Testing Tournament Theory in the Netherlands”

“

Testing Tournament Theory in the Netherlands”

Peter Slooff

University of Groningen

Faculty of Economics and Business

H.Colleniusstraat 15

9718 KR Groningen, The Netherlands

peet_slooff@hotmail.com

Study: International Economics & Business

Course: Master Thesis

Contents

ABSTRACT……….3

INTRODUCTION………4

LITERATURE REVIEW AND HYPOTHESES……….7

DATA AND METHODS……….19

CONCLUSION………32

REFERENCES……….35

1. Abstract

For the first time the theory of tournaments was tested in the Netherlands. A panel dataset on compensation levels of the executive board members of 100 Euronext-Amsterdam-listed firms over the years 2004 to 2007 was analyzed to determine the amount of consistency with the tournament theory. Two hypotheses were tested. First, according to theory, there should exist a positive relationship between the number of executives on the board and the wage premium associated with a move up to the top position in the hierarchy. The results obtained lead to the rejection of the hypothesis. Secondly, tournament theory suggests that there exists a significant relationship between the wage premium and the performance of a firm. Again, no significant results were found, thus it appears that Euronext-Amsterdam-listed firms do not implement tournament compensation schemes at the executive level.

1. Introduction

The compensation of executives is a topic that has been studied since the times of economists such as Adam Smith (Smith, 1776). In the last three decades, three models concerning executive compensation seem to be the most prevalent in the literature concerning this matter. The first model: the principal-agent theory/agency theory

describes the problems resulting from the fact that the principal, or shareholders of a firm, delegates duties to an agent, the chief executive officer (CEO), which leads to differences in amounts of risks endured. As a result, a CEO might behave rather egotistically and aim for self-enrichment instead of the maximization of the shareholders’ value. In order to prevent this and align the principal’s and agent’s incentives, compensation can be tied to corporate performance. (e.g. Jensen and Meckling, 1976).

The second model, dubbed ‘the managerial power approach’, contends that pay will be higher and/or less sensitive to performance in firms where managers have relatively more power. Managers are believed to be directed towards extracting maximum possible rent without creating an outrage. Furthermore, executives can exert more influence over the size of the firm than over the firm’s performance, as a result the executives at hand would rather see their pay to be based on size instead of performance. To study the empirical consistency of this model; the link between firm size and executive compensation is investigated. (Tosi et al, 2000).

The third model and topic of this research paper is the ‘tournament compensation scheme’. In the seminal paper ‘Rank-Order Tournaments as Optimum Labor Contracts’, which was published in 1981 and written by Lazear and Rosen, the tournament model was introduced and it pointed the focus of models designed to analyze executive compensation schemes in a new direction.

The authors introduced a rank-order payment scheme as a tool of analysis, which was not used before but which ‘seemed to be prevalent in many labor contracts’. For example, assistant professors are competing for a number of limited tenured positions, or vice-presidents are competing with each other for the top position as CEO.

Since the appearance of the Lazear and Rosen article, many theoretical adjustments have been made (Nalebuff and Stiglitz, 1983; Green and Stokey, 1983; Dye 1984, O’Keeffe et al., 1984; Rosen, 1986; Lazear 1989;; Krakel and Sliwka 2001; Hvid 2002) and numerous empirical investigations were conducted (Bull et al., 1987; O’Reilly et al., 1988;

Bognanno and Ehrenberg, 1988; Weigelt and Dukerich, 1989; Capelli and Cascio, 1991; Becker and Huselid, 1992; Lambert et al. 1993; Main et al., 1993; Knoeber and Thurman, 1994; Brown et al., 1996; Bloom, 1999; Eriksson,1999; Conyon et al., 2001; Sunde 2002; Ohrisson et al., 2004; Ivankovich, 2007). Numerous predictions made by the tournament theory were investigated. For example, the tournament theory predicts that there exists a positive relationship between the number of executives on a board and the wage gap between a CEO’s salary and the average salary of other executives.

The vast majority of empirical studies confirmed the propositions emanating from tournament theory after evaluating data from sports tournaments to data from firms in different international contexts.

However, as one author wrote: ‘empirically addressing the relevance of tournament theory and using enlarged data sets doing so, should rank high on the research agenda of economists interested in compensation issues’. More studies are needed, preferably in countries with different governance structures than in the US. An English and Danish study paved the way by finding results that were consistent with the tournament theory, thereby validating the theory outside the US (Conyon et al., 2001; Eriksson, 1999). Especially the Danish study is vital, due to the resemblance of the Dutch and Danish governance structure (Wieland, 2005). The Danish results provided a so far unused opportunity to study Dutch data.

This scientific demand is enhanced by the ongoing public discussion in the Netherlands, about the salaries of managers in the upper echelons. This discussion was further fueled by the current economic crisis. (NRC Handelsblad, April 2010).

2. Literature Review and Hypotheses

Since the publication of Lazear and Rosen’s article in 1981 the research area concerning tournament models has been expanded by numerous articles. The scholars in this field have expanded their paradigm from theory to empirical studies of data from countries all over the world. I have divided the review into different sections so that the reader can become familiar with the theory before I introduce the hypotheses concerning the two aspects of the theory that I would like to investigate in the Netherlands.

Foundation of the Tournament Theory

As was mentioned earlier, it all started with Lazear and Rosen’s paper. These authors introduced a model whereby compensation is tied to position, and whereby the top-seeded executive, or CEO, receives a salary that is substantially higher than any measure of his marginal product, or output. This is economically efficient, because the wage premium acts as an incentive for the other executives on the board to partake in a

tournament for the CEO position. In this tournament the contestants hand in part of their salary, which is assumed to be based on their marginal product. These parts are all added to the then larger first prize, or CEO salary, which enhances the participants’ effort to obtain that position. In addition, when it is arduous and costly for a principal to assess the efforts made by their employees, a tournament scheme might be compelling, because it, according to theory, maximizes the effort of the participants. As a result monitoring costs are saved.

On the one hand these theoretically based results are interesting as they were out of line with the main theories that were proposed until then, on the other hand one major drawback of the conclusions is that corporate managers are assumed to be either risk-neutral or risk-averse. Instead research has shown that corporate managers tend to take relatively high risks in order to achieve better results (Hoskisson et al.,1991)

reasoning as adopted by Lazear and Rosen: the main drawback of a piece-rate (equal to the marginal product of an employee) compensation scheme is that the input/effort of higher or lower level employees is not directly observable, or observable without incurring costs. A viable alternative for such a scheme is the ‘competitive compensation scheme’, whereby pay is based on performance relative to one’s peers rather than solely on output, because it doesn’t have the lack of flexibility of a piece-rate structure. This lack of flexibility is characterized by the high costs that are associated with adapting the piece rate to different sets of environmental variables. After performing numerous econometric procedures, the authors conclude that in a multi-agent (the agent being the employee) setting with large environmental uncertainty, a tournament scheme has the highest expected utility for the principal (the owner of the firm/manager).

Thus far, one of the more important assumptions of the theory is that higher effort levels automatically lead to higher output levels. Inevitably, one can argue that when the output of an employee is multi-dimensional, the output is not merely determined by the amount of effort, but also for example the efficiency of that effort.

which adopt piece-rate schemes, and receive the discounted value of their marginal product. This reduces their benefit of winning, or the difference between winning and loosing, because in a tournament scheme all contestants give up some of their marginal product so that the prize is big enough to work as an incentive mechanism. (Lazear and Rosen, 1981). In other words, the discounted marginal product of employees that lose in a tournament scheme is lower than that of employees of a contracting scheme.

The third additional point of critique made by Dye again concerned effort levels. He proposed that in a tournament all the contestants could collude in reducing their effort levels. Furthermore, agents could be induced to sabotage other contestants to obtain the first position.

However, the author does stress the positive effects of a tournament scheme in ‘hawkish’ companies.

Rosen, one of the two authors of the seminal paper published in 1981, proposed that the tournament prize, or the salary of a CEO, which is ultimately won in an organization, if set correctly, provides an incentive for the participants in the tournaments down the corporate ladder (Rosen, 1986). He thereby suggested, as did other authors before him, that multiple tournaments are held in a company whereby winners proceed to the next tournament and losers are out of the running for further promotions. However, in my opinion Rosen neglects an important aspect of career progression, namely that certain losers of earlier tournaments might still be eligible for promotion and could be described as slow starters.

According to Rosen the prizes that could be won in the consecutive tournaments should increase more than proportionate to rank, with an additional component for the overall winner, in order for the contestants to be striving for success at all stages in the

company’s tournament. If the first prize is not disproportionately large winners of pre-final tournaments could cease to partake as they find that their financial or status-driven goals are met. In addition, contestants in the final stage of the game could perceive the game to have ended unless extra weight is put on the first prize. In other words, if the prize differential of other executives pay to CEO pay is too small, there is little incentive for the survivors, or other executives, to compete for the first prize and as a result they could relish their past victories. However, if the top prize is effectively elevated it could lengthen the corporate ladder so that ‘no matter how far one has climbed, there is always the same length to go’ as Rosen argues. Nevertheless, in my opinion, this could also prove to be detrimental to the incentives of other executives, because as they have reached the final stage they, according to Rosen still have a long way to go, which could also lead to questioning the relevancy of earlier achievements.

student volunteers. The authors found that the theory explained the behavior of the students reasonably well across all tournaments the participants were involved in.

However, a large variance in effort levels was found for all rank-order tournaments. This could be explained by the more strategic nature of behavior of individuals that are

assessed by relative performance in tournaments. An interesting observation made during the experiments was that ‘disadvantaged’ individuals, or in earlier terminology dubbed ‘low-ability’ individuals (O’Keeffe et al., 1984), were exerting more effort when they knew they were competing with ‘advantaged’ individuals than was predicted in earlier articles describing the ramifications of certain theoretical assumptions. For example, Dye (1984) predicted sub-optimal effort levels when contestants knew they were competing with one or more contestants with greater abilities.

Hypotheses emanating from Tournament Theory

The empirical studies that followed tested tournament theory in different ways and used different sorts of datasets from sports tournament earnings to executives’ salaries.

One of the most influential empirical studies in the field was conducted by O’Reilly et al. (1988). The design of the study paved the way for studies that were to be conducted later on by introducing a testable hypothesis for the tournament model. The authors proposed that the gap between the CEO and Vice-Presidential salaries can be explained by the tournament model. In order to sharpen their research they mention the possibility of other determinants of executive compensation, namely: firm size, firm performance, industry (Ciscell and Carroll, 1980) and certain human capital characteristics such as tenure, age and education (Hogan and McPheters, 1980).

Subsequently, this hypothesis is tested by examining compensation data of executives of 105 American firms for the year 1984. Salary and cash bonus are used as measures of CEO and other executives pay, whereby the gap is used as a dependent variable. They actually find significant results that reject their hypothesis, and as a result have evidence to contradict the tournament theory.

Three other studies also examined the empirical validity of the proposed positive relationship between board size and the executive wage premium. First, Main et al., (1993) conclude after analyzing extensive data on executive compensation from over two hundred U.S firms from 1980 to 1984, that, at least in the upper echelons, there is a large amount of support for the tournament theory. The results suggest that the size of the prize increases with the number of contestants as is predicted by tournament theory.

Additionally, ‘Winners’ go from having a high salary in the tournament they win to a low salary in the consecutive tournament.

When using the average present value of CEO salary a result emerges that is even more consistent with tournament theory. Furthermore, the authors introduce the notion that ‘losers’ of earlier tournaments might still compete for the top job as they might have won tournaments they played subsequently.

Secondly, another study confirmed the predictions regarding this aspect of tournament theory in a paper that tested information on 2,600 executives in 210 Danish firms (Eriksson, 1999). For the first time Non-US data was analyzed, thereby providing an opportunity to assess the workings of the model in a country with a different corporate governance structure, which in addition resembles the Dutch system (Wieland, 2005). When controlling for other factors, the outcomes support the hypothesis, because in this dataset the higher the number of participants/other executives, the greater the first prize, or CEO salary.

stock market companies, with over 500 executives in the years 1997-1998. The authors follow an earlier study (Lambert et al., 1993) by using ‘total executive compensation’ as a measure, which consists of cash remuneration and stock-based remuneration. And introduce The Black-Scholes formula to calculate the present value of the grant of executive options (Murphy, 1999). The UK data shows consistency with the hypothesis that a larger board leads to a larger wage premium.

With the theoretical foundations of the framework and the empirical assessments of the hypothesis regarding board size emanating from it in mind I propose the following hypothesis:

1. After controlling for other factors, there exists a positive relationship between the number of contestants, or executives on the board, and the pay differential.

One might wonder whether there perhaps exists an endogenous relationship between the board size and the pay gap, in other words whether the causality between the two might easily be reversed. However, after an evaluation of studies examining the determinants of board size one can conclude that the number of executives is mostly determined by two factors, namely firm size and growth opportunities (e.g. Boone et al., 2005; Lehn et al., 2003). Furthermore, as will be shown after a performance of a Hausman test, there is no sign of an endogenous relationship between the regressors and the error term in the model investigating this hypothesis. In addition, no collinearity issues arise between firm size and board size. The model investigating this hypothesis and the Hausman and similar tests will all be explained more precisely in the subsequent sections.

The first hypothesis investigates a possible cause of the pay differential, in addition a number of other theoretical and empirical papers investigate the possible effects of pay differentials. One such effect is related to firm performance. Three studies have examined this relationship using firm data. The earlier mentioned UK study (Conyon et al., 2001) lead to the conclusion that the executive wage premium does not seem to have either a significant positive or negative link with firm performance.

which implies that managerial pay differentials provide useful incentives to improve the results of companies.

In addition, Main et al. (1993) establish the same results, because the expected significant and positive association between the wage differential and ROA is found.

The findings of Eriksson were predicted in earlier more theoretically oriented papers. For example, Rosen (1986) argued that a disproportionately large wage premium at the highest hierarchical level will act as an incentive for the contestants to counteract a tendency to relish in their past victories, which were necessary to reach the final stage of the tournament. As a result the managers will maintain high effort levels, leading to a higher individual performance which in turn should have a positive effect on the firm’s performance. The same line of reasoning is followed in other papers (e.g. O’Keeffe et al., 1984). However, earlier papers also predicted certain drawbacks associated with the implementation of a tournament compensation scheme, such as sabotaging other contestants, which in turn could lead to a lower overall firm performance (Dye, 1984)

To examine the relationship between pay spread and performance from another empirical angle some researchers investigated the effects of pay spread on the performance of athletes, thereby examining data from different sport disciplines. Note that in the investigated disciplines such as golf, tennis and car racing, there is very little incentive for the players to operate cooperatively. As a result, some caution is needed before drawing conclusions, based on those sports data, concerning the effects that certain tournament theory aspects have in a business environment. However, so-called ‘hawkish’ firms (Lazear, 1989) are characterized by a lesser need for cooperation at any level of the organization, therefore the results obtained after investigating sports data provide a good opportunity to draw conclusions regarding these type of organizations. Needless to say, this problem of limited applicability at the firm level is far less an issue when looking at other study’s results based on data from companies.

tournaments were used, because those data concerned incentives that players face, in the form of prize money, and the individual’s performance, or player’s score. The authors found, after controlling for player quality, that players perform better when the surplus of improving their rank by one is greater.

NASCAR, or National Association for Stock Car Auto Racing, was the subject of the subsequent paper discussing empirical findings on the effects of pay spread (Becker and Huselid, 1992). American auto racing is characterized by a variety in height and in

distribution of prizes and therefore provides a sound basis to test tournament theory. They found that greater differences in prizes clearly had an effect on the performance of the individual driver and on his recklessness. The effect concerning recklessness is quite interesting as the authors use it as a proxy to explain egocentric behaviour in firms. They reason as follows; when the prize of winning the tournament is too large an individual might begin to sabotage other contestants by driving rather recklessly, thereby

endangering other contestants. This tendency to sabotage adversaries was also predicted earlier in theoretical papers (Dye, 1984; Lazear 1989) Thus they concluded that the tournament model is supported by the data on both positive, or performance enhancing, and negative, or sabotaging, aspects.

‘The performance effects of pay dispersion on individuals and organizations’ is the title of a paper investigating the relationship between pay dispersion and performances of individuals and teams (Bloom, 1999). After examining Major League Baseball teams conclusions are drawn that contradict theory, namely that both individual and team performances were negatively influenced by a wider pay gap.

Data from tennis tournaments was also extensively investigated. Sunde (2003) concluded that players who reach the final and/or semifinal, and who are thus more competitive in that particular tournament are prone to exert more effort when the pay spread is larger. Similar results were obtained in another study (Ivankovich, 2007).

both positive and negative consequences with regard to firm performance. Furthermore, studies involving sports data mostly establish a positive relationship between the

implementation of a tournament pay structure and the efforts and performances of players. However, these results are mainly obtained when investigating data of individual sports, with an inherent lack of a need for cooperation, which is required at the corporate level at least to a certain extent. The only paper that investigated sports data with a focus on teams (Bloom, 1999) found a negative relationship. Moreover, a study investigating NASCAR data found evidence of sabotaging behaviour when the prize spread was increased.

Based on these mixed empirical findings and mixed theoretical predictions it is clear that divergent hypothesis can be derived. In a paper dealing with similar divergency problems two hypotheses were set up (Richard and Shelor, 2002). I would also like to propose two hypotheses, namely:

2.a: When controlling for other factors, there exists a positive relationship between the

pay differential at the executive level and firm performance.

2.b: When controlling for other factors, there exists a negative relationship between the

pay differential at the executive level and firm performance.

In addition to the derivation of these hypotheses I would like to discuss some of the explanatory variables that will be incorporated in the models that were designed to test the hypotheses. These variables are: ‘executive age’, ‘executive tenure’ and ‘number of executive officers’. The two other control variables that were adopted namely: ‘firm size’ and ‘industry’ will be discussed in the subsequent section. Moreover, the dependent variable of the first model: ‘executive compensation’, and the dependent variable of the second model: ‘firm performance’ will also be discussed in the subsequent section.

Executive age

1995) and it is proposed that this is mainly a result of compensating for higher

intellectual and managerial capabilities, because knowledge and experience increase with age (Hogan and McPheters, 1980; Hambrick and Mason, 1984; Hitt and Tyler, 1991). Note that I differentiate between CEO age and the average age of other contestants.

Consistent with prior research investigating the relationship between board characteristics and firm performance, I include CEO age and average age of other contestants as a control variable in the second model as it was found to be significantly related to firm performance (Wagner et al., 1984; O’Reilly et al., 1989). Some other studies found that management age and cognitive abilities can be correlated (Hambrick and Mason, 1984; Hitt and Tyler, 1991): the higher the age the more cognitive abilities. However, the exact opposite was proposed in other papers (Welford, 1977; Shock1962).

And some studies claim that the productiveness of young employees is a lot higher than that of older employees (Holstrom,1982; Fama, 1980). While Johnson in 1978 stated that older employees produce their output more efficiently. Van Olfen and Boone (2002) concluded that teams with a higher average age have more experience and will therefore be able to take advantages of alternations in the company’s environment. In conclusion, one can argue that there is sufficient evidence to include this variable into the second model as many studies concluded that age has an effect on performance, be it positive or negative.

Executive tenure

Measured as time in current or equivalent position. When executives on the board switched their main areas of responsibility, their tenure was assumed to be continuing. Hogan and McPheters (1980) proposed that the longer a CEO is in his position the more influence he could have on the remuneration committee. Moreover, the power of a CEO increases with his time on the board, and he could therefore more easily pursue his own interests rather than the interest of the stockholders (Hill and Phan, 1991). Prior empirical studies found evidence of a significant positive relationship between tenure and

Note that I also introduce the tenure of the other executives in the model.

In the second model tenure is adopted as a predictor variable because a number of articles found a significant and mostly negative relationship between tenure and performance. For example, a certain unwillingness to alter strategies characterizes individuals with higher tenures (Katz, 1982). This resistance to change was perpetuated even when a

modification of the chosen strategy would result in more beneficial firm performance (Staw and Ross, 1987). Furthermore, teams with higher tenure showed less capability to deal with new sources of information, due to the fact that they perceived relying on past experience to be more important than actively scanning the business environment for new clues and information (Katz, 1982). In addition, Finkelstein and Hambrick (1990) find a significant relationship between tenure of corporate officers and the performance of a company.

Number of executives

a CEO and non-executive officers, commonly referred to as a ‘one-tier’ board, whereas in the Netherlands a board consists of a CEO and executive officers. In addition, there is a board consisting only of non-executives that ‘supervise and oversee the company’. These two boards together characterize the ‘two-tier’ structure. In the two-tier structure the non-executives are more independent than in a one-tier structure (Maassen, 1999). However, in the research area investigating the effects of a tournament compensation scheme, the differences between a one-tier board and a two-tier board are not of such an importance that a replication outside the US is impossible, because the vast majority of the studies that examined US firm data compared the salary of the CEO to the salary of other executives, instead of non-executives. In addition, the consistency of the Danish

compensation data (Eriksson, 1999) with the tournament theory provides an incentive to investigate the theory in the Netherlands, due to the resemblance of the governance system in both countries (Wieland, 2005).

3. Data and Methods

I collected data on 100 Euronext-Amsterdam-listed firms over the years 2004 to 2007 using Amadeus and annual reports of the companies as sources. A list of the companies is presented in the appendix. The reason that I investigate Euronext-Amsterdam-listed companies is that Dutch compensation data has never been tested for the existence of a tournament pay structure. Moreover, Conyon et al, (2001) strongly urge scholars in the field of executive compensation to check for universal validity of the tournament model, because most research conducted so far has been conducted in the US.

Models

estimated. In this model it is assumed that the parameters are fixed for all time periods and are the same for all firms. However, when the assumptions are made that the

variances of the error terms are different for each equation and that the error terms in the different equations are correlated; a seemingly unrelated regression (SUR) is more efficient.

The third model that can be used to investigate panel data is the ‘fixed-effects’ model. This model of parameter variation specifies that only the intercept parameter varies, not the slope parameters, and the intercept varies only across individuals and not over time. Moreover, the error terms are assumed to be independent, with mean zero and constant variance. Given these assumptions all the behavioral differences between individual firms and over time will be captured by the intercept. In order to control for these firm specific differences an intercept dummy variable is included for each firm. This dummy variable is 1 for all time periods for that cross-section, and 0 otherwise. The coefficients of these dummy variables are equal to the firm intercepts and are called the ‘firm fixed effects’. These fixed parameters can be estimated directly by using a least squares estimator. A drawback of the fixed-effects model is that it cannot include variables that are constant for each individual across time. As a result these variables will be dropped.

The fourth model, or random effects model, also assumes that the individual differences are captured by the intercept parameters, however these differences are treated as random rather than fixed. This can be included in the model by specifying the intercept

parameters to consist of a fixed part that represents the population average, and random individual differences, or random effects (Ui), from the population average. These random effects are analogous to the error terms, and we make the same standard assumptions about them, namely that they have zero mean, are uncorrelated across individuals and have constant variance. As a result the combined error is Vit = Ui + εit. Thus the regression error consists of two components, one for the individual and the other for the regression. Moreover, the individual effects (Ui) are not correlated with the

are otherwise uncorrelated. Under these assumptions the generalized least squares (GLS) estimator is the minimum variance estimator for the random effects model.

The presence of random effects can be assessed by performing a Breusch-Pagan test, which will be explained in more detail later on. For now it is important to know that if random effects are present, so that σ²u > 0, then a random effects estimator is the preferred model for a number of reasons. First, it permits the adoption of variables that are time-invariant. Secondly the random effects estimator is a generalized least squares estimator, whilst the fixed effects estimation is least squares. As a result the GLS estimator has a smaller variance than the least squares estimator allowing for greater precision. Thirdly, the random effects estimator estimates the effects of the explanatory variables on y by using information on how changes in y across firms could be

attributable to the different x-values for those firms. In addition, information is used on the variation in x’s and y over time, for each individual. Contrarily, the fixed effects estimation only uses the latter.

A problem associated with the use of the random effects estimation is that the regressors could be endogenous. This occurs when the random error (Vit) is correlated with any of the right-hand side explanatory variables in the random effects model, as a result the least squares and GLS estimators of the parameters are biased and inconsistent, because it will attribute the effects of the error component to the included explanatory variables. This problem is common in random estimations, because the individual specific error component (Ui) may well be correlated with some of the explanatory variables. In the presence of such a correlation the fixed-effects estimation is still consistent, as it

models, or in other words: that there was no correlation between the error component and the regressors.

Model 1:

Vit = Ui +

ε

Whereby Y is the log of CEO pay minus the log of average executive officer pay, X2 is (log) firm size, X3 is age of CEO, X4 is average age of other executives, X5 is tenure of CEO, X6 is average tenure of other executives, X7 is the total number of executive officers on the board, X8 denotes the industry of a firm.

Furthermore, V denotes the combined error term of: U, which is the random individual effect and ε, which is the usual regression random error.

Moreover, ß1 is the intercept parameter. I make the following assumptions: the random individual differences (U) will have zero mean, are unrelated across individuals and have a constant variance.

Model 2:

Yit = ß1 + ß2X2it + ß3X3it + ß4X4it + ß5X5it +ß6X6it + ß7X7it + ß8X8it + ß9X9it + Vit

Vit = Ui +

ε

Herewith the Y is return on equity (ROE). X2 is (log) firm size, X3 the log of CEO minus the log of average executive officer pay, X4 is age of CEO, X5 is average age of other executives, X6 is tenure of CEO, X7 is the average tenure of the other executives, X8 is the total number of executive officers on board, X9 denotes the industry of a firm.

Yit = ß1 + ß2X2it + ß3X3it + ß4X4it + ß5X5it +ß6X6it + ß7X7it + ß8X8it + Vit

Measures

Executive compensation: cornerstone of this research area. I have computed ‘cash

compensation’ levels, or annual salary plus bonus measured in euros, for all CEO’s. In addition, I have computed average cash compensation levels for the rest of the executive board. Note that in some occasions I had to calculate the salary of the executives in euros using the average exchange rate for a particular year, because certain annual reports were reported using for example the British pound. Furthermore, almost all bonuses, hence salary levels, were related to performance measures for the prior year, this is congruent with earlier studies (Lambert et al., 1993).

I have omitted long-term compensation, such as stock options, because short-term compensation was found to be proportional to long term compensation (Jensen and Murphy, 1990). In line with prior research, the log of CEO pay minus log of contestants pay is used as a dependent variable in the model (1) investigating the effect of the number of executives on the pay gap. Furthermore, this same modified variable is used as an explanatory variable in the model (2) investigating the effect of pay spread in the boardroom on firm performance. An alternative way of measuring the premium is to calculate the logarithm of (CEO pay minus contestants’ pay), instead of: log (CEO pay) – log (contestants’ pay). However, because there are several observations whereby the average pay of the other executives is higher than the pay level of the accompanying CEO it becomes a mathematical impossibility to calculate the log of (CEO pay – contestants’pay). As a result, regressions ran with the latter as a dependent variable would not include the observations whereby the contestants pay is higher than that of the CEOs. As a result the parameters do not reflect all the observed data. On the other hand when log (CEO pay) – log (contestants’ pay) is adopted, these type of observations will be included.

Firm performance: an indicator widely used in this research area is Return on Equity

Firm size: defined as the log of total assets, measured in thousands of euros (Postma et al., 1999). Compensation equations universally control for measures of company size (e.g. Finkelstein and Hambrick, 1988; Jensen and Murphy, 1990). I also adopt it as a control variable in the second model.

Industry: I introduce dummies for industry as a measure of control, because earlier

studies observed industry-wide differences in top management compensation (Crystal, 1984). The industry dummies are also used as control in the second model.

Furthermore they are assumed to be constant over the investigated years. A list of industry characterizations is included in the appendix.

Results

Before running the regressions using the different models I will present some summary statistics and include a correlation matrix to show that there exists no multicolliniearity.

Summary statistics

Table 1: Descriptive statistics:

Variable N Mean St.Dev. Minimum Maximum

CEO age 400 53.09 6.70 35 76 VP age 400 49.53 5.59 33 67 CEO tenure 400 5.99 6.23 0.08 40 VP tenure 400 4.10 2.60 0.04 14 CEO salary 400 923,934.70 879,829 131,000 5,023,000 VP salary 400 532,261.50 395,003 67,000 2,517,998 Board size 400 3.42 1.50 2 12 ROE 397 12.10 31.56 -150 124.38

CEOs tend to be older, on average, than the other executives on the board. Furthermore, they seem to be longer in their position, as is shown by their average tenure, which is almost two years longer than other executives.

One can see from this table that the average cash compensation of CEOs is around 920.000 Euro and executive salary is 530.000 Euro.

Correlation matrix

Table 2: Correlation Matrix:

* = p<0,05 and ** = p<0,01

Variable LnPremium LnFirmSize ROE CEO age VP age CEO

Model 1:

In order to investigate whether the data is suitable for a random effects estimation two tests need to be run. First, a Breusch-Pagan test, to asses the presence of random effects. Secondly, a Hausman test is ran to test whether the random effects is consistent.

Breusch-Pagan test:

If Ui = 0 for every individual, there are no individual differences and no heterogeneity to account for. Hence, the pooled linear regression is appropriate and there’s reason to use either a fixed or a random model. To test for the presence of random individual

heterogeneity the following null hypothesis can be tested: H0: σ²u = 0 against the

alternative hypothesis H1 : σ²u > 0. If the null hypothesis is rejected, one can conclude that

there are random individual differences among the sample members, and as a result adopt the random effects model as the appropriate estimation.

Table 3: Breusch and Pagan Lagrangian Multiplier test for random effects

Estimated Results Var. Sd = √(Var.)

Ln Premium 0.1392 0.3731

ε

U 0.0635 0.2478

The chi-squared is 135.64 and the accompanying p-value is 0.0000.

Hausman test:

The Hausman test compares the coefficient estimates from the random effects model to those from the fixed effects model. If both estimators are consistent, which occurs when there is no correlation between Ui and the explanatory variables, then they should both converge to the true parameter values in large samples. However, if there does exist a correlation between Ui and the explanatory variables the random effects estimator is inconsistent and it converges to some other value that is not the value of the true parameters. Contrarily, the fixed effects estimator remains consistent because it

eliminates Ui as well as any other time-invariant variables and therefore converges to true parameter values whether the random effect Ui is correlated with the regressors or not. In case of correlation one can expect differences between the fixed and random effects estimates. The following hypotheses are proposed for when conducting the Hausman test:

H0 : Ui is uncorrelated with the explanatory variables and as a result there is no difference

between the fixed effects and random effects estimators.

This in turn advocates the use of the random effects model, because of the smaller variance that is associated with it, as compared to the fixed effects model, which leads to a more precise and efficient estimation.

H1: there does exist a correlation. Hence, the random effects estimator is inconsistent and

the fixed effects estimator should be used, or the model specification should be improved.

Table 4: Hausman test

Prob>chi2 = 0000....7747744455552222 = 33.33...44449999

chi2(666) = (b-B)'[(V_b-V_B)^(-1)](b-B)6 Test: Ho: difference in coefficients not systematic

B = inconsistent under Ha, efficient under Ho; obtained from xtreg b = consistent under Ho and Ha; obtained from xtreg lnfirmsize ....000011114424422266696994944 4 ...0.00303331111996996663333999 9 ---.-..0.00011117767766699949445455 5 ....0000333333733777777474434333 boardsize ---.-..0.00022222202200022232333333 3 ---.-..0.00000009909900022228888222 2 ---.-..0.00011211222999999599551511 1 ....0000115115552222882882252555 vptenure ---.-...000044644666555577077009099 9 ---.-..0.00404444444444411119999777 7 ---.-..0.00000200222111155155112122 2 ....0000005005550000116116636333 ceotenure ---.-...000000000000888833433441411 1 ---.-..0.00000002252255566661111888 8 ...0.00000001171177722272777777 7 ....0000000033833888666464454555 vpage ---.-...00000060066644442223233434 44 ---.-..0.00000200222999900800888333 3 ---.-..0.0000030033355551115155151 11 ....0000004004440000333434454555 ceoage ....000000008888112112262666666 6 ...0.00000007777004004443333222 2 ...0.00000001101100088838333333 3 ....0000000033233222222121161666 fe re Difference S.E. (b) (B) (b-B) sqrt(diag(V_b-V_B)) Coefficients

When the result of both the Breusch-Pagan and Hausman test are combined, a good foundation for the advancement of the usage of a random effects estimation emerges. Thus now I can proceed with such an estimation.

Table 5: Random effects estimation

rho ....5555222200040444444747757551511 (fraction of variance due to u_i)1

sigma_e ....22224444111818988998988888898999 sigma_u ....22225555222020000002022121161666 _cons ----....005005575777005005595997977 7 ....22622666666611811888888181 11 ----00.00...22221111 000.0..8.83883330000 ----....55755778788787777778788877 77 ....4464466464464656655959939333 ind08 ....111122022000888585555551511 1 ....111111111111222211111112122 2 1111..0..0009999 000.0..2.22727777777 --.--..0.009097997717111114114944999 ....3333338338888882822525515111 ind06 ....333322222222777676566553533 3 ....222211311333444422322337377 7 1111..5..5551111 000.0..1.11313033000 --.--..0.009095995555553537337477444 ....7777441441101006066767797999 ind05 ....000088188111555959899888888 8 ....111100800888999933433446466 6 0000..7..7775555 000.0..4.44545455444 --.--..1.113131331191990909009199111 ....2222995995515110100606676777 ind04 ....000099399333333737777778788 8 ....111144644666666622622663633 3 0000..6..6664444 000.0..5.55252422444 --.--..1.119194994404000004004644666 ....3333880880070776766060010111 ind03 ....000055255222111717477444444 4 ....111111011000888811911998988 8 0000..4..4447777 000.0..6.66363833888 --.--..1.116165665505002028228488444 ....2222669669939337377777717111 ind02 ....000044404000882882272774744 4 ....00900999000088688666999797 77 0000..4..4445555 00.00..6.66565553333 ----...1.11313373772727227477444 ....2222118118898992922828898999 ind01 ....222200400444444545155113133 3 ....111133433444999900400447477 7 1111..5..5552222 000.0..1.11313033000 --.--..0.005059559999995957557177111 ....4444668668888885855959979777 lnfirmsize ....000033313111996996636339399 9 ....00100111555511211222333535 55 2222..1..1111111 00.00..0.00303335555 ....0000000202232332322222322333 ....0000661661161660600505545444 boardsize ----....000000090999002002282882822 2 ....00100111666677577555444545 55 ----00.00...55554444 000.0..5.59559990000 ----....00400441411818688666666655 55 ....0020022323383818811010010111 vptenure ----....004004444444441441191997977 7 ....000000007777443443338885855 5 ----5555..9..9997777 00.00..0.00000000000 ----....0005058558898999998998888888 ----....0020022929989848844040060666 ceotenure ----....000000020222556556616118188 8 ....00000000444400100111666161 11 ----00.00...66664444 000.0..5.52552224444 ----....00100110100404344333333322 22 ....0000000505535303300909979777 vpage ----....00000002022299099080088383 33 ....00000040044422223333666767 77 ----0000....666699 99 000.0..4.449499922 22 ----..0..00101111121122121112222 ....0000000505535393399595545444 ceoage ....000000070777004004434332322 2 ....00000000333355255222444949 99 2222..0..0000000 00.00..0.00404446666 ....0000000000010113134334544555 ....0000113113393995955151191999 lnPREMIUM Coef. Std. Err. z P>|z| [95% Conf. Interval] corr(u_i, X) = 00 (assumed) Prob > chi2 = 000 00.0..0.000000000000000 Random effects u_i ~ GGaGGaauausuussssssisiaiiaaannnn Wald chi2(1111333) = 3 6669699.9..2.2202000 overall = 00.00...11191999444141 11 max = 4444 between = 0000....22232333333131 11 avg = 44.44..0.000 R-sq: within = 000.0...111212225585588 8 Obs per group: min = 4444 Group variable: iiiidddd Number of groups = 111010000000 Random-effects GLS regression Number of obs = 444040000000

z-statistics. In large samples it does not matter if critical values for tests come from one distribution or the other.

Note that there are no observations for Industry 07, or ‘Utilities’. Furthermore, I took Industry 09, or ‘Technology’ as the comparison group.

The outcomes show that the size of the executive team does not have a significant effect on the gap between CEO pay and pay of the other executives on the board. Moreover, the same result emerges when I control only for firm size and industry, as was done in numerous earlier papers discussing the matter. I have included those outcomes in the appendix. Hence the results are not consistent with my hypothesis that a larger board leads to a larger pay gap.

However, firm size appears to be positively and significantly related to the pay gap. In addition, the tenure of the other executive officers was found to have a significant

negative influence on the pay gap. Also the age of the CEO is positively and significantly related to the pay gap.

An overall R-squared of 0,19, which is rather low, implies that 19 % of the variability among the variables was explained.

Model 2

Table 6: Random effects estimation

rho ....5555999955505060066464494997977 (fraction of variance due to u_i)7

sigma_e 11118888....000505755773733737767666 sigma_u 22221111....888888988999999191191999 _cons ----666655.55..1.111996996606009099 9 222211.11...8888552552228882822 2 ----2222..9..9998888 00.00..0.00000003333 ----1111000808.88..0.002026226668888 ----222222.22..3.3363656655353353555 ind08 ----6666..4..4474777662662252556566 6 9999..5..555000000200222333434 44 ----00.00...66668888 000.0..4.49449995555 ----222255.55..0.00909699663633377 77 111122.22..1.1141434433838868666 ind06 8888....88888888222626866888888 8 11118888..2..222777799599556566 6 0000..4..4449999 000.0..6.66262722777 --2--226266.6.9..9949444445445955999 44444444..7..7707009099999979777 ind05 7777....00020222551551171779799 9 9999..3..333000066766777666565 55 0000..7..7775555 00.00..4.44545550000 --1--111111.1..2.2212151155757774444 222255.55..2.226266666616111 ind04 ----111177.77..1.11188288202200404 44 111122.22...555522228888555353 33 ----11.11...333377 77 000.0..1.17117770000 ----4414411.1..7.773733737775555 7777..3..3373773733434434333 ind03 ----111100.00..7.777664664434339399 9 9999..4..444666688888888888585 55 ----11.11...11114444 000.0..2.25225556666 ----222299.99..3.33232322330300077 77 7777..7..77979949424422828828222 ind02 8888....55155111444242122119199 9 7777....77677666333344344336366 6 1111..1..1110000 000.0..2.22727377333 --6--66.6..7.70770010118183883633666 22223333..7..7737330300202282888 ind01 1111....66566555555757077001011 1 11111111..5..555444455955991911 1 0000..1..1114444 000.0..8.88888688666 --2--220200.0.9..9979773738338788777 22224444..2..2282885855252272777 lnPREMIUM 2222....11711777333636066004044 4 3333....99599555444488888882822 2 0000..5..5555555 000.0..5.55858388333 --5--55.5..5.57557777778782882322333 9999....992992252550500303313111 boardsize ----....00400449499944344303300202 22 1111..3..333333322227777111313 33 ----00.00...000044 44 000.0..9.97997770000 ---2-2.22..6.666661661511555 2222....556556626226266363393999 vptenure 1111....22222222112112252558588 8 ....66066000666644744777333939 99 2222..0..0001111 000.0..0.00404444444 ...0.03003323225255959991111 2222....440440090999999292252555 ceotenure ....555544144111111515555558588 8 ....333322122111333311211223233 3 1111..6..6668888 000.0..0.00909299222 --.--..0.008088888868660604004844888 1111....117117707009099191161666 vpage ----....00900996966611311313311616 66 ....33333373377777779999444848 88 ----0000....222288 88 000.0.7..777777766 66 ----..7..775755858188119199797277222 ..5..55656656595599393343444 ceoage ----....33335555555566566550500404 44 ....22822888555588188111111 1 ----1111..2..2224444 00.00..2.21221113333 ----....99199115155858288229299966 66 ....2202200404454525522828898999 lnfirmsize 6666....66606000667667747449499 9 1111..2..222666677377333111919 99 5555..2..2221111 000.0..0.00000000000 444.4.1..1121222222828885555 9999....009009909006066464494999 roe Coef. Std. Err. z P>|z| [95% Conf. Interval] corr(u_i, X) = 00 (assumed) Prob > chi2 = 000 00.0..0.000000000000000 Random effects u_i ~ GGaGGaauausuussssssisiaiiaaannnn Wald chi2(1111444) = 4 6663633.3..1.1111111 overall = 00.00...22252555333636 66 max = 4444 between = 0000....33313111888888 88 avg = 44.44..0.000 R-sq: within = 000.0...000707779989988 8 Obs per group: min = 3333 Group variable: iiiidddd Number of groups = 111010000000 Random-effects GLS regression Number of obs = 333939979777

Again, I took Industry 09 as the comparison group.

No significant relationship between the premium and firm performance was established. Moreover, when adopting the real premium instead of the log of CEO pay minus the log of other executive pay as a predictor variable the same results emerge (see appendix).

However, the results suggest that firm size has a robust positive and significant

relationship with firm performance. In addition, the tenure of other executives appears to have a significant and positive effect on firm performance.

4. Conclusion

Implications

This paper investigated the existence of tournament pay structure in the Netherlands by examining data from 100 Euronext-Amsterdam-listed firms over the years 2004 to 2007. Data was collected using Amadeus and annual reports. Two models were designed to investigate two hypotheses. The first stated there should exist a positive relationship between the number of executive officers and the pay spread between the CEO pay and other executive officers’ pay. The second stated that there should exist a significant relationship between the pay gap, or premium, and the performance of a company. With regard to the hypotheses the following can be summarized.

Hypothesis 1 can be rejected on the basis of this dataset, because after controlling for firm size, industry, age of the CEO and executives, tenure of the CEO and executives and firm performance, no significant relationship was found between board size and the premium, which was used as the dependent variable. However, firm size was found to be significantly and positively related to the premium. This finding is quite consistent with earlier studies, because those found that a larger firm size leads to higher CEO pay (e.g. Finkelstein and Hambrick, 1988), which in turn could mean a larger premium exists. Furthermore, the average tenure of the other executive officers was found to have a negative and significant influence on the premium. Thus the longer other executive officers are on the board the lower the premium of becoming a CEO is. This could be explained by a theory focusing on CEO tenure that stated that the longer a CEO is in his position the more probable it will be for him to influence the salary decision making process (Hogan and Mcpheters, 1980). This same effect could apply to the other

executives on the board, thereby establishing a higher pay, hence lowering the premium. In addition, CEO age was found to be positively related to the wage gap, which is

Hypothesis 2 can also be rejected, since no significant relationship was found concerning the premium and firm performance. Yet a significant and positive relationship was found between firm size and firm performance. Grounded in the theory of economies of scale this relationship has been hypothesized in numerous scientific papers and a vast array of papers exist providing additional empirical evidence referring to this relationship (e.g. Haldi and Whitcomb, 1967; Sheperd, 1979).

Limitations

The results obtained rejected the accompanying hypotheses. Obtaining data from other countries in Europe might lead to different results. In addition, a panel set using different years might alter the results for Euronext-Amsterdam-listed firms. This could also be interesting in light of the economic crises, because, reasoning intuitively, those could have lead to lower wages or lower premiums at the top of the corporate ladder.

Furthermore, the measure used for firm performance is also subject of scientific

discussion, and there are some valid arguments in current literature to use other measures, such as return on assets (ROA).

The incorporation of other predictor variables, such as education or time with the firm, instead of time in position, might lead to a change in the outcomes.

In addition, it might be interesting to include long-term compensation in a study investigating the theory of tournaments in the Netherlands.

Also a study of data involving only ‘hawkish’ firms seems compelling as those are companies which could, according to theory, benefit the most from a tournament compensation scheme. A study investigating only such firms could prove to be

(Lazear, 1989) hinted at a definition with which an interested researcher could begin to work with: the lower the cost of sabotage, the more ‘hawkish’ a firm.

Moreover, if such a study assesses Euronext-Amsterdam-listed firm data it could provide part of the explanation for the results that I obtained, as such a study might show that such companies are simply not ‘hawkish’ enough, and therefore have no incentive to implement a tournament compensation scheme.

5. References

Barnum, P., Liden, R.C., Di Tomaso, N., 1995. “Double Jeopardy for Women and Minorities: Pay Differences with Age”, The Academy of Management Journal, 38 (3): 863-880

Becker, B.E., Huselid, M.A., 1992. “The Incentive Effects of Tournament Compensation Systems”, Administrative Science Quarterly, 37 (2): 336-350

Bhagat, S., Black, B., 2002, “The Non-correlation between Board Independence and Long-term Firm Performance”, Journal of Corporation Law, 27 (2): 231-274

Bloom, M., 1999. “The Performance Effects of Pay Dispersion on Individuals and Organizations”, The Academy of Management Journal, 42 (1): 25-40

Bull, C., Schotter, A., Weigelt, K., 1987. “Tournaments and Piece Rates: An Experimental Study”, The Journal of Political Economy, 95 (1): 1-33

Cappelli, P., W.F., Cascio, 1991. “Why Some Jobs Command Wage Premiums: A Test of Career Tournament and Internal Labor Market Hypotheses”, The Academy of

Management Journal, 34 (4): 848-868

Ciscell, D.H., Carrol, M.T., 1979. “Determinants of executive salaries: An econometric survey”, Review of Economics and Statistics, February: 7-13

Conyon, M.J., Peck, S.I., Sadler, G.V., 2001. “Corporate Tournaments and Executive Compensation: Evidence from the U.K”, Strategic Management Journal, 22: 805-815

Dutton, J.E., Duncan, R.B., 1987. “The Influence of the Strategic Planning Process on Strategic Change”, Strategic Management Journal, 8 (2): 103-116

Dye, R.A., 1984. “The Trouble with Tournaments”, Economic Inquiry, 22 (1): 147-149

Ehrenberg, R.G., M.L., Bognanno, 1990. “The Incentive Effects of Tournaments Revisited: Evidence from the European PGA Tour”, Industrial and Labor Relations Review, 43 (3): 74S-88S

Eisenhardt, K.M., Schoonhoven, C.B.,1996. “Resource-Based View of Strategic Alliance Formation: Strategic and Social Effects in Etrepeneurial Firms”, Organization Science, 7 (2): 136-150

Eriksson, T., 1999. “Executive Compensation and Tournament Theory: Empirical Tests on Danish Data”, Journal of Labor Economics, 17 (2): 262-280

Fama, E.F., 1980. “Agency Problems and the Theory of the Firm”, Journal of Political Economy, 88 (2)

Fullerton, R.L., Preston, R. 1999. “Auctioning Entry into Tournaments”, The Journal of Political Economy, 107 (3): 573-605

Graef, C.S., 1984. “Questions and Answers on Executive Compensation”, New Jersey: Prentice-Hall

Green, J.R., Stokey, N.L., 1983. “A Comparison of Tournaments and Contracts”, The Journal of Political Economy, 91 (3): 349-364

Hambrick, D. C., Mason, P.A. 1894. “Upper Echelons: The Organization as a Reflection of its Top Managers”, Academy of Management Review, 9: 193-206

Hill, S., 1982. “The Multiproduct Firm: Demand Relationships and Decision-Making”, Managerial and Decision Economics, 3 (2): 90-94

Hill, C.W.L., Phan, P., 1991. “CEO Tenure as a Determinant of CEO Pay”, The Academy of Management Journal, 34 (3): 707-717

Hitt, M.A., Tyler, B.B., 1991. “ Strategic Decision Models: Integrating Different Perspectives”, Strategic Management Journal, 12: 327-531

Hogan, T.D., McPheters, D., 1980. “Executive compensation: Performance vs. Personal characteristics”, Southern Economic Journal, 46: 1060-1068

Holstrom, B., 1982. “Moral Hazard in Teams”, Bell Journal of Economics, 13: 324-340

Hoskisson, R.E., Hitt, M.A. Hill, C.W.L. “Managerial Risk Taking in Diversified Firms: An Evolutionary Perspective”, Organization Science, 2: 296-314

Hvide, H.K., 2002. “Tournament Rewards and Risk Taking”, Journal of Labor Economics, 20 (4): 877-898

Hurd, M., 1990. “Research on the elderly: Economic status, retirement, and consumption and saving”, Journal of Economic Literature, 28: 565–637

Jackson, S., 1992. “Consequences of Group Composition for the Interpersonal Dynamics of Strategic Issue Processing”, Advances in strategic management, 8: 345-382)

Jensen, M.C., Meckling, W.H., 1976. “Theory of the firm: Managerial behaviour, agency costs, and ownership structure”, Journal of financial economics, 3: 305-360

Johnson, J.B., 1978. “Types of Organizational Control and Their Relationship to Emotional Well-Being”, Administrative Science Quarterly, 23 (2): 293-317

Katz, R., 1982. “The Effects of Group Longevity on Project Communication and Performance”, Administrative Science Quarterly, 27: 81-104

Knoeber, C.R., Thurman, W.N., 1994. “Testing the Theory of Tournaments: An Empirical Analysis of Broiler Production”, Journal of Labor Economics, 12 (2): 155-179

Krakel, M., Sliwka, D., 2001. “Risk Taking in Asymmetric Tournaments”. Bonn University, Economic Discussion Paper 33

Lambert, R.A., Larcker, D.F., Weigelt, K., 1993. “The Structure of Organizational Incentives” Administrative Science Quarterly, 38 (3): 438-461

Lazear, E.P., 1989. “Pay Equality and Industrial Politics”, The Journal of Political Economy, 97 (3): 561-580

Lazear E.P., S. Rosen., 1981. “Rank-Order Tournaments as Optimum Labor Contracts”, The Journal of Political Economy, 89 (5): 841-864.

Maassen, G.F., 1999. “An International Comparison of Corporate Governance Models”, Rotterdam School of Management, Ph.D. Series in General Management, 31

Main, B.G.M., O'Reilly III, C.A., Wade, J., 1993. “Top Executive Pay: Tournament or Teamwork?”, Journal of Labor Economics, 11 (4): 606-628

McConnell, P., 1993. “Accounting for Employee Stock Options: Crystal Ball Gazing”, New York: Bears Stearns & Co

McKnight, P.J., 2000. “CEO Age and Top Executive Pay: A UK Empirical Study”, Journal of Management and Governance, 4 (3): 173-187

Murphy, K.J., 1985. "Corporate performance and managerial remuneration.", Journal of Accounting and Economics, 7: 11-42.

Murphy, K.J., 1999. “Executive Compensation”, Handbook of Labour Economics, 3: 2485-2563.

Nalebuff, B.J., Stiglitz, J.E., 1983. “Prizes and Incentives: Towards a General Theory of Compensation and Competition”, The Bell Journal of Economics, 14 (1): 21-43

Olffen van W, and Boone, C. 1997. “The Confusing State of the Art in Top Management Composition Studies: A Theoretical and Empirical Review”, NIBOR, Netherlands Institute of Business Organization and Strategy Research.

O'Reilly III, C. A., Main, B.G., Graef S.C., 1988. “CEO Compensation as Tournament and Social Comparison: A Tale of Two Theories”, Administrative Science Quarterly, 33 (2): 257-274

O’Reilly, C.A., Caldwell, D.F., Barnett, W.P., 1989. “Workgroup Demography, Social Integration and Turnover”, Administrative Science Quarterly, 34 (1): 21-37

Orrison, A., Schotter, A., Weigelt, K., 2004. “Multiperson Tournaments: An Experimental Examination”, Management Science, 50 (2): 268-279

Postma, T., Ees, H., Gerritsen, H., 1999. “Top Management and Board Attributes and Firm Performance in the Netherlands”, Graduate School/Research Institute, Systems, Organisations and Management

Richard, O.C., Shelor, R.M., 2002. “Linking top management team heterogeneity to firm performance: juxtaposing two mid-range theories”, The International Journal of

Human Resource Management, 13 (6): 958-974

Rosen, S., 1986. “Prizes and Incentives in Elimination Tournaments”, The American Economic Review, 76 (4): 701-71

Shock, N., 1962. “The physiology of aging”, Scientific American, 206: 100-110

Staw, B.M., Ross, J., 1987. “Knowing when to pull the plug”, Harvard Business Review, 65 (2): 68-74

Sunde, U., 2003. “Potential, Prizes and Performance: Testing Tournament Theory with Professional Tennis Data” Bonn University, Discussion Paper 947

Tosi, H. L., Werner, S., Katz, J. P., Gomez-Mejia, L.R., 2000. “How much does performance matter? A meta-analysis of CEO pay studies”. Journal of

Management, 26(2): 301-339

Wagner, W. G., Pfeffer, J., O’Reilly, C.A. 1984. “Organizational Demography and Turnover in Top-Management Groups”, Administrative Science Quarterly, 29 (1): 74-92

Welford, A.T., 1977. “Serial reaction times, continuity of task, single channel effects, and age”, Attention and Performance VI, Erlbaum, Hilsdale: 79-97

Weigelt, K., Dukerich, J., 1989. “Reactions to Discrimination in an incentive Pay Compensation Scheme: A Game-Theoretic Approach”, Organizational Behavior and Human Decision Processes, 4: 26-44

6. Appendices

Model 1: using only firm size and industry as controls:

rho ...4.449499955555575577733338884844 (fraction of variance due to u_i)4

sigma_e ...2.2252555777722422444000000080888 sigma_u ...2.222555544944999777722292999 _cons ---.-..1.11111111111774774441111111 1 ....11811882822121511555111166 66 ----00.00...66661111 0000....55455444000 0 ----....44644668688787577555111177 77 ...2.242244455552262266969959555 ind08 ...0.00088788777111111111113133 3 ....1111112112202004044444444444 0000..7..7778888 0000....44443373377 7 --.--..1.111332332242449491991611666 ...3.333006006667777111414424222 ind06 ...3.33355555555888855055003033 3 ....2222117117777775756556661111 1111..6..6663333 0000....11110020022 2 --.--..0.000770770090994943443933999 ...7.777882882226666444444454555 ind05 ...0.00011411444666655055007077 7 ....1111110110060667679779991111 0000..1..1113333 0000....88889959955 5 --.--..2.222002002222227276776466444 ...2.222331331115555777777787888 ind04 ...0.0008888888855455444444545 55 ....114114844883833322223333 0000....66606000 00.00...55555515511 1 --.--..2.222002002212116163663333333 ...3.333779779992222555252222222 ind03 ...0.00066566555777766166116166 6 ....1111112112232331318118889999 0000..5..5559999 0000....55555585588 8 --.--..1.111554554434337379779499444 ...2.222885885559999000202272777 ind02 ...0.00044344333666677977993933 3 ....0000991991191992921221112222 0000..4..4448888 0000....66663353355 5 --.--..1.111336336646448482882922999 ...2.222223223338888444141151555 ind01 ...1.11155755777444411911998988 8 ....1111337337757554548448886666 1111..1..1114444 0000....22225525522 2 --.--..1.111112112212117170770600666 ...4.444227227770000111010020222 boardsize ...0.01001112222442442228883833 3 ....00100117177171911999555544 44 0000..7..7772222 00.00...44447777000 0 ----....000202212112127227477444 ...0.000446446661111333030070777 lnfirmsize ...0.03003334444001001117777777 7 ....00100114144545455444333399 99 2222..3..3334444 00.00...00001111999 9 ....0000000505555551512112222222 ...0.000662662225555222323323222 lnPREMIUM Coef. Std. Err. z P>|z| [95% Conf. Interval] corr(u_i, X) = 0000 (assumed) Prob > chi2 = 0000....0020022424474777 Random effects u_i ~ GGaGGaaauuusussssssisiaiiaaannnn Wald chi2(9999) = 11119999...0.0050555 overall = 000.0...11111111333399 99 max = 4444 between = 000.0...11117777777777 77 avg = 44.44..0.000 R-sq: within = 00.00...0000000000500555 Obs per group: min = 4444 Group variable: iiididd d Number of groups = 111010000000 Random-effects GLS regression Number of obs = 444040000000

Model 2 Breusch-Pagan test: Prob > chi2 = 0000..0..000000000000000 chi2(1) = 111177277222....33333333 Test: Var(u) = 0 u 4 44477779999....1111666688885555 22221111....88888888999999992222 e 3 33322226666....0000666688888888 11118888....00005555777733338888 roe 999999996666....2222888855559999 3 3331111....555566664444 Var sd = sqrt(Var) Estimated results:

roe[id,t] = Xb + u[id] + e[id,t]

Breusch and Pagan Lagrangian multiplier test for random effects

Hausman test:

Prob>chi2 = 0000....0000888888881111

= 11112222....44440000

Model 2: using premium instead of lnpremium as predictor variable Breusch-Pagan test: Prob > chi2 = 0000..0..000000000000000 chi2(1) = 111166766777....33330000 Test: Var(u) = 0 u 4 44466666666....3333999900008888 22221111....55559999666600008888 e 3 33322227777....0000222200005555 11118888....00008888333377771111 roe 999999996666....2222888855559999 3 3331111....555566664444 Var sd = sqrt(Var) Estimated results:

roe[id,t] = Xb + u[id] + e[id,t]

Breusch and Pagan Lagrangian multiplier test for random effects

Hausman test:

Prob>chi2 = 00.00...1111333399998888 = 9999....66666666

chi2(6666) = (b-B)'[(V_b-V_B)^(-1)](b-B) Test: Ho: difference in coefficients not systematic