1
University of Amsterdam
Risk and rewards: relating managerial
characteristics to compensation
Terence Speijer
Student number: 10465529
Master thesis 15 ECTS
MSc Business Economics
Specialization: Managerial Economics and Strategy
Supervised by: prof. dr. E.J.S. Plug
AbstractA large strain of literature on pay-for-performance focuses on the relationship between compensation and company performance. To offer new insights, I investigate the relationship between variable pay and risk-aversion as scarce empirical evidence exists. This study replicates the survey methodology of Graham et al. (2013) and targets Dutch managers and directors to compare results to US CEOs. I find that there is a positive relationship between risk-aversion and fixed pay which is economically meaningful. Other significant effects are found for agents who are averse to sure losses.
2
This document is written by Student Terence Speijer who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 Table of Contents
1 Introduction ... 4
1.1 The debate on bonuses ... 4
1.2 Research question ... 4
1.3 Remuneration packages... 6
1.4 Corporate Governance... 7
1.5 Incentives and risk ... 8
2 Literature review ... 9
2.1 Standard principal-agent model ... 9
2.2 Agency problems: how variable pay can induce more risk-taking ... 10
2.3 Remuneration contracts: how design can attract risk-averse agents ... 11
2.4 Empirical evidence on managerial traits and compensation ... 15
3 Methodology ... 17
3.1 Data description... 17
3.2.1 Measuring risk-aversion ... 22
3.2.2 Measuring optimism, time preferences and aversion to sure losses ... 23
3.2.3 Measuring volatility and cost of effort ... 24
3.3 Model and empirical strategy ... 26
3.4 Hypotheses ... 27 3.5 Survey delivery ... 28 4 Discussion... 29 4.1 Regression results ... 29 4.2 Limitations ... 35 5 Conclusion ... 37 Bibliography ... 38 Acknowledgements ... 40 Appendix ... 41
4 1 Introduction
1.1 The debate on bonuses
Over the past few years, it seems a trend is developing as newspapers report on renewed distress over executive pay. Critics say that executive pay has become too complex as there are many factors that determine an executive’s total compensation. In response to this renewed distress, the Dutch government adopted a new law1 which limits variable pay to 20% of base salary and is applicable to all financial enterprises in order to limit exposure to excessive risk-taking. The Dutch Minister of Finance claims that the bonus culture in the financial industry has contributed to excessive risk-taking in order to “maximize short-run profits”. In other words, he implies that a remuneration package that contains a high proportion of variable pay induces agents to take more risk. However, another possibility is that firms design their remuneration packages in such a way to attract agents with a certain appetite for risk which is the aim of exploration in this thesis. 1.2 Research question
According to the literature, incentives need to be balanced in a principal-agent model. Specifically, incentives may need to be weak within firms. In a standard principal-agent model (Baker, Gibbons and Murphy, 1994), it can be shown that it is optimal for a risk-averse agent to have a lower bonus2. This is also known as the incentive-intensity principle, which is derived in section 2.3, and states that there is a trade-off between incentives and risk-appetite. All else equal, the variable part of compensation should be lower if the degree of risk aversion is higher. Ross (2004) has shown that it is increasingly costly to incentivize agents with a higher degree of risk-aversion, given that their compensation package contains incentives.
Graham et al. (2013) stated that there has been scarce empirical evidence on the trade-off between risk and incentives which is why my thesis can contribute to the compensation literature. To be specific, the aim of my thesis is to investigate whether risk-averse managers have less variable pay following the methodology of Graham et al. (2013). Possible rationales for higher incentives (target compensation) can be explored taking risk-attitude and other factors into
1 In February 2015, the law “Reward policy financial enterprises” (translated from Dutch) was adopted by Minister of Finance Jeroen Dijsselbloem.
5 account. The contribution of my thesis to the literature is twofold: first, as far to my knowledge no other paper has empirically examined the incentive-intensity principle among the Dutch population which illustrates the importance of replicating the study of Graham et al. (2013) in order to test its external validity. As became clear from the adopted Dutch law, there is an urge to put stricter requirements on variable pay so the results of this thesis can help to understand how firms may attract, select and retain executives as part of their remuneration policy. Specifically, risk-tolerant executives might be attracted to firms offering high variable remuneration while risk-averse executives might be attracted to firms offering high fixed salary. In the context of this thesis, the aim is to investigate whether different remuneration contracts with higher variable pay attract less risk-averse managers, and not whether managers behave riskier due to higher variable pay. This is because a survey methodology is used which captures only one moment in time. Therefore, the extent to which risk-attitudes are aligned between different remuneration contracts can be compared. Second, I am contributing by introducing two new variables in the model of Graham et al. (2013). These are volatility and cost of effort which is in line with the incentive-intensity principle. I provide a more elaborate explanation in the methodology section.
Although the field of executive remuneration is well developed, behavioral effects such as risk-appetite provide unexploited opportunities for new research. Based on the above, I formulate the following research question: To what extent does the degree of risk-aversion depend on the variable component in a remuneration package?
I find empirical evidence for risk-averse agents being more likely to choose fixed remuneration so there is a positive relationship between risk-aversion and fixed pay. The results are robust with a non-context specific measure of risk-aversion but there are no significant effects for higher degrees of risk aversion. Furthermore, I find significant effects for aversion to sure losses.
For the interested reader, I provide more background information in the remainder of the introduction section on the following: how remuneration packages of Dutch CEOs are comprised, why (Dutch) companies need corporate governance and why an agent may face a trade-off between risk and incentives. The goal of these subsections is to provide intuitive information on the design of remuneration packages and how the choice of a remuneration package may depend on an agent’s risk-attitude. The rest of this thesis is set up as following: Section 2 discusses the most important principal-agent and compensation literature, Section 3 describes the data, explains the survey
6 mechanism and formulates the hypothesis, Section 4 presents the results and discusses the limitations and Section 5 concludes with the most important findings.
1.3 Remuneration packages
Section 1.3 describes remuneration packages as commonly observed for Dutch executives. The goal of this section is to provide more information on the complexity of remuneration packages which are publicly available. In order to explain remuneration packages, I have analyzed most Dutch companies that are publicly listed.
Every year public companies have to release an extensive annual report in which information is disclosed to the owner(s) of the firm: shareholders. Part of the information is on remuneration packages which discloses how executive board members and supervisory board members are paid. Moreover, the supervisory board typically discloses how a remuneration package is designed and may motivate why a certain remuneration package is chosen, this is also known as the remuneration policy. Such a remuneration package may consist of four elements: fixed wage, bonus, stock and options. The latter three are also referred to as variable remuneration. When analyzing annual reports of Dutch public companies, it is common to divide variable remuneration in two parts: short-term incentive pay (bonus) and long-term incentive pay (stock and options). Earlier I discussed that critics say that remuneration has become too complex. To elaborate on that point, variable remuneration can depend on many different factors. In a two-tier structure, the supervisory board determines these targets and may depend on both financial and non-financial targets. In a one-tier structure, the board of directors determines the remuneration scheme which effectively means that an executive director may set his own pay. This may cause a remuneration scheme to be biased.
It is common that actual pay differs from the remuneration policy, due to variable components. To determine the actual pay of executives, the current remuneration schemes are usually benchmarked against a peer group. Complications arise when one considers the long-term incentive variable pay. Most remuneration policies prescribe that executives are awarded shares and/or options conditionally. After two or three years, the company results are benchmarked against the relevant peer group. If performance has been sufficient, the conditionally awarded shares and/or options vest. But executives need to hold their awarded shares/options for another two years (holding period) to align interests with shareholders. Otherwise, executives have an incentive to maximize performance after three years and cash out at the expense of shareholders.
7 So, based on annual reports one can determine the pay-out ratio of a given executive. For example, a remuneration package for a given executive may be comprised as: 50 per cent base salary, 10 per cent short-term incentive pay and 40 per cent long-term incentive pay. Consequently, one can determine whether a trend can be observed over several years. However, it is not always clear how many shares/options vest for an executive in a given year as it may not be disclosed. Either one can defer the long-term incentive pay based on other information or long-term incentive pay cannot be determined. In order to judge whether an executive has an incentive to increase his/her payoff, it is important to identify long-term incentive pay. Then an analysis can be made to determine whether the pay-for-performance relationship leads to desirable outcomes.
1.4 Corporate Governance
One may wonder why corporate governance exists and why it is needed. As a response to the debate on bonuses, several organizations provide guidelines on corporate governance. Moreover, independent organizations and committees check whether corporations comply with pre-set rules of corporate governance. The goal is to align all interests involved and to provide more transparency. With respect to the Dutch Corporate Governance Code, it contains principles and best practice provisions that regulate relations between the management board, supervisory board and shareholders. The best practice provisions formulate a set of standards which govern the conduct of management board members, supervisory board members and shareholders3. It should be noted that listed companies may deviate from the best practice provisions and may justify why under certain circumstances. The Code works on a principal basis of either compliance or quality of explanation.
One of the results of the Dutch Corporate Governance Code is that public companies report on their remuneration policy. According to the Dutch Code, the supervisory board plays an oversight role and the remuneration policy must be aimed at creating long-term value for shareholders. However, according to the Code the reward structures have often been too complicated which leads to less transparency. Moreover, the Code has shown that companies most often do not comply with the best practice provisions concerning the reward structure4.
The Dutch Corporate Governance Code was first introduced in 2003 and amendments were made in 2008. During 2015, the Code was revised again by an appointed committee to ensure that
3 Dutch Corporate Governance Code: principles of good corporate governance and best practice provisions. 4 Proposal revision of the code: II.5 Remuneration: Cleaned up and simplified.
8 the Code is in line with current topics and remains relevant. Note that executive contracts are also subject to these regulations meaning their contracts have to comply with new rules. One of the topics addressed in the proposal revision is remuneration. A company’s remuneration policy should be simple and transparent while it should also promote long-term value creation. The committees state that management board members and supervisory board members have to be aware of the public sensitivity on remuneration and take their responsibility in this regard. In the context of risk and incentives, a clearer policy on disclosure of remuneration can help to understand why firms design their relevant remuneration packages. Because the revised Dutch Corporate Governance Code aims to create long-term value, remuneration packages should contain incentives that achieve this goal.
1.5 Incentives and risk
Generally, variable remuneration can give an agent incentive to maximize payoff as his/her payoff depends on performance of set targets. Economic theory states that a rational agent will always maximize payoff even if it is at the expense of another. Therefore, a rational agent will try to maximize performance by optimizing actions. If these actions are undesirable for the principal because they destroy value, they are classified as moral hazard. In other words, an agent might behave opportunistically when the agent is given incentives to maximize performance. Alternatively, the agent may be incentivized to maximize performance based on his risk-attitude given his remuneration package.
Effectively, an agent may face choosing between maximizing his/her own payoff or maximize value for the firm. Remember that a rational agent will always choose to maximize own payoff. However, there is extensive evidence from behavioral economics that characteristics may lead to outcomes different from predicted by economic theory. Section 2.4 discusses empirical evidence on the relationship between managerial characteristics and compensation. Consequently, an agent may choose not to maximize own payoff because he/she cares for the principal, for example. Concerning remuneration schemes, it is not so clear what motivates an executive to take relevant decisions that affect his/her total remuneration which is why the dynamics of decision-making of an agent facing remuneration incentives provides interesting areas for research. It could be that decision-making depends on an agent’s risk attitude because the agent has to increase risk-taking to maximize own payoff. In other words, a behavioral characteristic (risk-aversion) can determine whether a manager accepts a remuneration package that provides incentives because
9 variable pay is risky in the sense that it might not be obtained due to unachieved targets. Consequently, an agent might be averse to this riskiness and is more likely to choose fixed income. In short, the incentives faced might differ per agent based on his/her risk-attitude. As far to my knowledge, there is little evidence from the field of behavioral economics on the trade-off between risk and incentives concerning remuneration. Therefore, it is interesting to research whether bonuses attract managers with a certain appetite for risk because it may be less costly for firms to incentivize these managers to take risk (Ross, 2004). It is important to try to identify a mechanism because it can have implications for how firms design their remuneration packages. For example, if firms desire that agents take less risk, they can design remuneration packages with a high proportion of fixed pay to attract risk-averse agents.
2 Literature review
The literature review discusses the most important literature with respect to the relationship between variable pay and risk-aversion which is the aim of investigation of this thesis. Section 2.1 discusses a standard principal-agent model which lies at the basis of the relationship between risk and incentives and considers the conflict of interest that may arise in this model. The main contribution of my thesis is to examine to what extent variable pay depends on the degree of risk-aversion. If a relationship is identified, firms are able to implement less costly incentive schemes to attract agents who are risk-tolerant (Ross, 2004). To be complete, however, Section 2.2 discusses why an agent might be induced to take more risk given his or her remuneration package. Section 2.3 continues by discussing remuneration contracts and the most important corresponding principles. One of these principles, the incentive-intensity principle, lies at the basis of the empirical strategy and is derived formally. Finally, Section 2.4 considers empirical evidence on the relationship between variable pay and risk-aversion.
2.1 Standard principal-agent model
In the introduction, I briefly discussed the implications of a standard principal-agent model and what theory predicts. In this section, I will elaborate on how a principal-agent model may be modelled following Gibbons (2010). In a simple model, there is one principal and one agent. The basic idea is that the principal is too busy to do a given job so he hires an agent. However, the principal cannot monitor the agent perfectly. Consequently, the principal needs to write a contract which incentivizes the agent to maximize value for the principal. In other words, the agent needs
10 to receive a reward so he will provide effort. Gibbons (2010) uses the following definitions with respect to rewards, effort and incentives: “rewards are outcomes that people care about (not just money), effort means actions that people will not take without rewards and incentives is the link between rewards and effort”. If the principal desires that the agent maximizes value, the contract has to be efficient. However, a contract may also give the wrong incentives to the agent such that he will engage in moral hazard.
Moreover, Holmstrom (1979) has early recognized this problem and argues that information asymmetry causes the incentive problem. When the agent cannot be fully observed, a first-best solution is infeasible because unobservable individual actions cannot be contracted upon. One can create additional information systems or use other available information to assess the actions of the agent and effectively improve the contract. This has led to a principle known as the informativeness principle in contract design which states that any performance measure that contains information about the effort level chosen by the agent should be included in a compensation contract. In the context of executive remuneration, performance measures may use both financial and non-financial information to determine part of an executive’s remuneration. To elaborate on the notion that the principal cannot monitor the agent, a method for analyzing problems in a principal-agent model has been formalized by Grossman and Hart (1983). Some of the assumptions are that agents are risk-neutral and the principal is fully informed about the agent and about the firm’s production possibilities. According to them, the main incentive problem is that the principal cannot monitor the agent’s actions. As a result, their analysis concludes that incentive schemes are never optimal if both the principal’s and agent’s payoff are negatively related. They also argue that two factors determine the seriousness of the incentive problem. One determinant is the quality of information about the agent’s actions. The worse the quality, the bigger the loss in utility will be to the principal. Another determinant is the degree of risk aversion. No incentive problem arises when the agent is risk-neutral but an incentive problem does arise when the agent is risk-averse. However, it is ambiguous to what extent the principal incurs a loss in utility for different degrees in risk-aversion.
2.2 Agency problems: how variable pay can induce more risk-taking
Before discussing how remuneration packages can attract risk-averse agents, I first review literature on how incentives of variable pay can induce an agent to take more risk. Much of the debate can be traced back to the 1970’s when there was a growing literature on agency problems
11 (Alchian & Demsetz, 1972, Jensen & Meckling, 1976, Holmstrom, 1979). Companies responded to agency problems by awarding executives share options. According to Hall and Liebman (1997), 57 per cent of US CEOs received stock options in 1980 while this part of compensation increased to nearly 90 per cent by 1994. Their finding demonstrates how equity incentives have become an important part of CEO compensation. The idea is that stock options incentivize managers to ensure that stock performance is high on the long-term. However, it is often argued that stock options have an adverse effect as managers attempt to increase risk-taking to increase the share price when their options become exercisable. Consequently, shareholder value may be destroyed while the executive maximizes his/her payoff. This is one example when variable compensation does not align all interests in a principal-agent model as it causes outbursts among shareholders.
Pay-for-performance has been extensively researched in academics across many disciplines including accounting, finance and behavioral economics. Most notably, the agency problem was formalized by Jensen and Meckling (1976) who identified that the root of the agency problem lies at the separation of ownership and control. Their theory helps to explain why a manager of a firm does not maximize the total value of the firm but total value would be higher if he were the sole owner because then he is better incentivized to maximize firm value. It leaves to question whether managers that do not maximize firm value do maximize their own payoff at the expense of firm value. To ensure that the agent does not engage in moral hazard, the principal can make an agreement with the agent but she will incur agency costs. Their theory illustrates an interesting point: agency problems can be mitigated against agency costs such as the cost of monitoring. Their theory focuses less on compensation incentives and more on the incentives faced when the principal and agent have entered into a relationship. An amendment to the theory would include how appropriate compensation incentives can be best provided. Instead, they assume that individuals can structure appropriate incentives in a compensation contract such that the principal’s payoff is maximized despite the existence of uncertainty and imperfect monitoring. To summarize, there are different factors that determine the incentive problem such as information asymmetry, the quality of information and the degree of risk-aversion. In order to reduce the incentive problem, it becomes necessary to write a contract that provides the right incentives for the agent. In the next section, I review the literature on compensation contracts. 2.3 Remuneration contracts: how design can attract risk-averse agents
12
components. These are a base salary, an annual bonus tied to accounting performance, stock options, and long-term incentive plans. Jensen, Murphy and Wruck (2004) describe a well-designed remuneration package. It attracts and retains agents at the lowest cost and it also motivates agents to take actions that increase long-run value for the principal and avoid actions that destroy value. Furthermore, well-designed packages take the riskiness of the pay package into account. Bonuses typically depend on performance and are inherently volatile so when risk-averse agents are offered a basic remuneration package, they do not want to bear the expected costs associated with the riskiness of the remuneration package. The principal incurs the expected costs only if the expected benefits exceed costs.
Compensation contracts can be designed in many ways. Prendergast (1999) reviewed mechanisms such as piece rates, options, bonuses, promotions, efficiency wages and deferred compensation. This analysis of the costs and benefits of incentives in a compensation contract showed that there is little evidence that contracts are designed according to theory. In other words, contracts are not optimal in the sense that incentives do not induce firm value maximizing behavior. Economic theory provides an explanation by stating that there are differences in the quantity and quality of information available about the performance of an economic agent. In the previous paragraph, I referred to information asymmetry and mentioned the informativeness principle. To be complete on contract design, the related literature identifies two other principles. These are the incentive-intensity principle and the equal compensation principle. In the context of my thesis, I briefly review literature on the equal compensation principle while a more elaborate review is done on the incentive-intensity principle.
The equal compensation principle has been put forward by Holmstrom and Milgrom (1990). They formally analyze a principal-agent model which allows for an agent to carry out multiple tasks. The equal compensation principle states that the agent must spend the same marginal rate of time for all activities carried out. In terms of compensation, the agent must value all activities equally for he would otherwise shift attention to the task he values the most.
The incentive-intensity principle describes the trade-off between variable compensation and cost of effort, degree of risk-aversion and the precision of performance evaluation as prescribed by Prendergast’s (1999) model. Essentially, a trade-off between risk and incentives must be made when the agent is risk-averse. An agent is more exposed to risk when stronger incentives are provided. Therefore, the principal must increase the variable part of compensation
13 to ensure that the agent provides effort. However, the principal bears the cost of increasing compensation. As a result, the incentive-intensity principle states that variable compensation is lower when agents have a higher degree of risk-aversion. Furthermore, variable compensation is also lower when the cost of effort is higher or when the precision of performance evaluation is lower to ensure participation by the agent. In other words, there are three factors that determine the optimal bonus which are the degree of risk-aversion, cost of effort and volatility of the environment, and can be modelled as following: 𝑏∗= 1
1+𝑟𝜎2𝑐 where r is the degree of risk-aversion, σ2 is the precision of performance evaluation and c is the cost of effort. Through the remainder of this thesis, this relationship is referred to as the incentive-intensity principle.
On the contrary, Prendergast (2000) shows reasons why one might expect a positive relationship between risk and the provision of incentives. In the context of risk-neutral agents, he demonstrates a positive relationship under the following conditions: first, monitoring can replace output-based contracts. He states that output-based contracts are commonly used in agency theory so substituting those may affect agency theory when the marginal rate of cost of monitoring depends on environmental uncertainty. Specifically, he shows that input monitoring is less effective in uncertain environments. Second, he discusses sorting based on performance evaluation. Supervisors can distort evaluations to reward favored employees while they also need to allocate talented workers to certain tasks. In a risky environment, it is more likely that the supervisor’s evaluation about an agent is noisy. Noisy measures result in a lower marginal cost to the firm of distorted evaluations. In other words, the value of information in the context of sorting is low so it is less effective in uncertain environments. Third, he argues that investigations are less effective in uncertain environments. To arrive at the positive relationship between risk and incentives, he rejects the assumption that a principal can get costless signals and always monitors an agent. Instead he argues that endogenous events such as customer complaints can trigger investigations. In noisy environments, the agent knows that performance evaluation remains vague as no clear distinction between truth and false can be made. To summarize, the conditions of Prendergast (2000) are as following given that environments are uncertain: monitoring, sorting and investigations can lead to a positive relationship between risk and incentives as incentives are lower in riskier environments. Consequently, the principal needs to induce the agent to exert effort. Another reason for a positive relationship has been put forward by Holmstrom (1999) who shows that career concerns can increase an agent’s willingness to exert effort in order to generate a
14 reputation. In uncertain environments, career concerns may incentivize an agent less to exert effort because his labor supply is less informative to the principal. As a result, the principal has to increase incentive pay to ensure participation by the agent. Taken all together, standard agency theory states that there is a negative relationship between risk and incentives. On the contrary, Prendergast (2000) and Holmstrom (1999) have demonstrated reasons why a positive relationship may be expected. Therefore, these reasons can explain empirical results on the risk-incentive trade-off. However, it less clear how these reasons can affect testing hypotheses because uncertain environments can be broadly interpreted. For instance, uncertain environments can range from a supervisor’s distorted performance evaluation to worldwide economic recessions, both which can affect an agent’s variable pay. Therefore, I would have liked to see a clearer definition on uncertain environments.
So far I have discussed different contract designs while briefly mentioning that conflicts of interest such as moral hazard may exist in a compensation contract. Other problems may arise because of differences in the quantity and quality of information available. Such information problems typically come forward when performance measures are used to determine an agent’s compensation. In this case, the performance measures can be subjectively assessed or objectively determined. A combination of both is also possible. Prendergast (1999) notes that objective performance measures are verifiable ex-post while subjective performance measures are non-contractible and cannot be verified by a third party. The total pay that results from the agent’s effort can also be referred to as variable compensation. The main idea is that a firm’s remuneration policy aims to attract, motivate and retain the best-qualified agents. In order to do so, variable compensation must be designed such that the agent’s interest is aligned with the firm’s best interest. That is, the agent maximizes own payoff whilst increasing firm value. That being said, performance measures in a compensation contract determine whether a firm’s remuneration policy is able to serve both the agent’s best interest as the firm’s best interest. Baker, Gibbons and Murphy (1994) studied subjective performance measures in optimal incentive contracts. In their study, they distinguish implicit contracts (not enforceable by court) from explicit contracts (enforceable by court) and find that both are substitutes and complements depending on the circumstances. An explicit contract can replace an implicit contract if the objective measure(s) are close to perfect while they may be complements if the firm does not have the fallback option to renege on an implicit contract. Their formal analysis of performance measures has important implications. For
15 example, sufficiently efficient explicit contracts may use only use objective performance measures given that these are close to perfect. Typically observed performance measures of executive board members may depend entirely on financial metrics. As these are considered objective because they can be verified by a third party, their model predicts that they are close to perfect. However, a measure such as total shareholder return can be noisy as it depends on market performance. In other words, objective performance measures are imperfect which results in distorted incentives. On the other hand, their definition of an ideal performance measure is that “it would reflect an employee’s contribution to firm value, including both static externalities across business units and dynamic effects of current actions on long-run value”.
To put it in perspective, compensation contracts come in many different forms but they share the common idea that they are designed such that it can induce an agent’s behavior to maximize long-run value for the firm. This section has demonstrated that the optimal bonus depends partly on an agent’s risk-attitude and that a trade-off exists between risk and incentives. To better understand the relationship between risk and incentives, I review the empirical literature on managerial traits and compensation in the next section. In particular, I consider empirical evidence on the relationship between the degree of risk aversion and variable compensation. 2.4 Empirical evidence on managerial traits and compensation
There are many factors that determine executive remuneration. For instance, observable factors such as company performance and size can play a role. Also observable managerial characteristics such as job tender and age may help to determine a remuneration package. A large strain of literature examines managerial traits in relation to company performance (see Bertrand and Schoar (2002), Hackbarth (2008) and Kaplan et al. (2012)). However, there is scarce empirical evidence on managerial traits and variable compensation as managers may have some discretion over their remuneration package. Ceteris paribus, the incentive-intensity principle predicts that the variable part of compensation is lower when the degree of risk-aversion is higher. Graham, Harvey and Putri (2013) have found empirical evidence that supports this prediction. In order to obtain managerial traits, they use a survey with statements on life-attitudes to measure optimism while other questions gauge a person’s degree of risk aversion, aversion to sure losses and whether a CEO is impatient. They have found significant effects on the degree of risk-aversion and rate of time preference. For instance, a CEO who is highly risk-averse is more likely to receive a high salary while the marginal effect of being impatient leads to a decrease of having high variable
16 remuneration. While their study examines characteristics such as optimism and rate of time preference, the trade-off between risk and incentives is not fully explored. As I explained earlier, the incentive-intensity principle predicts that the optimal bonus is also dependent on the precision of performance evaluation and cost of effort. Their empirical study does not take these two factors into account.
When one examines the remuneration packages of executives, it becomes clear that a CEO typically earns more than other members of the executive board. A possible explanation is that observable firm characteristics are part of a firm’s remuneration policy which, in turn, reflect the suitable compensation for different positions. Moreover, Murphy and Zabojnik (2004) examined executive pay over a time period and point out two CEO trends. First, a rising trend in US CEO pay has been observed. Second, companies tend to hire outside CEO’s rather than promote internally. To reconcile these trends, they developed a new model in which general managerial ability is transferable to specific firm capital. The rising CEO pay trend can be explained by general managerial ability being priced while specific firm capital is unpriced because it is easily accessible nowadays. For instance, a manager does not have to spend a couple of years in a firm to acquire specific product information. So, there is an increase in the importance of general ability as opposed to firm-specific knowledge. In short, their model shows that general managerial ability is an important factor in determining executive pay.
There is a considerable amount of literature that examines the relationship between managerial characteristics and firm policies such as investment policies, debt policies and firm risk. For instance, see Bertrand and Schoar (2002), Coles et al. (2006) and Malmendier et al. (2011) for empirical evidence on manager fixed effects. However, there is less literature on the relationship between managerial characteristics and executive pay. Graham, Li and Qiu (2012) have examined the role of manager heterogeneities on executive pay. They found that time-invariant manager fixed effects explains a majority of the variation in executive pay which supports the view of Murphy and Zabojnik (2004). Time-invariant managerial fixed effects can be as broadly interpreted as potential managerial ability. However, it is not clear to what extent general managerial ability applies to time-invariant characteristics such as education, leadership and talent. Therefore, a more thorough investigation of these characteristics is needed to find out how they influence variation in executive pay, if any.
17 3 Methodology
3.1 Data description
This section discusses the data, summary statistics and correlations of covariates first, proceeds to the measurement of variables followed by the empirical strategy. Then, the hypotheses and survey delivery are discussed.
I gathered data by asking respondents to participate online in the survey which has led to a total sample size of 180. The survey was open for a month, took approximately 5 to 10 minutes to fill out and an example can be found in the appendix. The sample can be divided in two groups: a relatively high proportion of variable remuneration and a relatively high proportion of fixed salary. It is useful to include a second group for a more robust analysis. Data on remuneration as a percentage of base salary is gathered and behavioral characteristics are measured using statements on life-attitude and questions. The behavioral characteristics included are the degree of risk-aversion, optimism, rate of time preference, aversion to sure losses, cost of effort and (perceived) volatility. Linking remuneration to behavioral characteristics allows to investigate the incentive-intensity principle which was derived in section 2.3.
Table 1 on the next page shows the summary statistics to demonstrate how data can correspond to the incentive-intensity principle. In particular, the data of the Dutch sample is compared to the US sample of Graham et al. (2013). As can be seen, the summary statistics are comparable to the US sample as most people have a higher fraction of fixed pay, less than 10% is highly risk-averse and approximately 90% are males. Some differences appear with respect to the measurement of variable remuneration and the choice of control variables. First in the Dutch sample, variable remuneration is gathered by asking the target percentage of base salary while the US survey of Graham et al. (2013) asks the relative percentage as part of total remuneration. The reason is that peer review sessions5 with three independent managers had shown that asking variable remuneration as target percentage of base salary leads to less confusion from a respondent’s perspective. The mean of variable remuneration in the Netherlands is equal to 21.6% which implies that fixed salary represents the largest share of total remuneration and this is
5 With variable pay it is likely that total pay varies per year so that it becomes unclear which year is taken as a reference. Base salary varies less over years, so a reference to base salary results in more precise self-reported measures.
18 comparable to the mean of variable remuneration in the US sample which equals 36.7%. Since the target percentage of base salary is reported for variable remuneration, it is possible to calculate the relative proportion of fixed and variable remuneration as part of a total package which is discussed in section 3.3. Note that the maximum of variable remuneration reports 83.33% because of this definition. On the other hand, the minimum of fixed remuneration equals 16.67% so that total pay equals 100% (𝑡𝑜𝑡𝑎𝑙 𝑝𝑎𝑦 = 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒 𝑟𝑒𝑚𝑢𝑛𝑒𝑟𝑎𝑡𝑖𝑜𝑛 + 𝑓𝑖𝑥𝑒𝑑 𝑟𝑒𝑚𝑢𝑛𝑒𝑟𝑎𝑡𝑖𝑜𝑛). The table further mostly reports independent binary variables. The mean of binary variables can be interpreted as the relative proportion in the sample. Therefore, table 1 reports these as percentages. Comparable means are also found for individuals who are highly risk-averse, averse to sure losses and impatient. However, it seems that the percentage of very optimistic individuals is higher in the US than in the Netherlands. Other differences arise when control variables are compared. In the US sample, the average individual has a job tenure of more than 10 years while the Dutch have a job tenure of three to five years. Note that job tenure in the Dutch sample is a categorical variable. Other differences are that the Dutch sample uses a master degree as a control variable while the US reports a MBA degree. The reason is that MBA degrees are not as common in the Netherlands as the US. Taken all together, the summary statistics show comparable variables. Therefore, it is possible to compare the Dutch sample to the US sample but one may not make inferences based on the summary statistics.
19 T ab le 1 : P er son al charac te risti cs Summ ary stat ist ics D utch m anage rs and di re ctors Summ ary stat ist ics U S C EO s Obs . Me an S td. de v. Min. Me dian Ma x . Obs. Me an S td. de v. Min. Me dian Ma x . Va ria ble re mune ra tion (%) 180 21.6 17. 0 0 19.3 83.3 Va ria ble re mune ra tion (%) 793 36.7 26.2 0 35 100 Hig hl y risk -ave rse (%) 180 6.7 25.0 0 0 1 00 Hig hl y risk -ave rse (% ) 1008 9.8 29.8 0 0 100 Ve ry Optim ist ic (%) 180 40.0 49.1 0 0 1 00 Ve ry Optim ist ic (%) 992 80.2 39.8 0 100 100 S ure L oss (%) 180 10.0 30.1 0 0 1 00 S ure L oss (%) 861 8.5 27.9 0 0 100 Impatient (%) 180 35.0 47.8 0 0 1 00 Impatient (%) 996 32.9 47.0 0 0 100 Ma le (%) 180 90.6 29.3 0 100 1 00 Ma le ( % ) 1009 92.3 26.7 0 100 100 Age 180 49.2 7.4 22 51 64 Age 992 54.1 9.4 25 54 89 Te nure 180 2.8 1.1 1 3 4 Te nure 1011 10.4 8.4 0.5 8 56 Ma ster (% ) 180 47.7 50.1 0 0 1 00 MB A D egre e (%) 916 34.9 47.7 0 0 100 B usiness (% ) 180 31.7 46.6 0 0 1 00 F oc use d in F in. & A cc . (%) 986 16.0 36.7 0 0 100 Publi c (% ) 180 23.3 42.4 0 0 1 00 Publi c (% ) 785 11.5 N /A N /A N /A N /A P ersona li ty tra it s a nd b ac kg round info rma ti on of mana g ers and dire cto rs in t he Ne ther lands are c o mpar ed to US da ta a s re p orte d in Gr ah am et al. ( 2013) . F or c ompa rison re asons, t he sum mar y statis ti cs of the US s am ple do not r eport a ll va ri ables. Hig hl y risk -a ve rs e is a dumm y va ria ble tha t equ als one i f a pe rson p re fe rs to ha v e 100% c ha nc e to re ce ive curr ent i nc om e for li fe o ve r 50% c h anc e to re ce iv e twice c u rr ent income for li fe o r 50% c ha nc e to r ec eiv e 80% of curr ent i nc om e for li fe . Ve ry opti mi sti c is a dumm y va ri a ble c onst ru cted b y a ddin g six state ments that mea sure opti mi sm whe re ha lf a re opti mi sti c state ments a nd ha lf a re pe ssi mi sti c . S tate ments a re re cod ed a cc or ding ly . If a pe rso n sc or es 4 or hi g he r on a ve ra ge on a s ca le of 1 -5, tha t pe rson is c la ssif ie d a s ve ry optimist ic . Sure loss is a dumm y va ria ble w hic h mea sure s wh et he r a man age r or dir ec tor is av erse to a sur e loss . Impatient is a dumm y va ria ble tha t mea sure s wh ether a man age r or dir ec tor ha s a discount ra te gre ate r tha n 30% . On the le ft ha nd si de , t enure is a ca te g oric al var ia ble tha t m ea sure s how lon g a man ag er or dire cto r ha s be en in hi s/her posi ti on. In thi s c ase , i t i mpl ies a tenur e of 3 -5 y ea rs f o r the Dutc h sa mpl e. Ma st er is a dumm y v aria ble for e duc ati on, busi ne ss i s a dumm y va ri able f or a mana ge r’ s or d ire ctor’ s fi eld of stud y a nd publi c is a dumm y va ri able f or pu bli cl y li sted f ir ms.
20
Table 2 reports the correlations of both personality traits and background information of managers and directors in the Netherlands and US CEOs. Similar significant correlations are found in both samples. Some correlations are intuitive such as the positive correlation between the size of a firm and a firm being publicly listed. Apart from different significant correlations between the samples, the most interesting difference arises between the correlation of being very optimistic and aversion to sure losses. In the Dutch sample, very optimistic individuals are positively associated with being averse to sure losses while the US sample shows that impatient individuals are averse to sure losses. While both correlations are intuitive, the difference between the samples are not. It may be that managers and directors, who are averse to sure losses, overestimate the probability of getting a high return on investment in a project (Kahneman and Tversky, 1979). Optimistic US CEOs do not overestimate this probability and are less averse to a sure loss because they have more experience when it comes to making investment choices. Therefore, one may argue that US CEOs are more rational with respect to investment choices.
Comparing highly risk-averse to the other traits, it is noticeable that the correlations are negative and slightly significant for the correlation with very optimistic. However, the correlations are relatively small which is why no interaction variables are created. This also indicates that the degree of risk-aversion is an independent parameter unrelated to other managerial characteristics.
On another note, it may be that males are overrepresented in the Dutch sample because the proportion of males is equal to 90.6% while the CBS Netherlands reports that the proportion of Dutch managers is equal to 74%6. On the other hand, a similar proportion is observed for the US sample but both samples can only be compared to a certain extent. Looking at all variables, it seems that this sample can be representative for Dutch managers and directors because they receive variable remuneration, the mean age is 49 and 47.7% has completed a master’s degree.
Unfortunately, I was unable to find information that could confirm a representative population for variable compensation. However, the mean target percentages of both short-term and long-term variable remuneration of the sample are lower as compared to typical observed percentages of executive board members. For the target group, the mean target percentages are equal to 23.9% and 13.8%, respectively. This is lower than typical target percentages observed for
6 According to the CBS, there were 361.000 male managers in the Netherlands. For the female population, this is equal to 128.000
21 T ab le 2 : Cor re lation s Corre lati ons D utc h m anag ers and dire ctors Hig hl y risk -a ve rs e Ve ry Optim ist ic S ure L oss Impatient Ma ster Ma le Old P ubli c B usiness Hig hl y risk -ave rse Ve ry opti mi sti c -0.127* S ure L oss -0.015 0.181** Impatient -0.103 0.043 0.027 Ma ster -0.122 -0.032 -0.059 -0.119 Ma le 0.010 0.031 0.108 0.078 0.005 Old 0.116 0.052 -0.160** -0.125* -0.012 0.179** P ubli c 0.063 0.059 -0.053 0.091 -0.028 -0.226*** -0.058 B usiness 0.010 0.078 -0.068 0.026 0.114 -0.025 -0.061 0.133* B ig fir m 0.110 -0.037 0.118 -0.099 0.047 -0.175** -0.069 0.268*** -0.070 C orrelati ons US CEOs Hig hl y risk -a ve rs e Ve ry Optim ist ic S ure L oss Impatient Ma ster Ma le Old P ubli c B usiness Hig hl y risk -ave rse Ve ry opti mi sti c -0.013 S ure L oss 0.002 0.038 Impatient 0.039 -0.017 0.083** MB A -0.041 0.010 -0.073** -0.050 Ma le -0.004 -0.047 -0.006 -0.093 ** 0.048 Old 0.124** 0.014 -0.068** -0.040 0.020 0.081** P ubli c 0.015 0.001 0.007 -0.041 0.005 -0.000 0.020 F oc use d in fin. & a cc . 0.036 -0.014 -0.014 -0.082** 0.172** 0.064** -0.051 -0.012 > $1B f ir m -0.006 0.021 0.007 -0.011 -0.024 0.037 0.050 0.387*** 0.005 This t able show s the corr elations be twe en pe rson ali ty tra it s a nd b ac k g roun d infor mation of bo th t he Dutc h sa mpl e and the U S sa mpl e a s re porte d in G ra ha m et al. (2013 ). F o r compa rison re asons, not all US cor re la ti ons are re po rte d. Also, some control va ri ables dif fe r in how the y we re c onst ruc ted. F or inst anc e, siz e in the D utch sa mpl e is d enoted b y bi g fir m inst ea d of the va ria ble > $1B . B ig fir m is a dumm y va ria ble for fir ms that empl o y more than 1000 emp lo y ee s. All va ria bles ar e indepe nde nt and binar y . O ld is a dumm y va ri able fo r mana g ers or dir ec tors who ar e olde r tha n the me an ag e in t h e sa mpl e. S ig nific anc e a t the 10% leve l i s de noted b y *, 5% b y ** and 1% b y ***.
22 Dutch CEOs7 which verifies that the target percentages can be correct because CEOs typically earn excessively more. Moreover, Conyon et al. (2011) have shown that the pay mix of a Dutch CEO consists of 49% variable remuneration. But remember that variable remuneration is a self-reported measure in the survey and may deviate from policy values.
3.2.1 Measuring risk-aversion
It is generally hard to measure a person’s risk attitude which is why a survey can provide a solution. Barsky et al. (1997) have developed a valid measure of risk-aversion because it involves gambles over life time income. I follow the alternated version of Graham et al. (2013) which is a choice between having safe income for life or risky higher income for life. In particular, the respondent’s choice to measure the degree of risk-aversion (𝑟) can be modelled in the following tree map:
Respondents who answer 𝑟1 and 𝑟3 are classified as highly risk-averse while respondents who answer only 𝑟1 are classified as risk-averse. Dummy variables are used to classify the degree of risk-aversion because for choices 𝑟3, 𝑟4, 𝑟5 and 𝑟6 observations are missing due to redirected choices. Consequently, it is not possible to create a categorical variable of the degree of risk-aversion.
Dohmen et al. (2011) have shown that using a simple ordinal risk measure is statistically
7 Typical target percentages are accessible through annual reports of public companies. For the short-term target percentages typically range between 50%-100% while typical long-term target percentages range between 100%-300% for most AEX companies.
𝑟1: 100% chance that a new job pays current income for life
𝑟3: 100% chance that a new job pays current income for life
𝑟2: 50% chance that a new job pays twice current income for life and 50% chance that a new job pays 67% of current income for life
𝑟4: 50% chance that a new job pays twice current income for life and 50% chance that a new job pays 80% of current income for life 𝑟5: 100% chance that a new job pays current income for life
𝑟6: 50% chance that a new job pays twice current income for life and 50% chance that a new job pays 50% current income for life
23 valid to represent a risk attitude as if money is at stake. Although it is preferred to obtain a risk attitude that is context specific, they have shown that a general risk attitude is robust. Therefore, the survey obtains a general risk attitude by posing the following statement: “Generally, I am person who likes to take risk.” A Likert-scale of 1-5 is used with 1 being fully disagree and 5 fully agree. Persons who answer 1 or 2 are classified as risk-averse. Furthermore, Dohmen et al. (2011) have shown that such a general risk attitude can represent the same risk attitudes obtained in lottery experiments which typically involve gambles. I have included this measure to allow for a more robust analysis.
3.2.2 Measuring optimism, time preferences and aversion to sure losses
In order to measure these life-attitudes, the statements and questions of Graham et al. (2013) are followed and dummy variables are created in order to perform a logit regression.
Optimism is measured following the Life Orientation Test-Revised (LOT-R) designed by Scheier et al. (1994). It is a psychometric test that uses six statements on 5-point Likert scale to obtain optimism. According to Graham et al. (2013), it has been used extensively in the psychology literature which makes it a credible measure. The statements used are:
1. In uncertain times, I usually expect the best 2. If something can go wrong for me, it will 3. I’m always optimistic about my future 4. I hardly expect things to go my way
5. I rarely count on good things happening to me
6. Overall, I expect more good things to happen to me than bad.
As one can see, statements 1, 3 and 6 gauge optimism while the others gauge pessimism. To recode the values, statements 1,3 and 6 are numerically coded with a 5 if respondents answer ‘I agree a lot’ and 1 if they answer ‘I disagree a lot’. Statements 2, 4 and 5 are numerically coded with a 5 if respondents answer ‘I disagree a lot’ and 1 if they answer ‘I agree a lot’. The range of mean responses lies between 1 and 5. Accordingly, optimistic respondents are identified when they average 4 or higher. Considering the length of the survey, I have excluded four filter questions and try to measure volatility and cost of effort instead. The measurement of these variables is explained in the next section.
Time preference for gains measure whether a respondent is impatient. If respondents answer that (s)he would rather win €10.000 now than €13.000 a year later, (s)he is impatient
24 because it implies a discount rate of 30%. A dummy variable equals one if a respondent is impatient.
Aversion to sure losses is measured by posing the following question: suppose you lost €5 million in a project. Now you have the opportunity to invest an additional sum in this project. There is a 20% chance that the project generates €10 million and an 80% chance that the project generates nothing at all. How much are you willing to invest? If respondent answer at least €2 million, (s)he is averse to sure losses. A dummy variable equals one for aversion to sure losses, zero otherwise.
Firm characteristics will include variables such as size and industry and will serve as control variables.
3.2.3 Measuring volatility and cost of effort
In order to measure volatility and cost of effort, I use three statements on each which try to gauge these variables. As far to my knowledge, there were no other papers that try to measure volatility and cost of effort based on survey questions in the context of the incentive-intensity principle. Therefore, it is recommended that future research tries to academically validate these measures. Volatility relates to the uncertainty surrounding an agent. All else equal, an agent is more likely to receive higher variable compensation if the precision of performance evaluation increases. This means that if performance targets can be precisely measured, (that is, more certainty about the performance evaluation) agents are more likely to receive high variable compensation. Performance evaluations come in many forms and can be different for managers. Also, some managers might receive variable compensation that does not depend on performance targets. For instance, an agent may receive stock options that vest based on time conditions. Taking into account that not every person from the target group is subject to performance evaluation, I argue that uncertainty surrounding an agent may be representative of the precision of performance evaluation as the agent has to make a trade-off between uncertainty and variable compensation. The following three statements are used to measure volatility (𝜎):
𝜎1: “I do not have to worry about my financial situation over the next years” 𝜎2: “I find my personal economic circumstances unpredictable”
𝜎3: “I do not worry about things that are beyond my control”
Consider the first statement. Suppose that an agent has to choose between variable income and fixed income with variable income yielding the highest payoff, uncertainty about his financial
25 situation might influence his decision. A rational agent will always choose to optimize payoff but an agent might prefer certainty of receiving a lower fixed income rather than higher variable income. The second statement is used as a reference to the first statement to measure the consistency of people answering the statements as it may happen that questions or statements are misunderstood. The third statement is used to benchmark perceived uncertainty against general uncertainty. If all three are significantly correlated, these statements can be recoded to measure volatility. The first and last statement are numerically recoded with a 5 if respondent answer ‘I disagree a lot’ and 1 if ‘I agree a lot’. The second statement is numerically recoded with 5 if respondents answer ‘I agree a lot’ and 1 if ‘I disagree a lot’. If all statements are significantly correlated, volatility can be gauged. However, careful notion applies here because these statements have not been validated yet. Ideally, I would have included three more statements on uncertainty as has been done for optimism. Considering that the length of a survey plays an important role, I have decided to make the survey as short as possible.
Obtaining cost of effort through survey questions is not ideal and may not serve as a good measure. However, I try to contribute to the literature on the trade-off between risk and incentives by using the following three statements on cost of effort (𝑐):
𝑐1: “I expect to be rewarded reasonably for my actions”
𝑐2: “I care about the consequences of my actions in relation to my salary” 𝑐3: “I find it important to always exert the best effort I can”
As I discussed earlier, Gibbons (2010) defined effort as actions that people will not take without reward. Therefore, the first statement tries to capture this definition. The second statement tries to measure cost of effort in the sense that people who care about the consequences of their actions arguably may provide more effort than people who do not care about the consequences of their actions. As a result, people who care about their actions experience a higher cost of effort. Concerning the third statement, the idea is that it is indicative of how much effort people exert, all else equal. If it is relatively important for a person to always exert best effort, then the marginal cost of effort should be higher as compared to people who indicate that exerting best effort always is less important. For all statements, if respondent answer ‘I agree a lot’ the answer is numerically recoded as 5 and 1 if ‘I disagree a lot’. If all three statements are significantly correlated, answers can be recoded to use effort as a variable.
26 the degree of risk-aversion which is obtained in both specific and non-specific contexts. Risk-aversion equals one if a person is risk-averse, zero otherwise. In addition, measurements on volatility and cost of effort were included in the survey. Volatility equals one if a person perceives his or her environment to be volatile, zero otherwise. Cost of effort equals one if a person perceives his or her cost of effort to be greater than required, zero otherwise.
3.3 Model and empirical strategy
The research question partly addresses the theoretical framework of the optimal bonus and by conducting a survey, I try to measure the incentive-intensity principle. Moreover, the aim is to investigate whether a negative relationship holds between risk-aversion and variable pay by following the methodology of Graham et al. (2013) which uses a logit regression. In addition, multiple regressions are run in order to check for robustness of risk-aversion. Their regression examines the relationship between target compensation and risk-attitude and behavioral traits, such as time preferences and aversion to sure losses. As the survey measures capture only one moment in time, no causal inference can be made which is a limitation discussed in Section 4.2. Finally, the equation to be estimated is:
𝐻𝑖𝑔ℎ 𝑠𝑡𝑜𝑐𝑘, 𝑜𝑝𝑡𝑖𝑜𝑛𝑠 𝑎𝑛𝑑 𝑏𝑜𝑛𝑢𝑠𝑖 = 𝛽0+ 𝛽1𝐷𝑒𝑔𝑟𝑒𝑒𝑂𝑓𝑅𝑖𝑠𝑘𝐴𝑣𝑒𝑟𝑠𝑖𝑜𝑛𝑖 + 𝛽2𝑂𝑝𝑡𝑖𝑚𝑖𝑠𝑚𝑖+ 𝛽3𝐴𝑣𝑒𝑟𝑠𝑖𝑜𝑛𝑇𝑜𝑆𝑢𝑟𝑒𝐿𝑜𝑠𝑠𝑒𝑠𝑖+ 𝛽4𝐼𝑚𝑝𝑎𝑡𝑖𝑒𝑛𝑡𝑖 + 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝜀𝑖.
When the survey was conducted, I have asked questions on compensation, risk attitudes, behavioral traits, firm characteristics and background information. Most of these questions were mandatory to answer to ensure that enough observations were obtained. The statements are measured on a Likert-scale of one to five. The reason is that ordered logistic regression can then be used to categorize the outcomes based on the relative importance.
The target group of this survey is Dutch managers who receive variable compensation. To distinguish a high proportion of variable income from a low proportion, I first derive the relative proportion of variable and fixed remuneration as part of total pay. Note that variable remuneration (hereafter referred to as ‘variable’) is the sum of short-term incentives (bonus) and long-term incentives (stock and options) and that these are self-reported measures. Fixed remuneration can be interpreted as base salary and equals one because the survey asks to report ‘variable’ as a target percentage of base salary (hereafter ‘base’). Thus, the relative proportions are equal to: 𝑆𝑡𝑜𝑐𝑘, 𝑜𝑝𝑡𝑖𝑜𝑛𝑠 𝑎𝑛𝑑 𝑏𝑜𝑛𝑢𝑠 = 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒
(𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒+𝑏𝑎𝑠𝑒) and 𝑆𝑎𝑙𝑎𝑟𝑦 =
𝐵𝑎𝑠𝑒
27 remuneration is equal to 21.6% while the mean of fixed remuneration equals 78.4%. Accordingly, ‘high stock, options and bonus’ is a dummy variable for observations higher than or equal to 20% while ‘high salary’ is a dummy variable for observations higher than or equal to 78%. I use 20%, and not 21.6%, here because variable pay is relatively small so it is slightly relaxed to estimate the parameters more precise otherwise eleven observations would not be taken into account. Interesting to note is that Conyon et al. (2011) have analyzed executive pay across Europe and found that variable compensation makes up half of the average Dutch CEO pay mix which implies that the proportion of variable remuneration is more than twice as high as compared to Dutch managers and directions. Thus, substantial differences may arise as most respondents do not hold a CEO position.
The degree of risk-aversion can be expressed as either risk-averse or highly risk-averse as explained earlier. A dummy variable is used that equals one if a manager or director is highly risk-averse, zero otherwise. To be complete on the robust analysis, a regression with risk-aversion in a non-specific context is also run. Other traits are optimism, time preferences and loss aversion that equal one if one is very optimistic, impatient and averse to sure losses, zero otherwise.
3.4 Hypotheses
Based on the research question, the following two hypotheses are tested:
H0: There is no relationship between high stock, options and bonus compensation and highly
risk-averse.
H1: The relationship between high stock, options and bonus compensation and highly risk-averse
is negative.
To ensure that the obtained results can be verified, the independent variables are also benchmarked against a relatively high proportion of salary. The expected signs on all independent variables are reported in table 3 below. Concerning optimism, agents who are optimistic are expected to be more likely to opt for variable income because it can be argued that very optimistic agents have higher expectations of earning variable income. A more intuitive explanation is that very optimistic agents opt for variable income because they might be more positive about meeting performance targets that condition variable income. Aversion to sure losses is measured with a hypothetical gamble, which can be found in the appendix. If individuals want to invest more than 2 million euros, they are classified as averse to sure losses. As they already have lost money, sure loss averse individuals
28
can undertake more wasteful actions. Such wasteful actions might also occur when they are trying to maximize payoff by meeting performance targets. Therefore, the expected sign of aversion to sure losses on high stock, options and bonus is expected to be positive. Furthermore, impatient is measured as rate of time preference based on a gain question. The question poses whether an agent would rather win 10.000 euros now or 13.000 euros a year later. Persons who answer now are classified as impatient as it implies that their discount rate exceeds 30%. Variable remuneration can typically be split up in short-term incentives (bonus) and long-term incentives (stock and options) while fixed income is yearly, so impatient individuals are more likely to receive fixed salary. Therefore, the expected sign of impatient on high stock, options and bonus is negative.
Table 3: Expected signs
High salary, stock and options High salary
Coefficient Marginal effect Coefficient Marginal effect
Highly risk-averse - - + + Very optimistic + + - - Sure loss + + - - Impatient - - + + Cost of effort - - + + Volatility - - + + 3.5 Survey delivery
I have used Qualtrics as a survey medium and set it out via social media, colleagues, friends and family to ask managers and directors to fill out the survey which was closed after a month. In addition, I have contacted approximately fifty senior management recruitment firms and two have made it possible to create a sample size that is large enough to test the hypothesis and to test the most important independent variables. Approximately 93.5% of the respondents were delivered through these agencies who emailed their client base.
Two incentives were given to respondents: first, they were offered to receive a personal risk-profile based on (preliminary) results. If respondents chose this option, I would send them an email with a report describing their risk-profile and how they compare to other managers. In total, 105 people have requested such a risk profile which suggests that it is an effective incentive. Moreover, I only included this option after 40 respondents were gathered so the effectiveness could have been higher. However, a careful approach of the target group was needed because an
29 invitation to the survey can be seen as spam. A better approach for future research would be to offer personal feedback from the beginning. Second, respondents could win a gift card which was offered from the beginning but it seems that gift cards do not incentivize managers to fill out the survey.
4 Discussion
4.1 Regression results
Tables 4a and 4b report the regressions on managerial attitudes and compensation. Both tables provide a comparative look and both similarities and differences can be found. The sign of the coefficients is in both samples the same. This implies that the measurement of the dependent variables in the Dutch sample is similar to the US sample. Recall that Dutch respondents were asked to report variable remuneration as target percentage of base salary as opposed to a relative percentage of total remuneration. The latter can be inferred using the target percentages of base salaries. Differences arise when one considers significance of covariates because Dutch managers and directors who are averse to sure losses, are significantly more likely to choose variable remuneration. Specifically, the marginal effect suggests that sure loss averse managers are 27.6% more likely to opt for stock, options and bonuses. To the contrary, similar results are not found in the US sample but the degree of risk-aversion and rate of time preference do show significant effects. As I explained earlier, it may be that managers and directors overestimate the probability of getting a high return on investment (Kahneman and Tversky, 1979) while there is also a positive significant correlation with being very optimistic. Alternatively, simply being too optimistic may lead them to make riskier investment choices which can explain why they are more likely to choose variable remuneration. A common assumption in compensation literature is that firms undertake costly incentive schemes to induce managers to take risk. Agents who are averse to sure losses take riskier decisions despite known probabilities in this case. Therefore, it is less costly to incentivize these agents. On the other hand, individuals who are averse to sure losses can undertake more wasteful actions in order to meet performance targets which grants them variable remuneration if met.
This analysis was conducted using high stock, options and bonus as dependent variable and similar conclusions were reached using high salary as a dependent variable. To summarize, the signs of the coefficients show the same direction as the US sample while significant effects are
