Master Thesis Finance - PDF Free Download

Master Thesis Finance Anr: 120255 Name: Toby Verlouw Subject: Managerial incentives and CEO compensation Study program: Finance Supervisor: Dr. M.F. Penas

Managerial incentives: Does Stock Option Compensation Increase Managerial Risk Taking Activity? Toby Verlouw 3

Table of contents 1.Introduction...5 2. Theoretical background...7 2.1.Agency theory...7 2.2.Incentive effect of stock option- and other compensation...7 2.2.1.Stock vs stock options...8 2.3.Other compensation effects...9 3.Hypothesis development...11 3.1.Prior Literature...11 3.2.Hypothesis...12 4.Model explanation...14 4.1.Measuring incentive effects and option valuation...14 5.Sample collection, variable measurement and descriptive statistics...17 5.1. Sample description...17 5.2.Descriptive statistics...17 6.Results...20 6.1. Vega on R&D, CAPEX, Aqui and Booklev...20 6.1.1 Endogeneity in the model...23 6.2 Type of relationship...24 6.3. Vega grouping...26 6.4.Percentile grouping...27 6.5. Importance of determinants of option Vega...28 6.6. The influence of the banking crisis...30 7.Conclusions...32 8. Recommendations and limitations...33 9. References...34 10.Appendix...36 4

1.Introduction In today s business environment many companies are publicly traded. In the process of an IPO shares of the company are sold to the general public in order to generate capital. Every individual in possession of a share owns a part of the specific company. Although these shareholders do own the company, the Chief Executive Officer manages the firm. The CEO is responsible for the conduct of business; he is in charge of the daily decision making, finance and investments of the company. Almost every publicly traded company faces the fact of a separation in the ownership (shareholders) and control (CEO). Both parties will have different incentives known as the principal-agent problem. The shareholders main concern is to maximize his stock returns while the CEO value their personal status and prestige towards society. In order to reduce these agency conflicts it is argued that the managers' personal wealth should be tied to the company performance. One way of creating this dependence is by stock compensation. Over the last years we see a significant increase in the use of stock option compensation. The question that remains is whether these option compensation plans do indeed create the incentives that align the interests of the CEO and shareholders, as stock prices increase both the manager and shareholders benefit from this increase. However, options do not bear the downside risk in comparison to a normal stock. The holder of an option is not obliged to exercise the option at expiration. Therefore, when the stock price is below the exercise price of the option, the option will not be exercised. The value of that option is the same no matter if the stock price is only slightly or significantly lower than the exercise price. This characteristic of a stock option creates a positive relation between the option value and risk. (standard deviation of returns/volatility). As the mangers don not face the same consequences for bad investments, option compensation might create other incentives as desired. In this study I will provide evidence on the relationship between stock option compensation and managerial risk taking activity. The main characteristic of managerial compensation is the sensitivity of CEO wealth to a change in stock volatility, also called Vega. Besides, the sensitivity of CEO wealth to changes in stock returns, or Delta, is considered. According to the theory and earlier conducted studies CEO risk taking activity should increase in Vega and decrease in Delta. Obviously, as the number of option holding increase so does the Vega. The managerial risk taking activity is measured through four different variables namely; Research and Development, Capital expenditures, Acquisitions and Book leverage. Prior conducted studies use the Vega and Delta both as independent and dependent variables. Cohen et al.(2000) used the Vega as independent variable and found a positive relation between Vega and both leverage and volatility. On the other hand Vega is used as the dependent variable. 5

Guay(1999) found a positive association between Vega and firm risk, investment opportunities and R&D intensity. An obvious conclusion from these studies is the fact that the causation runs in two directions, Vega influences firm policy and vice versa. In summary, managerial compensation and risk taking activity are probably jointly determined. The main question addressed in this study is: does stock option compensation influence managerial risk taking activity. Obviously, this question implies causality in a certain direction. The literature provides evidence on the relationship between compensation and risk taking, the issue remaining is the underlying causal relationship. So is the positive correlation between R&D and Vega explained by Vega implementing high R&D expenditures or, Vega and R&D are both driven by other omitted factors? The remainder of this paper is organized in the following way. Part 2 provides an introduction and background on the basic principles of stock option compensation incentives. In part 3 the prior literature is discussed followed by the development of the hypotheses. Part 4 explains the model used for the calculation of option value and corresponding Vega and Delta. Part 5 provides a description of the sample and variable construction. Further, summary statistics are present. The first empirical results and regressions are discussed in part 6. In part 7 we take a closer look at the data providing additional evidence on the relation between Vega and risk taking measures. Part 8 concludes on the research. Finally limitations and recommendations are discussed in part 9. 6

2. Theoretical background 2.1.Agency theory During the 1960 s and 1970 s more economists were interested in the risk sharing among individuals and groups. The risk sharing problems arise when cooperating groups have different views towards risk (Arrow, 1971). The agency theory broadens the initial concept. In every business parties cooperate, having different interest and goals. The agency theory tries to describe the relationship between two parties, the principal and the agent, where the principal delegates work to the agent. The theory focuses on two problems perceived in any business. First, the conflict of interest between the principal and the agent. Intuitively speaking, as we assume that both agent and principal are utility maximizers the agent will not always act in the best interest of the principal. Second, the difficulty of measuring the agents performance. For the principal it is almost impossible to overcome these problems without incurring certain costs of bonding or monitoring. These agency problems are costly and can harm the business and should therefore be resolved. Companies have been searching for different ways to direct the agent, give the agent an incentive to act in the principals best interest. Much research has been done in the field of agency conflicts. Jensen and Meckling (1976) illustrate that in order to reduce agency conflicts managers wealth should be tied to firm or stock price performance. By using a compensation policy in which the managers wealth is tied to the firm, the slope of the relationship between managers wealth and stock price can be managed. In this way managers are rewarded for creating shareholders value. This gives the shareholders the power to induce the manager to take actions that increase equity value. However, just managing this slope of managers wealth to stock price performance is not sufficient, convexity must be managed as well to induce managers to make optimal investment decisions (Haugen & Senbet, 1981 and Smith & Stulz, 1985). Convexity is explained as the sensitivity of managers wealth to the volatility of equity value (Guay, 1999). This convexity and the relationship of wealth to volatility plays an important role in the measurement of stock option compensation plans. 2.2.Incentive effect of stock option- and other compensation Compensation plans are used to induce certain incentives to managers. The main focus is to align the managers interest to that of the shareholders. As mentioned before, compensation policies are used to manage the slope and convexity of managers wealth. Earlier conducted research shows that as managers wealth is related to firm performance risk taking incentive are created. For a risk averse 7

investor the appetite to invest in high risk projects would be significantly reduced if their personal wealth is tied to firm performance. Stock option compensation is one way to realize a relationship of managerial wealth to firm performance. One implication of stock option compensation is the fact that options have convex payoff schemes. In contrast to stock ownership, option payoff schemes are a non-linear function of stock returns. 2.2.1.Stock vs stock options Executive stock ownership and stock option compensation are often assumed to have equal incentive effects. A common practice in research is to combine stock ownership and stock option compensation as a single measure for stock based incentives (Jensen & Murphy, 1990 and Mehran, 1995). However, this is a major misconception, stock ownership and stock options have very different risk properties. This asymmetry in risk properties is important because it can explain different behaviour or decisions by managers. Stock ownership and stock option pay are both incentive schemes which tie the managers wealth to company performance and shareholders wealth, however in different ways. When managers are compensated with stocks their wealth changes in direct proportion to the return of the shareholders. These fluctuations in wealth are both positively and negatively. Stock options on the other hand bear different features. As the shareholders wealth increases when stock prices rise, so does the wealth of the managers. However, in the event of a drop in the firms stock price the manager experiences no drop in wealth. Stock options are only exercised when the stock price is above the exercise price, in any other case the payoff of the option is zero. So, both the ownership of stocks and stock options result in benefits in case of increasing stock prices, but, only stock ownership can result in direct losses on the current wealth of managers in case of a stock price decrease. Stock options do not bear the downside risk, so option holder is indifferent whether the underlying stock price is much or only slightly lower than the exercise price. Figure 1 Stock- and Stock option payoff 8

The difference between stock and stock option ownership or the slope and convexity of the wealth performance relationship can be further explained by the following example. As of December 31, 2010, the CEO of company X held 65.000 shares of stock, worth $156.000, and 62.000 stock options worth about $96.000. At the same time, CEO of company Y held 42.000 shares of stock worth $110.000 and 184.000 stock options worth approximately $1.550.000. By using the Black-Scholes (1973) formula the increase in CEO security value for an increase in firm stock price can be calculated. Both the CEO s wealth would increase with $35.000 for an increase in 5% in stock returns. The slope of the wealth performance relationship is the same for both CEO s. However, for a 5 percentage point increase in annual volatility the securities of CEO of firm X would increase in value by about $210.000 compared to an increase of only $65.000 for the CEO of company Y. So by compensation managers with stock options induce significantly greater risk taking incentives. Taking this story back to the agency theory, it is argued that stock based compensation provides an incentive for managers to invest in shareholder maximizing projects, thereby making them less risk averse. This view lacks emphasis on downside risk, which cannot be neglected. Kahneman & Tversky (1979) found that in the process of determining a decision makers risk aversion downside risk is as important as upside potential. Besides, the preference of investing in more risky alternatives is influenced by the magnitude of possible losses and gains. As mentioned before, stock options only face upside potential and no downside risk it can therefore be argued that option pay should result in risk seeking behaviour. 2.3.Other compensation effects Managerial incentives provided by option compensation are also influenced by other parts of the compensation plan. It is argued that compensation plans induce managerial opportunism, but is this really the case? Stock compensation is the easiest way to tie a CEO s personal wealth to firm performance. By providing managers with stock compensation, their wealth moves closely with the firm performance. In most cases bad CEO performance lead to a drop in stock prices. So it is in the CEO s own interest to achieve optimal firm performance. One remark, this incentive effect can only be achieved if the manager owns a substantial amount of company stocks. The magnitude of stock ownership should not be measured as the dollar value of stocks held by the CEO, neither as a percentage of the total cash compensation. What is really of interest is the percentage of 9

outstanding shares owned by the CEO. Only if a significant part of a manager his wealth is tied to the company, CEO s experience direct feedback effects from changes in stocks prices. What exact incentives provide the use of stock compensation? This can be explained with help of the portfolio theory. The modern portfolio theory (MPT) by Markowitz (1952) suggests that in the process of creating an investment portfolio assets should not be selected individually, on their own merits. Each investment assets has their own characteristic, a certain expected return for a given level of risk. Investing is a trade-off between these two measures; a higher expected return comes with an increase in risk. The MPT argues that diversification is the key in the creation of an investment portfolio. The aim is to create a portfolio of individual investment assets whose total collective risk is less than the risk of any individual asset. Therefore, in order to reduce firm-specific risk a risk averse manager will create a diversified investment portfolio where none of the individual investments are perfectly correlated. However, with respect to firm specific risk, CEO stock holdings are not diversified. Any change in stock price will result in direct changes in the personal wealth of the CEO. Stock compensation increases the Delta of the portfolio, leaving managers vulnerable to drops in stock returns and thereby decreasing their risk taking appetite. Opposite to stock compensation, cash salaries provide the CEO with diversification opportunities. Cash salary is managerial wealth untied to the firm and free to invest. As regulation forbids selling stocks granted from compensation managers would be wise to invest in other securities thereby diversifying their own portfolio. This diversifying behaviour can also be explained as hedging. The stock compensation provided by a firm is highly undiversified. Such under diversification comes with high firm specific risk for the CEO. In order to offset this risk, the CEO can invest in other companies or instruments to hedge the downside. According to Smith and Stulz (1985) managers that are able to hedge are more likely to take on variance increasing NPV projects. In other words, as cash salary provides diversification and hedging possibilities thereby reducing managerial risk aversion. 10

3.Hypothesis development 3.1.Prior Literature Earlier conducted studies provide considerable empirical and theoretical evidence on the role of compensation structures and managerial behaviour. Jensen and Meckling (1976) found that in order to reduce agency conflicts managers wealth should be tied to company performance. More recent studies by Guay (1999), Core and Guay (1999), Aggarwal and Samwick (1999), Holderness et al. (1999) and others did research on firm characteristic and compensation schemes. An important research conducted by Guay (1999), studies the determinant and magnitude of CEO wealth to equity risk. He found that stock options, but not common stockholdings, significantly increase the sensitivity of CEO s wealth to equity risk, ( Guay 1999) which is consistent with managers receiving incentives to invest in risky projects when the potential loss from underinvestment in valuable riskincreasing projects is greatest (Guay, 1999). Other studies focus on the relationship between different compensation schemes and firm performance. It is argued that as compensation schemes induce particular managerial behaviour, there should be some relation between company performance and compensation. In the beginning of the 21th century more studies attempt to focus more on the direct relationship between managerial incentives and their company policies. So whether options really increase the risk taking policy of a firm. Intuitively it is argued that managerial compensation should be convex in order to mitigate the effect of risk aversion and create more risk appetite. Guay (1999) and Ross (2004) illustrated that the managerial utility function is of importance too. Still there remains some debate in the role of stock option compensation. It seems obvious that stock option compensation provides managerial risk taking activity as options only provides upside potential and no downside risk. However the opposite can also be true. Ju et al. (2002) found that the magnitude of managerial risk taking depends on risk aversion and underlying investment technology. In some cases stock option compensation results in too little risk taking. Besides, Lewellen (2003) argues that in the money options can also discourage risk. Previous studies by Coles et al. (2006) provides evidence of a strong causal relation between managerial compensation structures and investment and debt policy for firms in the S&P 500, S&P midcap 400 and S&P smallcap 600, for the period 1992-2002. Rajgopal and Shevlin (2002) did research on managerial stock option compensation and their incentive to invest in risky projects, measured as exploration risk, for oil and gas companies for the period 1992-1997. They found that employee stock option risk incentives had a positive relation with future risk taking. 11

Other empirical work tries to explain the relationship of option holdings with firm financial policy. For example, Berger et al. (1997), Jolls (1998), Rogers (2002) explore the relationship of stock option holdings and company leverage, repurchases and derivative usage. Evidence is found that stock compensation and stock option compensation create different incentives for CEO s. By using options, in comparison to stock compensation, CEO s wealth is tied in a different way. There is an asymmetric risk effect; options do not bear the downside potential of stock ownership (Sanders 2001). Vega and Delta are widely used measures in recent studies. Research shows the existence of several relationships in firm policy and the sensitivity of CEO wealth to stock return volatility (Vega) Cohen et al. (2000), Rajgopal and Shevlin (2002) and Knopf et al. (2002). 3.2.Hypothesis Earlier conducted studies provide a good backbone for this research. The main question considered is whether stock option compensation result in more CEO risk taking activity. Since it is of great importance to state a reasonable measure for risk to insure robustness of the findings, especially in case of a cross sectional study among different industries and firms, we start with a proper definition of risk taking activity. Many different measures are used in earlier research in the field of managerial risk taking activity and stock option compensation. Since the CEO is head of the firm and responsible for the firms policy, an increase in risk taking should be reflected in changes in the firm policy. Risk taking activity is measured through different factors. Investments in Research and Development are typically viewed as one of the most risky investments. Results of these processes are highly uncertain and often quite expensive. If stock option compensation, measured in Vega, provides the CEO with risk taking incentive this should be reflected in the R&D investments. As mentioned by Coles et al. (2006); one way to increase risk would be to reallocate investment dollars away from tangible assets, such as capital expenditures, toward intangible assets, such as R&D. Therefore; H1: Stock option compensation (measured in Vega) is positively related to investments in R&D expenditures. Compared to R&D investments, capital expenditures are far less risky. Kothari et al.(2001) found evidence that Research and development investments generate more uncertain future returns. Investments in Capital expenditures can be seen as a relatively low risk investment, so; 12

H2: Stock option compensation (measured in Vega) is negatively related to capital expenditures. Another way in which managers can increase risky activity is through acquisitions. Research has shown that acquisitions and divestures are risky by their nature and acquisitions result in significant variance in returns. Furthermore it is suggested that acquisitions are functions of executive financial incentives (Sanders 2001). There is a great variance in the outcome of acquisitions; some produce very large gains while others result in losses. However, as options bear only upside potential and no downside risk managers are motivated to chase those potentially large gains notwithstanding the possible threat of great losses. Resulting in the following hypothesis; H3: Stock option compensation (measured in Vega) is positively related to firm acquisition activity. Leverage is part of the capital structure of the company. Capital structure theories argue that company value can be increased with the use of debt. The usage of leverage has both up and downside potentials. First, debt comes with a cost. This can be agency costs associated with claim dilution, asset substitution or underinvestment. More widely acknowledged costs are financial distress and bankruptcy costs. However, these costs are not affecting the personal wealth of the manager, these are firm specific costs. Empirical studies found that an increase in leverage is associated with an increase in stock value (Masulis, 1988). Options do not bear the risk of decreasing stock prices, its value will only increase if the stock price rises above the strike price. CEO s compensated with stock option will therefore aim to increase firms stock returns. H4: Stock option compensation (measured in Vega) is positively related to firm leverage. In summary, following prior research and theory I expect that a higher Vega will increase investments in R&D, lower the investment in capital expenditures, increase firm acquisition activity and a higher book leverage. If there exists a relationship between the Vega and risk taking activity this will probably be observed both cross sectional among firms and across time. 13

4.Model explanation 4.1.Measuring incentive effects and option valuation Core and Guay (2002) measure the incentive effect of stock option compensation by using the Black- Scholes model, modified by Merton (1973) to account for dividends. The aggregate value of total stock option in the portfolio is taken as option portfolio value. Option value = C = Se dt N (d 1 ) K e rt N (d 2 ). Where and C=Stock option value S= Price of the underlying stock K= exercise price of the option d= Dividend rate T= Time to maturity r= Risk free interest rate σ= volatility of stock returns (annualized) Explanation of the calculation of all the above stated parameters can be found in appendix A. The sensitivity of CEO wealth to stock price, Delta, is defined as the dollar change in CEO s wealth for a change in the stock price 1%. Delta measures the incentive to the increase stock price. The sensitivity to stock-return volatility, measured by Vega, is the change in CEO wealth in dollars for a one percentage point change in the annualized stock-return volatility. Incentives to increase risk are measured by Vega. Sensitivity to stock-price change of 1% is defined as: 14

0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 Master thesis Toby Verlouw 120255 Sensitivity to stock-return volatility (annualized) change of 0.01 is defined as: Graph 1 illustrates the importance of including variation in option characteristics into the estimates of the sensitivities Vega and Delta. 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Sensativity to stock price (Delta) Sensativity to stock return volatility (Vega) Graph 1. The Delta and Vega as a function of the price-to-strike ratio. The results are based on an underlying stock option with a maturity of 7 years for a stock price of $100. The dividend yield is 2.5% and the risk free rate is 7%. The annualized standard deviation of stock returns is set at 0.30. The exercise price of the option is determined by the price-to-strike ratio. Table of calculations can be found in appendix The graph plots the price-to-strike ratio against the dollar change per option. The price-to-strike ratio is calculated as the stock price divided by the exercise price of the option and measures to what extend the option is in or out the money. A price-to-strike ratio below 1 indicates that the option is out of the money while a value above 1 indicates the option is in the money. The graph clearly shows that both sensitivities vary significantly for changes in the price-to-strike ratio. We observe that as options move out of the money they become less sensitive to changes in stock prices while their sensitivity to volatility increases steadily. Contrary, options that are in the money are less sensitive to volatility and increasingly sensitive to changes in stock price. From this we can state that when options are deep out of the money increasing managerial risk taking can be expected. As the volatility of the underlying asset increases so does the likelihood that the option eventually ends up in the money. 15

The graph shows that Delta is increasing in price-to-strike ratio. As the option moves from deep out of the money to deep in the money sensitivity to stock price increases, exposing managers to possible substantial losses on their options. This seems to be consistent with the findings of Lewellen (2003) suggesting that in the money options could discourage risk taking activity. Managers preference for debt versus equity possibly declines after option grants or a stock price increase. In practice such events can be offset in two different ways, by granting new options or the cash out of options. When new options are granted the next year, last years incentive effects and current price should be taken into account when setting a specific price-to-stock ratio. Next, firms can allow managers to cash out their options before maturity. However, as option valuation models like Black and Scholes state that options exercised before maturity are valued higher than intrinsic value, this comes with a cost, and therefore not preferable. Usage of this model is consistent with earlier research by, Core and Guay (1999, 2002a), Cohen et all. (2000) and Rajgopal and Shevlin (2002). However, there remains some debate of whether the model is suitable for valuing managerial stock options. Since these stock options are non transferable and managers tend to be risk averse, these options are not held until maturity in most cases. However, Coles et al. (2006) did some further research by modelling how the holding period of stock option varies with volatility and found that the model holds. 16

5.Sample collection, variable measurement and descriptive statistics 5.1. Sample description The data is compiled from the Wharton Research Data services website, wrdsweb.wharton.upenn.edu/. I gathered data on the compensation and stock-based wealth for 155 corporate CEO s for the period 2004-2011. Execucomp provides data on salary, bonus and stock holdings for the corporate CEO. Firm specific data is obtained by using CRSP/Compustat merged database. Stock prices are gathered using the Center for Research in Security Prices (CRSP). I started by ranking the largest 1000 firms available in the Compstat database by market value for the year 2001. As there is a period of 3 years between the ranking date and measurement period the potential bias of only including successful firms is reduced. Firms for which the data is not available in the CRSP database are deleted. Next, firms are sorted on Global Industry Classification Standard (GICS) codes. Because of other reporting standards Financials and Utilities are removed. For some firms either the proxy statements are incomplete or data is not available for the whole period 2004-2011. Firms with no December ending fiscal year are dropped. Only companies reporting in US dollar currency are used. After filtering a sample of 155 US firms remains. 5.2.Descriptive statistics The focus of this research is on the influence of stock option compensation on CEO risk taking activity. To start, we have to calculate the option Vega and Delta. Vega is defined as the change in dollar wealth of the CEO for a 0.01 change in the annualized standard deviation of stock returns. Delta is the change in CEO wealth (dollars) for a change of 1% in the stock returns. The Vega and Delta are calculated by using the Black, Scholes and Merton (1973) formula following Core and Guay (1999), and widely used in earlier conducted studies. Details on the model are explained above and can be found in appendix A. First the Vega and Delta of a single option are calculated 1. In order to calculate the total change in wealth for the CEO we need to define the stock option portfolio. Information on the CEO stock option holding is gathered using Execucomp. Total option holdings are defined as the sum of total unexercised exercisable options and unexercised unexercisable options of the CEO. For the calculation of Vega we multiplied the single option Vega by the total option holdings of the CEO. Delta is calculated in a similar way. I multiplied the single option Delta by total option holdings of the CEO and added the change in total stock value for a 1% change in the stock price. The option parameters used in the Black-Scholes Merton model (1973) are taken as an annual average. One 17

option Vega and Delta are calculated for the accompanying year. Details on these calculations can be found in appendix B. Table 1 presents summary statistics on Vega, Delta and cash compensation. Furthermore firm characteristics, research variables and control variables are displayed. Consistent with earlier literature Core and Guay, 1999; Coles et al.,2006) Vega, Delta and cash compensation are winsorized at the 1 st and 99 th percentile. By winsorizing these variables outliers are not dropped out of the sample but take on a value equal to the specific percentiles. Mean (median) of Vega is $320.164 ($182.102), mean (median) Delta is $1065.876 ($465.797) and mean (median) cash compensation is $1496.866 ($1113.955). For Vega this implies that for a 0.01 change in annualized volatility the average change in CEO wealth is $320.164. Table 1 Summary statistics. Data on the executive compensation for the period 2004-2011 from Execucomp. Vega is the dollar change in the CEO s wealth for a 0.01 change in the annualized volatility of returns. Delta is the change($) in CEO wealth for a change of 1% in the stock price. The sum of salary and bonus is taken as the Cash compensation. All values are stated in dollars. Vega, Delta and Cash compensation are winsorized at the 1th and 99 th percentile. Mean Standard deviation 25 th percentile 50 th percentile (median) 75 th percentile CEO Characteristics Vega ($000s) 320.16 388.58 67.16 182.10 403.96 Delta ($000s) 1065.88 2274.87 206.88 465.80 994.58 Cash compensation ($000s) 1496.87 1185.65 875.38 1113.96 1600.00 Risk measures R&D 0.026 0.044 0.000 0.006 0.032 CAPEX 0.048 0.046 0.019 0.035 0.061 Aqui 0.024 0.055 0.000 0.003 0.023 Booklev 0.213 0.159 0.117 0.185 0.282 Control variables Firm size 8.941 1.270 8.092 8.936 9.684 Surplus cash 0.091 0.087 0.030 0.069 0.126 Sales growth 0.059 0.203 (0.002) 0.071 0.140 Return on 0.159 0.077 0.109 0.152 0.203 assets(roa) Net PPE 0.283 0.228 0.102 0.206 0.401 The risk measures I consider are (1)Research and Development (R&D), defined as the investments in research and development divided by the total assets of the firm; (2)CAPEX, which are the capital expenditures minus the sale of property, plant and equipment, scaled by assets; (3)Aqui, which is 18

defined as the cash outflow of funds used for and/or the costs relating to acquisition of a company scaled by assets; (4)Booklev, which is the long term debt scaled by assets. The control variables are included to represent forces that influence the Vega and Delta together with the risk measures and are based on earlier conducted studies. The following control variables are used, (1)Firm size, defined as the logarithm of sales; (2)Surplus cash, which is the amount of cash available for the financing of new projects divided by the total assets (Richardson, 2002); (3)Sales growth, the ratio of the sales of the current year to the sales in the previous year; (4)Return on assets (ROA), defined as the Earnings before interest, taxes, depreciation and amortization, scaled by assets; (5)Net PPE, which are the investments in property, plant and equipment divided by total assets. Furthermore, the risk aversion of the CEO s is controlled for. According to existing literature, discussed earlier, the CEO tenure and CEO cash compensation are a good proxy for risk aversion. However, CEO tenure is not included in this study since for every company in the sample there was no change in the employed CEO. Further details on the construction of all variables used are presented in appendix B. 19

6.Results 6.1. Vega on R&D, CAPEX, Aqui and Booklev In this section the results derived from the regression will be discussed. It is examined whether the Vega induces managers to take on a more risky investments policy. The dependent variables are Research and Development, Capital Expenditures, Acquisitions and the Book leverage of the firm. Following the theory we expect a positive relation between Vega and R&D, Acquisitions and Book leverage and a negative one towards capital expenditures. For the Delta measure we expect it to be exactly the other way around, a negative relationship between Delta and R&D, Acquisitions and Book leverage and a positive relation on CAPEX. Table two reports on the estimates from the regression of Vega and Delta on the risk taking measures. Following other studies we set R&D and Acquisitions equal to zero if missing from Compustat. Control variables, discussed earlier, are incorporated in the regression as additional explaining factors thereby increasing the total explained percentage by the model, R 2. Table 2 regression of risk taking measures on CEO wealth. The dependent variables are Research and Development, Capital Expenditures, Acquisitions and Book leverage. All these dependent variables are scaled by assets. Vega is explained as the change in dollar wealth of the CEO for a change of 0.01 in stock return volatility. Delta is the change in CEOs wealth for a change of 1% in stock returns. Control variables are explained in appendix B. p-values are reported in parentheses. Statistical significance is indicated by ***,** and * which is respectively a significance level of 1%, 5% and 10%. R&D CAPEX Acquisitions Book leverage Vega 0.0236*** 0.0030 (0.258) 0.0044 (0.367) 0.0163 (0.202) Delta -0.0075 (0.183) 0.0023*** -0.0005 (0.558) -0.0049** (0.016) Cash compensation -0.0017 (0.164) Firm size -0.0155*** Surplus cash 0.1358*** Sales growth 0.0066 (0.650) ROA 0.0374** (0.026) Net PPE -0.0223*** 0.0010 (0.275) -0.0080*** -0.0090 (0.432) 0.0287*** (0.007) 0.0969*** 0.1445*** 0.0012 (0.453) -0.0089** (0.011) -0.0850*** 0.0874*** 0.0016 (0.944) -0.3867*** 0.0071* (0.093) -0.0482*** -0.4569*** -0.0841 (0.104) -0.2384*** 0.1147*** Observations 1063 1063 1063 1063 R 2 20% 59% 5% 15% 20

Table 2 indicates a positive, and significant at the 1% level, coefficient of Vega on R&D (0.0236). From this measure we can argue that a higher Vega will result in higher R&D expenditures, consistent with the theory. Vega has an economically significant effect on Research and Development expenditures. For Delta we observe a negative value of -0.0075 indicating that an increase in Delta is followed by a decrease in Research and Development expenditures. However, this value is statistically not significant and therefore we cannot infer any value to this coefficient. Next, we consider the Vega on CAPEX. According to the theory the relation of Vega and Delta on capital expenditures should be negative and positive, respectively. A coefficient of 0.0030 for Vega on CAPEX is observed which is in contrast to the theory. Yet, the coefficient is statistically insignificant so nothing can be concluded from this measure. Looking at Delta, a positive value of 0.0023, statistically significant at the 1% level, is observed. So as Delta increases so do the capital expenditures. Further we observe an R-squared of 59% indicating that the capital expenditures are for 59% explained by the model used, which is quite large. Looking at the regression results of Vega on acquisitions the model estimated coefficients of Vega and Delta of 0.0044 and -0.0005 respectively. The positive coefficient indicates that as Vega increases so do the acquisitions of the company. Still, it is immediately observed that both Vega (0.367) and Delta (0.558) are insignificant, even at the 10% level, so no importance can be attached to these coefficients. The R2 of this regression is only 5%. From this it can be concluded that the costs related to acquisitions of the firm are minimally explained by the model variables. For the last dependent variable, book leverage, we expect a positive relation with respect to Vega. However, this positive relation is not widely accepted. There remains the discussion of whether the exits a positive or negative relationship. John and John (1993) state that firms with high leverage structure managerial compensation in a way that reduces Vega, so managers will choose low risk projects and shareholders bear lower costs of financial distress. I used book instead of market leverage because according to Welch (2004), market leverage can change passively. Not by active managerial choice but just because of the changes in the stock price. Besides, market leverage has a direct effect on stock price volatility thereby, perhaps, influencing CEO s incentives. The estimated coefficient in the model is 0.0163 indicating a positive relation. Yet, also this variable is statistically insignificant. The observed coefficient for Delta is -0.0049, significant at the 5% level. As Delta increases the book leverage of the company decrease, which is in line with the theory. From the results presented in table 2 it is easily observed that quite some coefficients of Vega and Delta are insignificant. In most cases it is possible to draw a (significant) relation between either Vega or Delta. In no case Vega and Delta are both significant explanatory coefficients on the CEO risk taking activity. Especially Vega, which is only significant for the research and development expenditures. Therefore, it is interesting to do some further research in the causation of these risk taking measures. 21

Robustness check As mentioned earlier, for the risk taking measures R&D and Acquisitions missing values in Compustat are set to zero. In order to check whether this choice has a major influence on the regression outcome robustness checks are performed for these variables. All missing values from Compustat are left out of the regression. This results in a decrease of 466 observations for R&D and 406 for Acquisitions. The remaining data is regressed and presented in table 3. On the left the new (robust) regression results are presented, earlier regression results are included for easy comparison. First we consider the research and development expenditures. For the robust regression we observe a small change in sign for the Vega coefficient, it decreases slightly. The change in Delta is more rigorous. The sign changed from a negative to a positive coefficient, implying that a higher Delta is associated with higher R&D expenditures. Still, the coefficient is insignificant. The same change occurs for the variable Acquisitions. The sign for Delta in the robust regression is positive (0.0008) whereas in the normal regression it was negative (-0.0005). Yet, both coefficients remain highly insignificant and nothing can be concluded from them. These changes in sign for both variables can be easily explained through the fact that for every missing value 0 was used. The average R&D and Acquisitions are larger in the robust regression. Table 3 (Robust) regression results of R&D and Acquisitions on CEO incentives. The dependent variables are Research and Development and Acquisitions. Dependent variables are scaled by assets. Vega is explained as the change in dollar wealth of the CEO for a change of 0.01 in stock return volatility. Delta is the change in CEOs wealth for a change of 1% in stock returns. Control variables are explained in appendix B. p-values are reported in parentheses. Statistical significance is indicated by ***,** and * which is respectively a significance level of 1%, 5% and 10%.. R&D (robust) R&D Acquisitions (robust) Vega 0.0174*** 0.0236*** 0.0044 (0.003) (0.553) Delta 0.00269-0.0075 0.0008 (0.279) (0.183) (0.647) Acquisitions 0.0044 (0.367) -0.0005 (0.558) Cash compensation -0.0016 (0.385) Firm size -0.0268*** Surplus Cash 0.1251*** Sales growth -0.0266 (0.266) ROA 0.1161*** Net PPE -0.0169 (0.262) -0.0017 (0.164) -0.0155*** 0.1358*** 0.0066 (0.650) 0.0374** (0.026) -0.0223*** 0.0008 (0.763) -0.0123** (0.024) -0.0917** (0.013) 0.1124*** 0.0260 (0.487) -0.0329** (0.027) 0.0012 (0.453) -0.0089** (0.011) -0.0850*** 0.0874*** 0.0016 (0.944) -0.3867*** Observations 597 1063 656 1063 22

R 2 24% 20% 4% 5% 6.1.1 Endogeneity in the model A positive association between Vega and stock return volatility indicates that Vega is used to implement high risk taking activity. Is this really the case? Or does it suggest that the association between Vega and volatility is driven by some other omitted variables? One issue remaining is endogeneity in the model. A model is said to be endogenous when there is a correlation between a variable and the error term. Endogeneity is caused as a result of measurement error, auto regression, omitted variables etc. The main question addressed in this study is whether stock option compensation influence managerial risk taking. Ideally if I could run an experiment where CEO compensation is changed randomly and see how the risk taking activity reacts to it, I am able to answer this question accurately. However, since I am dependent upon the available data this is not possible. This implies that the regression coefficients potentially suffer from this endogeneity problem and are biased. Therefore empirical methods are unlikely to capture the magnitude of the causation. Causality may be driven by omitted variables; in this case the endogeneity comes from a confounding variable. A confounding variable is a variable that correlates with both the independent and the dependent variable and can be explained in the following way. Say we have a linear regression y=α+βx i +γz i +u i. If we omit z i because we did not know about this factor affecting the variable y, z i will be absorbed by the error term ε. The actual estimated model becomes, y=α+βx i +ε i where ε= γz i +u i. If x is correlated with z and z affects y, then x is correlated with the error term and y is not only dependent on Alfa and Beta, but also on z and gamma. Reflecting this story on my study, R&D expenditures tend to be positively correlated with Vega, this can be explained that Vega implements higher R&D or some other factors both influence R&D and Vega. By omitting these factors the regression coefficient potentially suffers from endogeneity in the model. In order to reduce the endogeneity in the model we could make use of instrumental variables. An instrumental variable estimates the causal relationship when controlled experiments are not feasible. We could presume that the Vega and Delta are chosen to implement the second-best, value maximizing investment decisions. Therefore, optimal investment decisions should depend on predetermined general characteristics of the company. By using lagged values of Vega and Delta as instruments the likelihood that the regression results are biased from endogenous regressors is reduced. Instead of using lagged variables usage of predicted and residual estimates of Vega and Delta can be used. To further reduce the possibility of endogenous variables and isolate the effect of compensation on risk taking simultaneous equations models can be used. 23

-.2 0 0 Aquisition expenses.5.2.4.6 Book Leverage 1 1.5 0 -.2 R&D expenses.2.4.6.8 0 CAPEX.2.4.6 Master thesis Toby Verlouw 120255 6.2 Type of relationship In the earlier stated hypothesis development expectations are drawn regarding the relationship between stock option compensation and risk taking activity. Managerial incentives are measured through Vega and the risk taking activity is measured by the four aforementioned parameters. To get an easy and intuitive understanding of these relationships scatter plots are used. Below the 4 risk taking measures are plotted against the Vega. 0.5 1 1.5 2 Vega 0.5 1 1.5 2 Vega 0.5 1 1.5 2 Vega 0.5 1 1.5 2 Vega The red lines are the gross linear expectations regarding the relationships of Vega and the risk taking variables. The lines are just rough indication lines to predict positive or negative relationships. From these scatter plots it can be easily observed that, except for the CAPEX, the plots do not follow the expectations. The two-way graph of CAPEX on Vega shows some resemblance with the expectation. Furthermore, for all four of the plots most of the observations have a Vega of smaller than 0.5. About 980 of the total 1240 observations have a Vega smaller than 0.5. On flaw in these pictures is the linear expected relationship. From the stated hypothesis I expect R&D, Acquisitions and Book leverage to be an increasing function in Vega and Capital expenditures to be a decreasing function in Vega. However, this is not de case by definition. The relationship between the Vega and the risk taking variables can be a convex or concave utility function. Therefore 24

-.2 0 0 Acquisition expenses.5.2.4.6 Book Leverage 1 1.5 0 -.2 R&D expenditures.2.4.6.8 0 CAPEX.2.4.6 Master thesis Toby Verlouw 120255 it is interesting to take a closer look at the data, and test for this possible convexity. In order to test for these convex or concave relationships we use the curvilinear relationship regression in stata. Curvilinear regression makes use of several transformations of its variables to achieve its fit. A simple example of a curvilinear function is; Y =b0+b1x1+b2x2, for which X2=X1X1. The linear model is transformed to a quadratic model, thereby allowing for curvature. (For details on the regressions in STATA see appendix 3). Resulting in the following scatter plots. 0.5 1 1.5 2 Vega 0.5 1 1.5 2 Vega 0.5 1 1.5 2 Vega 0.5 1 1.5 2 Vega The red lines indicate the fitted variables of the regression. We observe only minor curvature in all of the scatter plots. For R&D we observe a slightly concave function while for both CAPEX and book leverage a minor convex function. R&D expenses increase in Vega till the Vega reaches the value of 1.25. This part is in line with the theory as it predicts R&D will increase in Vega. Theory tells us that CAPEX is expected to decrease in Vega. From the scatter plot we can tell that up to the point where Vega reaches a value of approximately 1 capital expenditures decrease in Vega. The fitted line in the relation of acquisitions and Vega is neither convex nor concave. From these results it is hard to state specific relationship properties. Further, it is quite interesting to check whether there exist differences between groups. The separation of the sample in groups is done in two different ways. First, the Vega is set into four equal 25

groups, with a width of 0.5. Easily, it can be concluded from the scatter plots that most observations fall in the first and second group, for which Vega is smaller than 1. The second way of grouping is done by using percentiles. 4 percentiles are taken, meaning that the sample is split up in four groups with equal number of observations. 6.3. Vega grouping As the sample is divided into four equal groups of Vega it is easily observed that most observations have a Vega smaller than 0.5, which was also found in the scatter plots. We further observe that for the groups with a Vega larger than 1 no single regression coefficient is significant. This can be explained by the fact that only 75 observations are included. Because of that we leave these coefficients for what they are. If we focus on the groups with a Vega smaller than 1 we see that, in comparison to Table 1, the Vega and Delta coefficients are significant for the same policy measures. The Vega coefficient is only significant in R&D and Delta in CAPEX and Book leverage. From this results it can only be stated what was already done in table 1. However, 988 observations fall in the first two groups; this causes the fact that most results are quite the same. It is therefore of interest to make another distinction, using percentiles. Table 4 Regression of policy measures grouped by Vega. The dependent variables are Research and Development, Capital Expenditures, Acquisitions and Book leverage. All these dependent variables are scaled by assets. Vega is explained as the change in dollar wealth of the CEO for a change of 0.01 in stock return volatility.. p-values are reported in parentheses. Statistical significance is indicated by ***,** and * which is respectively a significance level of 1%, 5% and 10%. R&D CAPEX Acquisitions Mean Vega Delta Mean Vega Delta Mean Vega Delta Vega<0.5 0.5<Vega<1 1<Vega<1.5 Vega>1.5 0.0230 0.0396*** (0.001) -0.0003 (0.717) 0.0479 0.0016 (0.856) 0.0021*** 0.0243-0.0146 (0.341) -0.0001 (0.913) 0.0334 0.0305 (0.122) -0.0014 (0.112) 0.0480 0.0208 (0.125) 0.0027*** 0.0248 0.0246 (0.480) -0.0013 (0.388) 0.0460 0.0057 (0.902) -0.0009 (0.896) 0.0442 0.0221 (0.381) -0.0014 (0.707) 0.0236-0.0539 (0.427) 0.0068 (0.479) 0.0337 0.0023 (0.956) -0.0126 (0.118) 0.0520 0.0107 (0.746) 0.0060 (0.329) 0.0258-0.0048 (0.947) -0.0057 (0.665) Book Leverage Mean 0.2164 0.1951 0.18 0.2377 26

# of observations Vega Delta 0.0243 (0.564) -0.0074*** (0.004) -0.0183 (0.772) 0.0013 (0.648) 0.0455 (0.677) 0.0085 (0.596) 836 152 44 31-0.0325 (0.725) 0.0014 (0.935) 6.4.Percentile grouping In the process of grouping the sample using percentiles I decided to use four percentiles. Meaning that the sample is divided into four equal groups with equal number of observations. In order to run the regressions I made use of dummies. First the observations are ordered from small to large values of Vega. For the 1th percentile only the first 25 percent of the observations are numbered 1 all other zero. For the second percentile the next 25% of the observations are numbered 1 and again all other zero, and so on for the last percentiles. In Table 5 I included the means of the policy measures. This for quick understanding and relationship matters. For example, looking at the Research and Development costs we see an increase in the average R&D costs. Capital expenditures are decreasing in the percentiles. The mean in the first percentile is 0.0537, decreasing in the second and third percentile to a value of 0.0480 in the 4 th. These results are in line with hypothesis 1 and 2. Furthermore, looking at the regression coefficients regarding R&D we observe significant values for Vega in the 1th and 4 th percentile and Delta is only significant in the 4 th percentile. Vega is positively related to R&D and Delta negatively. This is exactly in line with the theory, suggesting that more option holdings result in larger R&D expenditures, whereas larger stock holding result in a decrease in R&D costs. In CAPEX still only the Delta is significant for the last three percentiles, Vega is not significant in any of the percentiles. Acquisitions remain a difficult policy measure to explain. In any of the regressions I ran so far acquisitions can be explained by the Vega or Delta. All regression coefficients are insignificant. Apparently acquisitions cannot be explained from option and stock holdings. One explanation for this is the fact that acquisitions are not part of the everyday decision making. The process of acquiring a firm is long and complicated. Many factors, people and choices are involved. From beginning to end the process of firm acquisitions may take several years. In the 1th percentile Vega is highly negatively related to book leverage, and statistically significant. This in contrast with the theory which suggests that book leverage should be increasing in Vega. So for the smallest 25 percent of the observations (ranked by Vega), there exists a negative relation between Vega and book leverage. As Vega increases the book leverage decreases. This could 27

however be explained by the fact that in the first percentile some observations have a Vega value of 0, resulting from winsorizing. As a robustness check we drop out the observations with a Vega of 0, the resulting regression coefficient is -0.9118 with a p-value of 0.06. This still implies the negative relationship between Vega and book leverage for the 1th percentile. For Delta we observe a negative dependence with book leverage, statistically significant, in line with the theory. Table 5 Regression of policy measures grouped by percentile. The sample is divided into four percentiles each containing 25% of the observations. The dependent variables are Research and Development, Capital Expenditures, Acquisitions and Book leverage. All these dependent variables are scaled by assets. Vega is explained as the change in dollar wealth of the CEO for a change of 0.01 in stock return volatility.. p-values are reported in parentheses. Statistical significance is indicated by ***,** and * which is respectively a significance level of 1%, 5% and 10%. Percentile 1th 2th 3th 4th R&D Mean Vega CAPEX Acquisitions Book Leverage # of observations Delta Mean Vega Delta Mean Vega Delta Mean Vega Delta 0.0191 0.3446*** 0.0002 (0.799) 0.0537 0.1441 (0.127) 0.0013 (0.151) 0.0280-0.0811 (0.640) -0.0008 (0.641) 0.2178-1.1980*** (0.007) -0.0141*** (0.001) 0.0224-0.0065 (0.928) 0.0013 (0.502) 0.0465-0.0406 (0.509) 0.0036** (0.031) 0.0216 0.0092 (0.925) -0.0013 (0.624) 0.2242 0.5179 (0.133) -0.0194** (0.039) 0.0264 0.0165 (0.754) -0.0004 (0.797) 0.0435 0.0180 (0.532) 0.0030*** 0.0249-0.0287 (0.627) 0.0004 (0.805) 0.2088 0.0891 (0.482) -0.0027 (0.469) 0.0350 0.0115** (0.031) -0.0015** (0.035) 0.0480 0.0040 (0.273) 0.0024*** 0.0232 0.0102 (0.206) -0.009 (0.418) 0.2009 0.0120 (0.449) 0.0009 (0.670) 266 266 266 265 6.5. Importance of determinants of option Vega So far in this study we observed that in the regression results the Vega is only a significant variable in relation to the Research and Development costs. For every other policy variable the regression coefficients remain insignificant. Now let s take the study one step back. During the regressions we ran so far Vega and Delta are taken as the independent variables explaining the firm policy expenditures. Both of these variables are highly influenced by the number of option and stock holdings in the portfolio of the CEO. Vega depends on stock options and Delta on the sum of option and stock holdings. For now we take these highly influential parameters as independent variables. 28

Instead of using Vega and Delta, we use the number of option holdings and the total portfolio, number of options and stocks, as the independent variables. Regression these variables against our dependent policy variables we get the following outcome; R&D CAPEX Acquisitions Book Leverage Option Holdings 0.0040*** 0.0005 (0.390) 0.0021** (0.038) 0.0088*** (0.001) Total Portfolio -0.002 (0.530) 0.0008*** -0.0008** (0.031) -0.0011 (0.261) Cash -0.0018 0.0008 0.0010 0.0034 compensation Firm size -0.0126-0.0070-0.0086-0.0496 Surplus cash 0.1329-0.0126-0.0880-0.4728 Sales growth 0.0015 0.0316-0.0859-0.0884 ROA 0.0433 0.1011-0.0019-0.2510 Net PPE -0.0259 0.1442-0.0391 0.1128 When looking at the results we immediately observe that the coefficients for the Vega, although small are significant, except for the Capital expenditures. Option holdings are positively related to R&D, Acquisitions and book leverage. These results are far more in line with what the theory suggests. From these results the hypothesis stated in chapter 3 can be answered. However, it is too quickly to conclude in such way. Using these results the importance of the sensitivity of the CEO s wealth is omitted. According to earlier conducted studies and partly stated in the agency theory a managers' wealth should be tied to the firm. Therefore the main interest should be how sensitive managers wealth change to firm performance. This is what we did earlier in this study by using the Vega and Delta as independent measures. What we can conclude from this regression is the fact of the importance of the determinants of Vega and Delta. Coles et al. (2006) state that still very little is known about the determinants of the Vega. We do not go into detail of the specific determinants but do explain that the Vega and Delta determinants of single options cannot be neglected in the study of option compensation. Just focusing on the number of portfolio holding is far too straightforward. 29

6.6. The influence of the banking crisis In 2008 a global crisis in finance occurred. Due to this crisis stock markets collapsed. We discussed earlier that the regression results above are not quite significant. Could the collapse of the stock market play a role in the low significance of the results? The individual option and Delta measures are calculated using the Black-Scholes-Merton model (1973). The outcomes of the Vega and Delta are dependent upon several factors, explained in the model above. One of these determinants of Vega and Delta is the stock price. In comparison to the other determinants stock price is one of the most influential factors in the outcomes of Vega en Delta. During the stock exchange collapse in 2008 many listed companies faced large decreases in stock price. The volatility of stock prices were very high in that time. As these large changes in stock price directly influence the option Vega and Delta it is interesting to look whether this 2008 stock exchange collapse has influenced earlier results. Therefore we test again for causation between the policy variables and Vega and make a distinction in the time period. The total sample was taken from 2004 to 2011. For now we take the period until the crisis, so 2004 till 2007, compared to the period since the crisis, 2008 till 2011. First taking a quick look at the summary statistics of the Vega measure for these two periods we found that the standard deviation for the period 200-2007 is 0.3318 compared to a standard deviation of 0.4254 for the period 2008-2011. This larger standard deviation is, probably, caused by the large changes in the stock prices during the crisis. From the regression results we found no new understandings in the relationships. Vega remains positive significant towards R&D expenditures, where there is only a small change in the magnitude of relationship coefficient. The period before the crisis state a positive significant value for CAPEX. So from the period 2004-2007 we found a positive relationship between Vega and the capital expenditures. In line with the earlier regression results the Delta remains positive significant at the 1% level, same for both periods. Again acquisitions remain unexplained by the Vega and Delta measure. From this regression result we can only conclude that the crisis does have an impact on the Vega but does not change the regression results significantly. Only small differences are observed comparing both periods and which are quite the same with earlier results. 30

Table 6 Regression of risk taking measures on CEO wealth. The dependent variables are Research and Development, Capital Expenditures, Acquisitions and Book leverage. All these dependent variables are scaled by assets. Vega is explained as the change in dollar wealth of the CEO for a change of 0.01 in stock return volatility. Delta is the change in CEOs wealth for a change of 1% in stock returns. Control variables are explained in appendix B. p-values are reported in parentheses. Statistical significance is indicated by ***,** and * which is respectively a significance level of 1%, 5% and 10%. R&D CAPEX Acquisitions Book leverage 2004-2007 2008-2011 2004-2007 2008-2011 2004-2007 2008-2011 2004-2007 2008-2011 Vega 0.0372*** 0.0211*** 0.0112** -0.0005 0.0105 0.0024 0.0175-0.0038 (0.010) (0.880) (0.256) (0.689) (0.427) (0.818) Delta -0.0011-0.0006 0.0023*** 0.0022*** -0.0003-0.0009-0.0030-0.0057* (0.164) (0.428) (0.001) (0.766) (0.426) (0.249) (0.068) Cash compensation -0.0030-0.0023 0.0009 0.0005-0.0007 0.0024 0.0062 0.0154 Firm size -0.0098-0.0185-0.0129-0.0039-0.0100-0.0072-0.0546-0.0475 Surplus cash 0.1587 0.1218-0.0148-0.0055-0.1071-0.0734-0.5117-0.4374 Sales growth 0.0146 0.0018-0.0005 0.0385 0.1548 0.0557-0.2011-0.0105 ROA 0.0609 0.0183 0.1202 0.0795 0.0123-0.0017-0.3468-0.1336 Net PPE -0.0268-0.0187 0.1468 0.1431-0.0315-0.0446 0.1397 0.0906 Observations 456 607 456 607 456 607 456 607 R 2 22% 22% 63% 57% 6% 5% 20% 14% 31

7.Conclusions This study provides evidence on the causal relationship between CEO compensation and managerial decision making, specifically in the investment and debt policy. The main variable of interest in the CEO compensation structure is the sensitivity of the CEO wealth to stock volatility. Besides the sensitivity of CEO wealth to performance, measure in Delta, is considered. During the study we found that a higher Vega implements more risky investments. From the results we derived we can accept our first hypothesis, stating that stock option compensation is positively related to the investments in Research and Development. For the remaining hypotheses to little evidence is obtained to accept them. We cannot conclude that an increase in stock option compensation increases the acquisition expenses and book leverage, and decrease the capital expenditures. If we just focus on the number of options and stocks in the portfolio of the CEO, neglecting the fact of sensitivity, we observe positive relations with the number of options and the expenses in R&D, acquisitions and book leverage. If the number of options in the CEOs portfolio increases this will result in larger expenditures in Research and Development, Acquisitions and higher book leverage. According to earlier conducted studies and theory the relationship to Delta should run in the opposite direction. An increase in Delta should result in a decrease in the risky investments and debt policy. We found evidence that an increase in Delta results in an increase in the Capital expenditures and a decrease in book leverage. So if a CEO his wealth is more sensitive to changes in the stock price he will probably invest more in riskless capital expenditures and lower the book leverage of the firm. The sample used runs from 2004 till 2011. Since in 2008 a large collapse on the stock market occurred we tested for the effects of the crisis on our results. During the collapse of the stock exchange we observed many stock prices changes. Since the Vega measure of an option in highly dependent upon the stock price this resulted in an increase in the standard deviation of the Vega. Still, influencing the Vega, the resulting causation does not significantly differ in the period before and during the crisis. Last, it is important to mention that the resulting regression coefficients are probably biased. For the regression to have unbiased estimates the independent variables Vega and Delta should be exogenous. Since I am not able to run a controlled experiment the Vega and Delta might be endogenously determined. From this I am not able to give a conclusive answer to the main research question. 32

8. Recommendations and limitations Much prior research has been done in the field of compensation structure and the effects on the firm and employees. These studies differentiate on the way the causation runs. Cohen et al. (2000) and Guay (1999) uses the Vega as the independent variable. These studies found a significant positive relationship between the Vega and both firm leverage and stock return volatility. Other studies used the Vega and Delta as depend variable and found evidence of a relationship between Vega and firm characteristics (Guay, 1999, Bizjak et al., 1993 and Core and Guay, 1999). From this it can be easily concluded that the causation runs in both directions. Compensation plans influence firm policy and vice versa. In this research the Vega and Delta are used as independent variables. The study focus on how stock option compensation influences firm policy and therefore neglects the fact of how firm policy influences the compensation structure. Only one direction of causation is considered in this study and thereby a limitation. The resulting coefficient estimates of the ordinary least squares may be biased as the variables probably suffer from endogeneity. In order to reduce this problem lagged values of Vega and Delta should be used. If we go even one step further, instead of using an OLS regression for the estimation of the parameters, the data could be approached with simultaneous equations models. 33

9. References Aggarwal, R., Samwick, A., 1999. The other side of the tradeoff: the impact of risk on executive compensation. Journal of Political Economy 107, 65-105. Agrawal, A., Mandelker, G., 1987. Managerial incentives and corporate investment and financing decisions. Journal of Finance 42, 823-837. Bizjak, J., Brickley, J., Coles, J., 1993. Stock-based incentive compensation and investment behavior. Journal of Accounting and Economics 16, 349-372. Carpenter, J., 2000. Does option compensation increase managerial risk appetite? Journal of Finance 55, 2311-2331. Coles, J., Daniel, N., Naveen, L., 2006. Managerial incentives and risk-taking. Journal of Financial Economics 79, 431-468. Core, J., Guay, W., 2001. Estimating the value of employee stock option portfolios and their sensitivities to price and volatility. Journal of Accounting Research Vol. 40, No. 3. Core J., Guay, W., 2002a. Estimating the value of employee stock option portfolios and their sensitivities to price and volatility. Journal of accounting research 40, 613-630. Guay, W., 1999. The sensitivity of CEO wealth to equity risk: an analysis of the magnitude and determinants. Journal of Financial Economics 53, 43-71. Hall, B., Liebman, J., 1998. Are CEOs really paid like bureaucrats? Quarterly Journal of Economics 113, 653-691. Haugen, R., Senbet, W., 1981. Resolving the agency problems of external capital through options. Journal of finance 36. 629-647. Holderness, C., Kroszner, R., Sheehan, D., 1999. Were the good old days that good? Changes in managerial stock ownership since the great depression. Journal of Finance 54, 435-469. Jensen, M., Meckling, W., 1976. Theory of the firm: managerial behavior, agency costs and ownership structure. Journal of financial economics 3, 305-360. Jensen, M., Murphy, K., 1990. CEO incentives: it s not how much you pay, but how. Harvard business review, No. 3, 138-153. John, T., John, K., 1993. Top-management compensation and capital structure. Journal of Finance 48, 949-974. Ju, N., Leland, H., Senbet, L., 2002. Options, option repricing and severance packages in managerial compensation: their effect on corporate risk. Working paper. University of Maryland. Kahneman, D., Tversky, A., 1979. Prospect theory: An analysis of decision making under risk. Econometrica 42, 262-291. Lewellen, K., 2003. Financing decisions when managers are risk averse. Working paper, MIT. 34

Masullis, R. The debt/equity choice. Cambridge Massachusetts: Ballinger Pub. Co., 1988. May, D., 1995. Do managerial motives influence firm risk reduction strategies? Journal of Finance 50, 1291-1308. Mehran, H., 1995. Executive compensation structure, ownership, and firm performance. Journal of financial economics 38, 163-184. Rajgopal, S., Shevlin, T., 2002. Empirical evidence on the relation between stock option compensation and risk taking. Journal of Accounting and Economics 33, 145-171. Sanders, W., 2001. Behavioral responses of CEO s to stock ownership and stock option pay. Academy of management journal 44, No. 3, 477-492. Sanders, W., Hambrick, D., 2007. Swinging for the fences: the effects of CEO stock options on company risk taking and performance. Academy of management journal 50, No. 5, 1055-1078. Smith, C., Stulz, R., 1985. The determinants of firms hedging policies. Journal of finance and Quantitative analysis 20, 391-405. Tufano, P., 1996 Who manages risk? An empirical examination of risk management practices in the gold mining industry. Journal of finance 51, 1097-1137. 35

10.Appendix Appendix A For measuring the sensitivity of stock options to volatility the Black and Scholes model, modified by Merton (1973) to account for dividends, was used. The measurements are based on the valuation of a European call option; therefore option value is calculated using: Option value = C = Se dt N (d 1 ) K e rt N (d 2 ). Where and S is the price of the underlying stock, N is the cumulative probability function of the normal distribution, K is the exercise price of the option, r the risk free interest rate, d expected dividend rate, T the time to maturity of the option (years) and σ the annualized volatility of stock returns. In the process of valuing European call options all the aforementioned parameters should be available. However, for some companies values of these parameters are missing. Therefore it is necessary to estimate these missing values, in the following way: S = Underlying stock price on 31th of December, for the period 2004-2011 K= Exercise price of the option measured as the weighted average price of exercisable options. For about 5 percent of the options this value is not available. The exercise price for these options is set as the weighted average exercise price of the stock options outstanding at year-end. Stock options outstanding refer to the number of options that have been granted to date, which have not been exercised or cancelled. T= Remaining time to maturity of the option, in years. I use the expected life of the option used in fair value calculations, available in the CRSP/Compustat merged database. σ = Annualized volatility, estimated as the standard deviation of daily returns over the year, for period 2004 till 2011, multiplied by the number of trading days in each year to the power 0.5 (t in days, t 0.5 ). r = Risk free interest rate. I used the average risk free rate used in the fair value calculation of stock options, available in the CRSP/Compustat merged database. This estimate excludes the stock purchase plans or any other non-option plans when reported separately from the option assumption section. The risk free rate could not be obtained for about 15 percent of the options. Since the r in 36

the Black Scholes Merton formula is not of great influence i used the average risk free rate, set at 3.25% for the missing values. d = Expected dividend rate over the life of the option. The expected dividend rate is calculated as Ln(1+dividend rate), where the dividend rate is the annual dividend per share for the period 2004-2011 divided by the year-end stock price. 37

Appendix B This appendix provides further explanation on the variables used in this study. CEO compensation data is compiled using Execucomp. Firm specific data is gathered using the CRSP/Compustat and Compustat databases. Stock return data is taken from CRSP. Vega and Delta measures The Vega used in this study is calculated in the following way. We start by calculating the single annual option Vega; For every parameter in this formula only one single annual value is taken.. This value measures the sensitivity of the option to a change of 0.01 in volatility, called V. In order to calculate the total change of CEO wealth for a change of 0.01 in volatility we multiply the single option Vega with the total option held by the CEO; V*# of options. This value of Vega is used in all the regressions in this study. The Delta is calculated in a similar way. First the Delta of a single option is calculated; This single option Delta, D, is multiplied by the number of options held by the CEO; D*# of options. Since the Delta is the dollar change in CEO wealth for a 1% change in stock price we should also include the stock holdings of the CEO. Therefore, stock holdings (# of stock) and stock prices (S) are gathered. The change in the stock wealth is calculated as 0.01*S*# of stock. By summing the change in option wealth and stock wealth; (D*# of option)+(0.01*s*# of stock), we find the Delta. This Delta measure is used in the study. Risk measures R&D=Research and Development expenditures/total assets. CAPEX= (Capital Expenditures Sale of Property, Plant and Equipment)/total assets. Acquisitions= Acquisitions/total assets. Book leverage=long term debt/total assets. Control variables Firm size= Logarithm(sales). Surplus cash= Cash/total assets. Sales growth=logarithm(sales (n)/sales(n-1)). 38

Return on assets (ROA)= EBITDA/total assets. Net PPE= Net investments in Property, plant and equipment/total assets. Cash compensation= Salary + Bonus 39

-.2 0 CAPEX.2.4.6 0 R&D expenditures.2.4.6.8 Master thesis Toby Verlouw 120255 Appendix C Scatter plot details 0.5 1 1.5 2 Vega 0.5 1 1.5 2 Vega 40

0.5 Book Leverage 1 1.5 -.2 0 Acquisition expenses.2.4.6 Master thesis Toby Verlouw 120255 0.5 1 1.5 2 Vega 0.5 1 1.5 2 Vega Stata function description 41

R&D. regress rd c.vegawinthous##c.vegawinthous Source SS df MS Number of obs = 1232 F( 2, 1229) = 15.94 Model.060543226 2.030271613 Prob > F = 0.0000 Residual 2.3344758 1229.001899492 R-squared = 0.0253 Adj R-squared = 0.0237 Total 2.39501902 1231.001945588 Root MSE =.04358 rd Coef. Std. Err. t P> t [95% Conf. Interval] vegawinthous.0382627.0084582 4.52 0.000.0216687.0548567 c. vegawinthous# c. vegawinthous -.0152352.0052325-2.91 0.004 -.0255008 -.0049696 _cons.0174377.0019962 8.74 0.000.0135213.021354. linktest Source SS df MS Number of obs = 1232 F( 2, 1229) = 16.22 Model.061587774 2.030793887 Prob > F = 0.0000 Residual 2.33343125 1229.001898642 R-squared = 0.0257 Adj R-squared = 0.0241 Total 2.39501902 1231.001945588 Root MSE =.04357 rd Coef. Std. Err. t P> t [95% Conf. Interval] _hat -.1581952 1.571498-0.10 0.920-3.241311 2.924921 _hatsq 20.31634 27.39065 0.74 0.458-33.42128 74.05395 _cons.0153627.0212474 0.72 0.470 -.0263225.0570479. ovtest Ramsey RESET test using powers of the fitted values of rd Ho: model has no omitted variables F(3, 1226) = 1.94 Prob > F = 0.1211. predict p (option xb assumed; fitted values) (2086 missing values generated). scatter rd p vegawinthous, msym(o i) con(. l) aspect(1) sort. 42

CAPEX. regress capex c.vegawinthous##c.vegawinthous Source SS df MS Number of obs = 1232 F( 2, 1229) = 1.53 Model.006580029 2.003290015 Prob > F = 0.2163 Residual 2.63797999 1229.002146444 R-squared = 0.0025 Adj R-squared = 0.0009 Total 2.64456002 1231.002148302 Root MSE =.04633 capex Coef. Std. Err. t P> t [95% Conf. Interval] vegawinthous -.0151956.0089912-1.69 0.091 -.0328353.0024442 c. vegawinthous# c. vegawinthous.0096649.0055622 1.74 0.083 -.0012476.0205774 _cons.0502981.002122 23.70 0.000.046135.0544613. linktest Source SS df MS Number of obs = 1232 F( 2, 1229) = 1.64 Model.00705976 2.00352988 Prob > F = 0.1935 Residual 2.63750026 1229.002146054 R-squared = 0.0027 Adj R-squared = 0.0010 Total 2.64456002 1231.002148302 Root MSE =.04633 capex Coef. Std. Err. t P> t [95% Conf. Interval] _hat -3.806704 10.18249-0.37 0.709-23.78369 16.17028 _hatsq 48.76656 103.1441 0.47 0.636-153.5915 251.1246 _cons.1180877.2512584 0.47 0.638 -.3748551.6110306. ovtest Ramsey RESET test using powers of the fitted values of capex Ho: model has no omitted variables F(3, 1226) = 1.21 Prob > F = 0.3031. drop p. predict p (option xb assumed; fitted values) (2086 missing values generated). scatter capex p vegawinthous, msym(o i) con(. l) aspect(1) sort 43

Acquisition expenses. regress aqui c.vegawinthous##c.vegawinthous Source SS df MS Number of obs = 1232 F( 2, 1229) = 0.08 Model.000479559 2.000239779 Prob > F = 0.9254 Residual 3.79982175 1229.0030918 R-squared = 0.0001 Adj R-squared = -0.0015 Total 3.80030131 1231.003087166 Root MSE =.0556 aqui Coef. Std. Err. t P> t [95% Conf. Interval] vegawinthous -.003935.010791-0.36 0.715 -.0251058.0172359 c. vegawinthous# c. vegawinthous.0026291.0066757 0.39 0.694 -.0104679.0157261 _cons.0251255.0025468 9.87 0.000.0201289.030122. linktest Source SS df MS Number of obs = 1232 F( 2, 1229) = 0.34 Model.002129889 2.001064945 Prob > F = 0.7086 Residual 3.79817142 1229.003090457 R-squared = 0.0006 Adj R-squared = -0.0011 Total 3.80030131 1231.003087166 Root MSE =.05559 aqui Coef. Std. Err. t P> t [95% Conf. Interval] _hat 55.71146 74.91235 0.74 0.457-91.25878 202.6817 _hatsq -1088.15 1489.067-0.73 0.465-4009.545 1833.245 _cons -.6868883.9420272-0.73 0.466-2.535048 1.161271. ovtest Ramsey RESET test using powers of the fitted values of aqui Ho: model has no omitted variables F(3, 1226) = 0.69 Prob > F = 0.5608. predict p (option xb assumed; fitted values) (2086 missing values generated). scatter aqui p vegawinthous, msym(o i) con(. l) aspect(1) sort 44

Book Leverage. regress booklev c.vegawinthous##c.vegawinthous Source SS df MS Number of obs = 1232 F( 2, 1229) = 2.33 Model.111793655 2.055896827 Prob > F = 0.0978 Residual 29.4916543 1229.023996464 R-squared = 0.0038 Adj R-squared = 0.0022 Total 29.603448 1231.024048292 Root MSE =.15491 booklev Coef. Std. Err. t P> t [95% Conf. Interval] vegawinthous -.0551116.0300629-1.83 0.067 -.1140918.0038686 c. vegawinthous# c. vegawinthous.0235567.0185979 1.27 0.206 -.0129304.0600438 _cons.2236298.0070951 31.52 0.000.2097099.2375496. linktest Source SS df MS Number of obs = 1232 F( 2, 1229) = 2.58 Model.123967939 2.061983969 Prob > F = 0.0759 Residual 29.47948 1229.023986558 R-squared = 0.0042 Adj R-squared = 0.0026 Total 29.603448 1231.024048292 Root MSE =.15488 booklev Coef. Std. Err. t P> t [95% Conf. Interval] _hat 16.70305 22.04659 0.76 0.449-26.55006 59.95617 _hatsq -37.64571 52.84171-0.71 0.476-141.3157 66.02424 _cons -1.633695 2.295254-0.71 0.477-6.136745 2.869355. ovtest Ramsey RESET test using powers of the fitted values of booklev Ho: model has no omitted variables F(3, 1226) = 0.38 Prob > F = 0.7675. predict p (option xb assumed; fitted values) (2086 missing values generated). scatter booklev p vegawinthous, msym(o i) con(. l) aspect(1) sort 45