I. Introduction

Executive compensation has attracted significant attention from economists and business experts since the early 1990s. News stories have questioned the fairness of executive pay practices and articles have been published about inappropriate pay practices, spurring pressure for oversight and regulation (e.g., Boyd 1994; Hall and Murphy 2003; Jensen et al. 2004). As delegates of shareholders, board members are entitled to monitor and control managers to minimize agency costs. However, boards of directors (BOD) have not been very effective, partly because they have been captured by chief executive officers (CEO) and because of the directors’ own agency problem; just as no reason exists to presume that managers automatically seek to maximize shareholder value, no reason exists to expect a priori that directors will also do so. Bebchuk and Fried (2003) refer to this phenomenon as the director–agency problem.

For instance, most outside directors have no significant vested interest in the firm and, therefore, need not be very careful when deciding on company expenditure amounts, including executive pay (Baker et al. 1988; Cyert et al. 2002; and Brenner and Schwalbach 2009). Furthermore, they may not want to hurt their personal or business relationships over stingy decisions about compensation. If outside directors are appointed owing to personal ties, they may even fear losing their board seats.

Additionally, it is noteworthy that most outside directors are executives, active or retired, at other firms in the United States and in many other advanced countries, including the United Kingdom, Australia and the Netherlands. In other words, executives from other companies make executive compensation decisions in those countries. This situation may significantly affect the executive compensation decision-making process. That is, directors’ status as executives at other firms can be another source of the director–agency problem.

Numerous governance experts describe their concerns about this additional agency problem. For example, Jensen et al. (2004) recommend that boards limit the number of CEO-directors who are outside directors but serve as CEOs for other firms, as such outside directors tend to “subconsciously (if not consciously) view the board through the eyes of a CEO.”1 However, solid empirical support of their concerns is relatively rare, except for much of the “social network” literature that examines the influence of connections between board members and executives on the level of CEO compensation (e.g., Hwang and Kim 2009; Bizjak et al. 2009; Larcker et al. 2005; and Hallock 1997).

In contrast, business practitioners often claim that this concern could be imaginary and that their policy suggestions would cause over-regulation of BOD operations. Instead, they emphasize that having other top executives as directors on the board provides the BOD with numerous advantages related to its evaluation of business strategies, voting on major operational proposals, and the scrutinizing of financial and accounting reports. Their claims are partly supported by positive stock price reactions to director appointments when the appointee is an active CEO (Fich 2005).

The empirical work in this paper provides supplemental evidence of whether, and to what extent, the concurrent executive employment of outside directors distorts pay decisions at a given company. We collected a consecutive two-year compensation dataset of the S&P 1500 firms from Compustat Execomp. Using the Board Analyst database, each company listed in the S&P 1500 was matched to its outside directors’ primary companies for which they serve (or served) as executives. We then calculated the average CEO pay of matched outside directors’ primary companies. Based on this constructed sample, we tested whether the average level of CEO pay at directors’ primary companies affects the level of CEO pay at a given company. If this finding is insignificant, we may not be able to state that the characteristics of outside directors’ primary companies directly distort the executive pay decisions at a given company.

In our analysis, the test result was statistically significant. Using the fixed effects model with a panel data analysis, which is identical to the first-difference model in the given consecutive two-year sample, we observe a strong link between CEO pay at outside directors’ primary companies and CEO pay at a given company; a $1.00 increase in CEO pay at outside directors’ primary companies leads to an approximate $0.22 increase in CEO pay at a given firm. We also observe that a 1% increase in CEO pay at outside directors’ primary companies is associated with a 0.13% increase in CEO pay at a given firm.

We obtained these test outcomes even after controlling for the size (and the entrenchment) associations between outside directors’ primary companies and each given company. Taking all of the interlocking cross-directorship observations out of the samples made little difference in the test outcomes. The observed strong link may be generated by several other sources. For instance, the positive association between CEO pay levels is well supported by a psychological phenomenon known as the anchoring effect (Tversky and Kahneman 1974), which holds that outside directors may use the CEO pay levels at their primary companies as firsthand reference points when making CEO pay decisions at a given firm. It is also plausible that the homogeneity and cohesiveness of top executives, termed the inner circle, may affect the observed positive association between CEO pay levels.

Our empirical findings are noteworthy from a policy perspective because the additional agency problem within the supposedly “independent” BOD is difficult to control; despite the fact that we establish board independence through various imposed regulations and shareholder activism, this inherent characteristic of concurrent employment among board members continues to exist in the corporate governance systems of many advanced countries, unless the government imposes direct regulations on the board structure.

The paper is organized into the following sections. Section II describes the data sources and the procedure used to construct the dataset. Section III covers the methodologies used in the analysis. Section IV presents the test results. Section V discusses the implications of the findings, and Section VI concludes the paper.

II. Dataset and Variables

In this section, we describe the data sources and the process of constructing the dataset and explain the key variables used in the analysis. The Compustat Execomp (also known as ExecuComp) and Board Analyst databases were used as sources of the board and ownership characteristics and the economic information pertaining to the selected firms. Compustat Execomp, which began in 1994, provides data on compensation for the top five executives and basic economic performance metrics, primarily for S&P 1500 companies. Board Analyst, which began in 2001, is a source of comprehensive, objective corporate governance and compensation information for more than 10,000 U.S. companies.

B. Variables

The dependent variable is CEO pay, which is the CEO’s total compensation, including their salary, bonus, restricted stock grants, stock options, and other benefits. The variable of primary interest is director company CEO pay, which represents the average CEO pay in the matched directors’ primary companies.

Table 1 (Variable Definitions) and Table 2 (Summary Statistics) describe all of the control variables used in the test. Below, we define the key control variables and explain the correlation of these variables with the dependent variable, CEO pay.

TABLE 1

VARIABLE DEFINITIONS

Table 1 presents definitions of the variables used in this paper. [CE] indicates that the data source is Compustat Execomp and [BA] indicates that the data source is Board Analyst.

Table 2

SUMMARY STATISTICS

Table 2 presents summary statistics pertaining to the variables in the constructed dataset, which is a two-year panel dataset composed of firms with a fiscal year ending between June of 2002 and May of 2004. We provide only pooled statistics.

The following eleven board and ownership characteristics were collected for the test: independent directors, chairman CEO, compensation chair appointment, director appointment by CEO, busy directors, aged directors, board size, insider ownership, institutional ownership majority, entrenchment index, and CEO tenure.

The first four variables represent the degree of board independence: independent directors, chairman CEO, compensation chair appointment, and director appointment by CEO.5 The next three variables indicate the degree of board effectiveness: busy directors, aged directors, and board size.6 Boards that are more independent and more effective are expected to be negatively correlated with the level of CEO pay (Core et al. 1999).

The next two variables, insider ownership and institutional ownership majority, represent the ownership structure. Insider ownership is the expected percentage of shares held by top management and directors. Institutional ownership majority indicates whether a majority of outstanding shares are held by institutions. Larger insider ownership levels result in fewer agency problems and are thus expected to be negatively correlated with the level of CEO pay (Jensen and Meckling 1976). Institutions are expected to participate more actively in corporate governance issues. Thus, majority ownership by institutional investors may prevent excessive CEO pay.

To reflect the overall entrenchment of the management team, we use the entrenchment index of Bebchuck et al. (2009), which is composed of four “constitutional” provisions, including a supermajority requirement for mergers and a staggered board, along with two “takeover readiness” provisions, specifically poison pills and severance agreements. Stronger entrenchment may imply a weaker governance structure and thus may result in higher CEO pay.

Finally, CEO tenure is expected to be negatively correlated with CEO pay. Jensen et al. (2004) argue that the BOD almost invariably pays “too much” for newly appointed CEOs.7

In addition to the board and ownership characteristics, the following eight performance variables were devised for the test: market cap, operating income change, operating income growth, ROE, relative performance, ROA, stock market return (three-year), and stock volatility. Among them, market cap is used to reflect the size of the company. Operating income change, which is the year-to-year percentage change in operating income, represents short-term performance. Operating income growth, which is the three-year growth rate of operating income, represents long-term performance. ROE (ROA), which is net income divided by total common equity (total assets), represents the profitability of the firm’s operations. Relative performance indicates the degree to which a company outperformed or underperformed its industry in the stock market over the five previous years. These performance variables are expected to be positively correlated with CEO pay.

Finally, to represent the macroeconomic condition in each year, the variable of the SP index yearly average is constructed, representing the average of 12 monthly averages of the S&P 500 index. To reflect the industry effect (Finkelstein and Hambrick 1995),8 we also include the two-digit SIC code, consisting of the first two digits of the four-digit SIC code.9

III. Methodologies

In this section, we present the methodologies used for the panel data analysis. Primarily, we conduct the Breusch-Pagan Lagrange Multiplier (LM) test for random effects (Breusch and Pagan 1980). The null hypothesis of the test is that the variance of the unobserved firm-specific fixed effects is zero. Because the null is rejected in the test for various possible specifications, we confirm that a simple pooled OLS estimator is not efficient. (For instance, refer to the LM test result at the bottom of Table 3 for regression 3.) Either the random (or the fixed) effects models or a pooled OLS with cluster robust standard errors would be more appropriate.

TABLE 3

EFFECTS OF DIRECTOR COMPANY CEO PAY

Table 3 presents the test results of the pooled regression with cluster robust standard errors (1–2), the random effects model (3), and the fixed effects model (4). The dependent variable is CEO pay. The key variable of interest is Director Company CEO Pay. The Breusch and Pagan LM test and Hausman’s specification tests are reported at the bottom. In each regression, the following economic variables are controlled, together with Market Cap and ROA: Operating Income Change, Operating Income Growth, ROE, Relative Performance, Stock Market Return, Stock Volatility, and SP Index Yearly Average. Standard errors are reported in parenthesis. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.

A. Regression Specification for the Hypothesis

First, we search for suitable regression specifications to examine the relationship between CEO compensation in a given company and average CEO compensation in the directors’ primary companies, presenting the empirical test outcomes for a pooled regression with cluster robust standard errors (Rogers 1993). The test outcomes are then compared with their alternatives — the random effects (RE) and the fixed effects (FE) models. We start with the following regression specification:

In this regression, CEOpay_it is the level of CEO compensation in the given company (CEO pay), DirCEOpay_it is the level of average CEO compensation in the directors’ companies (director company CEO pay), B_it is the set of the eleven board and ownership characteristic variables, and E_it denotes the set of the eight performance variables, including market cap and ROA. DI_i is an industry dummy obtained from the two-digit SIC code, which is used to control for the industry fixed effect. S_t is the S&P 500 index yearly average, which is included to reflect the macroeconomic conditions for each year.

The residuals for a given year may be correlated across different firms according to a concept known as “time effects” or “spatial correlation.” To handle time effects, year dummies (DY_t) are included in the regression. On the other hand, the residuals for a given firm may be correlated across years, a concept referred to as “firm-specific effects” or “temporal correlation.” This unobserved firm-specific effect is denoted by c_i in the given specification. Finally, u_it represents the idiosyncratic error.

As a primary estimation, we use a pooled OLS regression with cluster robust standard errors. Regression 1 in Table 3 presents the test result, which shows a significant coefficient for the variable of interest, DirCEOpay_it, where a $1.00 increase in CEO pay at the directors’ primary companies is associated with an approximate $0.12 increase in CEO pay at the given firm.

The first concern with this initial specification is that the coefficient of the variable of interest DirCEOpay_it may not capture the direct relationship between the two CEO pay variables. This concern arises out of the director selection procedure; large firms tend to hire outside directors from other large firms, and the size of director firms is strongly correlated with the level of CEO pay at the director firms. Panel 1 in Figure 1 indicates the potential seriousness of this concern: larger firms are more likely to hire directors from other larger firms. Thus, even without the strong direct effect of DirCEOpay on CEOpay, the coefficient for the variable of interest, α, could be significant in the test results. For example, DirCEOpay could simply be a proxy for the size of the directors’ companies. To control for this possible bias, we include the size variable of director firms, director company market cap, as a control variable, representing the average market capitalization of all of the matched directors’ companies.

Similarly, the coefficient for the variable of interest DirCEOpay_it may reflect the positive correlation between the entrenchment in the directors’ primary companies and that in the given company. Panel 2 in Figure 1 shows the potential seriousness of this concern: more strongly entrenched firms tend to hire outside directors from other entrenched firms. To control for this bias, we also include the entrenchment variable of director firms director company entrenchment as a control variable, representing the average entrenchment index of all of the matched directors’ companies.

FIGURE 1

CHARACTERISTICS OF DIRECTORS’ PRIMARY COMPANIES

Panel 1 presents a scatterplot of director company market cap versus market cap and its fitted line. Panel 2 presents a scatterplot of director company entrenchment versus entrenchment index and its fitted line.

Thus, we arrive at the following adjusted specification, in which DirComp_it represents a set of two additional control variables, director company market cap and director company entrenchment:

The pooled OLS regression with the adjusted specification is displayed in the results for regression 2 in Table 3. This reconfirms the significant coefficient α for the variable of interest, DirCEOpay_it, which is estimated to be 0.14.

IV. Empirical Analysis

In this section, we summarize the test results from the previously described panel data analysis. Table 3 indicates the test results for the given hypothesis using the pooled OLS (with cluster robust standard errors) and its alternative RE/FE models. In all of the regressions, the coefficient for director company CEO pay is positive and significant at the 1% level. However, the Hausman specification test demonstrates the unacceptability of the pooled OLS and random effects models against the fixed effects model. The unbiased FE(FD) estimation in regression 4 suggests that a $1.00 increase in CEO pay at the outside directors’ primary companies results in an approximate $0.22 increase in CEO pay.

As a robustness check, we also report the elasticity estimates in Table 4 using the logarithm transformations of CEO pay and director company CEO pay. The three regressions present the elasticity of the two compensation variables in the pooled OLS, RE, and FE models. (The Breusch-Pagan LM test for random effects and Hausman’s specification tests are reported at the bottom.)

TABLE 4

ROBUSTNESS CHECK (ELASTICITY)

Table 4 presents the test results of the pooled regressions with cluster robust standard errors (1), the random effects models (2), and the fixed effects models (3). The dependent variable is CEO pay (Log). The key variable of interest is Director Company CEO Pay (Log). The Breusch and Pagan LM test and Hausman’s specification tests are reported at the bottom. In each regression, the following economic variables are controlled together with Market Cap and ROA: Operating Income Change, Operating Income Growth, ROE, Relative Performance, Stock Market Return, Stock Volatility, and SP Index Yearly Average. Standard errors are reported in parenthesis. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively. [*] indicates significance at the 10% level according to a one-sided test.

The elasticities estimated using the three different regressions are found to be very close to each other, at approximately 0.12~0.13. According to the unbiased FE(FD) estimate, it is interpreted that a 1% increase in CEO pay at the outside directors’ primary companies is associated with a 0.13% increase in CEO pay at a given firm. Except for the FE model, the coefficients of the variable of interest, director company CEO pay (Log), are positive and significant at the 5% level. The coefficient of the FE model is significant at the 10% level when using a one-sided test, which is acceptable in the given analysis because we conjecture that the negative association between the two compensation variables is unreasonable.

V. Discussion

The positive association between CEO pay at outside directors’ primary companies and CEO pay at a given company may stem from either 1) the direct positive association between the size of the director firms and that of the given firm, or 2) the direct positive association between the degree of entrenchment in the director firms and that of the given firm; larger firms tend to hire outside directors from other large firms and more entrenched firms tend to hire outside directors from other entrenched firms. However, as discussed in Section III, the positive effect (observed in Tables 3 and 4) of the change in CEO pay at outside directors’ primary companies on the change in CEO pay at a given company is not easily attributable to the direct positive associations between the director firms and the given firm because we include both director company market cap and director company entrenchment as control variables in the regression analysis.

On the other hand, some may argue that the observed positive effect is attributable to cross-directorship, in which outside directors favor CEOs who serve as outside directors at their own primary companies (e.g., Hallock 1997). However, it is noteworthy that while reciprocal CEO interlocks exist occasionally, they are not very common practice in the business world as it exists today.11 In our limited sample of 556 firm entries, we observe only several interlocking relationships. Indeed, we find that removing those few entries from the sample makes little difference in the test outcomes. Therefore, without hesitation, we conclude that cross-directorship is not a major driving force behind the observed positive effect in the above analysis.

Among other possible sources that may generate the observed effect, we pay close attention to the following two phenomena: 1) the psychological phenomenon known as the anchoring effect, and 2) the societal phenomenon termed the inner circle. Tversky and Kahneman (1974) suggest that individuals make estimates by starting from an “initial value” (anchor) that is adjusted to yield the final answer. Because adjustments are typically insufficient, different starting points yield different estimates that are biased toward the initial value. According to this theory, CEO compensation in outside directors’ primary companies may serve as a firsthand reference point for directors’ decisions on CEO pay. Once their primary companies pay higher CEO compensation, they are more likely to choose a higher value for CEO pay in a given company and to accept a CEO’s request to increase his or her salary. Thus, all else being equal, CEO compensation in a given firm may reflect the level of CEO compensation at the directors’ primary companies. This anchoring effect may be a primary source of the observed sensitivity between the change in CEO pay at the directors’ primary companies and the change in CEO pay at a given company.

Top corporate executives tend to be a relatively homogeneous, cohesive collection of individuals, which Useem (1984) calls the inner circle.12 Outside directors who are top executives at other firms find it difficult to be fully objective about issues related to executive compensation. For instance, the pay consultant Watson Wyatt conducted a survey of two separate groups — directors and institutional investors — on the subject of the U.S. executive pay model.13 The survey reveals that the two groups have very different views of the current pay system; 65 percent of directors and only 22 percent of institutional investors believe that the stock-based pay model in the U.S. has contributed to superior corporate performance; 90 percent of institutional investors think that executives at most companies are overpaid, compared to only 60 percent of directors. This homogeneity of the top executives and the concurrent executive employment of outside directors may contribute to the observed positive association between CEO pay levels as far as they can stimulate implicit collusion among executives. The more cohesive the top executives (outside directors) are in a specific industrial sector (or in a specific regional market), the more closely linked CEO pay levels of the relevant companies are.

As supplementary evidence to support this inner circle argument, we checked whether a higher number of outside directors who are active CEOs in other firms results in the board granting a higher level of CEO compensation. A higher number of CEO-directors on the board may help the CEO in a given firm propose suggestions that are favorable to his interests and position; CEO-directors are more likely to appreciate the CEO’s efforts and contributions and, thus, willingly increase his salary. Let the variable ratio of CEO-directors represent the ratio of CEO-directors to the total number of outside directors on the board (c.f., Tables 1 and 2). At first glance, the scatterplot and the fitted line in Figure 2 imply a possible positive association between CEO pay and ratio of CEO-directors. Table 5 reports the estimation results of the pooled OLS (with cluster robust standard errors), the random effects model, and the fixed effects model. According to the fixed effects estimation, the unbiased estimate of the coefficient for the variable of interest is 31.5, indicating that an increase of one percentage point in the ratio of outside directors who are active CEOs in other firms results in an approximate $31,500 increase in CEO pay.14

FIGURE 2.

CEO PAY AND THE RATIO OF CEO-DIRECTORS

The figure presents a scatterplot of ratio of CEO-directors versus CEO pay with its fitted line.

TABLE 5

SUPPLEMENTARY ANALYSIS (RATIO OF CEO-DIRECTORS)

Table 5 presents the test results of the pooled regression with cluster robust standard errors (1), random effects model (2), and fixed effects model (3). The dependent variable is CEO pay. The key variable of interest is the ratio of CEO-directors. The Breusch and Pagan LM test and Hausman’s specification tests are reported at the bottom. In each regression, the following economic variables are controlled, together with Market Cap and ROA: Operating Income Change, Operating Income Growth, ROE, Relative Performance, Stock Market Return, Stock Volatility, and SP Index Yearly Average. Standard errors are reported in parenthesis. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.

In the given sample, the average number of outside directors on the board is 9.3, indicating that one additional CEO-director on the board is equivalent to a 10.8% increase in the ratio of CEO-directors. Therefore, the replacement of a non-CEO outside director with a CEO-director may result in an approximate $0.34 million increase in CEO pay, a significant amount considering that the average CEO pay in the sample is approximately $6 million. This observation supports the argument that CEO-directors tend to view the BOD (either consciously or subconsciously) through the eyes of a CEO.

There may be other valid factors influencing the observed strong link between CEO pay at outside directors’ primary companies and CEO pay at a given company. Further explorations are left for future research.15

VI. Conclusion

Since the early 1990s, many articles have been published on executive pay (Boyd et al. 2012). Some academic publications have focused on the optimal compensation contract and the pay–performance relationship when debating stock-based pay and soaring executive pay (e.g., Jensen and Murphy 1990a, 1990b; Hall and Liebman 1998; Hall and Murphy 2003; Aggarwal and Samwick 1999; Bebchuk and Fried 2003). Others have focused on the relationship between board characteristics and CEO compensation, concentrating on corporate governance weaknesses and BOD independence (e.g., Core et al. 1999; Jensen 1993). The primary concern in the literature has been connected to the agency problem caused by the management team, or what is termed a “captured” BOD.

However, the literature has not put much weight on the agency problem that may be caused purely by the BOD, including a concurrent employment issue of board members which may seriously affect their decision processes as related to the level of CEO compensation. As a complement to the “social network” literature that emphasizes the impact of cross-directorship on the level of CEO compensation, the empirical work in this paper provides a new approach to the issue of concurrent employment on a board.

The unbiased fixed effects estimation confirms a strong link between CEO pay at outside directors’ primary companies and CEO pay in a given company, even after controlling for both the size (and the entrenchment) associations between the directors’ primary companies and the given firm. Furthermore, we find that the number of cross-directorship relationships is very small in the sample here; thus, the “social network” argument does not support the observed strong link. Instead, we do not rule out the possibility that the link results from unintended psychological bias, often known as the anchoring effect. It is also plausible that the pay inefficiency originates from the homogeneous and cohesive collection of individuals (the inner circle) who serve as top executives as well as outside directors.

The degrees of psychological bias and pay inefficiency will be greater when outside directors are more negligent or less considerate. Thus, these factors cannot be excuses for the violation of the directors’ duty to care. Moreover, this may introduce improper incentives for CEOs, as they may be tempted to recruit outside directors who serve for companies paying greater CEO compensation or from their own inner circle.

This additional agency problem — caused by the BOD and not by the management team — will be difficult to control as long as concurrent executive employment on boards continues to exist. As suggested by some governance experts, including Jensen et al. (2004) and Brenner and Schwalbach (2009), this conflict may be alleviated through specific legal rules that make boards more accountable to shareholders, such as strengthening the procedural rights of shareholders who vote to elect directors.