1
Executive compensation during the financial crisis:
are CEO’s penalized for bad luck?
Joost van Peer (6079857) November 25th, 2015
Master thesis in Business Economics, specialization Finance Supervisor: Dr. F. Lopez de Silanes Molina
2 Statement of Originality
This document is written by Student Joost van Peer 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 Acknowledgements
After having spent the past couple of years of my life at the University of Amsterdam, it is time to finish this chapter by writing my master thesis. This will be the closing of my master in Business Economics with the specialization Finance.
First of all, I would like to thank my supervisor Dr. F. Lopez-de-Silanes for all his support and patience in the last couple of months. After we set up a tight planning, he always provided me with very useful feedback and advice, even during the summer break. I couldn’t imagine having a better supervisor.
Also I would like to thank Dr. J.E. Ligterink who is responsible for the coordination of the master thesis as the head of the finance department. He arranged for me to finish my master thesis after I had taken a gap year within my master.
Last but not least, I want to thank my parents and my girlfriend. Although I have not been the nicest person to have around the last couple of months due to the work and stress that came along with my thesis, they have always been supportive and motivated me to work on my thesis.
4 Abstract
According to the optimal contracting view, executive compensation is the solution to the principal agency problem. Shareholders need to incentivize executives by rewarding them based on their relative performance in order to exclude rewarding for factors that are beyond the executives’ control. This paper examines if executives are rewarded for factors that are beyond their control – better known as pay for luck - and whether this effect changes as a result of the financial crisis. The mean industry performance is used as proxy for luck. From the results it appears that executives are rewarded for luck and the effect is almost equal as the effect of pay for performance. This is in line with the findings of Bertrand and Mullainathan (2001), who explain this phenomenon based on the managerial power view. The theory suggests that executives use their power to take control over their own pay-setting process to obtain a compensation packages that rewards them for positive luck while minimizing the influence of bad luck. Using the financial crisis as an indicator for bad luck, this paper will test for a decrease in pay for luck. For total cash compensation there appears to be a
significant decrease in pay for luck as a result of the financial crisis. Also for most of the individual compensation components the results showed a decrease in pay for luck, but these results were not statistically significant. These findings are in line with the managerial power view that assumes that executives have taken control over their own pay-setting process.
5 Table of content
1. Introduction 6
2. Literature Review
2.1 Principe Agent Theory 8
2.2 Optimal Contracting view 9
2.3 Managerial Power view 11
2.4 Other views 14 3. Research Methodology 3.1 Data description 15 3.2 Variable description 15 3.3 Hypothesis 17 3.4 Regression model 18 4. Results 5.1 Descriptive statistics 21
5.2 Pay for performance 24
5.3 Pay for luck 25
5.4 The impact of the financial crisis on pay for luck 28
5. Discussion and conclusion 36
List of references 38
Appendix 40
Table 1 Descriptive statistics raw sample Table 2.1 Descriptive statistics 2001 - 2006 Table 2.2 Descriptive statistics 2007 - 2013 Table 3 Pearson correlation matrix
6
1. Introduction
Executive compensation has always attracted a lot of attention in the past, however the public outrage on the excessive compensation and the doubtful incentives given to CEO’s during the financial crisis has brought this topic back to the agenda. The fact that executive
compensation hasn’t decreased significantly regardless of the financial crisis suggests executives are rewarded irrespectively of their performance.
Originally, executive compensation is considered to be the solution to the principal agent theory; a situation where two people or entities have different interests and where the actions of the agent are unobservable to the principal. Since the principal has no direct control over the agent, the principal needs to provide the agent with incentives to align their interests. In the context of executive compensation, the CEO represents the agent and the principal is represented by the shareholders. While shareholders want to maximize the shareholder value, the CEO has no financial interest in doing so. Using incentives in executive compensation could align those interests.
Building on the principal agent theory by Berle and Means (1932), two widely used theories have been developed. The optimal contracting view by Murphy (1986) states that compensation is the solution to the principal agency problem by the use of pay for
performance measures. Executives should be rewarded based on their relative firm
performance as this aligns the interest of the agent with that of the principal. Shareholders would benefit from the increase in firm value through their stock holdings while the executive will benefit as their compensation will increase along with firm value. This theory is based on relative performance as this prevents pay for luck, where luck stands for factors beyond the CEO’s control. On the other hand, the managerial power view considers executive
compensation as part of the principal agency itself. Bertrand and Mullainathan (2001) assume CEO’s gain control over the pay-setting process which results in compensation packages that are not optimal for the shareholders. Under the managerial power view, the CEO will use his power to make sure that he will be rewarded in case of good luck while he will minimize any punishment in case of bad luck. This results in the existence of (asymmetrical) pay for luck.
Since then, a lot of papers have validated the existence of pay for luck. However, most of the papers focused mainly on the pay for luck when luck is good. Also most of the research have been conducted before the start of the financial crisis. As there should be a clear
distinction between the pay for luck when luck is good and the pay for luck when luck is bad according to the theory, it would be interesting to have a closer look at years when luck was
7
bad. The recent financial crisis provides us with the unique opportunity to examine this specific assumption as it can be used as a factor on firm’s performance which is completely beyond the CEO’s control, a factor of luck. Also it is a negative factor (bad luck), so it enables us to conduct the comparison.
The aim of this paper is to test if the pay-for-luck sensitivity is consistent during the financial crisis. First, the existence of a pay for general performance based on stock return will be tested on the entire sample. Then, using the instrumental variable regression as used by Bertrand and Mullainathan (2001) there will be a test for the existence of pay-for-luck. Finally, the regression will be adjusted to test for the change in pay-for-luck as a result of the financial crisis. The period from 2001-2006 will be used as the pre-financial crisis period and the period from 2007 – 2013 will be considered as the financial crisis. The models will be tested on a sample that contains all firms in the S&P 500.
To provide some background in the executive compensation, this paper will start with a discussion of the existing literature and empirical research in chapter 2. First, the principal-agency theory - which could be considered as the basis for executive compensation - will be explained, followed by the two most common theories about this topic; the optimal
contracting view and the managerial power view. The empirical research of others will help in formulating the hypothesis. In chapter 3, the methodology that will be used to test the
hypothesis will be discussed. Also the dataset, the variables and the regression models are explained in this chapter. The outcome of the models will be discussed in chapter 4 where all the results will be published and interpreted. Finally, chapter 5 will provide a summary of the research and will come to a conclusion based on the results from the previous chapter.
8
2. Theoretical Background
In this section, the literature about executive compensation will be discussed. First, I will start with a short introduction of the principal-agency theory, as it is the basis for the topic of this paper. The principle agent theory explains the conflict of interests due to information
asymmetry between the principle and the agent. Many academics came up with theories how to best align those interests, the two most common theories will be discussed in detail. In the fourth paragraph of this chapter, we will briefly discuss some alternative theories that
researchers came up with.
2.1 Principal-agent theory
The principal agent theory occurs when two people or entities have different interests where one of them, the agent, can take actions that are unobservable to the other, the principle, but which will impact the principle. This information asymmetry arises since the actions of the agent cannot be measured or this process could be too costly. Consequently, the agent could be motivated to act in his own interest rather than those of the principal, especially when actions that are beneficial to the principal are costly or time consuming to the agent. The principal-agent theory is developed initially by Berle and Means (1932). The theory is further developed by Jensen and Meckling (1976).
In the case of executive compensation, the CEO represents the agent and the shareholders act as the principal. It is in the interest of the shareholders that the CEO maximizes the firm value, but the CEO has no (financial) interest in doing so. Since shareholders cannot observe (all) the actions of the CEO, they can’t simply reward him for taking actions in their interest. This could cause the CEO to take actions that are not beneficial for the shareholders like undertaking large acquisitions that expand their
responsibility (empire building), utilizing firm’s assets for personal goals (e.g. abuse of the company yet) and take actions to entrench themselves in their current position. The decrease in firm value caused by such behavior is called ‘agency costs’ as it is the result of the agency problem.
One of the drivers of the conflict of interests, is the difference in the risk-preference of the shareholder and the executive. CEO’s are typically risk averse as they have the risk of becoming unemployed if a firm performs bad – or in worst case- goes bankrupt. On the other hand, shareholders are able to hedge some of the risk by diversifying their investments so they
9
are willing to take on more risk. In order to overcome the conflict of interests, shareholders need to incentivize the executives in a way that their interests get aligned.
2.2 Optimal Contracting View
To get the CEO incentivized to maximize the shareholder value, the shareholders need to arrange the executives compensation so that is also in the CEO’s best interest to maximize shareholder value. They need to align the CEO’s interest with their interest. Since the actions of the CEO are hard to observe, rewarding executives based on firms’ performance is the solution according to Murphy (1986). This view is called the optimal contracting view. Murphy states that it is most optimal to reward executives based on their performance, whereby filtering out effects that are beyond the executives’ control. Including these effects, also called ‘luck’, would have no added benefit for the shareholder as the CEO has no control over luck. Motivating him on luck has no incentive effect, on the contrary it will actually cost the shareholder extra money since the risk-averse CEO will demand compensation for bearing extra risk as his compensation would be based on an effect outside of his control. This
assumes the fixed pay (salary component) should be raised.
As in most firms it is impossible, nor very costly to monitor the executives’ actions, Holström (1979) also argues that firm performance is the best way to measure these actions. According to Holström, rewarding an executive based on firm performance is optimal since it – in contradiction to the single payoff – mitigates the potential moral hazard problem. This arises in situations where the executive gets involved in a risky situation knowing that he is
protected against the risk since his salary is fixed, while the other person – the shareholder – will incur the cost.
In 1990, Jensen and Murphy did some research to test whether CEO’s were rewarded based on the firms’ performance. They discovered a relatively small relation between CEO wealth and shareholder wealth, which decreases over time. They find that for each $1000 dollar increase in firm value, the CEO wealth increases by $3,25 dollar. They conclude that this pay for performance sensitivity is too low to be consistent with the principal-agent theory. Although bonuses represent over 50 percent of the total compensation, they don’t seem to be sensitive to firm performance and compensation seem to fluctuate very little. Jensen and Murphy find that the year-to-year variability of CEO compensation is comparable to that of a sample of 10.000 randomly chosen workers. Although the pay-for-performance sensitivity is very low, they believe rewarding executives based on performance is the optimal. They argue
10
that public disapproval (outrage) and political forces impose a constraint on generous compensation when firms perform well. Consequently, equilibrium in labor market will prevent large penalties when performance is bad and as a result, the pay-for-performance decreases which is suboptimal according to the optimal contracting view.
However, while Jensen and Murphy find a relatively low pay-for-performance
sensitivity, Hall and Liebman (1998) argue that the sensitivity is much higher. They show that both the level of CEO compensation and the sensitivity of CEO compensation to firm
performance have risen dramatically in the period from 1980 to 1994. They explain the difference in their findings compared to other research papers, and especially to that of Jensen and Murphy by two reasons. First, Jensen and Murphy used data in the period from 1969 to 1983, which excludes the enormous increase in stock option grants during the 1980’s and 1990’s. As stock options are perfectly correlated with firm performance (as measured in stock performance), the increase in stock options grants has effected the pay-for-performance sensitivity in a positive way. Secondly, Jensen and Murphy use the dollar change in CEO wealth instead of a percentage change in CEO wealth. This can be misleading as for a $10 billion dollar company, an increase of only 5 percent would increase CEO wealth with a little over $1.6 million dollar. Although the absolute sensitivity seemed low, in percentage terms, the $1.6 million dollar will probably be a significant part of the executives total
compensation. Thus, for billion dollar companies, the sensitivity may seem small only because it is implicitly measured. As a result, according to Hall and Liebman CEO’s seem to be rewarded based on firm performance in accordance to the optimal contracting view.
Rewarding an executive based on firm performance without taking into account the part that is due to luck requires an extra measure. Gibbons and Murphy (1990) came up with the use of the relative performance evaluation (RPE). RPE separates luck from performance by not looking at the absolute performance of an executive, but at their relative performance, i.e. their performance relative to the performance of other executives in a representative peer group. Designing compensation packages based on RPE creates incentives for executives to increase shareholder value while they have less risk of fluctuations in their due to factors beyond their control.
According to the optimal contracting view, compensation packages should be designed to provide executives with efficient incentives to maximize shareholder value. Focusing on pay-for-performance seem to be most optimal, under the condition that luck is filtered out of the performance. The optimal compensation package could either be the result of bargaining between the board of directors and executives, or from market forces that
11
stimulate them to set such an optimal compensation contract even without bargaining. As discussed in this chapter, there is empirical evidence for a pay-for-performance. However it is questionable whether the pay-for-performance measures are corrected for luck.
2.3 Managerial Power View
As discussed in the previous papers, the contracting view assumes that executive
compensation is the solution to the principal-agent problem. In rewarding executives based on firm performance their incentives get aligned with those of the shareholders, i.e. maximizing shareholder value. According to the theory, this only holds under the condition that
compensation is not based on luck, where luck is defined as factors or shocks beyond the CEO’s control. Problem with the optimal contracting view is that the discretionary parts of compensation – salary and bonus – are especially weakly linked to performance. Murphy (1999) find no significant correlation between a CEO’s non-equity parts of his compensation and the RPE. Similarly, Blanchard, Lopez-de-Silanes, and Schleifer (1994) find that cash compensation increases when firm performance increases for reasons that cannot be attributed to the executive. It seems that executives are benefitting from factors beyond their control.
Bertrand and Mullainathan (2001) find that executives’ compensation responds to ‘lucky shocks’. They have used the oil price, exchange rates and the mean industry
performance as a proxy for luck as all of these measures are outside the control of a single CEO. They find that “CEO pay is almost as sensitive to a lucky dollar as to a general dollar”. Their result also holds for discretionary components of pay - salary and bonus. They explain this phenomenon by the skimming view, where the CEO uses his power to skim off the gains in case of good luck. Also, pay for luck seem to be smaller in better governed firms as a CEO will have less power in a better governed firm. This theory weakens two of the most common explanations of pay for luck: “paying for luck is optimal” and “filtering out luck is
impossible”.
The skimming view – later known as the managerial power view - explains that CEO’s gain effective control of the pay-setting process because of entrenchment (e.g. dependent directors, staggered boards) and because of the complexity of the pay process. Their pay level then becomes constrained by an unwillingness to draw shareholders’ attention and public outrage. Pay for performance is part of the skimming view as if performance is good, shareholders will less likely criticize on the generous compensation package of the CEO. Thus in times of good performance – despite whether the good performance is caused by luck or by performance –
12
CEO’s will ensure to increase their compensation which also explains the existence of pay-for-luck in the skimming view context. After all, one could say that poorly governed firms fit the skimming view while better governed firms fit the optimal contracting view.
The constrain on compensation caused by public outrage is illustrated by Johnson, Porter, and Shackell (1997). They showed that executives that received negative media attention on their compensation schemes received relatively small pay increase in the years after and their pay-for-performance sensitivity was increased.
Bebchuk and Fried (2003) argue that executives gain control over their compensation package through the lack of independence of directors and the failing power of market control. The board of directors arise because shareholders have very little influence over the executives compensation. The board of directors is designed to act on the shareholders behave. The optimal compensation package could either be the result of bargaining between the board of directors and executives, or from market constraints that stimulate them to set such an optimal compensation contract even without bargaining.
However, it is assumed that directors – as they are representing the shareholders – are seeking to maximize shareholder value. In reality, directors will not automatically act on the interest of the shareholder. To explain this, we need to have a closer look at the way directors are (re-)- nominated. Although board elections by slate arouse the impression that
shareholders have some influence in the election of directors, typically the director slate proposed by a firm is the only one offered. This assumes that in order to become a director, you need to be on the firm’s slate. Once a director is on the board, the CEO plays an
important role in re-nominating directors as he is most of the time also chairman of the board of directors. It may seem clear that directors that wish to be re-appointed because of the attractive compensation, the provided prestige that comes along with being part of the board of directors and the valuable business and social connections, will favor the CEO since he has a big role in the (re-)nominating process. Among other things, directors that are critical towards the executive compensation will be less likely invited to join other firms’ boards and therefore they will harm their position in the labor market for directors. Even if directors do not value a board seat, they have little personal interest in arguing the CEO or other board members. Typically, they have only nominal equity interests in the firm and they have limited time which forces them to rely on information provided by the firms HR and compensation consultants (Baker e.a., 1988). All of whom have interest in favoring the CEO.
Besides the board of directors, also market forces are assumed to contribute to the optimal compensation. Markets forces impose constraints on what directors will agree and
13
what executives will ask them to approve, however these constraints are questionable. According to Bebchuk, Fried and Walker (2002), the constraints are not tight enough
resulting in compensation contracts that can be far from optimal. Having a closer look at the market for corporate control – the treat of takeover, diverse defense mechanisms provide executives with extra power. Staggered boards prevent hostile bidders to gain control over a board of directors for at least a year and helps executives to block hostile bids that are attractive to shareholder. Also golden parachutes provide executives to benefit personally from an eventual takeover. It is proven that CEO’s of firms with stronger takeover protection receive higher average compensation that is also less sensitive to performance (Bebchuk and Fried, 2003). In other words, CEO’s that gained more power trough entrenchment, are able to reward themselves by increasing compensation and making it less dependent on performance. This process directly refers to the managerial power view which assumes that executives have the power to increase their own pay. This is also in line with the findings of Bertrand and Mullainathan (2001) that managers in poorly governed firms are more able to use their power to influence their own compensation package.
Also the paper by Garvey and Milbourn (2004) is consistent with the managerial power view. They discovered that the average executive loses approximately 25-45% less total compensation in situations where luck was bad compared to the gains made in case of good luck. This implies that the pay for luck is significantly higher when the benchmark is up, suggesting that executives are able to influence their own compensation package. This is consistent with the view that important determinants of the executive compensation are not chosen as part of an ex ante contract agreement, but rather as a way to transfer wealth from shareholders to executives ex post. The findings of Leone and Zimmerman (2004) are in contrast with these findings and state that the compensation is twice as much sensitive to negative stock returns than it is to positive stock returns. However they only looked at the cash component. According to Garvey and Milbourn (2004), payment for luck is not necessarily a bad thing as long as firms are persistent. If a firm decides to reward it’s executives based on firm performance, it should do so in both good and bad states. In this paper, we will point out whether firms are indeed persistent in the use of their determinants for the executive compensation.
14
While the optimal contracting view argues that executives should not be rewarded for luck and the managerial power view assigns the existence of pay-for-luck to the power executives gained over the pay-setting process, there are also some other theories. In this chapter, we will discuss some of these alternative theories regarding executive compensation
Oyer (2004) has a similar view on executive compensation as Murphy (1999), although he argues for relative executive compensation instead of basing the executive compensation on relative performance evaluations (RPE). Their view is based on the participation constraint. If a firm refuses to index the compensation package to that of other executives, executives could be attracted by other firms that offer them a better compensation package. This does not have to firm-specific as for the function of CEO management – and strategic skills are more valuable than industry knowledge. Especially during periods of excessive growth when there is an increase in demand for skilled executives, relative performance would be an outcome.
A more recent study done by Gopalan, Milbourn and Song (2010) explains why it is optimal that executives are rewarded for their own performance but also for the luck part due to sector or market forces. They state that executives are able to set their exposure to the industry through their choice of strategy. The pay for luck thus provides them to respond correctly to movements within the industry. In accordance to their view, they find that the pay-for-luck sensitivity is higher in firms where the executive has greater flexibility in changing the firm’s strategy.
Hoffman and Pfeil (2010) give another explanation for the pay for luck. They state that the pay for luck can be incorporated in the optimal contracting view, since tying the executives’ compensation to luck is also optimal for the firm for efficiency reasons. If a project is generating more cash due to a ‘lucky factor’, terminating the project would be inefficient. Rewarding an executive based on luck prevents him from terminating the project. This theory holds under the condition that the extra costs of pay-for-luck are less than the cash flow generated by the project due to the luck.
It seems there is an ongoing discussion about the way executives should be optimally compensated. According to the optimal contracting view, the compensation should be based on firm performance where luck should be filtered out. The managerial power view argues that executives gained power over the pay-setting process and that they are also rewarded for luck resulting in suboptimal compensation arrangements for the shareholders. Finally, there are some theories that explain that incorporating pay-for-luck is indeed optimal for both executives as shareholders.
15 3. Research Methodology
In this section the methodology will be explained. It contains a description of the data, a definition of the variables, the hypothesis and the regression models that are used for the actual research. There will be an explanation of why the specific variables are being used and how they are calculated. The hypothesis will be formulated based on the literature and
empirical findings of others. The final model to test for a change in pay for luck over time is derived from the research of Bertrand and Mullainathan (2001).
3.1 Data Description
For this research, data on American firms that are listed on the S&P 500 will be used. Since the main focus in this paper is on the financial crises, the data will be selected within the time frame 2001 – 2013. The period from 2007 – 2013 will be distinguished as the financial crises (Taylor, 2009).
The data on CEO’s will be gathered from Standard and Poor’s Execucomp. This database provides us with detailed information about the composition of CEO compensation and also CEO characteristics like their age, tenure etcetera. For the determination of the performance of the companies, the CRSP database is used. This database provides annually stock prices for all stocks listed on the S&P 500. Both databases will be merged into one database providing per specific company both their stock performance as well as the
associated compensation package of their CEO. After eliminating observations with missing data, the final dataset contains 12,884 observations. Fiscal year stock returns are used as Compustat also uses fiscal years. If the fiscal year stock return is missing, calendar-year return will be used instead.
3.2 Variable Description
In this paragraph all the variables that are used will be explained. After the general variables are explained, the proxy of luck will be discussed. At last, the control variables used in the regressions will be explained.
For firm performance, the stock market return will be used. According to the principal agent theory, executives need to be rewarded based on the interests of the principal. In this
16
case, the shareholders are considered as the principal so the stock market return will best reflect their interest. This is in line with the research of Jensen and Murphy (1990).
As CEO Compensation, the total annual compensation is used. The total compensation consists of base salary, bonus, total value of restricted stock granted, total value of stock options granted, long-term incentive payouts and other annual compensation. Other annual compensation could indicate the existence of perquisites, tax reimbursements, or other
personal benefits. In contradiction to most of the literature, stock options granted are included in the total compensation package. They are calculated by using the Black-Scholes formula. Due to a change in disclosure requirements in option grants, from 2006 onwards firms need to report the fair value of the option grants using an option pricing model. Consequently, to prevent missing observations, the option grants in the sample are replaced by their reported fair value starting from 2006. Also for the restricted stock grants, the reported fair value is taken from 2006 onwards.
For the proxy of luck, the mean performance of the industry is used. If an
industry is performing great and faces increasing profits, a CEO should not be rewarded for the average increase of the industry but only for his relative performance (Bertrand and Mullainathan 2001). The mean performance of the industry is a factor beyond the CEO’s control and thus it can be considered as a factor of luck. In this paper, the equal-weighted mean industry stock return will be used, where a firm’s industry is given by the firms in the same 2-digit Standard Industrial Codes (SIC) in the database. Also other proxies for luck are used in the literature, like the change in oil price or the change in exchange rate, but since our database contains all different types of firms of which some have no direct link to the oil price or to a specific exchange rate, it seems best to use the mean industry performance. Bertrand and Mullainathan (2001) showed that all tree proxies of luck generate similar findings so the choice for a specific proxy will not influence the findings.
Firm size is included as a firm specific control variables. According to Jensen and Murphy (1990) the pay for performance decreases with firm size. They explain this by the fact that bigger firms have to deal with stricter governance, making it harder for a CEO to influence his own remuneration package and therefore this variable also influences the CEO pay.
Also some CEO specific variables are added as control variables. Hermalin and Weisbach (1998) provided a theory that suggests that CEO’s with longer tenures have greater influence over the board of directors. Indirectly a CEO can set his own targets and thus influence the pay for performance sensitivity. CEO tenure is calculated as the difference
17
between the fiscal year end of the current year and the year the CEO was appointed according to the database. Also the age of a CEO seems to be correlated to his pay according to
McKnight e.a. (2000). They discovered that the effect of age on salary decreases over time. Mcknight explains this as a CEO increases his wealth during his career, his need for cash decreases. Instead of cash they will tend to focus more on long term incentives as restricted stock grants and options grants. Both CEO tenure and CEO age are included as control variables.
3.3 Hypothesis
In this paragraph, the hypothesis about the persistence of the pay for luck sensitivity during the financial crisis will be discussed. The hypothesis will be based on economic theory and empirical research done by others.
Literature has proven that CEO’s are payed for luck almost as much as they got payed for their performance (Bertrand and Mullainathan, 2001). Attaching pay to luck implies that the same should be done in times of bad luck, firms should be persistent (Garvey and
Milbourn, 2004). This is also the main question of this thesis: Is the pay-for-luck sensitivity persistent during the financial crisis?
Basically the main question is preceded by two sub questions. First there will be determined if executive compensation is correlated with the general performance of a company. Then there will be tested if executive compensation is also correlated with luck (e.g. factors that are beyond the control of a CEO). Finally the main question will be tested to see whether the pay-for-luck sensitivity during bad luck differs from the sensitivity in times of good luck.
Based on the principle agent theory, it is expected that executive compensation is positively correlated with firm performance. Shareholders are expected to motivate the CEO to act in their interest by making the compensation dependent on the profit of the shareholder. Also both the optimal contracting view, as the managerial power view argue that executive compensation should be based on firm-performance.
According to Bertrand and Mullainathan (2001), CEO compensation is as sensitive to general performance as it is to luck. Also Gopalan, Milbourn and Song (2010) and Hoffman and Pfeil (2010) confirm the existence of pay for luck. Therefore, for the second hypothesis it is also expected that executive compensation is positively correlated with luck.
18
Based on the skimming view where CEO’s gain control over their own remuneration package it is expected that the pay-for-luck decreases during the financial crisis. If a CEO is indeed able to influence the remuneration process, it would be unlikely that CEO
compensation would decrease in line with bad luck as much as it gains with good luck. This is consistent with the findings of Garvey and Milbourn (2004). As a result, for the main
hypothesis it is expected that the pay-for-luck sensitivity is not persistent during the financial crisis. The pay-for-luck sensitivity is expected to decrease.
3.4 Regression Model
Now the hypotheses have been formulated based on previous empirical research, they will be tested based on our own empirical research. For each hypothesis there will follow an
explanation of the model that is used and how the corresponding results should be interpreted. The first hypothesis, ‘Executive compensation is positively correlated with firm performance’, will be tested using a simple Ordinary Least Squares (OLS) model including control variables for firm size, CEO age and the tenure of the CEO;
(1) 𝑌𝑖𝑡 = 𝛽0 + 𝛽1* 𝑃𝑒𝑟𝑓𝑖𝑡 + 𝛾𝑖+ 𝜒𝑡+ 𝑎𝑥 * 𝑋𝑖𝑡 + 𝑒𝑖𝑡
Where 𝑌𝑖𝑡 is equal to the total compensation for CEO i at time t. As described in chapter 3.2, the total compensation is defined as salary, bonuses, equity rewards (both restricted stock and stock options), long-term incentive payouts and other annual compensation.
For the measure for performance 𝑃𝑒𝑟𝑓𝑖𝑡, stock market returns are used. The variables 𝛾𝑖 and 𝜒𝑡 are the firm fixed - respectively time fixed effects. These are included to control for
unobservable firm heterogeneity and to take into account a common trend over time like an increase in executive compensation. And 𝑋𝑖𝑡 represents the CEO and firm specific control
variables; firm size, CEO age and CEO tenure. A positive value for 𝛽1, would indicate that executive compensation is positively correlated with firm performance.
For the second hypothesis, which states that ‘executive compensation is positively correlated to luck’, the model by Bertrand and Mullainathan (2001) is used. They have
constructed a two stage model, using luck as an instrumental variable for firm performance. In the first stage, the performance of a firm is predicted based on luck, so that changes in
performance that are out of CEO’s control (e.g. due to luck) are isolated. Luck is represented as the mean industry performance as explained in chapter 3.2. In the second stage, the
19
sensitivity of the CEO pay to the predicted performance will be tested. For the first stage, this model is used:
(2) 𝑃𝑒𝑟𝑓𝑖𝑡 = 𝛽0 + 𝛽1 * 𝑙𝑢𝑐𝑘𝑖𝑡 + ԑ𝑖𝑡
The dependent variable 𝑃𝑒𝑟𝑓𝑖𝑡 is the performance measure. 𝑙𝑢𝑐𝑘𝑖𝑡 is the proxy for luck,
which will be the mean industry performance. It is calculated as the equal-weighted mean industry stock return where industries are distinguished by their unique 2 digit SIC-code. The second stage is performed by regressing the predicted performance on the total CEO’s pay, this will show if CEO’s are rewarded for luck.
(3) 𝑌𝑖𝑡 = 𝛽0 + 𝛽𝑙𝑢𝑐𝑘 * 𝑃𝑒𝑟𝑓̂𝑖𝑡 + 𝛾𝑖+ 𝜒𝑡 + a𝑥 * 𝑋𝑖𝑡 + ԑ𝑖𝑡
The dependable variable 𝑌𝑖𝑡 represents the total compensation for CEO i at time t. However, to see if the pay for luck sensitivity differs within total compensation, the regressions are also conducted on each individual compensation components. 𝑃𝑒𝑟𝑓̂𝑖𝑡 is the predicted performance
of a firm based on luck, taken from the first stage regression. 𝑋𝑖𝑡 stands for the control variables and 𝛾𝑖 and 𝜒𝑡 represent the firm fixed – and time fixed effects. In this equitation,
𝛽𝑙𝑢𝑐𝑘 measures the sensitivity of pay to luck. Based on the optimal contracting view, luck should be filtered out so 𝛽𝑙𝑢𝑐𝑘 should be equal to zero (Murphy 1986). However, a positive 𝛽𝑙𝑢𝑐𝑘 would be in line with our hypothesis that executive compensation is positively correlated with bad luck.
The two-stage Instrumented Variable (IV) model will also be used for the third hypothesis; is the pay-for-luck sensitivity persistent during the financial crisis? Since this hypothesis
requires to track the change in a sensitivity over time, a dummy will be created. The dummy equals 1 in the period 2007-2013 (the period that represents the financial crisis in this research). Since we want to see if the effect of pay-for-luck is persistent, there will be an interaction dummy added to the original model.
20
Where 𝐷1 is the crisis dummy and ‘𝐷1 * 𝑃𝑒𝑟𝑓̂𝑖𝑡‘ is the interaction term for the change in
sensitivity of pay to luck during the financial crisis. It is expected to see a decrease in the pay-for-luck sensitivity, this implies an expected negative value for 𝛽2. Except for the interaction term, the regression is similar to the one used for testing the second hypothesis.
Another approach would be to simply duplicate the model that was used to test the second hypothesis but to use two different datasets. First, the model is tested on a dataset with data in the timeframe 2001 – 2006, the period that we note as the pre-financial crisis period. This would result in a 𝛽𝑙𝑢𝑐𝑘 that provides us with the pay-for-luck sensitivity. Then, we will test the model again on a dataset containing the financial crisis, 2007 – 2013. Again the 𝛽𝑙𝑢𝑐𝑘
will show the pay-for-luck sensitivity during the financial crisis. Testing for a difference between both betas would indicate whether the pay-for-luck is persistent during the financial crisis. If the F-statistic is significant, the null-hypothesis that both coefficients 𝛽𝑙𝑢𝑐𝑘 are equal
21 4. Results
In this section, the results will be reviewed. First, the adjusted dataset will be compared with the original dataset to see whether it is still representative for the firms within the S&P500. Then the dataset will be analyzed through a table with descriptive statistics. Any remarkable trends of the variables will be explained. After that, the data will be split into a pre financial crisis period from 2001 - 2006 and a financial crisis period from 2007 until 2013. Both
datasets will be compared and any significant changes in variables will be discussed. Once the data is set, the existence of pay for performance and pay for luck will be tested according to the hypotheses. This will be done for each component of compensation individually to see whether the sensitivity of pay for performance and/ or luck differs within the compensation package. Finally, using the model as described in chapter 4, there will be tested if the pay for luck sensitivity changes as a result of the financial crisis.
4.1 Descriptive statistics
Before the results of the regressions is interpreted, we will discuss the variables and their trends. Next to the table with summary statistics that is published in this chapter, more details can be find at the appendices in table 1 – 3.
Comparing the dataset that is used in the regressions with the raw dataset, there are no significant changes in the variables. Therefore we can conclude that despite deleting some of the observations the dataset is still representative for the sample of firms.
Based on the summary statistics in table 2.1, the average CEO in our sample earns $776,00 in salary. Next to that, the average bonus is $479,000. In terms of equity, the average CEO receives $1,521,000 in restricted stock grants and $1,540,000 in option grants. Note that all the components except salary have a standard deviation that is more than double the mean so the spread within the sample is very large. On average, a CEO had a yearly compensation of $5,377,000. The difference between the sum of the individual components and the total compensation is due to and other long-term incentive payouts and other annual compensation such as tax reimbursements. Furthermore, the average CEO is 56 years old and has a tenure of approximately 7 years. During the sample period of 2001 – 2013, the average annual stock return of the firms in the dataset is 17.1 percent. With a mean industry performance of 17.9 percent, the average firm slightly underperforms the industry. Based on accounting measures,
22
the average return is 11.6 percent. Eventually the average firm in the dataset has total assets worth almost $19 million.
Besides the high standard deviations that indicates a wide spread within the sample, it is remarkable that almost all the variable means are higher compared to the median. Looking at the compensation components, it is also noteworthy that in most cases the minimum observation is equal to zero. These observations can be linked to each other since there are apparently many firms that don’t pay bonus, stock grants or option grants at all resulting in a higher average than median. These findings indicate the existence of outliers, more precisely a right skewness. To correct for this, we will take the natural logarithm of all the variables in the regressions.
Table 2.1: Descriptive statistics 2001 - 2013 (sample used for regressions)
Variable Obs Mean SD Min Median Max
Salary 12,884 775.7 380.6 0 732.2 5,600
Bonus 12,884 478.7 1,649.1 0 0 76,951
Restricted stock grants 12,884 1520.5 3003.9 0 327.1 56,818.2 Option grants (Black Scholes) 12,884 1539.8 4,154.0 0 433.1 243,558.1 Total compensation 12,884 5,377.2 6,700.6 0 3399.9 243,558.1 Age of CEO (years) 12,884 55.7 6.9 28 56 86
CEO tenure 12,884 7.5 6.8 0 5 47
Stock return 12,884 0.171 0.620 -0.983 0.116 25.08 Accounting return 12,884 0.116 0.118 -2.671 0.112 1.250 Mean industry return 12,884 0.179 0.391 -0.735 0.145 6.885 Total assets ($ millions) 12,884 19.49 104.22 0.005 2.44 2,415.69
Since this paper focusses on the impact of the financial crisis, the dataset is split in two based on the financial crisis. All the data from 2001 – 2006 is considered as the pre-financial crisis dataset and consequently, the data from 2007 – 2013 is considered as the financial crisis dataset. The summary statistics for both periods can be found in the appendix in table 2.2 and 2.3
When comparing the summary statistics between the pre-financial crisis dataset and the financial crisis dataset, a few variables show a remarkable difference. While the total compensation is approximately the same in both datasets, the bonus, stock grants and option grants differ significant. During the financial crisis, the average bonus given to CEO’s decreased by almost 75 percent. The mean decreased to zero which indicates that many firms did not reward the CEO with a bonus at all. It seems that bonuses are used to reward CEO’s for (market) performance, however there is a possible better explanation. Since restricted stock grants more than doubled in both periods, it seems that firms are trying to retain their CEO
23
during the financial crisis. While a cash bonus rewards a CEO one-off and independent of future performance, stock grants reward a CEO at the moment itself but will also reward him for future performance (i.e. increase in share price). Also restricted stock comes most likely with restrictions that prevent a CEO from selling the stock within a few years after they were granted. This could be a way to keep the CEO attached to the firm. However, according to this explanation you would also expect an increase in option grants which is not the case. In contrast, option grants decrease by over 50 percent and the median decreases to zero. The only possible explanation that I can come up with is that restricted stock grants are preferred over option grants by CEO’s in times of financial crisis since they always contain value (considering the firm doesn’t file for bankruptcy) while options can be worthless if they become out-of-the-money. Looking at the returns, both the average firms’ stock return and the mean industry performance decreased during the financial crisis although the decrease in stock return is slightly with a little less than two percent. You would expect a sharper increase due to the financial crisis, although in 2013 the economy was already partially recovered from the worst part of the financial crisis. Including only the fiscal years 2007 and 2008, the years in which the financial crisis started, results in a negative stock return as well as a negative mean industry performance. These result are not included in the tables.
The Pearson correlations are presented in table 3. As the natural logarithm of all the variables will be used in the regressions due to outliers, they are already taken into account in the table instead of the regular variables. From the table it appears that all the compensation components, as well as total compensation are positively correlated and highly significant. Also stock return and mean industry performance are positively correlated with each other and significantly different from zero at a 1 percent level. This is not surprising as the mean industry performance is based on the stock return of the individual firms within that specific industry. CEO age is positively correlated with almost all compensation components except option grants. This is in line with the literature that states that older CEO’s get rewarded more on average (McKnight e.a., 2000). Also the fact that CEO age is negatively correlated with options can be explained by the probability that a CEO prefers less risky compensation like cash (salary and bonus) when he gets older and approaches his retirement (Gibbons and Murphy, 1992). The correlation of CEO tenure with the compensation components is remarkable. All the coefficients are negative and significant at a 5 percent level which implies that the longer a CEO is in his function, the less he will receive as compensation. This strokes with the literature that states that CEO’s with longer tenure have greater influence over the pay-setting process and consequently have higher compensation packages. I cannot think of a possible explanation
24
for this. The compensation data shows mixed results with the returns. Total compensation correlates positively and significant with stock return and accounting return. This implies that there is a pay for performance relation which would confirm our first hypothesis. On the contrary, option grants and salary are negatively correlated with stock return and the mean industry performance suggesting that CEO pay in salary and options decreases when the stock returns and/or the mean industry returns is up. A possible explanation for this is that CEO’s get rewarded in other components since there is a positive significant correlation between total compensation and the returns. Bonus on the other hand shows a positive correlation with all the returns. Finally, in general there is no correlation between CEO age and CEO tenure and the returns as one could already expect based on literature.
4.2 Pay for performance
To test the main hypothesis, we first need to test the two sub questions that were formulated in chapter 4.3. To test if CEO’s are rewarded for general performance, equitation 1 will be used.
(1) 𝑌𝑖𝑡 = 𝛽0 + 𝛽1* 𝑃𝑒𝑟𝑓𝑖𝑡 + 𝛾𝑖+ 𝜒𝑡+ 𝑎𝑥 * 𝑋𝑖𝑡 + 𝑒𝑖𝑡
Although tests pointed out to include both firm fixed effects and time fixed effects in the regression, the regression is also done without the effects to see the difference in result. Table 4 presents the results of both regressions with column 1 presenting the regression without firm – and time fixed effects and column 2 presenting the regression including both time – and firm fixed effects. Time fixed effects where included via the use of year dummies.
The coefficient for stock return is significant at a 1% level and has a value of 0.0619. Since the natural logarithm for total compensation is used, this implies a 0.619 percent increase in total compensation for each percentage increase in stock return. The negative coefficient for CEO age suggests that as a CEO gets older, the compensation will decrease. This doesn’t seem logical, and since the coefficient is not significantly different from zero we cannot interpret this coefficient. For both CEO tenure and firm size there is a positive coefficient that is significant at a 1% level. These findings are in line with the literature. Including time – and firm fixed effects resulted in a significant coefficient for CEO tenure.
25 Table 4: Pay-for-performance Dependent variable Ln total compensation Intercept (1) 5.8043*** (0.2755) (2) 5.8180*** (0.4296) Stock return 0.1051*** (0.0135) 0.0619*** (0.0110) Ln CEO age -0.1120 (0.0710) -0.1439 (0.1051) Ln CEO tenure 0.0122 (0.0098) 0.0700*** (0.0119) Ln firm size 0.3425*** (0.0047) 0.3329*** (0.0186) Firm fixed effects No Yes Time fixed effects No Yes
N 12,551 12,551
R² 0.3069 0.0976
1. Sample period is 2001 – 2013
2. The p-values can be found between parenthesis
3. The first column presents the regression results without including firm fixed – and time fixed effects, the second column presents the regression results including these effects.
4. *** statistical significant at 1% level, ** statistical significant at 5% level, * statistical significant at 10% level.
Based on the positive and highly significant coefficient for stock return, the hypothesis that CEO’s are rewarded for general performance can be confirmed. For every percent increase in stock return, the compensation of a CEO will increase with 0.619 percent. This is in line with the findings of Hall and Liebmann (1998) who discovered a significant pay for performance sensitivity.
4.3 Pay for luck
After the existence of pay for performance is proven, it is time to come to the main topic of this paper: pay for luck. The existence of pay for luck will be tested using a 2-stage instrumental variable regression as done by Bertrand and Mullainathan (2001).
In the first stage, the performance of a company will be predicted by the mean industry performance as a proxy for luck.
26
From the results in table 5.1 it appears that there is a positive relation between the mean industry performance and the stock return. The coefficient indicates that stock return is highly related to the mean industry performance, i.e. factors beyond the CEO’s control. In line with the results of Bertrand and Mullainathan (2001), this seems to prove there is a pay for luck. The coefficient is significant at a 1% level.
Table 5.1: Pay for luck – stage 1 regression Dependent variable Stock return
Intercept 0.0304***
(0.052) Mean industry return 0.7885***
(0.0121)
N 12,884
R² 0.2476
1. Sample period is 2001 – 2013
2. p-values can be found between parenthesis
3. *** statistical significant at 1% level, ** statistical significant at 5% level, * statistical significant at 10% level.
Based on the coefficients in table 5.1 we can predict the stock return that is due to the mean industry performance. As shown in equitation 3, it is possible to isolate luck from the overall stock performance.
(3) 𝑃𝑒𝑟𝑓̂𝑖𝑡 = 0.0304 + 0.7885 * 𝑙𝑢𝑐𝑘𝑖𝑡
The dependent variable 𝑃𝑒𝑟𝑓̂𝑖𝑡 shows the predicted stock return by luck. Consequently, the difference between the actual stock return and the predicted stock return by luck reflects the return that is due to skill.
As the predicted return due to luck is known, the second stage is performed by regressing the predicted performance on the total CEO’s compensation, this will show
whether a CEO is rewarded for the performance predicted by luck, i.e. pay for luck. Since it is expected that pay for luck differs among the compensation components, the regression is performed on every single compensation component individually. Also total cash
compensation and total equity compensation are added as dependent variables as total cash compensation represents the part of the compensation that is not directly linked to firm
27
performance through the stock price while the total equity compensation is completely linked. Adding those combinations could provide us with more insights about pay for luck for these specific two types of compensation.
(4) 𝑌𝑖𝑡 = 𝛽0 + 𝛽𝑙𝑢𝑐𝑘 * 𝑃𝑒𝑟𝑓̂𝑖𝑡 + 𝛾𝑖+ 𝜒𝑡 + a𝑥 * 𝑋𝑖𝑡 + ԑ𝑖𝑡
The regression results can be found in table 6 in this chapter. First, the regression including total compensation will be discussed as the hypothesis is based on total compensation. Afterwards the other compensation components will be discussed. For every compensation component, also the regression as stated in equitation 1 is performed to also provide us with a elasticity for pay for general performance. In this way, it is easy to compare the pay for general performance and the pay for luck.
28 Table 6: Pay-for-luck
Total compensation Salary Bonus Option grants Stock grants Total cash comp. Total equity comp. General Luck General Luck General Luck General Luck General Luck General Luck General Luck
Intercept 5.8180*** (0.4296) 5.8281*** (0.4304) 4.6585*** (0.2990) 4.6503*** (0.2911) 3.5254** (1.1002) 3.4397** (1.1085) 5.6069*** (0.7074) 5.6268*** (0.7080) 3.5793*** (0.8394) 3.5340*** (0.8400) 5.1660*** (0.3704) 5.1574*** (0.3708) 6.1827*** (0.5808) 6.1626*** (0.5813)
Pay for performance 0.0619*** (0.0110) - -0.0074 (0.0074) - 0.1895*** (0.0249) - -0.0622** (0.0186) - -0.0557** (0.0176) - 0.0451*** (0.0094) - -0.0493*** (0.0143) -
Pay for luck - 0.0647*** (0.0335) - 0.0081 (0.0225) - 0.1391* (0.0741) - -0.1239** (0.0517) - -0.0560 (0.0531) - 0.0798*** (0.0287) - -0.0248 (0.0440) Ln CEO age -0.1439 (0.1051) -0.1311 (0.1053) 0.1246* (0.0711) 0.1229* (0.0711) 0.0751 (0.2730) 0.1590 (0.2747) -0.2514 (0.1739) -0.2650 (0.1739) -0.0414 (0.2029) -0.0467 (0.2031) 0.1378 (0.0906) 0.1473 (0.0906) -0.4927*** (0.0143) -0.4357** (0.1430) Ln CEO tenure 0.0700*** (0.000) 0.0686*** (0.0119) 0.0409*** (0.0080) 0.0410*** (0.0080) 0.1111*** (0.0281) 0.1046*** (0.0283) 0.1003*** (0.0194) 0.1025*** (0.0195) 0.0792*** (0.0209) 0.0806*** (0.0209) 0.0286*** (0.0102) 0.0275*** (0.0102) 0.0962*** (0.0158) 0.0973*** (0.0158) Ln firm size 0.3329*** (0.0119) 0.3248*** (0.0186) 0.1460*** (0.0125) 0.1476*** (0.0125) 0.2793*** (0.0450) 0.2496*** (0.0452) 0.3523*** (0.0310) 0.3580*** (0.0309) 0.3839 (0.0360) 0.3929*** (0.0359) 0.1438*** (0.0159) 0.1393*** (0.0159) 0.3688*** (0.0250) 0.3767*** (0.0249) N 12,551 12,551 12,483 12,483 5,285 5,285 7,533 7,533 7,107 7,107 12,504 12,504 10,516 10,516 R² 0.0976 0.0953 0.0740 0.0740 0.1069 0.0943 0.0557 0.0549 0.1755 0.1741 0.1060 0.1047 0.0944 0.0932 1. Sample period is 2001 – 2013
2. Total cash compensation is the sum of salary + bonus and total equity compensation is the sum of option grants + stock grants
3. The column called “general” presents the regression result for the pay for general performance, the column called “luck” presents the regression results for the pay for luck.
4. ‘Pay for performance’ presents the elasticity for the stock return with respect to the dependent variable at the top of the column.
5. ‘Pay for luck’ presents the elasticity for the proxy for luck; mean industry performance, with the dependent variable at the top of the column. 6. Firm fixed effects and time fixed effects are included in all regressions.
7. The p-values can be found between parenthesis
8. *** statistical significant at 1% level, ** statistical significant at 5% level, * statistical significant at 10% level.
29
Since all the variables in the regression are yearly returns or the natural logarithm, the coefficients can be interpreted as elasticity. For total compensation the pay for general performance is 0.0619 which implies that compensation rises with 0.0619 percent when stock return increases with one percent. On the other hand, the pay for luck elasticity for total compensation is 0.0647. Both coefficients are highly significant at a 1% level. These findings indicate than the pay for general performance is about the same as the pay for luck which is in line with the findings of Bertrand and Mullainathan (2001) that CEO pay is just as sensitive to a lucky dollar as to a general dollar.
Considering the salary, the pay for general performance elasticity is -0.0074 while the pay for luck elasticity is 0.0081. However, as expected these coefficients are both not significant. Salary is a fixed compensation components so it would be remarkable to find a relation to the stock return. Therefore salary has nothing to do with pay for general performance nor pay for luck.
For the bonus component, we find a highly significant coefficient of 0.1895 for the pay for general performance elasticity and a slightly less significant coefficient of 0.1391 for the pay for luck. Comparing the elasticity’s to those corresponding to other compensation components it seems that the pay for luck sensitivity is the highest in bonus. For every percentage increase in mean industry performance, i.e. a factors beyond the CEO control, compensation rises with almost 0.14 percent. This is in line with the research from , Blanchard, Lopez-de-Silanes, and Schleifer (1994) who found that cash compensation increases when firm performance increases for reasons that cannot be attributed to the executive. It seems that bonus is used to reward a CEO based on general performance but also for a part on lucky performance.
As already notified in the Pearson correlations, both option grants and stock grants provide some remarkable results. For option grants the pay for performance elasticity and the pay for luck elasticity are -0.0622 respectively -0.1239. Both coefficients are significant at a 5% level. The negative coefficient means that CEO’s get relatively less options granted when the stock return or the mean industry performance increases. A possible explanation is that granting options seems more beneficial for the CEO when the stock price is relatively low (considering the exercise price of the option is equal to the stock price) as options become valuable when stock return increases. Also option grants can be used to retain a CEO in periods where the market as a whole is performing badly since the option grants most likely come with restrictions that prohibit the CEO to sell them within a certain period of years. When a CEO leaves the company before the restricted time period is met, most often the options become worthless. Concluding, it seems that options are more beneficial for both the firm as the CEO
30
in times of negative stock return and/or negative mean industry performance. Therefore the pay for performance and pay for luck elasticities are negative, but significant.
Also for stock option grants the pay for performance and pay for luck elasticity is negative with values of -0.0557 respectively -0.0560. Although the latter is not significant. Comparable to the situation with option grants, the negative coefficient is not what is expected based on the literature. For stock grants, the same explanation as for option grants could hold. However, stock grants always have a certain value (considered the firm doesn’t file for bankruptcy) and they belong to the CEO also when he leaves the firm within a short period after the grant. Although this is more beneficial for the CEO, it is less attractive for firms to use stock grants.
The total cash compensation and the total equity compensation doesn’t provide any new insights. Just as salary and bonus, the total cash compensation shows a positive pay for both performance and luck of 0.0451 respectively 0.0798, with both coefficients being highly significant. For the total equity compensation, we found a negative value of -0.0493 for pay for performance and a value of -0.0248 for pay for luck. In line with the findings for option grants and stock grants, executives appear to be granted relatively less equity compensation when stock return or mean industry performance is up.
As a result, it can be concluded that there is also a pay for luck besides a pay for performance. Also these effects are approximately equal when considering the total compensation. When looking at individual compensation components, roughly all the coefficients for both pay for performance and pay for luck are of the same sign: both positive or both negative, although the elasticities could sometimes differ in magnitude. Overall, we can conclude that there is a pay for luck and that the effect is approximately equal to the pay for performance. For total compensation and for bonus payments, the second hypothesis that states that executives are rewarded based on luck can be confirmed.
4.4 The impact of the financial crisis on pay for luck
The main purpose of this paper is to test the relative change in the pay for luck sensitivity as a result of the financial crisis. It is proven that executives are rewarded based on performance and more importantly, that they are rewarded for luck. To test whether this effect changes over time as a result of the financial crisis, equitation 4 will be used. We will perform a similar 2-stage instrumental variable regression as used for testing the pay for luck, but with the addition of an interaction term that measures the effect due to the financial crisis.
31
(5) 𝑌𝑖𝑡 = 𝛽0 + 𝛽𝑙𝑢𝑐𝑘 * 𝑃𝑒𝑟𝑓̂𝑖𝑡 + 𝛽2 * 𝐷1 * 𝑃𝑒𝑟𝑓̂𝑖𝑡 + 𝛾𝑖+ 𝜒𝑡 + a𝑥 * 𝑋𝑖𝑡 + ԑ𝑖𝑡
The results of this regression are presented in table 7.1. The variable of interest is the interaction dummy, as this is the measurement for the change in pay for luck during the financial crisis period.
Table 7.1: Pay-for-luck including a dummy for the financial crisis period
1. Sample period is 2001 – 2013
2. Total cash compensation is the sum of salary + bonus and total equity compensation is the sum of option grants + stock grants
3. ‘Pay for luck’ presents the elasticity for the proxy for luck; mean industry performance, with the dependent variable at the top of the column.
4. The variable ‘interaction dummy’ refers to the term ‘dummy * predicted performance based on luck’. The dummy equals 1 during the financial crisis (2007 – 2013).
5. Firm fixed effects and time fixed effects are included in all regressions. 6. The p-values can be found between parenthesis
7. *** statistical significant at 1% level, ** statistical significant at 5% level, * statistical significant at 10% level.
Total compensation
Salary Bonus Option grants
Stock grants Total cash compensation Total equity compensation Intercept 5.8376*** (0.4306) 4.6511*** (0.2913) 3.4437** (1.1086) 5.6114*** (0.7080) 3.5305*** (0.8410) 5.1390*** (0.3709) 6.1443*** (0.5815)
Pay for luck 0.0261 (0.0591) 0.0047 (0.0398) 0.1680* (0.0891) -0.0433 (0.0848) -0.0454 (0.1305) 0.1567*** (0.0506) 0.0420 (0.0787) Interaction dummy 0.0567 (0.0715) 0.0050 (0.0482) -0.0929 (0.1591) -0.1278 (0.1067) -0.0127 (0.1426) -0.1129* (0.0613) -0.0968 (0.0974) Ln CEO age -0.1308 (0.1053) 0.1230* (0.0711) 0.1563 (0.2748) -0.2670 (0.1739) -0.0465 (0.2031) 0.1465 (0.0906) -0.4357*** (0.1430) Ln CEO tenure 0.0686*** (0.0119) 0.0410*** (0.0080) 0.1050*** (0.0283) 0.1026*** (0.0195) 0.0806*** (0.0209) 0.0277*** (0.0102) 0.0975*** (0.0158) Ln firm size 0.3242*** (0.0185) 0.1476*** (0.0125) 0.2498*** (0.0452) 0.3594*** (0.0309) 0.3930*** (0.0360) 0.1405*** (0.0159) 0.3777*** (0.0250) N 12,551 12,483 5,285 7,533 7,107 12,504 10,516 R² 0.0954 0.0740 0.0944 0.0551 0.1742 0.1050 0.0933
32
Although most of the coefficients are not significant, it is striking that almost all of the coefficients for the compensation components are negative which means that the pay for luck has decreased during the financial crisis. This statement holds for bonus, option grants and stock grants. Also the total cash compensation and the total equity compensation show a negative value for the interaction term. The results are in line with the results from Garvey and Milbourn (2004) who also found a lower pay for luck when luck was negative.
The coefficient for the total cash compensation of -0.1129 is significant at a 10% level. This suggests that during the financial crisis the total cash compensation increased (decreased) 0.1129 percentage less (more) for every percentage increase (decrease) in the mean industry performance compared to the years prior the financial crisis. In other words, the pay for luck in the total cash compensation seems to decrease during the financial crisis. On the other hand, the coefficient for total compensation shows a positive value of 0.0567 indicating that the pay for luck for total compensation would increase during the financial crisis. However, this coefficient is not significant, the chance that the coefficient is equal to zero is 43 percent.
Overall, we can conclude that the statistical results provide us with mixed results about the change in pay for luck due to the financial crisis. Although most of the interaction terms indicate a decrease, the pay for luck in total compensation appears to have increased. The lack of significance also complicates the interpretation. Based on the only significant result, we can conclude that the pay for luck in total cash compensation has decreased which would confirm the hypothesis.
In order to extend the research to get possibly more significant results, another model will be used. Instead of creating a dummy variable and adding an interaction term to the original model, the original regression from equitation 4 will be performed on two different datasets by splitting the original dataset. One dataset represents the period before the financial crisis (2001 – 2006) and the other dataset represents the financial crisis (2007 – 2013). This will provide two different coefficients for the pay for luck sensitivity. We will test for any difference between these coefficients. The result of these regressions are presented in table 7.2 of this chapter.
