in the position to supervise the CEO, and senior management’s domination in the board of director can lead to the collusion and transfer of shareholder wealth. Based on this, the research of Mehren (1995) finds that equity-based compensation is more massively used in firms with more outside directors, whilst firms with more inside directors use more cash compensation influenced by top management. The collusion between inside director and top management makes firms adopt less equity-based compensation, which obtains benefits for the collusive group at the expense of shareholder’s interests.

(1991) indicates that since the outside directors are substantially hired and can be evaluated and removed by CEOs, they are more often than not ineffective in setting the fair level and structure of CEO compensation. Also, boards usually rely on the compensation consultants to determine executive compensations, who are also usually hired by CEOs. A similar view comes from the research of Mangel and Singh (1993), where they conclude that there exist the incentives for collusion between top management and board of directors to the detriment of outside shareholders. These factors may make CEO compensation customized for their personal preference.

2.2 Hypothesis Development

According to Tilore’s three-tier “principal-supervisor-agent” model, direct monitor from dispersed shareholder without specialized knowledge is impractical and highly costly so directors are entrusted by shareholders to monitor and incentivize senior managers’ behaviors. They cannot form direct productivity but can report information on senior managers’ efforts and productivity to shareholders thus bringing indirect but vital effect on the performance of managers and the whole company. Hence, the board of directors functions mainly through the incentive mechanism. From the perspective of compensation, directors can indirectly affect firm performance by intervening the arrangement of CEO’s pay so it is feasible to test the relation between the director and senior manager by researching on the effect of director’ s compensation on CEO’s compensation and then on firm performance.

In this paper, I expect that different compositions of director’s compensation have different influences on CEO’s compensation, thus forming different economic outcomes.

Specifically, the demand for the extent of a combination of director supervision and CEO incentive varies with the extent of complexity of firm operation. Brick et al. (2006) and Cornelli et al. (2013) find that where the firm business is more complicated and thus need both director’s and CEO’s higher-level skills and efforts, there will be a positive relation between director’s pay and CEO’s pay. Furthermore, according to Tilore’s three-tier agency model, it is directors and CEOs that control the quality of information about productivity whereas shareholders do not take charge but and make decisions based on the productivity information that directors report. So it is probable that the more complex the firm operation is, the more difficult it is for directors to

supervise CEO’s behaviors and under this circumstance it will be beneficial to increase director’s supervising compensation. Further, in a complex organization, monitoring and incentivizing works from directors are more needed to ensure less deviation and more compliance of executives’ behaviors, thus reaching a higher-level output. So the diligent directors will be willing to introduce a more intensive incentive compensation to their CEOs in order to motivate them to realize the organizational objectives. Or as Brick et al. (2006) says, this possible positive relation between director compensation and CEOs compensation “could be due to unobserved firm complexity which requires higher levels of skill or effort by the CEO and directors.”

In this situation, there is mainly a supervising relation between directors and CEOs, so it is reasonable that when the supervising compensation of directors is higher, CEO’s incentive compensation will be higher correspondently, and because of the combination of supervising and incentive, firms will witness a better performance in the future. Hence, it brings the first hypothesis.

H1: If director’s supervising compensation is higher, firm performance will be

better. CEO compensation has a mediating effect between director supervising compensation and firm performance.

Meanwhile, there may be a collusion relation between the board of directors and CEOs and “sharing high pay” is the main approach of the collusion. There is evidence that “executive compensation is higher when the board is relatively weak or ineffectual vis-a-vis the CEO” (Bebchuck and Fried, 2003, p.7). In particular, CEO compensation is higher when more of the outside directors have been appointed by the CEO, which could cause them to feel gratitude or obligation to the CEO. Also, CEO pay is 20 to 40 percent higher if the CEO is the chairman of the board (Bebchuk and Fried, 2003). Also according to Tilore’s three-tier agency model, since shareholders do not control the quality of productivity it is possible for directors and CEOs to form “coalition” by means that directors report CEO’s productivity information that is beneficial to CEO’s compensation, thus producing a kind of considerable “coalition rent” which is then shared by directors and CEOs, the conspirators.

Theoretically, CEO compensation and firm performance should be consistent and CEO compensation contracts should be set to maximize firm value (Frank, 1984). However, in the previous empirical studies of the relation between CEO pay and firm

value, no obvious and significant outcome exists. According to another study, “the empirical relation between the pay of top-level executives and firm performance, while positive and statistically significant, is tiny. On average, each $1000 change in shareholder wealth corresponds to an increase in this year’s and next year’s salary and bonus of only two cents” (Baker, Jensen, and Murphy, 1988, p.611). This blurred relation may probably indicate some invisible factors regarding CEO compensation. Also, Jensen and Murphy (1987) find the discrepancy between actual compensation contracts and economic theory based ones and they explain it by suggesting that theoretical approaches miss the consideration of the political factors which may characterize the process of setting executive pay. CEOs always have considerable influence on the board. For example, “CEO influences the composition of the board first, and sets the tone of what's considered on the agenda, what information is available, how issues are dealt with in committee or by the full board, and who is put on what committee” (Lorsch, 1989, p.82)

From the perspective of directors, they are sometimes reluctant to behave against CEO’s tone. “Many boards tend to develop a clubbable, if elusive, characteristic of organizations which place internal harmony and fitness before such attributes as objectivity and independent judgment” (Mueller, 1979.).

Therefore, I anticipate that the higher the non-supervising compensation granted to directors, namely, the excess compensation, the higher the possibility will be that directors and CEOs may reach some forms of “coalition” and share rents. In this situation, there is mainly a collusion relation between directors and CEOs. This compensation not connected with supervising demand or supervising complexity not only makes no contribution to the efficiency of supervising CEO but may shed negative effect on firm performance. As what Brick et al. (2006) pointed out, if cronyism occupies the majority of the relation between directors and CEOs, there will be a negative connection between director and CEO’s excess compensation and firm’s future performance. Here comes my second hypothesis according to this thought.

H2: If director’s non-supervising excess compensation is higher, firm

performance will be worse. CEO compensation has a mediating effect between director non-supervising excess compensation and firm performance.

3.

R

M

3.1 Data Source and Sample Selection

The executive and director compensation data is obtained from Execucomp, while other information regarding companies’ financial positions and outcomes and stock market position is mainly from Compustat After excluding firms with data missing and matching records amongst different database, 1089 firms listed in the US stock markets are finally settled. The main sample year is 2014, while the standard deviation of ROA is calculated based on data from 2009 to 2013 and the dependent variable ROAt+1 is

2015 data. Cross-sectional method is employed in this study where these 1089 firm records covering nearly all the industries for one year in America are used to examine the overall observational industry level. For this study, it is not necessary to process time series analysis or comparison because there is no considerable change related to firm compensation such as the change of overall level, political regulation, or other external environment before or after the sample year. Collecting multiple-year data in this study can be just for enriching the sample size. But after trying to use multi-year data the sample size is smaller than choosing one-year data because of the elimination of null and unmatched records within multiple years. Table 1 describes the main dependent and independent variables in the paper.

Table 1

Variables definitions and descriptions

Name Description

Ln(CEOComp t)

The natural logarithm of CEO total compensation for the year 2014, comprised of Salary, Bonus, Other Annual, Total Value of Restricted Stock Granted, Total Value of Stock Options Granted (using Black-Scholes), Long-Term Incentive Payouts, and All Other Total, by Compustat-Execucomp.

Ln(DirComp t)

The natural logarithm of Director total compensation for the year 2014 as reported in SEC Filings, by Compustat-Execucomp.

Ln(DirComp_S t)

The natural logarithm of Director supervising compensation for the year 2014.

Ln(DirComp_E t)

The natural logarithm of Director excess compensation for the year 2014.

Ln(CEOComp_due t)

The natural logarithm of CEO compensation derived from director due work for the year 2014

3.2 Decomposition of Director Compensation

In light of the work of Brick et al. (2006), a regression model is constructed to measure the director compensation in response to the need and difficulty of director’s supervising works. In this model, the part explained by the need and difficulty of director’s supervising is regarded as supervising compensation, whilst the residual value is the excess compensation irrelevant to supervising works.

𝐿𝑛(𝐷𝑖𝑟𝑝𝑎𝑦

) = 𝛼 + ∑ 𝛽

𝑀𝑜𝑛𝑖𝑡𝑜𝑟

𝑘

+ 𝜀

Where, i represents firm and t represents year; 𝑳𝒏(𝑫𝒊𝒓𝒑𝒂𝒚𝒊,𝒕) is the natural logarithm

Ln(CEOComp_excess t)

The natural logarithm of CEO compensation derived from the collusion with directors.

Ln(Sales t) The natural logarithm of sales of the sample firms in 2014.

Ln(Asset t)

The natural logarithm of total asset of the sample firms till the end of 2014.

Q t

Tobin's Q of the sample firms of 2014. The market value of common stock plus the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities).

ROA t

Return on asset of the sample firms of 2014 defined by the operating and non-operating income before provisions for income taxes, interest and related expenses, and minority interest.(presented in percentage)

ROA t+1

Return on asset of the sample firms of 2015 defined by the operating and non-operating income before provisions for income taxes, interest and related expenses, and minority interest.(presented in percentage)

Standard deviation of ROA

The standard deviation of annual percentage corporate return on assets for the five years from 2009 to 2013.

Leverage t The ratio of total liabilities to total assets.

% Equity owned by CEO The percentage of the company's shares owned by the named executive officer.

of chairman of board compensation, defined as the total compensation including the cash compensation and granted stocks and options; 𝑴𝒐𝒏𝒊𝒕𝒐𝒓 is the part of compensation that can be explained by all the factors affecting the need and difficulty of director’s supervising, a compensation that is called supervising compensation (𝐿𝑛(𝐷𝑖𝑟𝑝𝑎𝑦_𝑀)); 𝜺_𝒊,𝒕 is the residual value which represents the director’s excess compensation.

According to the works of existing researches (Linn and Park, 2005; Brick et al. 2006; Bryan et al. 2000), several factors contribute to the need and difficulty of director’s supervising.

(1) Firms with high investment opportunities tend to compensate more for director’s supervising activities. The reliance on stock-based compensation is due, in part, to the high level of information asymmetry (and the low level of liquidity) that is typical of high-growth firms, which increases the complexity of monitoring managers. In this paper, Tobin’s Q Ratio calculated as the market value of common stock plus the book value of total liability, divided by the book value of total assets, is used to measure firms’ investment opportunities.

(2) Director cash compensation is positively related to leverage of the firm since increased debt may indicate that the firm may require more monitoring as its debt-paying ability is weakening and equity is eroding.

(3) The percentage of equity owned by the CEO exerts negative influence on director cash and total compensation because the increased consistency of CEO’s and shareholder’s interests resulting from a larger proportion of firm equity owned by the CEO may require less monitoring.

(4) There is a positive relation between firm size and director compensation since the complexities of investing and operating decisions for large diversified firms make the boards’ monitoring function extremely difficult (Eaton and Rosen, 1983). In this paper, the natural logarithm of sales is used to measure firm size.

Hence, in the decomposition of director compensation part, the fair supervising compensation is measured by the need and the difficulty of director’s supervising works, which in this case, is the regression of firm’s investment opportunities, firm’s leverage, firm’s sale, and the percentage of equity owned by CEO. And the residual value of the regression model is reasonably the unfair excess compensation for director.

compensation and Table 3 shows the correlations between the director compensation and its determining factors. Most of them are significantly correlated as estimated. With the majority of them existing positive correlations, one of them, the percentage of equity owned by CEOs, also as estimated before, shows a negative and significant correlation with director compensation. This is mainly because the director’s monitoring work is less needed due to the CEO bonding to the shareholders.

This table provides the descriptive statistics of the sample data. The matched compensation data and firm’s data are obtained from Execucomp and Compustat data sets for 1089 firms after deleting the null and duplicated records. In this table, Ln(DirComp t) is the natural logarithm of director

compensation (cash and equity) for 2014; Ln(CEOComp t) is the natural logarithm of CEO total

Table 2 Descriptive statistics Quantity Mean Std. Deviation Percentiles 25% 50% 75% Ln(DirComp t), in thousands of dollars 1089 7.320 0.712 7.000 7.000 8.000 Ln(CEOComp t), in thousands of dollars 1089 9.370 0.873 9.000 9.000 10.000 ROA t 1089 3.506 11.140 1.111 3.724 6.972 ROA t+1 1089 4.200 17.451 2.000 6.000 10.000 Standard deviation of ROA 1089 4.890 7.020 1.000 3.000 6.000 Ln(Sales t), in thousands of dollars 1089 7.470 1.623 6.000 7.000 9.000 % Equity owned by CEO 1089 2.490 6.000 0.000 1.000 2.000 Leverage t 1089 0.602 0.254 0.449 0.601 0.760 Q t 1089 3.515 5.511 0.882 1.299 3.36450 Ln (Asset t) 1089 8.160 1.730 7.000 8.000 9.000

compensation for the year 2014; ROA t is the return on asset of the sample firms of 2014 defined by

the operating and non-operating income before provisions for income taxes, interest and related expenses, and minority interest (presented in percentage); Standard deviation of ROA is the standard deviation of annual percentage corporate return on assets for the five years from 2009 to 2013; Leverage

t is the ratio of total liabilities to total assets; Ln (Asset t) is the natural logarithm of sample firms’ total

asset till the end of the year 2014; % Equity owned by CEO is the percentage of the company's shares owned by the named executive officer; Q t is Tobin's Q of the sample firms of 2014 which is the market

value of common stock plus the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities).

Table 3

Correlations between the director compensation and its determining factors

Ln(DirComp t) Ln(Sales t) % Equity owned by CEO Leverage t Q t ROA t Ln(DirComp t) 1 0.566 -0.289 0.183 0.293 0.077 Sig. (2-tailed) (0.000) (0.000) (0.000) (0.000) (0.011) Ln(Sales t) 1 -0.157 0.261 0.39 0.172 Sig. (2-tailed) (0.000) (0.000) (0.000) (0.000)

% Equity owned by CEO 1 -0.129 -0.083 0.006

Sig. (2-tailed) (0.000) (0.006) (0.846) Leverage t 1 0.06 -0.155 Sig. (2-tailed) (0.046) (0.000) Q t 1 0.102 Sig. (2-tailed) (0.001) ROA t 1

This table provides the correlations between the director compensation and its determining factors of the sample data. The matched compensation data and firm’s data are obtained from Execucomp, Compustat, and IBES data sets for 1089 firms after deleting the null and duplicated records. In this table, Ln(DirComp t) is the natural logarithm of Director total compensation for 2014 as reported in

SEC Filings; Ln(Sales t) is the natural logarithm of sales of the sample firms of 2014; Q t is Tobin's Q

of the sample firms of 2014 which is the market value of common stock plus the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities); ROA t is the return on asset of the sample firms of 2014 defined by the

and minority interest (presented in percentage); Leverage t is the ratio of total liabilities to total assets; %

Equity owned by CEO is the percentage of the company's shares owned by the named executive officer.

To decompose the director compensation into supervising compensation and excess compensation, the model is introduced within which the proportion explained by the independent variables is the reasonable compensation while the unexplained residual can be proxied for the unreasonable excess part. In this article, forward selection approach is employed to determine the final effective independent variables for the model, starting with the testing for the most correlated candidate to examine the P-value. After the whole selection, Ln(Sales t), % Equity owned by CEO, and Q t are

settled but Leverage t and ROA t are eliminated with the P-value of 0.674 and 0.719

respectively. The selection process and the subsequent outcomes are as follows:

Table 4

Regression of director compensation

Dependent Variable: Ln(DirComp t)

(1) (2) (3) (4) (Constant) 5.420 5.592 5.654 5.654 （66.811）*** （68.900）*** （67.802）*** （67.802）*** Ln(Sales t) 0.253 0.238 0.225 0.225 (23.707) *** (22.678) *** (19.887) *** (19.887) *** % Equity owned by CEO -0.025 -0.025 -0.025 (-8.665) *** (-8.631) *** (-8.631) *** Q t 0.010 0.010 (3.049) *** (3.049) *** Leverage t 0.010 (0.421) ROA t -0.008 (-0.360) Adjusted R2 _{0.322 0.361 0.366 0.366} F-statistics 562.048 336.168 228.784 228.784

N 1089 1089 1089 1089

This table presents the estimated coefficients from the regression on director compensation. The dependent variable is the natural logarithm of director total pay defined as the combination of Salary, Bonus, Other Annual, Total Value of Restricted Stock Granted, Total Value of Stock Options Granted (using Black-Scholes), Long-Term Incentive Payouts, and All Other Total, by Compustat-Execucomp. The estimated independent variables are: Ln(Sales t) is the natural logarithm of sales of the sample firms

of 2014; Q t is Tobin's Q of the sample firms of 2014 which is the market value of common stock plus

the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities); ROA t is the return on asset of the sample

firms of 2014 defined by the operating and non-operating income before provisions for income taxes, interest and related expenses, and minority interest (presented in percentage); Leverage t is the ratio of

total liabilities to total assets; % Equity owned by CEO is the percentage of the company's shares owned by the named executive officer.

Numbers in the parentheses are T-values. ***. Significant at the 0.01 level.

Hence, after eliminating the ineffective variables with high P-values in the group (3), final variables are settled and presented in the group (4). Since the constant and each predictor are determined with respective coefficients, the residual can be calculated for each record of the data. Using this way, the director compensation is decomposed into the reasonable supervising part and unreasonable excess part. Then for both of the parts, mediating analyses are followed to examine the different influences that different forms of director compensation can shed on the different levels of CEO compensations, then on the subsequent firm performances.

3.3 Mediating Variable Model

In the paper’s setting, CEO compensation is the mediating variable, which means director compensation has the impact on CEO compensation and then CEO compensation influence firm performance. Three regressions need to be built to test the mediating role of CEO compensation (MacKinnon et al. 2007):

𝑅𝑂𝐴_𝑖,𝑡+1 = 𝛼₁+ 𝑐𝐿𝑛 (𝐷𝑖𝑟𝐶𝑜𝑚𝑝_𝑖,𝑡) + ∑ 𝛽_𝑘𝐶𝑜𝑛𝑡𝑟𝑜𝑙_{𝑘,𝑖,𝑡}

𝑘

𝐿𝑛 (𝐶𝐸𝑂𝐶𝑜𝑚𝑝𝑖,𝑡) = 𝛼2+ 𝑎𝐿𝑛 (𝐷𝑖𝑟𝐶𝑜𝑚𝑝𝑖,𝑡) + ∑ 𝛽𝑘𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑘,𝑖,𝑡 𝑘 + 𝜀𝑖,𝑡 𝑅𝑂𝐴_𝑖,𝑡+1 = 𝛼₃+ 𝑐′𝐿𝑛 (𝐷𝑖𝑟𝐶𝑜𝑚𝑝_𝑖,𝑡) + 𝑏𝐿𝑛 (𝐶𝐸𝑂𝐶𝑜𝑚𝑝_𝑖,𝑡) + ∑ 𝛽_𝑘𝐶𝑜𝑛𝑡𝑟𝑜𝑙_{𝑘,𝑖,𝑡} 𝑘 + 𝜀_𝑖,𝑡

Where, in the first equation, c measures the overall effect of director supervising compensation on firm performance; in the second equation, a reflects the effect of director supervising compensation on CEO compensation; while in the last regression, c’ is the direct effect of director pay on firm future performance and b represents the direct effect of CEO pay on firm future performance. Similarly, if another three equations are built, it is possible to test how director excess compensation can influence CEO compensation and then firm future performance. Also following MacKinnon et al. (2007), to prove the existence of mediation of the CEO compensation in this case, the coefficients must meet the following conditions: (1) in the first equation, the coefficient c must be significant; if so, (2) in the second equation, the coefficient a must be significant; if so, (3) then in the last equation, a significant coefficient b is the premise of the existence of the mediation, where if the coefficient c’ is insignificant then there is full mediation, or there is nonfull mediation.

3.3.1 Mediation test of CEO compensation between director supervising

compensation and firm performance

In the first mediation test, the main purpose is to examine the existence and the significance of the CEO compensation’s mediation effect on the relation between reasonable director supervising compensation and firm performance. As the hypothesis indicates, to reach the firm objectives, the diligent and competent directors are willing to introduce a more intensive incentive compensation to their CEOs as the motivation. Or due to the firm or operation complexity, a higher level of skills and efforts of directors and CEOs is needed. Hence, the estimation will be that there is the mediation effect of CEO compensation on the relation between director supervising compensation and firm performance. And the higher the director supervising compensation is, the better the firm performance will be.

the results of the regressions are presented. Table 5 reports the results of the regression of next-year firm performance on director supervising compensation and other control variables. Column (1) in this table contains the basic model, where firms’ ROAs of the year 2015 are regressed against director supervising compensations for the year 2014. The coefficient 3.833 of the independent variable is positive and significant, indicating that firm performance increases with the increase of director supervising compensation. To test the sensitivity of the previous result, additional explanatory variables are then introduced in the regression, the outcome of which is presented in column (2). It shows that the coefficient 7.391 of Ln(DirComp_S t) is more positive and significant,

indicating that firms usually have better performances when their director supervising compensations are high. Hence, in the first step of mediation test, the coefficient of the independent variable is significant.

Table 6 shows the result of the second step of mediation test which examines the relation between the mediating variable and the independent variable. In this regression, column (1) only contains the basic variables where CEO compensation is regressed against director supervising compensation. The coefficient here is positive and significant and after adding other explanatory variables the new coefficient of the independent variable is still positive and significant, which is exhibited in column (2).

In Table 7, the final step of the mediation test is processed, which regresses the dependent variable against independent variable and mediating variable. In column (1), I simply test the basic model and it shows that the coefficients of both the independent variable and the mediating variable are positive and significant. After considering other explanatory variables the examined variables are more significant, which meets the expectation.

Therefore according to the mediation model presented above, all criteria are met thus there does exist the mediation effect and the mediating variable is CEO compensation. But since the coefficient of independent variable Ln(DirComp_S t) is

also significant, there is no full mediation, which means only part of the relation between firm performance and director supervising compensation is mediated by CEO compensation. Directors do not directly make better or worse firm performance because

they do not engage in the daily operation which is exactly CEO’s job. But directors can shed significant influence on firm performance by regulating, monitoring, and incentivizing CEO’s behaviors. Apart from taking the responsibilities, directors may engage in the collusion or conspiracy with top management to seek their private benefit at the expense of shareholders’ interest. These different choices can be reflected in director compensation. The increase of director supervising compensation resulted from the difficulty and the need of monitoring works on senior management in this case brings an increase to CEO compensation because of the more intensive incentive plan made by directors and the increasing difficulty of operation. Then incentivized managers bring the better firm performance by diligent works and optimized activities more consistent with firm value. Obviously, through this test, the first hypothesis “If

director supervising compensation is higher, firm performance will be better. CEO compensation has a mediating effect between director supervising compensation and firm performance” is confirmed.

Table 5

Regression of firm performance on director supervising compensation

Dependent Variable: ROA t+1

(1) (2)

(Constant) -30.385

(-3.373)***

Ln(DirComp_S t) 3.833 7.391

(3.765) *** (5.207) ***

Standard deviation of ROA -0.477

(-5.985) ***

Leverage t 4.686

(2.108)**

Ln (Asset t) -2.559

% Equity owned by CEO 0.202 (2.216)** Q t 0.168 (1.605) Adjusted R2 _{0.012 0.060} F-statistics 14.172 12.510 N 1089 1089

This table presents the results of the regression of next-year firm performance on director supervising compensation and other explanatory variables. The dependent variable is the sample firms’ ROA of the year 2015 and the predictors are: Ln(DirComp_S t) is the natural logarithm of director supervising

compensation (cash and equity) for 2014; Standard deviation of ROA is the standard deviation of annual percentage corporate return on assets for the five years from 2009 to 2013; Leverage t is the ratio of

total liabilities to total assets; Ln (Asset t) is the natural logarithm of sample firms’ total asset till the end

of the year 2014; % Equity owned by CEO is the percentage of the company's shares owned by the named executive officer; Q t is Tobin's Q of the sample firms of 2014 which is the market value of

common stock plus the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities).

Numbers in the parentheses are T-values. **. Significant at the 0.05 level.

***. Significant at the 0.01 level.

Table 6

Regression of CEO compensation on director supervising compensation

Dependent Variable: Ln(CEOComp t)

(1) (2) (Constant) 2.345 3.964 (7.614) *** (11.244)*** Ln(Dircomp_S t) 0.964 0.487 (22.866)*** (8.757)*** ROA t 0.005 (0.013) Leverage t -0.077

(-0.869)

Ln (Asset t) 0.217

(12.395)***

% Equity owned by CEO 0.003

(0.864)

Q t 0.012

(2.992)***

Standard deviation of ROA 0.014

(4.334)***

Adjusted R2 _{0.385 0.425}

F-statistics 114.542 116.002

N 1089 1089

This table shows the results of the regression of CEO compensation on director supervising compensation and other explanatory variables. The dependent variable is the natural logarithm of CEO total compensation for 2014, comprised of Salary, Bonus, Other Annual, Total Value of Restricted Stock Granted, Total Value of Stock Options Granted (using Black-Scholes), Long-Term Incentive Payouts, and All Other Total, by Compustat-Execucomp and the predictors are: Ln(DirComp_S t) is

the natural logarithm of director supervising compensation (cash and equity) for 2014; ROA t is the

return on asset of the sample firms of 2014 defined by the operating and non-operating income before provisions for income taxes, interest and related expenses, and minority interest (presented in percentage); Standard deviation of ROA is the standard deviation of annual percentage corporate return on assets for the five years from 2009 to 2013; Leverage t is the ratio of total liabilities to total assets;

Ln (Asset t) is the natural logarithm of sample firms’ total asset till the end of the year 2014; % Equity

owned by CEO is the percentage of the company's shares owned by the named executive officer; Q t

is Tobin's Q of the sample firms of 2014 which is the market value of common stock plus the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities).

Numbers in the parentheses are T-values. ***. Significant at the 0.01 level.

Table 7

Regression of firm performance on director supervising compensation and CEO compensation

(1) (2) (Constant) -24.743 -38.343 (-3.239)*** (-4.040) *** Ln(Dircomp_S t) 3.422 6.396 (2.761) *** (4.362) *** Ln(CEOComp t) 0.426 2.011 (0.582) *** (2.602) ***

Standard deviation of ROA -0.501

(-6.266) ***

Leverage t 4.929

(2.221)**

Ln (Asset t) -2.996

(-6.294) ***

% Equity owned by CEO 0.196

(2.148)** Q t 0.142 (1.356) Adjusted R2 _{0.011 0.065} F-statistics 7.251 11.748 N 1089 1089

This table exhibits the results of the regression of next-year firm performance on director supervising compensation and CEO compensation. The dependent variable is the sample firms’ ROA of the year 2015 and the independent variables are: Ln(DirComp_S t) is the natural logarithm of director

supervising compensation (cash and equity) for 2014; Ln(CEOComp t) is the natural logarithm of CEO

total compensation for 2014, comprised of Salary, Bonus, Other Annual, Total Value of Restricted Stock Granted, Total Value of Stock Options Granted (using Black-Scholes), Long-Term Incentive Payouts, and All Other Total, by Compustat-Execucomp; Standard deviation of ROA is the standard deviation of annual percentage corporate return on assets for the five years from 2009 to 2013; Leverage

asset till the end of the year 2014; % Equity owned by CEO is the percentage of the company's shares owned by the named executive officer; Q t is Tobin's Q of the sample firms of 2014 which is the market

value of common stock plus the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities).

Numbers in the parentheses are T-values. **. Significant at the 0.05 level.

***. Significant at the 0.01 level.

3.3.2 Mediation test of CEO compensation between director excess

compensation and firm performance

Following the test of mediation effect of CEO compensation between director supervising compensation and firm performance, this part focuses the director excess compensation, specifically, the mediation effect of CEO compensation between director excess compensation and firm performance. As explained before, besides the monitoring work, directors may engage in the collusion with top management to share the rents. Executives can have extensive influence on the tone of the board, such as setting the agenda and determining the composition. Also, directors are sometimes reluctant to behave against CEO’s tone either to prioritize the internal harmony or to guarantee the mutual interests. This collusion can be reflected in the director excess compensation when actual compensation exceeds the one they deserve.

Shielded, overlooked, or connived by directors, CEO may make more arbitrary decisions or take more self-serving behaviors at the expense of firm value. Therefore, although similar in the basic pattern of mediation effect described as “director compensation – CEO compensation – firm performance” chain, the direction of effect, as predicted, will be opposite. Supervising compensation boosts firms future performance whilst excess compensation erodes firm resources. The mediation tests are hereby presented.

Table 8 exhibits the results of the first step of the mediation test, where the relation between director excess compensation and firm performance is examined. Column (1) is based on the basic model where only the key independent variable is added. The coefficient -3.095 is negative and significant, and obviously, different from the counterpart in the mediation test concerning director supervising compensation, the negative relation indicates that the increase of excess compensation would worsen firm performance. When involving other explanatory variable in column (2), the

corresponding coefficient is still negative and significant, strengthening the prior outcome.

The second step of mediation test is to examine the relation between the independent variable and the mediating variable. In column (1) of Table 9, the coefficient 0.517 is positive and significant, and after adding more explanatory variables in column (2) the significance still exist. The results of the last step which regresses the dependent variable against independent variable and mediating variable is presented in Table 10, where the coefficients of both variables are significant. Therefore, the mediation effect of CEO compensation on the relation between director excess compensation and firm performance does exist and it is not the full mediation, either. The second hypothesis is then confirmed: If director non-supervising excess

compensation is higher, firm performance will be worse. CEO compensation has a mediating effect between director non-supervising excess compensation and firm performance.

Table 8

Regression of firm performance on director excess compensation

Dependent Variable: ROA t+1

(1) (2)

(Constant) 4.295 13.039

(8.131)*** (4.194) ***

Ln(Dircomp_E t) -3.095 -2.269

(-2.838) *** (-2.095)**

Standard deviation of ROA -0.464

(-5.733) ***

leverage t 3.879

(1.729)*

Ln (Asset t) -1.231

(-3.185) ***

% Equity owned by CEO 0.076

(0.859)

(2.821) ***

Adjusted R2 _{0.006 0.04}

F-statistics 8.051 8.559

N 1089 1089

This table presents the results of the regression of next-year firm performance on director excess compensation and other explanatory variables. The dependent variable is the sample firms’ ROA of the year 2015 and the predictors are: Ln(DirComp_E t) is the natural logarithm of director excess

compensation (cash and equity) for 2014; Standard deviation of ROA is the standard deviation of annual percentage corporate return on assets for the five years from 2009 to 2013; Leverage t is the ratio of

total liabilities to total assets; Ln (Asset t) is the natural logarithm of sample firms’ total asset till the end

of the year 2014; % Equity owned by CEO is the percentage of the company's shares owned by the named executive officer; Q t is Tobin's Q of the sample firms of 2014 which is the market value of

common stock plus the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities).

Numbers in the parentheses are T-values. *. Significant at the 0.1 level.

**. Significant at the 0.05 level. ***. Significant at the 0.01 level.

Table 9

Regression of CEO compensation on director excess compensation

Dependent Variable: Ln(CEOComp t)

(1) (2) (Constant) 9.355 7.001 (367.874)*** (58.757) *** Ln(Dircomp_E t) 0.517 0.449 (9.843) *** (10.885) *** ROA t 0.006 (3.243) *** Leverage t -0.077 (-0.883) Ln (Asset t) 0.277

(18.821) ***

% Equity owned by CEO -0.006

(-1.733)*

Q t 0.024

(6.157) ***

Standard deviation of ROA 0.011

(3.403) ***

Adjusted R2 _{0.081 0.445}

F-statistics 96.876 125.763

N 1089 1089

This table shows the results of the regression of CEO compensation on director excess compensation and other explanatory variables. The dependent variable is the natural logarithm of CEO total compensation for 2014, comprised of Salary, Bonus, Other Annual, Total Value of Restricted Stock Granted, Total Value of Stock Options Granted (using Black-Scholes), Long-Term Incentive Payouts, and All Other Total, by Compustat-Execucomp and the predictors are: Ln(DirComp_E t) is the natural

logarithm of director excess compensation (cash and equity) for 2014; ROA t is the return on asset of

the sample firms of 2014 defined by the operating and non-operating income before provisions for income taxes, interest and related expenses, and minority interest (presented in percentage); Standard deviation of ROA is the standard deviation of annual percentage corporate return on assets for the five years from 2009 to 2013; Leverage t is the ratio of total liabilities to total assets; Ln (Asset t) is the natural

logarithm of sample firms’ total asset till the end of the year 2014; % Equity owned by CEO is the percentage of the company's shares owned by the named executive officer; Q t is Tobin's Q of the

sample firms of 2014 which is the market value of common stock plus the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities).

Numbers in the parentheses are T-values. *. Significant at the 0.1 level.

***. Significant at the 0.01 level.

Table 10

Regression of firm performance on director excess compensation and CEO compensation

(1) (2) (Constant) -16.791 -13.358 (-2.853)*** (-2.105)** Ln(Dircomp_E t) -4.260 -3.940 (-3.763) *** (-3.490) *** Ln(CEOComp t) 2.254 3.746 (3.597) *** (4.756) ***

Standard deviation of ROA -0.496

(-6.173) ***

Leverage t 4.381

(1.970)**

Ln (Asset t) -2.276

(-5.157) ***

% Equity owned by CEO 0.098

(1.112) Q t 0.195 (1.893)** Adjusted R2 _{0.017 0.059} F-statistics 10.540 10.714 N 1089 1089

This table exhibits the results of the regression of next-year firm performance on director excess compensation and CEO compensation. The dependent variable is the sample firms’ ROA of year 2015 and the independent variables are: Ln(DirComp_E t) is the natural logarithm of director excess

compensation (cash and equity) for 2014; Ln(CEOComp t) is the natural logarithm of CEO total

compensation for 2014, comprised of Salary, Bonus, Other Annual, Total Value of Restricted Stock Granted, Total Value of Stock Options Granted (using Black-Scholes), Long-Term Incentive Payouts, and All Other Total, by Compustat-Execucomp; Standard deviation of ROA is the standard deviation of annual percentage corporate return on assets for the five years from 2009 to 2013; Leverage t is the

ratio of total liabilities to total assets; Ln (Asset t) is the natural logarithm of sample firms’ total asset till

the end of the year 2014; % Equity owned by CEO is the percentage of the company's shares owned by the named executive officer; Q t is the Tobin's Q of the sample firms of 2014 which is the market

value of common stock plus the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities).

Numbers in the parentheses are T-values. **. Significant at the 0.05 level.

***. Significant at the 0.01 level.

3.3.3 Comparison between two different mediations

The prior two panels indicate the existences of mediations with respect to director supervising compensation and director excess compensation. Further research is performed to examine how much proportion the mediations are out of the total effects in different panels. Hereby the three regressions are reviewed:

𝑅𝑂𝐴𝑖,𝑡+1 = 𝛼1+ 𝑐𝐿𝑛 (𝐷𝑖𝑟𝐶𝑜𝑚𝑝𝑖,𝑡) + ∑ 𝛽𝑘𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑘,𝑖,𝑡 𝑘 + 𝜀𝑖,𝑡 𝐿𝑛 (𝐶𝐸𝑂𝐶𝑜𝑚𝑝_𝑖,𝑡) = 𝛼₂+ 𝑎𝐿𝑛 (𝐷𝑖𝑟𝐶𝑜𝑚𝑝_𝑖,𝑡) + ∑ 𝛽_𝑘𝐶𝑜𝑛𝑡𝑟𝑜𝑙_{𝑘,𝑖,𝑡} 𝑘 + 𝜀_𝑖,𝑡 𝑅𝑂𝐴_𝑖,𝑡+1 = 𝛼₃+ 𝑐′_{𝐿𝑛 (𝐷𝑖𝑟𝐶𝑜𝑚𝑝} 𝑖,𝑡) + 𝑏𝐿𝑛 (𝐶𝐸𝑂𝐶𝑜𝑚𝑝𝑖,𝑡) + ∑ 𝛽𝑘𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑘,𝑖,𝑡 𝑘 + 𝜀_𝑖,𝑡

Where, in the first equation, c measures the overall effect of director supervising compensation on firm performance; in the second equation, a reflects the effect of director supervising compensation on CEO compensation; while in the last regression, c’ is the direct effect of director pay on firm future performance and b represents the direct effect of CEO pay on firm future performance.

The absolute value of “a b/c” is cited to calculate the degree of mediation, which in this case, represents the proportion of the effect of CEO compensation on the relation between director compensations and firm performance.

Table 11

Director supervising compensation – CEO compensation -- firm

performance

Director excess compensation – CEO compensation -- firm performance a 0.488 0.462 b 2.011 3.746 c 7.391 -2.269 Mediation proportions 13% 76%

In the first column of Table 11, it shows that within the total effect of director supervising compensation 13% of it is caused by the mediation effect of CEO compensation, whereas when it relates to director excess compensation, the corresponding figure is 76%. Obviously, the mediation effect of CEO compensation on the relation between director excess compensation and firm value is overwhelmingly more significant, indicating that the harm of the collusion between directors and senior management may exceed the benefit from their due diligence. CEO’s behavior may play a more important role in the collusion at the expense of shareholder’s interests.

3.4 Additional Analyses

Since director compensation can be decomposed and has been already done, it is rational that CEO compensation has different proportions due to the different relationships between CEOs and their directors. These decomposed CEO compensations reflecting various CEO behaviors can thus be expected to have direct influences on firm performance. If the expectation is met and the influence is significant, the mediation role of CEO compensation in the link between director compensation and firm performance can be strengthened.

In this part, I try to decompose the CEO compensation according to the different derivations from the director supervising works and the collusion with directors proxied by director supervising compensation and excess compensation respectively. First, I regress CEO compensation against explanatory variables without the involvement of

director supervising compensation. Then director supervising compensation is added as an independent variable in the second regression, thus the difference with the first regression will be regarded as the proportion of CEO compensation influenced by director’s due works. Using the same method, the part of CEO compensation derived from the collusion with directors can be reached.

Further, the difference of the effects of CEO compensations derived from different behaviors on firm performances is examined by regressing the next-year ROA against different CEO compensations adding relevant explanatory variables. Column (1) in Table 12 exhibits the results of the regression on the variables without director supervising compensation which is then included in column (2). Using the method mentioned above, the CEO compensation influenced by director supervising works can be calculated. Similarly, the proportion of CEO compensation derived from the collusion with directors is then obtained based on the result presented in Table 13.

Table 12

CEO compensation derived from director due works

Dependent Variable: Ln(CEOComp t)

(1) (2) (Constant) 6.865 3.964 (55.028) *** (11.244) *** Ln(Dircomp_S t) 0.487 (8.757) *** ROA t 0.006 0.005 (2.861) *** (0.013) Leverage t -0.112 -0.077 (-1.217) (-0.869) Ln (Asset t) 0.297 0.217 (19.338) *** (12.395) ***

% Equity owned by CEO -0.005 0.003

(-1.512)* (0.864)

Q t 0.021 0.012

Standard deviation of ROA 0.014 0.014

(4.208) *** (4.334) ***

Adjusted R2 _{0.385 0.425}

F-statistics 114.542 116.002

N 1089 1089

This table shows the comparison between the regressions of CEO compensation on director supervising compensation and explanatory variables without director supervising compensation. The dependent variable is the natural logarithm of CEO total compensation for 2014, comprised of Salary, Bonus, Other Annual, Total Value of Restricted Stock Granted, Total Value of Stock Options Granted (using Black-Scholes), Long-Term Incentive Payouts, and All Other Total, by Compustat-Execucomp and the predictors are: Ln(DirComp_S t) is the natural logarithm of director supervising compensation

(cash and equity) for 2014; ROA t is the return on asset of the sample firms of 2014 defined by the

operating and non-operating income before provisions for income taxes, interest and related expenses, and minority interest (presented in percentage); Standard deviation of ROA is the standard deviation of annual percentage corporate return on assets for the five years from 2009 to 2013; Leverage t is the

ratio of total liabilities to total assets; Ln (Asset t) is the natural logarithm of sample firms’ total asset till

the end of the year 2014; % Equity owned by CEO is the percentage of the company's shares owned by the named executive officer; Q t is Tobin's Q of the sample firms of 2014 which is the market value

of common stock plus the book value of total liability, divided by the book value of total assets (it assumes that the market value of liabilities is equal to the book value of liabilities).

Numbers in the parentheses are T-values. *. Significant at the 0.1 level.

***. Significant at the 0.01 level.

Table 13

CEO compensation derived from the collusion with directors

Dependent Variable: Ln(CEOComp t)

(1) (2)

(Constant) 6.865 7.001

(55.028)*** (58.757) ***

Ln(Dircomp_E t) 0.449