Master Thesis. Does Executive Compensation Influence Credit. Default Swap Spreads?

Transcription

1 Master Thesis Does Executive Compensation Influence Credit Default Swap Spreads? By Lisa Senders ANR s Finance Department Supervisor: Dr. Alberto Manconi Second reader: Dr. Paul Sengmüller October 2012

2 Preface This Master Thesis concludes my master degree in finance and my student life in Tilburg, leaving me with mixed feelings. On the one hand, I am very proud and happy to have finished this thesis, but on the other hand this also means I have to say goodbye to the wonderful time I had in Tilburg. In the past five years I not only had the opportunity to learn about (business) economics and especially finance, likewise I was able to develop myself in a professional manner by doing several committees and a full-time board year at MAK, a committee and two study tours at Asset Accounting & Finance and by representing students of Tilburg School of Economics and Management at the Faculty Council ECCO. Finally, the moment comes to write the master thesis and I honestly have to admit that I have liked most of the endless hours working with Stata and writing the thesis itself. It was challenging and every little step forward was satisfying. Of course, I also experienced moments of frustration. This brings me to thanking everyone that has encouraged me in these moments and during the rest of my study period. First, I would like to thank my supervisor, Dr. Alberto Manconi. I am very thankful for your dedication and enthusiasm throughout the process. The weekly updates, on which I always received quick replies, really motivated me to finish something every week. My friends will all confirm that I am very grateful for you being my supervisor, which I told them more than once. Moreover, I would like to express my gratitude to my family for supporting and encouraging me throughout my career at Tilburg University. And last but not least, I would like to thank my friends for supporting me, all the coffee breaks in the library that kept me going, and the faith they had in me. Lisa Senders October, 2012 ii

3 Abstract Although executive compensation and credit default swaps (CDS) are both hot topics in finance literature, the influence of executive compensation on CDS spreads has never been investigated. Therefore this study aims to do so, with a focus on the effect of option-based compensation. I find strong robust results that managerial compensation in stock options leads to higher CDS spreads due to the increased incentive for managers to take risks. Managers paid in stock options take more risk by making larger investments in general, and typically larger investments in intangible assets than tangible assets. Additionally, I find that the effect of option-based compensation on CDS spreads is stronger in industries with higher information asymmetry and in industries with lower product market competition. Keywords: Credit default swaps, executive compensation, managerial risk-taking iii

4 Table of Contents Preface... ii Abstract...iii Table of Contents... iv 1. Introduction Theory development Executive compensation Pay-performance relationship Information Asymmetry Product market competition Managerial risk-taking Credit default swaps Data and Methodology Data collection, variables and sample construction DataStream Compustat Execucomp CRSP Compustat Methodology Model development Sample selection bias Hypothesis testing Results Descriptive statistics iv

5 4.2 Probit regressions The complete regression model Executive compensation CDS spread Concept check Information asymmetry Product market competition Managerial risk-taking Conclusion References Appendix Table 1: Variables List Table 2: Summary Statistics Table 3: Correlation Matrix Table 4: Probit Regression Table 5: Regression model structure and robustness explanatory variable Table 6: Regression model robustness dependent variable Table 7: Regression model - robustness concept Tabel 8: Information Asymmetry Table 9: Product Market Competition Table 10: Managerial risk-taking Graph 1: Sample Selection Bias v

6 1. Introduction Excess risk-taking at financial institutions affects more than just creditors; it affects depositors, taxpayers, and potentially the financial system as a whole. Bolton, Mehran, and Shapiro, 2010 Executive compensation is a hot topic in the academic literature. 1 Recently, and in particular in light of the events of the recent financial crisis, a debate has raged about whether or not executive pay structure encourages excessive risk-taking (Balachandran, Kogut, and Harnal, 2010). In principle, boards of directors design managerial compensation to mitigate the agency problem, creating value for the shareholders. Conventional compensation structures tie the manager s wealth to stock returns or stock return volatility. A commonly used compensation structure includes a stock option component. This kind of structure is known to induce managerial risk-taking, and may thus increase firm risk (Guay, 1999). The finance literature has focused on investigating the contribution of executive compensation to firm risk. The contribution of executive compensation to credit risk as priced by Credit Default Swap spreads, however, to the best of my knowledge has never been investigated. 2 Therefore in this study I aim to fill this gap in the literature. Thus, I aim to investigate the relationship between executive compensation and Credit Default Swap spreads. Credit Defaults Swaps, in short CDSs, are a hot topic in the academic literature 3 as well, mainly for the tremendous growth of the CDS market in the past decades and the ability of CDS spreads to price credit risk. A CDS provides insurance against the risk of a credit event, commonly the default of a company. The higher the riskiness of a firm, the higher the premium is that the CDS holder pays to the issuer for insuring a principal amount (Hull, Predescu, and White, 2004). It is interesting to see how executive compensation structures that affect 1 For example see: Bolton, Mehran, and Shapiro (2010), Coles, Daniel, and Naveen (2006) and Guay (1999). 2 This is what I infer from all publicly available information. Comparable studies are Coles et al. (2006) that investigate the impact of executive compensation on firm risk and Balachandran, Kogut, and Harnal (2010) that investigate the impact of executive compensation on the probability of default. 3 For example see: Blanco, Brennan, and Marsh (2005), Longstaff, Mithal, and Neis (2005) or more recent Breitenfellner and Wagner (2012) 1

7 managerial behavior and thus their risk-taking, can be related to the riskiness of the firm reflected in the CDS spread. Hence, the main research question in this study is: What is the impact of executive compensation on the CDS spread of a company? As managers compensated with larger components of stock-options are known to face increased incentives for risk-taking, I focus on option-based compensation structures instead of other pay structures with less obvious effects on managerial incentives or potentially subject to manipulation, such as restricted stocks and bonuses. The literature 4 on executive compensation focuses mainly on the pay-performance relationship as predicted by the agency theory (Jensen and Meckling, 1976). Other subjects of research in this field include what the influence is of different levels of information asymmetry, the firm s competitive environment and the investment policy on this relationship. 5 To give a more complete picture on the relationship between option-based compensation and CDS spreads, I investigate the influence of these three factors on this relationship. To answer the research question and the effect of these factors, I design a set of testable hypotheses. The literature on executive compensation suggests that option-based compensation should be associated with a higher CDS spread, which forms the central hypothesis for my thesis. Subsequently, the other hypotheses broaden the perspective on this relationship, not only with the just mentioned factors, also by investigating the relationship under different conditions of information asymmetry and product-market competition. This research design represents an innovation, compared to other studies in the field of executive compensation, in which I focus on CDS spreads as the main dependent variable. One key issue is that the set of CDS reference entities is not exogenously determined, influencing the characteristics of the firms in the data set. CDS reference entities tend to be larger, stable firms that are likely to have relatively high credit-ratings (Blanco, Brennan, and Marsh, 2005). In other words, they are safer firms. Failing to adequately control for these characteristics could thus lead to underestimate the impact of the risk-taking incentives that managers derive from their compensation structure on the CDS spread. To address this issue, I resort to the procedure proposed by Heckman (1979), mitigating this sample selection bias. 4 For example see Coles, Daniel, and Naveen (2006), Guay (1999), Jensen and Murphy (1990) amongst others. 5 See Myers and Majluf (1984), Alexander and Zhou (1995), Aggarawal and Samwick (1999), or Schmidt (1997). 2

8 My findings can be summed up as follows. I find a strong, positive relationship between managerial incentives as given by option-based compensation and CDS spreads. The effect is not only statistically significant, but also, and more importantly, economically meaningful: one standard deviation increase in option-based compensation for CEOs results in an increase of 4.43% in the 5-year CDS spread, relative to its mean value. This is consistent with the view that the CDS spread incorporates the effect of the risk-taking incentives contained in managerial compensation. Additional results provide suggestive evidence about how managers undertake riskier policies. CEOs with larger option-based components of compensation make larger investments in general, and in particular in intangible assets. In addition, the relationship between optionbased compensation and CDS spreads is stronger in industries with higher information asymmetry, where it is harder to monitor the management, and in industries with lower product market competition, where managers are less pressured to undertake safe firm policies. The relationship between executive compensation and CDS spread is not necessarily stronger for CEOs than the average management team member. These findings contribute to the knowledge of both executive compensation and Credit Default Swaps. This is highly useful for audiences concerned with managerial (excessive) risktaking, including investors, debt holders, boards of directors, compensation consultants, and financial analysts. Moreover, this study enables me to contribute to the academic literature. The remainder of this thesis is organized as follows. In section 2 I give the theoretical background on executive compensation and CDSs on which I build the hypotheses. In section 3 I cover the data that I used, and I explain the methodology. In section 4 I present the results. And finally in section 5 I provide the conclusion. 3

9 2. Theory development In this chapter I will analyze the literature on executive compensation and credit default swap spreads and formulate the hypotheses for this research. 2.1 Executive compensation The payment of executive has been subject of discussion ever since Holmstrom (1979) showed that the compensation of managers should be tied to the firms performance to act in the shareholders interest and increase equity value, relating the agency problem to the executive compensation. The literature on executive compensation that has evolved focuses mainly on the pay-performance relationship as predicted by the agency theory (Jensen and Meckling, 1976). 6 Other interesting subjects of research in this field include what the influence is of different levels of information asymmetry, the firm s competitive environment and the investment policy on this relationship. This paragraph covers these subjects, which I use to formulate hypotheses Pay-performance relationship The idea behind increasing the sensitivity of the manager s wealth to the share price is that the manager is induced to either work harder or more effectively, since the manager shares in the gains and losses of the firm. However, as Coles, Daniel, and Naveen (2006) state, there is another effect of tightening the payment of the mangers to the performance of the firm. The managers are exposed to more risk, when increasing the sensitivity of the manager s wealth to the stock price. Modern portfolio theory would suggest that these managers have undiversified positions, which they should try to diversify by selling (part of) their position in the firm or other shares which could reduce the unsystematic risk they are exposed to. Additionally, given the fact that managers can be risk averse, it is likely that managers will forgo risky investments with a positive net present value. Thus, one effect of equity-based compensation 7 can be riskreduction or even value destruction. Thus, managing the slope between managers wealth and stock price might not be sufficient to control for the agency problem. According to Smith and Stulz (1985), the payment 6 Studies on the pay-performance relationship include Coles, Daniel, and Naveen (2006), Guay (1999), Jensen and Murphy (1990) amongst others. 7 When referring to equity-based compensation in this study, the payment of managers in stocks (not stock options) is meant. 4

10 in stock options leads to a higher sensitivity of compensation to firm performance, since stock options are increasing in value with stock return volatility. According to Hall and Liebman (1998), the use of stock options in the compensation of managers has been increasing rapidly in the 1990 s, resulting in the growth of the sensitivity of compensation to firm performance. They add to this statement that it is possible that firms not only increased the payment of managers in stock options for this reason, an alternative reason can be that the payment in stock options is less visible. This is desirable with the growing public opposition to high payment levels, while boards are possible competing for executive talent or beholden to their CEOs. Summarizing, the payment in stock options give clearer incentives for managerial risktaking than payment in stock. Option-based compensation is therefore expected to increase the riskiness of the firm. Since firm risk increases the chances of a possible default, I expect that the CDS spread will increase as well. From this I can infer that the compensation in options leads to higher CDS premiums, summarized in hypothesis 1: Hypothesis 1: There is a positive association between option-based compensation and CDS spreads of a firm. What is not explicitly noted in this statement is whether the executive compensation refers to the CEO of the company or other executives. Chava and Purnanandam (2009) investigated the difference between decisions of the CEO and other managers, such as the CFO. They find that the CEOs incentives are significant in decisions concerning capital structure and cash holdings, while the CFOs incentives are more important in the debt-maturity structure and accrual decisions of the company. This corresponds to Bertrand and Schoar (2003) that find considerable heterogeneity across managers. From these findings I infer that it is very likely that the behavior of CEOs is quite homogeneous, compared to the other more diverse management team members. Therefore, when comparing the impact of the payment of CEO and the payment of the average management team member on the CDS spread, I expect that the CEOs will show a stronger effect. I formulated this in hypothesis 2: Hypothesis 2: The option-based compensation of CEOs has a stronger impact on CDS spread than average management team option-based compensation. Additionally, the pay-performance relationship as discussed above suggests that option-based compensation has a positive impact on the firm performance and it also has an impact on the 5

11 share price itself. Since there always should be a risk-return trade-off, I infer that option-based compensation has a positive impact on the share price as well. Price-to-book value ratio being a well known indicator for firm performance, I expect that option-based compensation has a positive impact on this ratio, which allows me to check if the pay-performance relationship exists in this study as expected. This brings me to the following hypothesis: Hypothesis 3: Option-based compensation has a positive impact on price-to-book value ratio Information Asymmetry As the agency problem states there exists a misalignment of interest between the principal (shareholder) and agent (manager), since the agent might undertake actions that maximize his own wealth at the cost of the principal. This situation is created by the existence of information asymmetry. The manager has information and knowledge about the firm, which the shareholders do not have. As monitoring is very costly, the compensation contracts which tighten the manager s wealth to the firm are designed (Jensen and Meckling, 1976). A following interesting question in my research is what the impact of the information asymmetry within the firm is on the relationship between option-based compensation and the CDS spread? Information asymmetry increases the agency problem and therefore the need for compensation contracts that mitigate the problem (Myers and Majluf, 1984). Managers that operate in firms with relatively high information asymmetry, 8 are more free to act in own interest. Since the option-based compensation contracts, increases the sensitivity of the manager s wealth to the volatility of the stock returns, it is of the manager s own interest to take more risk. From this I infer that the effect of option-based compensation on CDS spreads will be stronger in industries with relatively high information asymmetry. This is stated in the following hypothesis: Hypothesis 4: The impact of option-based compensation on CDS spreads is stronger if there is high information asymmetry. 8 High information asymmetry can be interpreted with the existence of high monitoring costs. 6

12 2.1.3 Product market competition What the agency theory typically fails to recognize, is that the managers operate in an environment of strategic interactions between firms, an imperfect competitive market (Aggarawal and Samwick, 1999). It would be optimal to design a compensation contract that is passed on a relative performance evaluation of the manager that filters out the industry shocks. This should enhance the optimal strategic product market choice of the manager. However, literature is inconclusive about the use and effect of relative performance evaluation. Though, it may be stated that higher product market competition induces managers to increase their effort, to reduce the probability of liquidation and is accompanied with reduced profit margins, which might reduce the managerial effort (Schmidt, 1997). Furthermore, increasing competition generates additional information, which mitigates the agency problem at least partly. When regarding this effect in the framework of this study, it is ambiguous what the effect of option-based compensation is on CDS spreads in high product market competition. However, the situation of low product market competition can be interpreted: the profit margins are relatively fat and information scarce compared to the high product market competition industries. From this I infer that the effect of option-based compensation on CDS spreads will be stronger in industries with low product market competition than in industries with high product market competition, since the managers have more leeway to take risk under these circumstances. Hypothesis 5: The impact of option-based compensation on CDS spreads is stronger in highly competitive markets Managerial risk-taking Besides the circumstances in which the managers take risks and how this is related to the compensation contracts and agency problem, there is wide range of literature investigating the decision-making of the manager. The decisions of the manager can be related to behavioral aspect as for example the prospect theory 9 or managerial characteristics such as the age or the 9 The prospect theory states that managers do have a loss-aversion (Kahneman and Tversky,1979) 7

13 education of the manager. 10 Furthermore, the decisions itself can be analyzed as Coles et al. (2006) do. They investigate the investment policy of the manager in relation to managerial compensation. Their results suggest that managers that are induced to take more risk increase the R&D investments and leverage of the firm and decrease the capital expenditures. I expect that these parities will also hold in this study, which brings me to the following hypothesis: Hypothesis 6: Managers compensated with stock options take risks via their investment decision: a. By investing more b. By investing in risky assets (intangibles) c. By investing less in less risky assets (tangibles) 2.2 Credit default swaps Credit derivatives are an exciting innovation in financial markets. - Hull, Predescu, and White, 2004 Credit derivatives are contracts on the creditworthiness of a firm, designed to trade and manage credit risks. CDSs are the most popular credit derivative, accounting for around half of the credit derivatives market (Longstaff, Mithal, and Neis, 2005). This is how a CDS contract works: the CDS buyer makes periodic payments to the seller and in return receives compensation when a credit event occurs before maturity. This compensation equals the difference between the par value of the underlying bond or loan and its market value after default. The periodic payment is a fee also known as the CDS spread or premium. Typically this fee is quoted in basis points per $100 notional amount of the reference obligation (Longstaff, et al., 2005). CDS spreads price credit risk of the reference entity, which is measured by rating agencies, which base their quality ratings on a number of financial ratios (Bodie, Kane, and Marcus, 2011). Furthermore, CDS contracts are over-the-counter (OTC) traded products, which typically are restricted to the big institutions with relatively high credit ratings. So, the CDS market trades mostly the credit risk of the less risky firms. Given this information, I expect that 10 An example of a study in this field is Bertrand and Schoar (2003). They find that managers from earlier birth cohorts are on average more conservative and that managers with an MBA degree on average take more aggressive strategies. 8

14 the firms selected on the criteria of having a CDS spread will have low firm risk compared to the firms without CDS spread. When defining firm risk as the volatility of the daily stock returns, 11 I expect this volatility to be lower for the firms with CDS spreads. This can be summarized in the last hypothesis: Hypothesis 7: Firms for which CDS contracts are available are associated with lower firm risk than firms without CDS contract. 11 Coles, Daniel, and Naveen (2006) define firm risk as mentioned here. The same definition will be used in this study. More information about the variable firm risk is given in the appendix, table 1. 9

15 3. Data and Methodology In this chapter I describe the data and methodology of this study. In the first section I cover the data collection, the sample selection and variable construction. In section 2 I elaborate on the methodology used in this study, starting with correction for a sample selection bias and subsequently I describe the regression models and tests applied to test the hypotheses. 3.1 Data collection, variables and sample construction The data sample used in this study is constructed with data extracted from different sources, which are merged to form the complete data sample. The final data set consists of CDS spreads, executive compensation data, stock data and firm and industry characteristics for the time period January 1999 until December This is an optimal time period, since the CDS data starts in However, the executive compensation data and some of the financial statement data start in 1998, because lagged variables are used. Given that yearly variables are included in this study, the period ends at the last finished year, The universe of firms included in this study is US based, which is the most comprehensive database available. CDS spread data is retrieved from DataStream, data on the executive compensation is retrieved from Compustat Execucomp, stock data is retrieved from the Center for Research in Security Prices (CRSP) and data on firm and industry characteristics are retrieved from Compustat. Compustat Execucomp, Compustat and CRSP are accessible via Wharton Research Data Services. 12 These databases are the most comprehensive and widely used databases in the finance literature. Below I will elaborate on the data used and variables constructed per database, starting with the CDS data DataStream Thomson Reuter s DataStream is an extensive database including CDS data that have a monthly interval. In this study, CDS5 will be the most relevant variable. This spread with a five year maturity is the most liquid spread amongst the available CDS spreads. 13 The variable CDS5 is 12 For more information see 13 Cds1 to cds10 are available, ten different financial instruments with a maturity of one to ten years respectively. However CDS5 is commonly used in empirical work. See for example Blanco et al. (2005) and Longstaff et al. (2005) 10

16 after outlier treatment 14 at the 5%-percentile ready to be used in the data analysis. CDS5 is the main dependent variable in this study Compustat Execucomp The Execucomp database contains over 2800 companies, listed at the Standard & Poor's 1500 plus. The executive compensation data is on a yearly basis. The variables option-based compensation (OBCO) and equity-based compensation (EBCO) 15 are constructed with the items total compensation (TDC1), restricted stock grants (rstkgrnt), option awards valued using the Black and Scholes option valuation method (option_awards_blk_value). More details on the construction of the variables can be found in the appendix, table 1. Option-based compensation and equity-based compensation give respectively the percentages of the manager s compensation in stock options of the company and the manager s compensation in restricted stock grants to the total compensation. The study focuses on option-based compensation, which will be the main explanatory variable in this study. Compustat Execucomp gives data on multiple managers per firm, which allows me to determine compensation on both CEO-level and on management team-level. The compensation of CEOs is extracted using the variable CEOANN, indicating the CEO per firm per year. The compensation on management team-level is extracted by calculating the average compensation of all the managers per firm per year listed in the database. The focus is on the compensation of CEOs equivalent to hypothesis 2. The compensation variables will be matched to the CDS spread data of one year later, to attenuate possible simultaneity 16 concerns CRSP CRSP maintains a large collection of stock prices, returns and trading volumes for the companies trading at the AMEX, NASDAQ and NYSE stock exchanges. With stock prices and the number of outstanding shares I construct the variable market value, which is used to 14 The outlier treatment used is dropping the observations below the 5 th percentile and above the 95 th percentile. 15 Equity-based compensation will only be used to check robustness of the results and is therefore not reported later on. 16 There could be a simultaneity problem for the compensation data and CDS spread data, both variables can influence each other. 11

17 determine the book-to-market value ratio and the price-to-book value ratio. Other variables from this database included in this study are trading volumes, variances of daily stock prices 17 and bid- and ask prices Compustat Balance sheet data from Compustat are used to construct variables on the firm and industry characteristics, which are mostly controlling variables in this study. Firm characteristics include size, return on assets, cash holdings ratio, investment index, investment in tangible or intangible assets, book value, and leverage ratio. I approximate product market competition with the industry concentration for every industry per time period. The industry is defined by the Standard Industry Classification code (SIC). The industries will be defined by the first digit of this code, grouping firms in the same industry. 3.2 Methodology This section elaborates on the methodology used in this study. First the main regression model is presented, then the procedure used to control for the data bias and subsequently I will elaborate on the methodology used to test the hypotheses Model development This study will focus on the effect of executive (option-based) compensation on CDS spreads. OLS -regression with these variables and a set of control variables is a suitable approach to test this relation, where executive compensation the explanatory variable is of the dependent variable CDS spread. The regression equation is as follows:, (1) The explanatory variable compensation is lagged to minimize possible simultaneity concerns, as mentioned in paragraph To investigate this relationship between CDS spreads and executive compensation as complete as possible, I estimate a set of different regressions. However, before elaborating on these additional regressions, it is important to acknowledge that there is a limitation to the just constructed model. This study intents to make inferences 17 The logarithm of the daily stock return volatility is a proxy for firm risk, following Coles et al. (2006). 18 This is a proxy for Information Asymmetry, following (Amihud and Mendelson, 1989) 12

18 about the universe of firms in the US, while only firms having a CDS spread are included in this regression equation. 19 This bias is known as the sample selection bias and Heckman (1979) was the first to derive a procedure mitigating the problem. The next section will elaborate on the bias and procedure Sample selection bias In this paragraph the sample selection problem will be explained based on Heckman (1979) and Greene (2000), so that I gradually come to the methodology that I used in this study to control for this bias. Suppose that we want to estimate parameters of a regression for a population of firms, which can be formulated as follows: (2) Where is the dependent variable, are the explanatory variables and the error term. Under the classical regression conditions, Ordinary Least Squared (OLS) estimates. However, in this study I consider a subsample of firms which are selected by the condition of having a CDS available. The selection bias problem arises due to an incidental truncation of the sample (Greene, 2000). If the selection choice per firm is called, the selection equation is: Where (3) (4) In this study is the dummy variable indicating whether the firm has a credit default swap. is a vector of factors that influences whether the firm has a credit default swap and is the error term. The sample can be summarized in the outcome equation combining the regression of interest and the sample selection according the selection equation: 0 0 (5) 19 As far as I know, there are no other papers in this field using this procedure. However I am the first to combine CDS data with executive compensation data in this framework. Not to confuse with the study of Balachandran, Kogut, and Harnal (2010) which have a similar investigation, but explain probability of default instead of CDS spreads. A study in the field of CDS spreads using the procedure given by Heckman (1979) is Pires, Pereira, and Martins (2008). 13

19 The outcome equation can be elaborated as follows: (6) OLS regression of on will give consistent estimates of, if the errors and are independent, since the last term simplifies to 0. However, when the two errors terms are correlated, we need to obtain. Greene (2000) provides: = (7) Where is the inverse of Mill s ratio, indicating the probability that a firm is selected. I will refer to this ratio as hazard rate. The hazard rate is a ratio of the probability density function (pdf) divided by the cumulative density function (cdf) for a standard normal variable,. and Φ Φ (8) is assumed to be 1 when estimating with a probit regression of the selection equation. Then and Φ Φ (9) When substituting equation 7 into 6, we find the model as proposed by Heckman (1979): (10) I will refer to this model as the sample selection model. I can control for the sample selection problem by interpreting the self-selection as an omitted variable that controls for the factors that influence the selection. Heckman (1979) proposed a two step procedure: - Estimate the selection equation (eq. 3) to obtain estimates of with a probit regression. - With these estimates the inverse of Mills ratio ( ) can be calculated (eq. 9). 14

20 - OLS regression of equation 10, with added as explanatory variable compared to the regression of interest (eq. 1), will give consistent estimates of and. One condition for this procedure is that all explanatory variables in the outcome equation are included in the selection equation, plus at least one more variable that strongly correlates with the probability of the CDS being available ( 0, where can be referred to as the identifying variable (Li and Prabhala, 2007). The variable trading volume will be the identifying variable, since it can be expected that the firms having a CDS spread are large and widely traded firms, for having relatively high credit ratings (Blanco et al., 2005) Hypothesis testing In this paragraph I will combine the regression model as given in paragraph and the sample selection model in paragraph to construct the complete model. Subsequently, I will build on this model to determine the methodology to test the hypotheses. The first step is combining the regression model (eq. 1) with the sample selection procedure. The selection equation can be defined as follows:,,,,,,, (11) Where, refers to the expected value of the indicator variable for the presence of a CDS spread for company i at time t. Compensation refers to the variable indicating the lagged managerial compensation in either equity or options, Trading volume is the identifying variable and the Controls refer to the set of control variables that correspond to the set of control variables in the outcome equation. I estimate the coefficients with a probit regression. The first hypothesis that I will analyze, is hypothesis 7 using the regression results of this probit regression. Subsequently, the hazard rate (eq. 9) can be calculated as follows: 20,, Φ, (12) Where is the probability density function, Φ the cumulative standard normal and, the coefficient that has just been calculated with the probit regression of equation 11. The hazard 20 One note that has to be made here, is that the hazard rate is calculated using the variable option-based compensation as explanatory variable in the probit regression and a set of control variable. When regressing with other variables, for example equity-based compensation as explanatory variable, a different hazard rate is calculated. 15

21 rate will be added to the set of control variables in the outcome equation, so that equation (10) will then be altered to:,,,, (13) Where, denotes the CDS spread for company i at time t. For the composition of control variables in equation 11 and 13, I will follow Coles et al. (2006). The effect of the hazard rate and the control variables can be analyzed, when adding them in steps to the explanatory variable compensation. This is the main model of interest in this study. With this model I test the relation between executive compensation and CDS spreads as formulated in hypothesis 1. Furthermore, I check the robustness by changing the explanatory variable compensation and dependent variable CDS spread. Two different explanatory variables are available, namely the compensation for management team members and the CEOs. I test Hypothesis 2 on the compensation of CEOs versus the compensation of the average management team member by comparing the regression results. Subsequently, I analyze ten different CDS spreads as dependent variable for the robustness. Next to replacing the dependent variable with other variables that should measure the same, the dependent variable can be substituted by the price-to-book value ratio as indicator of performance, to test whether the pay-performance parity holds in this dataset. This enables me to answer hypothesis 3. After having constructed the complete regression model and having checked the robustness of the explanatory and dependent variable, there are still some interesting tests to run. Hypothesis 4 on information asymmetry and hypothesis 5 on the product market competition, should place the relationship between CDS spreads and executive compensation in a broader perspective. To test hypotheses 4 and 5, I split the sample in high and low groups, by the median values of the proxies for information asymmetry and product market competition. Subsequently, a chi-square test 21 should indicate whether there exists a significant difference between the samples. 21 The chi-square test allows comparing the coefficients of two regressions. The null hypothesis states that the coefficients are equal. 16

22 Finally, I analyze hypothesis 6. What is it that managers do, to take extra risk? Following Coles et al. (2006) again on this subject, I investigate the investment policy of the managers with balance sheet items on capital expenditures, tangible asset and intangible assets. 22 I regress the different proxies on the explanatory variable and control variables, so that I can examine the impact of option-based compensation on the investment policy of the manager. Thus, I established the following equation:,,, (14) Where, represents the investment proxy per firm per year. Additionally, in all regressions I apply clustered standard errors, to correct for possible heteroskedasticity and serial correlation of the error terms. When using panel data, it is very likely that the observations are moving together in time (Petersen, 2009). Therefore the cluster date will be used. 22 The item capital expenditures is a proxy for investment level of the manager. A proxy for tangible assets is the item property, plant and equipment and for intangible assets the item R&D expenditures. All the proxies will be divided by lagged total assets, so that the investments will be relative to the size of the firm. 17

23 4. Results In this chapter, I present the results of this study. All regressions and tests as discussed in the methodology are discussed separately. Table 2 of the appendix tabulates the descriptive statistics, table 3 gives a correlation matrix for the most relevant variables and tables 4 to 10 tabulate the regression results. 4.1 Descriptive statistics First, I will touch upon the descriptive statistics tabulated in table 2 of the appendix. To begin with the executive compensation data, the average CEO in this sample receives 19% of his compensation in stock options, and the average management team manager similarly receives 17% compensation in stock options. Looking at the CDS data, I observe that the liquidity of the CDS contracts with a 5 year maturity (CDS5) is reflected in the largest number of observations. The mean value of the CDS5 spread is much lower than its maximum value, with a standard deviation that is about as large as the mean value. This indicates a skewed distribution to the left. The indicator variable dcds gives information about the distribution of firms being a CDS reference entities and counts around 39 hundred observations, of which almost 25 hundred observations are expected to have the CDS5 spread. This corresponds to the mean value of the variable dcds which is Continuing with the variables on firm characteristics, I observe that the variables net income (ni) and return on assets (roa) have a negative minimum value. Thus, bad performers are not excluded from this sample. Last step before starting with the regressions, I analyze the correlation matrix for a large number of variables, tabulated in table 3. The variable firm risk (FR) has correlation coefficients larger than 0.6, with trading volume (TV) and information asymmetry (IA). Therefore, I calculated the variance inflation factors (VIF) for these variables and the results indicate that there are no collinearity problems with these variables. 18

24 4.2 Probit regressions Table 4 in the appendix shows the probit regression, which is the first step in the sample selection model. The regression results for two different explanatory variables 23 are tabulated to show the robustness of the results. First, I examine whether the model is doing what it is supposed to do: estimating the effect of the explanatory variable on the probability of a firm being a CDS reference entity, while controlling for the set of variables in the model and the identifying variable. The model has a good fit, implied by the pseudo R-square of 46.0% and 45.8% for columns 1 and 2 respectively. Furthermore, the identifying variable trading volume (TV) is significant at a 1%- level, which is critical in this procedure. Trading volume has a positive coefficient which is economically 24 meaningful: one standard deviation change in trading volume results in an increase in the indicator variable dcds of 28.81% and 31.61% relative to the mean value for the regressions in column 1 and 2 respectively. These results suggest that trading the expected positive impact has on the probability of the firm being a CDS reference entity. So, the model seems to be doing what it is supposed to. Consequently, I continue with evaluating whether resorting to the Heckman procedure (Heckman, 1979), is necessary: is the sample of CDS reference entities deviating from the total set of observations in terms of firm risk? The coefficient of firm risk is negative and statistical significant at the 10%-level. One standard deviation change in firm risk results in a decrease in the indicator variable dcds of 15.23% and 12.79% relative to its mean value for the regressions in column 1 and 2 respectively. From this economic effect I infer that the firm risk is significantly lower for the sample with CDS reference entities. Additionally, graph 1 graphically shows this relationship. These results corresponds to hypothesis 7, and enable me to conclude that this hypothesis can be adopted. Last, I evaluate the coefficient of option-based compensation. This coefficient is negative and statistically significantly at the 1% level. This implies that option-based compensation has a 23 The variables option-based compensation for the CEO and for the average management team member are used, OBCO and avobco. 24 The economic significance is calculated by multiplying the estimated coefficient of the explanatory variable with one standard deviation of this explanatory variable. This is a one standard deviation change in the dependent variable. I will refer to the percentage change in the dependent variable relative to the average value, obtained by dividing the one standard deviation change by the mean value of the dependent variable times 100%. 19

25 negative impact on the probability of a firm being a CDS reference entity. Associating optionbased compensation with firm risk, this result is not interfering with hypothesis The complete regression model I constructed the regression model by adding the hazard rate and control variables to the explanatory variable. These steps are tabulated in table 5 of the appendix. Before evaluating the effect of the explanatory variable, I analyze the fit of the model in this paragraph. To begin with, the role of the hazard rate in the model. The hazard rate is statistically significant at the 10%-level in column 2 and at the 1%-level in column 3 and 4. The direction of the coefficient is not really relevant; it is most important what the effect of the hazard rate is on the coefficient of option-based compensation. It increases the impact of optionbased compensation on the CDS spread. 26 Continuing with examining the control variables, the effect of option-based compensation on the CDS spread decreases when controlling for other factors that have influence on the credit risk of the firm. Despite this diminishing positive effect, 27 the explanatory power of the model increased quite a bit. The R-squared of the complete model as shown in column 3, is 46%. When further elaborating on the control variables, I argue that the set of variables form an optimal set of control variables in this model. Other compositions of control variables are not included in this paper to avoid obscurity. The included control variables are a proxy for firm size, book-to-market ratio, return on assets, cash holdings ratio, leverage ratio, proxy for product market competition, a proxy for firm risk and date dummies. This set of control variables is similar to those of Coles et al. (2006). All control variables are statistically significant. Book-to-market value ratio, return on assets and firm size have a negative effect on the CDS spread, while the other control variables show a positive effect. Firms having a high book value compared to market value, bigger firms 25 See Guay (1999) 26 An explanation for the hazard rate s change in direction could be that the hazard rate correlates with some of the date dummies. Since the date dummies are addressing a possible omitted variable bias, I choose not to adapt the model. 27 The effect of option-based compensation on the CDS spread will be analyzed in the next section, so that hypothesis 2 can be confirmed or rejected. 20

26 in terms of firm size, 28 and firms generating high returns on every dollar invested in assets, are typically stable the firms, not associated with large risks. Therefore it does make sense that these indicators of the firm s condition are associated with lower CDS spreads. Information asymmetry and product market competition is discussed in detail in the next paragraphs. The proxy for firm risk and the leverage ratio are variables for which the positive association with the CDS spread makes a lot of sense. Leverage increases the firm s risk, as widely discussed in finance literature. 29 The positive association between firm risk and the CDS spread corresponds to the findings of Ericsson et al. (2009) amongst others. The positive association between the cash holdings ratio of a firm and its CDS spread is less obvious. An explanation could be that firms that hold large amounts of cash in the company are firms which do not get money easily through other channels for being relatively risky, but need this cash for investments. This is for example the case for innovative technology firms (Léautier, 2007) Executive compensation First I evaluate the explanatory variables option-based compensation, this variable is statistically significant in the columns 2, 3 and 4, thus with at least the hazard rate as control variable. More important is the economic significance to make inferences. A one standard deviation change in option-based compensation results in a change of -0.62%, 11.75%, 4.43% and 4.99% in the CDS spread relative to its mean value. This positive and economically meaningful effect of option-based compensation on the CDS spread corresponds to hypothesis 1. For robustness, I compare the results of the full regression model on option-based compensation for the CEO, OBCO (column 3) with the full regression model on option-based compensation for the average management team member, avobco (column 4). The regression results are very similar. The coefficient of avobco is somewhat higher than the coefficient of OBCO, and has the same statistical significance. The chi-square test indicates that there is a significant difference between these explanatory variables, though I cannot conclude that the effect of OBCO on the CDS spread is stronger than the effect of avobco. Therefore, I reject hypothesis Firm size is approximated by the natural logarithm of total assets. 29 For example see Hamada (1972), Merton (1974) or Zingales (1998) 21

27 4.3.2 CDS spread I also evaluate the robustness of the results, by investigating the dependent variable the CDS spread. In table 6 of the appendix the regression results for ten different CDS spreads are tabulated. Only four of the ten CDS spreads show statistically significant results in relation to option-based compensation, namely the CDS1, CDS3, CDS5 and CDS7. Option-based compensation is positively directed towards these four CDS spreads. The control variables in these four regressions are statistically significant and directed in the same way. 30 Moreover, the economic effects of option-based compensation on these variables indicate the robustness of the regression results on CDS5. One standard deviation change in option-based compensation results in a change of respectively 13.02%, 9.60% and 4.56% in the CDS spreads CDS1, CDS3 and CDS7 relative to its mean value. However, with these results it is not possible to make inferences about the maturity of the CDS spread in relation to managerial compensation. 31 Compared to the CDS5 spread, CDS1 spread and CDS3 spread do have shorter maturities and option-based compensation has a bigger economic effect on these CDS spreads. CDS7 has a longer horizon; still option-based compensation has a bigger coefficient in this model Concept check Last, I evaluate the concept of the regression model, by investigating whether the payperformance relationship holds, as formulated in hypothesis 3. Table 7 shows the regression results of the base regression model next to price-to-book value ratio. 32 To begin, the coefficient of option-based compensation is in this model positive and statistically significant at the 5%-level related to price-to-book value ratio. One standard 30 There is one exception: the proxy variable for information asymmetry is not significant in the regression model with CDS7, however directed in the same way as in the other regression models. 31 One possible explanation for this inconclusive result could be that the non-default component in CDS spreads is time-varying as stated by Longstaff et al. (2005). 32 In this model the control variables hazard rate and market-to-book value ratio are omitted. The hazard rate is not necessary in this model, since there is no selection bias occurring like in the base regression model. The market-tobook value ratio is calculated as the yearly market value divided by the yearly book value of a firm, while the priceto-book value ratio is calculated as the monthly market value divided by the yearly book value. Since these variables are measuring the same, this control variable is omitted. The variable based on monthly market values is chosen above the yearly market value to be compared to the dependent variable CDS spread, since this variable is monthly as well. 22