Methodology

I consider both annual compensation and cumulative compensation of CEOs. CEOs receive their compensation in a variety of forms. Each pay component received by the CEO during a particular year will be taken into account. I construct pay variables as a proportion of total CEO compensation, which is the sum of annual salary, bonuses, the estimated values of stock options and restricted shares, and the pension increment. In addition, as a robustness check, I also employ an alternative proxy for annual compensation, which is the value of each pay component scaled by the firm‟s total sales. Previous studies on CEO compensation did not consider pension element of pay because that information was not easily available in the early period. However, the disclosure of pension data is now mandatory following the introduction of the Directors‟ Remuneration Report Regulations (2002), which allows us to collect full compensation data to accurately estimate each pay component and analyse the impact of pensions. I focus on the defined benefit (DB) pension, as only DB pensions are a potential liability for a firm and therefore represent inside debts. I hand collect the actuarial value of the defined benefit pensions as reported in firm annual reports. Since defined benefit pension value is reported as cumulative number, I estimate the amount of new pension awarded in a particular year as a year-to-year change in accumulated pension.

24 In addition to the annual monetary amounts of compensation, I look at the total amount of equity-like (stock options and shares) and debt-like pay (pension) accumulated by a CEO during his or her tenures. This has potentially even stronger implications for firm policies, since CEOs are much more likely to be motivated by changes in their total wealth rather than changes in the value of their annual compensation. For stocks and options, I conduct this analysis by using the number of shares grants because monetary values can vary based on valuation assumptions. I therefore define new compensation variables by considering the number of stock options, unrestricted shares (ownership) and restricted shares held by a CEO as a proportion of total number of shares outstanding. These new definitions are also useful to check the robustness of my findings.

The yield spread of a corporate bond is used to measure the cost of debt. Following prior literature (e.g. Anderson, Mansi and Reeb, 2003; Ertugrul and Hegde, 2008), it is estimated as the difference in yield to maturity between a firm‟s bond and a UK government bond with a comparable maturity. The spread is expressed in basis points.⁴ When a firm has multiple bonds outstanding in a year, I use the market value weighted average yield spread. This procedure allows me to use a single representative bond yield per firm year.

I perform ordinary least square (OLS) regression to measure the effect of CEO compensation on the cost of debt. The yield spread of corporate bonds is used as the dependent variable, and the CEO compensation components are used as the explanatory variables. Following prior studies examining yield spread (e.g. Ortiz-Molina, 2006;

4 For a few corporate bonds with a maturity longer the longest maturity of government bonds, the yield spread is compared with the longest available maturity of the latter.

25 Ertugrul and Hegde, 2008; Devos, Prevost and Rao, 2008), the estimated regression model is written as follows:

Spread _i,t = α₀ + β Compensation _i,t-1 + ∑ λ Bond Characteristics _i,t +∑ δ Firm Characteristics i,t + ∑ ζ Industry dummies i,t

+ ∑ υ Year dummies i,t +εi,t. (1)

The regression specification considers a lagged relationship because bondholders adjust the bond price once information on compensation becomes publicly available. All bond-specific information is therefore collected three months after the end of a fiscal year.⁵ We can see how bondholders react to the latest CEO compensation information.

Although OLS regression is popular and widely used, it is not short of limitations. OLS is sensitive to outlier. OLS estimator is biased and inconsistent if multicollinearity and individual effects exist among independent variables (e.g. Greene, 2007). In this study, the possible individual effects are that different firms may apply unique remuneration packages for their CEOs. Hence the variation among CEO compensations may be contaminated by unmeasured individual firm characteristics (unobserved firm heterogeneity). To address such a concern, I also use fixed effect regression, focusing on within variation of CEO compensation for individual firms.

Prior literature suggests several bond and firm characteristics that can also influence the yield spread of bonds (e.g. Ortiz-Molina, 2006; Ertugrul and Hegde, 2008). These factors are included as control variables in regression. The bond characteristics are bond

5 UK Publicly traded firms are required to publish their annual reports within four months after the end of the fiscal year. I also randomly check the date of annual report release in Thomson Banker.

26 rating, duration and bond size. For the bond rating variable, I convert each rating category into a numerical scale. Following Klock, Mansi and Maxwell (2005), I assign the lowest Moody‟s rating D a value of 1, and then, as the bond rating increases, the numerical rating changes by an increment of 1, up to a value of 22 for the highest Moody‟s rating, Aaa. I further convert this rating into a rating residual to control for all information other than compensation that can affect bond rating (spread) and that is not captured by other control variables used in the regression. The residual is estimated from the regression, where the dependent variable is bond rating and the independent variables are the various compensation components. The duration of the bond is used to control for differences in bond maturity and coupon rate. The bond size is used to control the impact of liquidity on yield spread. A large bond size suggests higher liquidity and therefore a lower cost of debt. Similar to Ortiz-Molina (2007), I use the relative bond size (as the fraction of a firm‟s total assets) instead of the absolute bond size. The firm characteristics that can affect yield spread include firm size, debt ratio, profitability, market-to-book ratio and firm risk. These firm characteristics are found to be informative in explaining the cost of debt (e.g. Ortiz-Molina, 2007; Ertugrul and Hegde, 2008). The exact definitions of all these variables are presented in Table 2.1.

The regression model also incorporates industry and time factors.