Firm performance and corporate governance
mechanisms as determinants of CEO compensation?
A panel data analysis of the influence of firm performance and corporate governance mechanisms on CEO compensation, using Dutch data.
Author:
H.T.E. Melman
Supervisor:
Dr. T.T.T. Pham
Oktober 2009
University of Groningen Faculty of Economics and Business Business Administration, MSc Finance
A panel data analysis of the influence of firm performance and corporate governance mechanisms on CEO compensation, by using Dutch data.
Abstract
In this paper the impact of firm performance on CEO compensation is analyzed. In addition, the effect of several corporate governance mechanisms on CEO pay are examined. More specific, CEO tenure, board size, Anglo-American board membership, and CEO ownership will be linked to CEO compensation. Using panel data for 97 Dutch listed firms for the period 2004-2007, multiple regression analyses have been performed. In total 13 variables were studied, including six control variables. In most regressions, the evidence indicates that firm performance has a small positive effect on CEO pay, but in most cases not significant. Although, a few results show a negative link. These result are partly in line with prior worldwide research, however, the outcomes depict an opposite relationship compared to previous research done in the Netherlands.
In general, Anglo-American board membership, board size, and CEO tenure have the tendency to increase CEO remuneration. On the other hand, CEO ownership negatively influences executive pay.
Hinse Melman Hinsemelman@hotmail.com Student number: 1677950
JEL classification: G30; G34; J33; M52
Keywords: CEO compensation; firm performance; board size; CEO ownership; Anglo-American board membership; CEO tenure; corporate governance; the Netherlands
1. Introduction
The topic of this study is the effect of firm performance on CEO compensation and also study the impact of corporate governance mechanisms on CEO compensation in Dutch listed firms.
Executive pay always receives a lot of attention from the media. Occasionally, this discussion reaches the political arena. This especially holds when excessive CEO compensations are not tied to firm performance. A good example to illustrate this is the case of Anders Moberg, former CEO of Ahold, who received a lot of criticism from the society as a whole. Essentially, this was attributable to the announcement of a large compensation payout to Moberg. This announcement came around the same time when the company was in bad financial shape, which was largely due to one of the biggest bookkeeping scandals in Dutch history.1 More recently during the credit crisis which turned into a recession, the proposed CEO of ING, Patrick Flynn received 100.000 stocks at his installment. This action was highly criticized by the minister of finance, for the reason that ING was not in good shape due to the credit crisis. Moreover, ING even obtained government support in preventing bankruptcy.2
Although the effect of firm performance on CEO compensation has been extensively investigated by various eminent authors like Jensen and Murphy with their paper about performance pay and top-management incentives (1990), the evidence found is still not persuasive. The agency problem is one of the most prominent theory in this field of research, and suggests that there is a positive association between firm performance and CEO pay. Nevertheless, the results in the literature refute each other. Several studies including the well-known study of Jensen and Murphy, found a positive link between firm performance and CEO compensation. However, a number of studies document no relationship, or even a negative relationship between firm performance and CEO pay. Typically this type of research is performed for Anglo-American countries and in a much lesser extent for a majority of European countries. This also holds for the relationship between the corporate governance mechanisms and CEO compensation.
The objective of this study is to extent the literature and fill the gap for non-Anglo-American countries, by investigating how firm performance and several corporate governance mechanisms influence CEO remuneration in Dutch listed companies. To my knowledge, there is no previous research performed by linking CEO compensation to firm performance and some corporate governance mechanisms in the Netherlands, by using a panel data analysis. The paper of Randøy
1 http://www.volkskrant.nl/economie/article149665.ece/Nasleep_boekhoudschandaal_Ahold
2 http://www.brabantsdagblad.nl/algemeen/economie/kredietcrisis/article4688209.ece
and Nielsen (2002) analysis the company performance, the corporate governance mechanisms, and CEO compensation in a similar way. However, these authors employ a simple cross-sectional Ordinary Least Square (OLS) regression model, with a sample of Norwegian and Swedish companies. Duffhues and Kabir (2007) where the only authors that examined the pay for performance relationship in the Netherlands. They also apply a simple cross-sectional OLS regression model, but do not include any corporate governance variables. Their period of research stretched from 1998 until 2001, and failed to detect a positive pay for performance relationship for Dutch listed companies. In most regression results they found that firm performance has a negative impact on CEO remuneration. However, after their period of research some important regulations were established in the Netherlands. One of the main objectives of these regulations was to counteract unjustified excessive CEO compensations, and make CEO compensation more dependent on company performance. In addition, corporate governance mechanisms play a more dominant role. These introduced regulations were intended to influence Dutch CEO compensations. Besides extending the literature and filling up the gap for non-Anglo-American countries, it is moreover interesting to see how the corporate governance variables influence executive pay in the Netherlands, and how the pay for performance relationship has changed overtime. Two important regulations were introduced after the time scope of Duffhues and Kabir, these will be mentioned below.
Since April 2002, all Dutch listed companies are obliged to disclose the compensations of the board of directors in their annual reports. At that time regulators thought this legislation would slow down the fast uprising of larger compensations with the idea that directors would be more modest and take their own compensation more seriously into consideration. This means that each CEO should judge whether or not their obtained remuneration is acceptable. In the short run, the opposite turned out to be true. For example, managers in the same industry want to have a similar compensation in comparison to their peers. In reality this is what actually happened, the largest compensations in an industry forces up the smaller compensations, because CEOs with smaller compensation packages are guided by CEOs with larger compensation packages in the same industry.3 Since 2004, the ‘Code Tabaksblat’ has become effective in the Netherlands. One of the intentions of this code of good conduct was to oppose higher demands of compensation. The code also fortified the influence of the shareholders’ meeting. Listed firms are obliged to report in what extent they comply with the ‘Code Tabaksblat’ in their annual report. When firms do not follow this code of good conduct, they are forced to give an explanation why they do not comply.
3 http://www.intermediair.nl/artikel.jsp?id=63763
Furthermore, a committee has been established to verify if listed companies act in accordance with this corporate governance code.
Data has been collected for 97 Dutch companies, which were listed on the Amsterdam Stock Exchange, during the period 2004-2007. The effect of firm performance on CEO pay will be studied. Return on assets (hereafter: ROA), stock return, and Tobin’s Q will be used as firm performance measures. In addition, the effect of several corporate governance mechanisms on CEO compensation will be studied. More specifically (1) board size, (2) CEO ownership, (3) Anglo-American board membership, and (4) CEO tenure will be linked to CEO compensation.
This thesis is organized as follows. The following section reviews the relevant previous literature.
In the third section the data is described and the fourth section deals with the methodology. The empirical results are reported in section five. Lastly section six summarizes the main findings, and conclusion are drawn.
2. Literature review
The literature review provides an overview of the existing theoretical concepts. After that, some empirical evidence about the influence of firm performance and the corporate governance mechanisms on CEO pay will be presented. Finally, the hypotheses will be formulated.
2.1 Theoretical framework
Two main theories emerge by reviewing the literature, with respect to the impact of firm performance on CEO pay. The first recognized theory is the agency problem, also sometimes referred to as the principal-agent problem. This agency problem arises from a conflict of interest between the principal and agent. The agency theory suggest some remedies to align the interests of the principal and agent. One of the main remedies, is to pay an incentive based compensation.
This will reduce the agency costs. There are several approaches to create higher compensations, for example, constructing performance-based bonuses. Another possibility could be the altering of the salary payments on an annual basis, depending on last year’s performance. According to this theory the CEO will be more motivated by creating a more incentive based compensation, and will act in greater extent in the interests of the principal. In other words, an increase in the performance of a company should be reflected in an increase with regard to the level of incentive based compensation that a CEO receives.
Bebchuk and Fried (2003) label the pay arrangements as a (partial) remedy to the agency problem, as the “optimal contracting approach”. Under this approach, boards are supposed to design compensation packages to provide managers with efficient incentives to maximize shareholder value. An additional theory offered by Bebchuk and Fried can shed more light on the impact of firm performance on CEO pay. They entitle their new theory as the managerial power theory. It states that executive pay is not only an useful device for dealing with the agency problem, but is also an element of the agency problem itself. The authors argue, based on earlier research, that managerial power and rent extraction are likely to play a key role in the proposed compensation scheme. For example, a CEO with a strong reputation and a high demand for his unique skills can place him in a special situation with significant bargaining power. The CEO can take advantage of his special position and influence the relevant boards to acquire a larger compensation package, regardless of firm performance. However, Bebchuk and Fried acknowledge that the managerial power theory is not a complete replacement of the optimal contracting approach. Therefore, the managerial power theory should be seen as an extension to the optimal contracting approach.
The effect of different corporate governance variables on CEO compensation is the second subject of study. Several papers provide evidence that some corporate governance variables influence CEO remuneration. The corporate governance variables are closely related to the agency theory. According to the agency theory, the best way to reduce the agency costs is to make CEO pay dependent on firm performance. However, the agency theory also offers some indirect remedies to alleviate the agency costs and can reduce the CEO compensation (Randøy and Nielsen, 2002). This can be done by creating a smaller board of directors and increasing CEO ownership. The human capital theory states that any job requirements that limit the supply of CEO candidates would increase CEO pay (Oxelheim and Randøy, 2005). Within a large board of directors, the CEO needs to possess certain qualities to maximize firm value. For smaller boards these qualities are less relevant. For example, the CEO should possess excellent communication skills to reach consensus within a larger board of directors. This is especially necessary when the CEO needs to make prompt decisions that are crucial for the company. An increase in CEO ownership can reduce CEO remuneration, as well. Allen’s (1981) line of reasoning states that the CEO implements a lower compensation package as an intentional strategy. In particularly, the CEO wants to keep the stockholders satisfied, thereby not creating any conflicts. These conflicts can eventually result in possible hostile takeovers or potential proxy contests. The power theory of organizational stratification (developed by Weber, 1964), claims the opposite effect is more accurate. This theory states that power can take a variety of forms. One of these forms is to
exercise power in order to increase wealth. For illustration, a CEO that can exercise some power due to a certain amount of ownership, can raise its own compensation.
Besides corporate governance mechanisms which reduce the agency costs, opposite mechanisms that enlarge agency costs are at hand. Hence, these mechanisms can raise the CEO remuneration.
Two of these mechanisms will be studied, more specifically, Anglo-American board membership and CEO tenure will be considered. Oxelheim and Randøy (2005) argue that the corporate governance culture of a specific country plays a major role. More precise, Oxelheim and Randøy claim that independent outside Anglo-American board members bring their corporate governance culture of their home country within the company. CEOs operating in a board of directors among Anglo-American board members, can be exposed to conflicts between both corporate governance cultures. This can create new corporate governance mechanisms, influenced by both corporate governance cultures. I argue that the human capital theory can also be applied. Recalling, the human capital theory states that any job requirements which limit the supply of CEO candidates would increase CEO pay. In an environment where the CEO has to deal with two corporate governance cultures, reaching consensus can be harder. Additionally, by taking this theory in consideration one would expect that Anglo-American board membership has a positive effect on CEO compensation. Finally, by applying the power theory of organizational stratification there should be a positive relationship between CEO remuneration and CEO tenure. A corresponding remark of Weber: “If you respect someone or view him as your social superior, then he will potentially be able to exercise power over you”. I argue that a CEO who works for many years at the same listed company in the same position, will receive some form of social status. Otherwise, shareholders would have selected a new CEO. This acquired social status is a form of social power, which can be exercised in the salary negotiations to obtain a larger remuneration.
2.2 Empirical literature
The evidence for each relationship will be discussed separately. Starting with the evidence about the effect of firm performance on executive pay. Subsequently, evidence about the influence of the corporate governance variables on executive pay is presented in the same sequence as described in table 1. This table briefly summarizes the main empirical literature. As can be seen most studies are based on U.S. companies. This is also my general impression by reviewing numerous papers. This underlines the fact that not much research is done for non-Anglo- American countries. In addition, some relationships provide conflicting evidence. Therefore, it is interesting to see which relationships will hold in the Netherlands.
Table 1: Overview of the empirical literature Authors Region Time period CEO components used Results Link: firm performance & CEO pay Jensen and Murphy (1990) U.S. 1974-1986 Salary, bonus, stock options, & Positive relationship, although very small. performance related dismissals Leonard (1990) U.S. 1981-1985 Salary & bonus Find no evidence that executive pay and firm performance are related. Kato and Kubo (2003) Japan 1986-1995 Salary & bonus Firm performance has a significant positive effe CEO pay. Randøyand Nielsen (2002) Norway & Sweden 1996-1998 Salary, bonus (cash and stock), Insignificant impact of firm performance on CE stock options pay. Duffhues and Kabir (2007) Netherlands 1998-2001 Salary, bonus, pension, stock Fails to detect a positive pay-performance & options, otherrelationship. Link: board size & CEO pay Core et al. (1999) U.S. 1982-1984 Salary, bonus, some stocks Find a significant positive relationship between t & options variables. Randøyand Nielsen (2002) Norway & Sweden 1996-1998 Salary, bonus (cash and stock), Find a significant positive relationship between t stock options variables. Link: CEO ownership & CEO pay Allen (1981) U.S. 1975-1976 Direct & total compensation, Confirm a negative relationship. & aggregate income Core et al. (1999) U.S. 1982-1984 Salary, bonus, some stocks Find a negative relationship between these two & options variables. Randøyand Nielsen (2002) Norway & Sweden 1996-1998 Salary, bonus (cash and stock), Also find a significant negative relationship. stock options Link: Anglo-American board membership & CEO pay Oxelheim and Randøy (2005) Norway & Sweden 1998Salary, bonus (cash and stock), Found that Anglo-American board membership h stock option grants positive impact on CEO pay. Link: CEO tenure & CEO pay Randøyand Nielsen (2002) Norway & Sweden 1996-1998 Salary, bonus (cash and stock), Find no significant relationship between these tw stock options variables. Attaway (2000) U.S. - - Finds a positive relationship between these two variables.
Jensen and Murphy (1990) were one of the numerous authors, who investigated the effect of firm performance on CEO compensation, hereafter also referred to as the pay-performance relationship. Jensen and Murphy document a decline in the pay-performance relationship since the 1930s. Although, they discover a significant positive relationship, they state that the compensation paid for performance is relatively small. The authors conclude that the results are somewhat inconsistent with the implications of the formal agency models of optimal contracting, since incentive pay would been expected to play a more important role. Randøy and Nielsen (2002), also investigate this pay-performance relationship. They find no significant relationship between the compensation of CEOs and firm performance. An even more contradicting evidence of the principal-agency theory is from Duffhues and Kabir (2007), who included Dutch listed firms in their sample. These authors found no systematic evidence of a positive pay-performance relationship. More precise, a large number of regressions show a statistically significant negative relationship. This is partly in line with the research of Leonard (1990), who found that corporate success has no significant impact on executive pay. Hence, he revealed that a majority of the changes in company performance are unrelated to the simultaneous changes in CEO compensation. However, the results of Kato and Kubo (2003) are consistent with the principal- agency theory. In their research they found that ROA as performance-based measure has a positive and significant effect on CEO compensation. Furthermore, they include a stock market performance measure which turns out to be less sensitive, and is statistically insignificant. Almost all studies indicate that there is no strong support in favor of the pay-performance relationship.
Therefore, firm performance may not be a strong determinant of CEO remuneration. This evidence shows that the managerial power theory of Bebchuk and Fried may also hold.
Now, empirical evidence about the impact of corporate governance variables on CEO compensation will be presented, starting with board size. Yermack (1996), discovers that small board of directors are more effective in a company. Additionally, these companies demonstrate more beneficial financial ratio’s. A further interesting finding is that CEO incentives, derived from the received compensation, are higher within companies that include smaller board of directors. Furthermore, the CEOs of companies with smaller board of directors get paid less. Core et al. (1999), were one of the first who examined the effect of board size on CEO compensation.
Substantive cross-sectional association was found between the size of the board of directors and CEO remuneration. More precise, board size had a large positive effect on CEO compensation.
Core et al. collected a sample consisting of 495 observations over a three-year period for 205 listed U.S. companies, and included firms from a variety of different industries. This result was
confirmed by Randøy and Nielsen (2002), who used a sample of Norwegian and Swedish companies. These results are in accordance with the human capital theory, and seems to be a mechanism to alleviate CEO compensation. This is in line with the research of Agarwal (1981), who finds that job complexity and executives’ human capital characteristics are in large extent responsible for the variance in executive compensation.
One of the first studies that addresses the impact of CEO ownership on CEO compensation, is a study by Allen (1981). The study uses a one-year sample, with the 218 largest U.S. industrial corporations. Allen concludes that executive pay is directly related to the power of the CEO (through stocks) within the firm. Furthermore, he found a negative relationship between these two variables, that is, an increase in CEO ownership leads to a smaller CEO compensation. More recent studies frequently indicate that this negative relationship exists (see Core et al., 1999;
Randøy and Nielsen, 2002). These results support the line of reasoning formulated by Allen (1981), that lowering the compensation as a result of an increase in CEO ownership can be seen as an intentional strategy.
In addition, Anglo-American board membership can have an effect on CEO compensation. For illustration, the average CEO pay in the U.S. is much larger in comparison to its Continental European counterparts, although Europe is catching up, there are still very large differences.4 The paper of Conyon and Peck (1998), shows evidence that the board structure influences top management compensation. Oxelheim and Randøy (2005), examined the Anglo-American financial influence on CEO compensation for Norwegian and Swedish listed companies. They found that Anglo-American board membership has a positive impact on CEO pay in non-Anglo- American firms. This relationship is akin to the reasoning of Oxelheim and Randøy, and the human capital theory.
Finally, a long CEO tenure can also influence CEO remuneration. Hill and Phan (1991) recognize that a long tenure can elevate the power of the CEO. Tosi et al. (2000) performed a meta-analysis with respect to CEO compensation and could not explain a large variance. Tosi et al. argued that a part of this unexplained variance could be explained with CEO tenure. Up to this date, this relationship has not been frequently researched and shows some mixed results. Randøy and Nielsen (2002), assume a negative relationship. They argue that social pressure of stakeholder groups will reduce CEO power or will remain unchanged with a longer CEO tenure. The authors
4 http://www.usatoday.com/money/companies/management/2008-06-29-europe-ceo-pay_N.htm
found no significant evidence in favor of their argument. In contrast, Attaway (2000) found a positive relationship. This can imply that the power theory of organizational stratification is more appropriate. But a marginal note has to be placed within this discussion: extra social power is not necessary a bad thing. It could very well be that the CEO might create a large increase in firm profitability due to more experience.
2.3 Hypotheses
As mentioned before, the research objective of this paper is to extent the literature and fill the gap for non-Anglo-American countries, by investigating how firm performance and several corporate governance mechanisms influence CEO remuneration in Dutch listed companies.
I follow the agency theory that states that the agency costs can be reduced by making CEO compensation reliant on firm performance. In this way causality is running in only one direction.
So, I investigate the effect of firm performance on executive pay and not the other way around.
This is a well accepted way in the literature of testing this specific relationship (see Jensen and Murphy, 1990; Kato and Kubo, 2003; Duffhues and Kabir, 2007). I assume a positive relationship between financial performance and CEO compensation, in Dutch companies that are listed on the Amsterdam Stock Exchange. The results in the literature are mixed. As mentioned before Duffhues and Kabir (2007) use Dutch listed companies, which show no evidence in favor of the optimal contracting approach. But, they collected data for the years 1998-2001. I collected data after these periods, in which legislation and rules (for example the ‘Code Tabaksblat’) were introduced concerning CEO compensation and corporate governance mechanisms.
To my knowledge, there is no previous research performed regarding the impact of board size and CEO ownership on CEO compensation in the Netherlands. Similar to other papers, I argue that a positive relationship exists between board size and CEO remuneration in Dutch companies that are listed on the Amsterdam Stock Exchange. This relationship is in line with the human capital theory. Furthermore, I state that a negative relationship between CEO ownership and CEO compensation exists in Dutch companies, that are listed on the Amsterdam Stock Exchange. This link is the opposite of what the power theory of organizational stratification suggests.
Nevertheless, this stated relationship is in consensus with the results found in the literature.
Finally, I argue that Anglo-American board membership and CEO tenure are contributing factors to an increase in CEO compensation. More precisely, there is a positive relationship present
between Anglo-American board membership and CEO compensationin Dutch companies that are listed on the Amsterdam Stock Exchange. In addition, I proclaim that a positive relationship exists between CEO tenure and CEO compensation in Dutch companies that are listed on the Amsterdam Stock Exchange. This relationship is in line with the power theory of organizational stratification. So, I argue that all five variables (firm performance and corporate governance mechanisms) can have an impact on CEO remuneration.
3. Data
3.1 Data
The database used in this research consists of 97 Dutch companies, which are listed on the Amsterdam Stock Exchange. The database consist of yearly observations for the periods 2003- 2007. All industries are included in the sample, except for utility companies. This is because, there are no utility companies listed on the stock exchange in the Netherlands during the period of research. I use panel data, for the reason that a simple cross-sectional data set would limit the amount of data points.
The majority of the data is extracted from the database REACH. REACH is a modular database which consists of financial information about Dutch companies and institutions. A large quantity of data, particularly information concerning the independent variables and control variables, are derived from this database. The website www.bestuursvoorzitter.nl is used to extract information about the dependent variable, that is CEO compensation. The total compensation package is divided into five components, which are also extracted from this website. The website displays information about the received compensation packages of CEOs in Dutch listed firms, and these packages are divided in several components. All this information is published by the Dutch Investors Association. The independent variable, stock performance, is collected from the website www.iex.nl. This website possesses information about the stock performance of each individual stock. Stock performance is the stock price at years-end minus the stock price at the beginning of the year divided by the stock price at the beginning of the year, expressed as a percentage.
Almost all companies have missing data. Some CEO compensations and information about certain control variables were missing. This information was manually collected by using annual reports. This also applies for the independent variables, CEO ownership, Anglo-American board membership, and board size.
The database REACH showed 111 Dutch companies that were listed in the Netherlands during the period 2003-2007. Due to the lack of financial information, namely missing annual reports or
required information which was not disclosed, 13 companies were excluded from the sample. One company was excluded for the reason that it showed a lot of extreme values on various measures.
In five cases, firms were established during the period of research. I included these new listed companies to create more data points. So, all available information of these firms was included.
Finally, I end up with a sample of 97 companies and 376 company-year observations.
3.2 Definition of the variables
CEO compensation is the dependent variable in this study, which is defined as the total compensation paid to the CEO in a fiscal year at time t. Severance pay is the only major component that is excluded. This payment is usually based on several years, and this payment can be out of proportion in comparison to the previous years, which can make the relationship less reliable. In other words, total CEO compensation is the sum of the salary component, bonus payment, pension contribution, shares & options, and other payments. The shares and options are valued at the date of issuance. The options are valued with the help of the Black and Scholes option-model. Finally, other income is the component that could not be assigned to the four categories mentioned before. Usually this are expenses made by the CEO that reasonably belong to the position the CEO serves in his company. Take for instance the CEO of Unilever, his other compensation consisted among other things of a car remuneration, entertaining allowance, medical insurance, and private use of chauffeur driven cars. All these different components of CEO compensation, and all other variables are described in greater detail in appendix A, table A1.
Should the independent variables be lagged one year with respect to the dependent variable? It could be the case that a fraction of the bonus, stock, and option payment is determined based on certain performance objectives during the same year in which the performance is determined. In other words, this fraction of the received compensation package can depend on the performance in the same year. In addition, this can also hold for the corporate governance variables. Some authors followed this approach (Randøy and Nielsen, 2002; Duffhues and Kabir, 2007). However, it can also be argued that the salary payment and partially the bonus, stock, and option components are determined on past performance. One of the main reasons is that these parts are not constantly being assessed. This means that these CEO remuneration components are not judged each week or every month whether or not they are acceptable, but more on a yearly basis.
Besides the firm performance variables, the other corporate governance independent variables should also be lagged (Oxelheim and Randøy, 2005). This implies that these corporate governance variables do not influence CEO compensation instantly. According to Kerr and Bettis
(1987), salary payments distributed over year t are based on the board perceptions of firm performance in the previous year, also known as t-1. These authors state that if you do not account for the lag with respect to salary, the relationship between company performance and CEO compensation can be lost or out of sight. In addition, bonus payments, granted stocks and options that are received in one year can be based on long-term goals, and in this way partially depend on past performance. Therefore, testing will be done by lagging the independent variables with respect to the dependent variable, and a second test by including no lags. In view of the fact that the arguments for including a lag or excluding a lag seem both plausible, and both ways are frequently done in the literature.
I use three different performance measures. The first measure is an accounting-based measure, namely ROA. The second measure applied, is a market-based measure (stock performance). The third performance measure is Tobin’s Q, which is a mix of an accounting-based and market-based measure. These three different measures of firm performance are frequently used in the literature (see Mehran, 1995; Randøy and Nielsen, 2002). The second explanatory variable is board size.
The third independent variable is CEO ownership. CEO ownership is a dummy variable and will be given a one if a CEO holds more then five percent of the stocks in the company, and a zero otherwise. The five percent is a widespread selected threshold level. Behind this level, the CEO may have extra influential power to control the decision making process. In the period of research, it was compulsory in the Netherlands to cite the name of the firm or individual who owned more than five percent of the stocks in the firm. I recognize that in some countries slightly less or even more stock ownership is needed in order to influence identical decisions. However, I select this threshold level for the reason that it is often applied in the literature (see Allen, 1981; Randøy and Nielsen, 2002). Anglo-American board membership is the fourth independent variable. This is a dummy variable and receives a one if an Anglo-American board member is present in the board of directors or in the supervisory board and receives a zero otherwise. Finally, the last explanatory variable is CEO tenure.
With the inclusion of the following control variables I control for general firm characteristics and industry-specific factors. This may influence the outcomes of the regression analysis, and therefore, prevent highly biased results. The first control variable included is debt ratio. Debt ratio can serve as a disciplining device for a company to avoid paying high CEO compensations.
Because a high leverage ratio generates less free cash flows, that eventually will lead to lower compensations (Jensen, 1989). Therefore, I expect a higher debt ratio implies a lower
compensation. Considering debt ratio as a control variable with respect to CEO pay is often seen in the literature (see Randøy and Nielsen, 2002; Duffhues and Kabir, 2007). Market capitalization, sales, and total number of employees are the measures for firm size, which are used as second control variable. Kostiuk (1990) studies the relationship between firm size and executive compensation in great depth. He founds that the relationship between these two variables are relatively stable over time and are present in multiple countries. Another remarkable research is from Tosi et al. (2000), who included 16 firm size measures and 30 performance measures. They discovered that firm size accounts for more than 40 percent of the variance in CEO remuneration, and firm performance contributes for less than five percent of the variance in total CEO pay. I expect that on average CEOs in larger firms receive larger compensations in comparison to smaller firms. This positive relationship is confirmed by a large amount of empirical papers (see Kostiuk, 1990; Zhou, 2000). The third control variable is firm age. Firm age is occasionally included as control variable in this type of research (see Randøy and Nielsen, 2002; Oxelheim and Randøy, 2005). The expectation is that older firms pay more to their CEO, which corresponds to the work of Brown and Medoff (2003). They point out four arguments why older firms pay more. Finally, I include an industry dummy which will take certain specific industry characteristics into account. The companies are divided into nine industries, classified according to the Euronext website. These different industries are listed in panel A of table A3 (appendix B).
Table 2 provides an overview of the expected signs of the key variables.
Table 2: Overview of expected sigs key variables
Variable Expectation Source Independent variables
Firm performance Positive Jensen and Murphy (1990), Kato and Kubo (2003) Board size Positive Core et al. (1999),
Randøy & Nielsen (2002) CEO ownership Negative Allen (1981), Core et al. (1999),
Randøy and Nielsen (2002) Anglo-American board membership Positive Oxelheim and Randøy (2005)
CEO tenure Positive Attaway (2000) Control variables
Debt ratio Negative Jensen (1989)
Firm size Positive Kostiuk (1990), Zhou (2000) Firm age Positive Medoff (2003)
3.3 Summary statistics
In this part some general figures are presented. Table 3 presents the general statistics concerning the dependent variable. Panel A shows that in the entire sample period the mean compensation amounts to € 1,12 million. However, the median of this sample is only € 0,61 million. This means that many of CEOs get paid less than one million Euros, and some companies pay their CEO well over one million Euros. Another notable fact is the increase of CEO compensations each year, even after adjusting for inflation (see panel B). However, an analysis of variance (ANOVA) test is conducted to determine if the means significantly differ on an annual basis. The result indicates that the means are not statistically different. The largest compensation paid in this sample is € 11,51 million. This payment was collected by Davis in 2004, former CEO of Reed Elsevier N.V.
A negative obligation of € -0,10 million had to be paid by the CEO of Nedap N.V. in 2005. This had to do with his retirement scheme. The distribution is skewed and displays a leptokurtic distribution, which is commonly seen in financial time series. Therefore, the natural logarithm is taken to improve the normal distribution. Two observations drop out, because the computation of the natural logarithm cannot be done for negative values or values that amount to zero.
Table 3: Total CEO compensation, excluding severance payment (2004-2007) Panel A
2004 2005 2006 2007 2004-2007 Mean 964.975 1.050.510 1.158.042 1.306.857 1.122.596 Median 528.394 603.000 647.500 672.400 610.342
Maximum 11.511.082 6.834.508 7.439.910 8.545.000 11.511.082 Minimum 30.000 -104.000 76.000 90.000 -104.000 Std. Dev. 1.468.281 1.316.979 1.476.198 1.725.086 1.505.794
Skewness 5 2 3 2 3 Kurtosis 31 8 10 9 14 Observations 92 93 94 97 376
Source: Dutch Investors Association
Panel B
2004 2005 2006 2007 Mean 993.156 1.068.369 1.158.042 1.286.277 Median 543.825 613.251 647.500 661.811
The mean and median figures in this table are the same as in panel A, however they are adjusted for inflation. The base year is 2006. Inflation percentages of the Netherlands are derived from the website of the Central Bureau for Statistics (CBS).
In table A2 (appendix B) some general descriptive statistics are presented regarding the five different components of total CEO compensation. On average, the largest component of the CEO compensation package consists of shares and options. However, the median of the fixed remuneration (salary) is higher in comparison to the compensation received in shares and options.
This indicates that many executives receive less compensation in the form of shares and options, and more as a fixed payment. All CEOs receive a fixed remuneration, almost every CEO receives a payment in the form of a variable compensation (bonus) and a contribution to his pension plan.
Compensation consisting of shares and options is allocated in slightly less than half of the observations, other income is assigned in only one-fourth of all observations. Once more, the natural logarithm of these different remuneration components are taken. Since these variables possess high levels of Kurtosis and do not demonstrate a symmetric distribution around the mean.
Table A3 (appendix B) is divided into several panels. These panels show how CEO compensation is distributed across different control variables and independent variables. For example, panel A illustrates the distribution of companies across nine industries. This panel shows the number of companies per industry, and reports the mean and median values regarding the compensation in a specific industry. One side note should be placed, the mean and median compensations may not represent a reliable estimate for a given industry. While some industries contain a few large companies or observations, others contain many small companies. The other panels are to a great extent in line with the anticipated relationships, except for debt ratio and CEO tenure that exhibit a reverse link. Not much value should be attached to these results. Since, these relationships are not being controlled for specific effects, like firm size.
Table A4 is included in appendix B and shows some general statistics about the independent variables and control variables over the whole sample period. More ANOVA tests are performed to find out if the means of the firm performance variables differ on an annual basis. The outcomes for all firm performance variables imply that the means differ each year, more than would be expected by chance alone. Table 4 reports the correlations between all variables used in this research, except for time and industry dummies. The correlations between the independent variables suggest no strong presence of multicollinearity, with the largest correlation of 0,46 between board size and Anglo-American board membership. However, the firm size measures are highly correlated with the dependent variable. This is in line with other research (see Boyd, 1994;
Oxelheim and Randøy, 2005).
Table 4: Correlation matrix 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. This table shows the correlations between the dependent variable, independent variables, and the control variables. The first variable is the dependent variab the six following variables are the independent variables, and the last five variables are control variables. Correlations above the 0,70 are expressed in bold.
1. LN(Total compensation) 1 2. Performance 1 (ROA) 0,191 3. Performance 2 (Stock) -0,010,181 4. Performance 3 (Tobin’s Q) 0,190,140,071 5. Board size 0,640,07-0,120,161 6. CEO ownership -0,18-0,16-0,050,05-0,081 7. CEO tenure -0,060,090,08-0,03-0,080,201 8. Anglo-Am. Board mem. 0,530,01-0,020,160,460,03-0,091 9. Debt ratio -0,05-0,18-0,13-0,040,11-0,01-0,030,031 10. LN(Employees) 0,820,160,020,150,63-0,16-0,030,360,111 11. LN(Sales) 0,860,22-0,030,120,66-0,17-0,020,420,100,951 12. LN(Market cap.) 0,910,25-0,010,210,69-0,17-0,110,510,010,850,901 13. LN(Firm age) 0,170,08-0,08-0,120,02-0,070,07-0,05-0,060,190,220,071
4. Methodology
4.1 Model specification
The two equations used to test the formulated hypotheses are related to the equation applied by Randøy and Nielsen (2002), with respect to the variables. In addition, they attempt to investigate a similar topic. Therefore, their model is suitable for testing my hypotheses. In equation 1 the independent variables are lagged with respect to the dependent variables, and in equation 2 there are no lags included. I am aware of the fact that endogeneity problems may exist by including no lags. However, this way of testing is frequently done in the literature (see Duffhues and Kabir, 2007; Randøy and Nielsen, 2002; Oxelheim and Randøy, 2005). In addition, CEO remuneration partly consists of a variable component which is often based on short-term and long-term objectives (according to the ‘Code Tabaksblat’). I use a model that captures the short-term pay- performance effect, because I include contemporaneous or one-year lag variables. As a result of this, the total pay-performance effect may not be grasped. Due to the lack of data and controlling for long-term objectives that can run up to 30 years or more, the amount of observations are reduced by including more lags, and only seizes a part of the long-term objectives. Hence, just the contemporaneous and the one-year lag variables will be included.
The main idea of the used equations is essentially based on the model of Randøy and Nielsen, with some small adjustments. Since, I apply a panel OLS regression method I add some time dummies. These time dummies (δ) are included to control for time specific shocks. The time dummies make it possible to see how the key variables change over time. These dummies have the same value for cross-sectional units at a certain time, but possess different values for different cross-sections. Another important distinction between the model of Randøy and Nielsen and model 1, are the lagged independent variables. Randøy and Nielsen only examined the contemporaneous impact of the explanatory variables on CEO remuneration and did not lag the explanatory variables with respect to CEO remuneration.
In addition, the same equations will be used for a more in depth investigation with regard to the different CEO components. This can yield additional insights with respect to the regression estimates on total CEO compensation and can strengthen the results. The equations remain identical with the exception of total CEO compensation as dependent variable, which is replaced by the different CEO components. The regressions are performed with the help of the following two equations:
i t j
i t
i t i
t i
t
i t i
t
i t i
t i
t
controls Industry
age Firm Ln
size Firm Ln ratio
Debt tenure
CEO
membership board
American Anglo
ownership CEO
Boardsize e
performanc Firm
on Compensati Ln
, ,
, ,
1 ,
1 , 1
,
1 , 1
, ,
_ )
_ (
* 9
) _ (
* 8 _
* 7
_ 6
_
_ _
* 5
_
* 4
* 3 _
* 2 07 06 05 0 ) (
ε
λα
α α
α
α α
α α
δ δ δ α
+ +
+ +
+
+ +
+
+ +
+ + +
=
−
−
−
−
−
(1)
i t j
i t
i t i
t i
t
i t i
t
i t i
t i
t
controls Industry
age Firm Ln
size Firm Ln ratio
Debt tenure
CEO
membership board
American Anglo
ownership CEO
Boardsize e
performanc Firm
on Compensati Ln
, ,
, ,
,
, ,
, ,
,
_ )
_ (
* 9
) _ (
* 8 _
* 7 _ 6
_
_ _
* 5 _
* 4
* 3 _
* 2 07 06 05 0 ) (
ε
λα
α α
α
α α
α α
δ δ δ α
+ +
+ +
+
+ +
+
+ +
+ + +
=
(2)
The merger of these two equations make it possible to compare the lagged variables with the contemporaneous variables or vice versa. However, I choose to keep the equations separated.
There are several arguments for doing this. Obviously high correlations exists between the independent variables by including lagged as well as contemporaneous variables. This multicollinearity problem may not provide valid results. Furthermore, the amount of observations will be reduced, causing less data points. Hence, the regression estimates can become less reliable.
This way of testing is in line with the research of Duffhues and Kabir (2007), who also use separate equations.
The accurate definitions of the variables used in both equations are provided in table A1 (see Appendix A). This table is divided into 3 types of variables, (1) the dependent variables, (2) the independent variables, (3) and the control variables. For three different variables in both regression equations, the natural logarithm is taken. First of all the natural logarithm of compensation is taken as mentioned before. This is common in the literature about CEO remuneration (see Hall and Liebman, 1998; Conyon and Peck, 1998). In addition, the natural logarithm is taken for all firm size measures. These measures are market capitalization, sales, and total number of employees. Finally, the natural logarithm is taken of firm age. This is in line with the paper of Oxelheim and Randøy (2005). These measures are usually heavily skewed and possess large values of kurtosis. This improves the normal distribution. This is necessary, because one of the underlying assumptions of OLS is that the disturbances are normally distributed. By using the Bera-Jargue normality test, I conclude that the residuals are normally distributed in both specifications, by including only one firm size measure. But, the null hypothesis of normality is rejected by including all firm size measures.
4.2 Estimation strategy
To begin with, a simple OLS estimation will be performed by arranging the data in a pooled work
file. Due to the high correlation between several firm size measures, and the high correlation between the explanatory variable board size and the firm size measures, an extra test will be carried out. In this way the present multicollinearity can be reduced. In addition, the White cross- section standard errors are used to allow for general contemporaneous correlation between the firm residuals. This makes the regression robust for possible covariance’s across cross-sections.
After this, the data will be organized as a panel work file. Subsequently, a panel OLS method will be applied. Advantages of panel data over cross-sectional and time-series data are the increase in data points that raises the capacity to capture greater complexity (e.g. to construct and test more complicated behavioral hypotheses) (Hsiao, 1986). Furthermore, panel data typically contain more degrees of freedom and enjoys more sample variability. This leads to an efficiency improvement of the econometric estimates (Hsiao et al., 1995). Kato and Kubo (2003) who investigate firm performance and CEO compensation in Japan, also applied this method with a 10-year panel database. I include a fixed or random effects model. The fixed effects model controls for the unobserved effect, also known as unobserved heterogeneity. In this way, time independent effects are imposed by including an additional variable that captures all the unobserved time-constant factors that are correlated with the independent variables. The random effects model is also usually capable to control for the unobserved effect, but is more appropriate when the unobserved effects are uncorrelated with each explanatory variable in all time periods (Wooldridge, 2000a). A Hausman specification test will be applied to determine which effects model should be used.
5. Results
In this part the OLS regression results will be presented. First, OLS estimations will be performed by arranging the data in a pooled work file. Subsequently, a panel OLS regression will be carried out by classifying the data in a panel database. Then some tests will be performed by breaking down total CEO compensation into several components. Finally, I will perform some robustness checks to strengthen my analysis. The regression outcomes will be presented in tables 5-8 and are briefly described.
Table 5 shows pooled regression estimates with total CEO compensation as dependent variable.
In models 1-3 the independent variables are lagged with respect to the dependent variable. In models 4-6, the explanatory variables are not lagged with respect to total CEO compensation. In all these models White cross-section standard errors are used to allow for general
contemporaneous correlation between the firm residuals. In each separate model only one firm performance measure is included, to meet the normality assumption requirements. In addition, by including all different firm size measures the high correlations between these measures can bias the regression estimates. Therefore, I include the natural logarithm of the total number of employees as firm size measure. This specific measure is selected, because it exhibits the lowest correlation with respect to board size. In this way the specification bias can be minimized.
However, the results do not change much by including one of the other size measures. For that reason these tests are not stated in this paper. The correlation between board size and the natural logarithm of the total number of employees is still high, as well as the correlation between board size and Anglo-American board membership. As a result an extra robustness check is carried out.
This check excludes board size for each model in order to observe the effect of this variable exclusion. This robustness check is not mentioned in this paper, because the results appear in great extent to be similar to the results presented in table 5. Therefore, there is no reason to assume that the inclusion of board size will bias the regression results.
The first thing to notice looking at table 5, is that all performance measures show positive signs.
Only Tobin’s Q as performance measure is significant by including lagged independent variables.
These positive results match the stated hypothesis for a pay-performance relationship, and can be seen as evidence in favor of the agency theory. This is also in line with the paper of Jensen and Murphy (1990) who found a small positive significant relationship. Randøy and Nielsen (2002) found a lot of insignificant results. This is also consistent with my outcomes.
Board size is positively related to CEO compensation and significant in all specified models.
Backing up the view that a CEO within a larger board of directors obtains a larger compensation.
CEO ownership is highly significant in all regression estimates, although it somewhat less significant by including no lagged explanatory variables. This negative link corresponds to a voluminous body of empirical research. Moreover, it could be that the reasoning of Allen (1981) also applies to the Netherlands. This means that a Dutch CEO implements a lower compensation as an intentional strategy. As expected, Anglo-American board membership has a large significant impact on CEO remuneration in all regression results. The last explanatory variable is CEO tenure, this variable shows a mixed representation of the impact on CEO pay. Model 1 and 2 show a negative link, and the other models show a positive link. These positive links are in accordance with the stated hypothesis. These negative links and small coefficients can indicate that the reasoning of Randøy and Nielsen (2002) can be more appropriate. Who state that CEO remuneration will be reduced or unchanged due to social pressure of stakeholders groups. It
should be mentioned that all these links are insignificant, and the magnitudes of the coefficients are very small. This can indicate that CEO tenure does not play a large role in setting up compensations for Dutch CEOs. In general, the estimated coefficients of all variables in the lagged and contemporaneous models have roughly the same size and significance levels.
The average R-squared indicates that the regressed models are capable to explain approximately 83 percent of the variations in the dependent variable. Although the R-squared is slightly higher with the inclusion of the lagged explanatory variables. Furthermore, all control variables included in the models show the anticipated signs and are generally highly significant.
Table 5: Pooled regression estimates with White’s cross-section coefficient covariance method Lagged relationship Contemporaneous relationship
Model: 1 2 3 4 5 6 Variable Coefficient Coefficient Coefficient Coefficient Coefficient Coefficient
(t-Statistic) (t-Statistic) (t-Statistic) (t-Statistic) (t-Statistic) (t-Statistic)
Constant 4,603 4,595 4,586 4,654 4,650 4,654
(89,86)*** (88,11)*** (86,35)*** (88,75)*** (87,32)*** (84,60)***
ROA 0,038 0,030
(0,65) (0,33)
Stock return 0,018 0,005
(0,78) (0,19)
Tobin’s Q 0,009 0,002
(2,35)** (0,60)
Board size 0,016 0,017 0,017 0,019 0,019 0,020
(1,98)** (2,14)** (2,08)** (2,35)** (2,37)** (2,44)**
CEO ownership -0,093 -0,094 -0,101 -0,082 -0,084 -0,086
(-2,91)*** (-2,93)*** (-3,15)*** (-2,59)** (-2,57)** (-2,65)***
Anglo-American 0,195 0,194 0,189 0,186 0,185 0,183 board membership (6,84)*** (6,77)*** (6,64)*** (6,00)*** (5,97)*** (5,87)***
CEO tenure -0,000 -0,000 0,000 0,001 0,001 0,001
(-0,06) (-0,08) (0,00) (0,57) (0,62) (0,59)
Control variables
Debt ratio -0,089 -0,094 -0,094 -0,101 -0,102 -0,102 (-6,89)*** (-7,49)*** (-8,39)*** (-7,72)*** (-7,08)*** (-7,54)***
LN(Employees) 0,307 0,306 0,304 0,288 0,289 0,287
(21,73)*** (21,38)*** (21,24)*** (17,89)*** (18,13)*** (17,91)***
LN(Firm age) 0,049 0,050 0,052 0,053 0,054 0,052 (1,93)* (2,02)** (2,02)** (2,10)** (2,14)** (2,05)**
R-squared 0,83 0,84 0,84 0,82 0,82 0,82
N 372 372 371 362 362 361
The table presents regression results where the estimation method is ordinary least squares. The dependent variable is the natural logarithm of the total CEO compensation. Firm performance is measured by ROA, stock return, and Tobin’s Q. Firm size is measured by taking the natural logarithm of the total number of employees. Each regression model controls for time and industry dummies, but these are not reported for the sake of brevity. The t-statistics are reported in parentheses. *** indicates significance at the 1%, ** significance at 5%, and * significance at 10% level.
The panel least square regression outcomes with fixed effects are presented in table 6. Based on the Hausman test the random effects model is rejected. Hence, the fixed effects model is selected.
The fixed effects model controls for the unobserved heterogeneity. Again, firm size is measured by taking the natural logarithm of the number of employees, in order to diminish the specification bias. The natural logarithm of the number of employees is replaced by other size measures and board size is excluded, to test for the robustness of the results. Again, no large variation is found between the different regression estimates. By comparing the pooled results with the panel results the positive pay-performance relationship still exists. In this case the lagged stock return variable has a significant positive impact on CEO pay. However ROA (model 1 and 4) and Tobin’s Q (model 6) do not indicate a positive link with regard to CEO pay. Hence, this contradicts the stated pay for performance hypothesis. Although, these negatively related variables have on average small coefficients and are far from significant. A possible explanation for the different signs with regard to the firm performance measures is how these variables are measured. ROA and partially Tobin’s Q are accounting-based measures. Accounting-based measures reflect firm performance over the past financial year, and can reflect past performance over a longer period of time. Market-based performance measures reflect the present value of the future flows of income.5 These firm performance measures that have a negative effect on CEO pay are not uncommon in the literature. Duffhues and Kabir (2007) discovered a few positive relationships, nevertheless, in many cases their regression results show that firm performance has a negative impact on CEO remuneration. These inconsistent results are a possible indication in favor of the managerial power theory offered by Bebchuk and Fried (2003). They state that managerial power and rent extraction are expected to play a key role in the proposed compensation package.
The panel estimations for board size and CEO tenure report roughly similar results as presented in table 5. Board size maintains the anticipated sign, but lost its significance. On the other hand, CEO tenure has a positive influence on CEO pay in each regression estimate, and even becomes significant by the inclusion of no lagged explanatory variables. This suggests that an increase in social power due to a longer CEO tenure will lead to a higher CEO remuneration. In addition, Anglo-American board membership and CEO ownership show the same expected sign as in table 5, with the inclusion of the lagged independent variables. Only the significance level of Anglo- American board membership disappeared. But the signs of these two variables are the opposite when looking at the models that include contemporaneous variables. Nevertheless, these particular estimates are far from significant. Still it can be questioned why these results refute
5 More detailed information about the performance measures and the differences between these measures can be found in the book: integrated performance management: a guide to strategy implementation.
