Journal of Management and Governance 3: 117–136, 1999. © 1999 Kluwer Academic Publishers. Printed in the Netherlands. 117 The Governance of Directors’ Pay: Evidence from UK Companies ANDREW BENITO1 and MARTIN J. CONYON2 1 Department of Economics, University of Warwick and Oxford Economic Research Associates (OXERA), Oxford, UK; 2 Warwick Business School, University of Warwick, UK and The Wharton School, University of Pennsylvania, USA Abstract. We examine the determination of directors’ compensation in UK quoted companies between 1985 and 1994. The primary innovation contained in the paper is the focus on the governance mechanisms that determine pay outcomes. Our results indicate that: (i) directors’ compensation is positively related to pre-dated shareholder returns and company size with the quantitative effect of the latter dominating the former. (ii) We find that the pay-for-performance link has become quantitatively stronger over our sample period. (iii) There has been positive adherence to the principles of the Cadbury report, but these variables play little statistical role in shaping the direct compensation of top directors. 1. Introduction The determination of executive compensation has emerged as an issue of considerable academic and media interest.1 A central theme in the UK debate is whether directors’ pay is adequately tied to measures of corporate performance. Much of the evidence assembled so far indicates that if a link between pay and performance can be established then its magnitude is quantitatively quite small relative to the effect of company size (see Gregg et al., 1993; Conyon, Gregg and Machin, 1995; Conyon and Peck, 1998). Furthermore, Gregg et al. (1993) contend that even this small link has become de-coupled during the early 1990s. Since there is a perception in the UK that director compensation is inadequately tied to corporate performance, and the relationship may have been further weakened in the recent past, this has heightened the debate about appropriate institutional top pay-setting mechanisms. If the pay contracts offered to senior executives do not provide sufficient incentives for management to pursue long-term shareholder interests, then the role of the board in monitoring and compensating the senior executives assumes great importance. It has been in this vein that recommendations concerning governance practices have been made, most notably by the Cadbury Committee Report (1992). However, data on actual top pay-setting institutions and structures in the UK are rare, such that the effectiveness of the board is not empirically well understood for British companies. The substantial 118 ANDREW BENITO AND MARTIN J. CONYON degree of change in top pay-setting structures that we are able to document places a premium upon assessing their impact upon the levels of pay awarded to executives in UK companies. Moreover, these pay awards have been viewed as a symptom of the perceived problems associated with governance practices. There are, therefore, two broad goals to our research. First, we revisit the standard test carried out in the executive compensation literature. Namely, we examine the link between directors’ pay and the stock-market performance of their companies. There are two novel features to this aspect of the paper. First of all, we use a large sample of companies (in excess of 1000) over a relatively long time period from 1984 to 1995. Using panel data econometric techniques, we establish the relationship between directors’ cash compensation and shareholder return. Second, we also examine whether there is any time series variation in the directors’ pay and company performance relationship. Prior research, as indicated above, has tended to conclude that the relationship between pay and performance has become “weaker” over time in the UK. Since there has been comparatively little evidence on this supposed effect, we document whether or not the relationship between directors’ pay and company performance has become stronger or weaker between 1985 and 1994 in our large sample of data. The econometric results of our study in relation to the pay-performance relationship are relatively straightforward to summarise. After controlling for size and macro-economic effects, we are able to isolate a robust positive relationships between directors’ cash compensation and pre-dated stockholder return. Furthermore, our results indicate that the relationship between pay and performance has become quantitatively larger over the time period under investigation. This contrasts with prior UK research but is consistent with recent US investigations. However, we also note that, in general, the quantitative effect of the size variable on directors’ cash compensation is much larger than that of the performance variable. Our second broad goal is to examine the effects, if any, of boardroom governance variables on directors’ pay. This is a relatively unexplored area in the executive pay literature. We have access to unique survey data that indicates whether companies have adopted certain key governance policies and when they did so. We have such data for a sub-sample of 211 companies. We examine the effects of separating the posts of CEO and chairman, and the adoption of remuneration and nomination committees on directors’ pay. These variables might be thought important for directors’ pay outcomes for a variety of reasons and we can ask various questions of our data. Do companies that adopt remuneration committees in the UK have lower levels of pay? Do companies that combine the roles of CEO and chairman have higher levels of pay? The survey instrument was pivotal, as such information about the governance arrangements in UK companies is not typically in the public domain prior to the mid-1990s. Our results relating to the effects of boardroom governance variables turned out to be less than robust. We find that there is little role for these variables to jointly influence directors’ cash compensation. THE GOVERNANCE OF DIRECTORS’ PAY 119 The rest of the paper is organised as follows. Section 2 outlines the relevant theoretical considerations provided by agency theory and discusses potential influences of corporate governance structures on directors’ pay. Section 3 describes our estimating strategy and contains a description of the data. This is followed in Section 4 by our estimation results. Finally, in Section 5 we offer some concluding remarks. 2. Theoretical Considerations In this section, we outline important insights emerging from principal agent theory, which we use to motivate and interpret our subsequent empirical research. Such agency models have been used extensively to motivate other empirical work on executive pay (see Jensen and Murphy, 1990). However, we do not attempt to provide a comprehensive review of the optimal contract literature but instead highlight some of the important themes from it that have typically been used by executive compensation researchers in their empirical work. The key theoretical issue that we wish to highlight from the agency literature, which provides the coherent principle when discussing the variables we use in our data analysis below, is the so-called “informativeness principle” (see Murphy, 1998; Holmstrom, 1979). The informativeness principle argues that when setting contracts for CEOs, the board of directors (or shareholders) will relate compensation to measures or standards that provide useful information in determining the actions taken by a CEO. As Murphy (1998) makes clear, this approach allows for a rich set of performance variables to be introduced into the compensation/reward contract including stock returns, accounting returns, and relative performance measures, but generally, any signal or information that might be useful to shareholders in understanding CEO actions. In particular, our research also focuses on the role of boardroom governance variables in explaining directors’ compensation. Again, agency considerations suggest that the absence of certain boardroom institutional arrangements, such as remuneration or compensation committees, can lead to agency costs since executives may have an opportunity to influence their own pay at the expense of shareholder interests. Overall then, this theoretical consideration section provides an agency-theoretic background within which to motivate and interpret our subsequent empirical results. 2.1. D IRECTORS ’ PAY AND COMPANY PERFORMANCE One of the key hypotheses in the empirical literature on the determination of directors’ pay is the supposed positive relation between compensation and company performance (see Murphy, 1998; Hall and Liebman, 1998). It is worth, therefore, outlining the theoretical principal agent considerations that yield this prediction prior to providing tests of it using British data. The ideas introduced here 120 ANDREW BENITO AND MARTIN J. CONYON are based on Murphy (1998). In the typical principal-agent model (i.e. the so-called “hidden action” or “moral hazard” model), the risk averse CEO (agent) is assumed to take some action a that affects output x(a) where the output itself is stochastic due to a (typically) random noise component. The CEO has a utility function u(w, a) where utility depends positively on compensation, w, and negatively on actions, a. The agent receives compensation w(x, z) indicating that compensation depends on output (i.e. performance) and a set of other observable variables in the compensation contract denoted z. The risk neutral shareholder (principal) on the other hand wants the agent to undertake actions a, but cannot directly observe the actions that the CEO takes. Since actions are not perfectly observable, a contract relating CEO income directly to effort is not enforceable. Instead, the optimal contract maximises the shareholders objective (i.e. x − w) subject to an incentive compatibility constraint (i.e. the CEO of his or her own volition chooses the action to maximise own utility) and a participation constraint (i.e. the CEO receives at least the reservation utility). As Murphy (1998) comments, “The fundamental insight emerging from the traditional principal-agent models is that the optimal contract mimics a statistical inference problem: the payouts depend on the likelihood that the desired actions were indeed taken”. Murphy (1998) refers to this as the “informativeness principal”. It suggests that CEO compensation is related to stock-based measures of performance, x, not only because this is desired by shareholders, but because higher stock returns signal information about the actions, a, taken by the CEO. Moreover, the informativeness principle provides a clear rationale for other performance indicators within the CEO compensation contract. Such non-stock-market based measures of performance will be useful to the extent that they convey information to shareholders as to whether the CEO pursued the desired activity or not. Indeed, Murphy (1998) comments, “In fact, if these other measures constitute a “sufficient statistic” for the CEO’s actions, stock based measures need not be used at all”. The important point emanating from agency theory and the “informativeness principle” is that other signals of managerial effort are potentially important in designing the compensation contract. Other important signals that may be important for shareholders (which we consider below) are relative performance evaluation and the role of director monitoring through remuneration committees. For example, companies that adopt remuneration committees potentially introduce important governance features that act as informative aspects of CEOs’ action and effort.2 The empirical literature that uses agency theory to motivate statistical analyses of the relationship between executive compensation and shareholder returns is reviewed by Murphy (1998) and in the context of the UK by Conyon et al. (1995). The typical result, produced in many studies, is that the correlation between executive pay and stock returns is small or weakly determined. The results are usually based on linear regression models catering for idiosyncratic company effects. In addition, Gregg et al. (1993) argues that the link between pay and performance may THE GOVERNANCE OF DIRECTORS’ PAY 121 not be constant over time. They present evidence for a sample of UK companies and find that there was a significant relationship between top pay and performance between 1983 and 1988, but that there is no such relationship between 1988 and 1991. In the light of this, we also allow for time-series heterogeneity in the pay-performance link in the work reported below. Overall, then, agency theory generally predicts that pay and performance should be positively correlated and we test this in our data reported below. 2.2. R ELATIVE PERFORMANCE EVALUATION Recently, empirical researchers (again using the insights of agency theory) have also stressed the importance of relative performance evaluation in executive compensation contracts (see Gibbons and Murphy, 1990). This is also an issue we shall address below in our empirical work and so we highlight its theoretical importance in agency models. Since shareholder return is a noisy signal for managerial effort, including indicators on how benchmark companies are performing yields additional information to shareholders (i.e. consistent with the informativeness principle). In this way, including shareholder return of benchmark companies in the pay contract can signal to shareholders information about managerial effort. If, for example, shareholder returns in similar companies increased but the returns in the CEO’s own company declined, then this can be useful information to shareholders when rewarding (or penalising) the CEO. Specifically, Gibbons and Murphy (1990, pp. 36–37) argue, in the context of empirical models of CEO pay-setting, that after controlling for the company rate of return (absolute performance) then CEO pay awards should be negatively related to industry and market measures of return. We test for this effect in our empirical models below. 2.3. C OMPANY SIZE In addition to these standard agency considerations, which predict that company performance should enter directly into the executive compensation equation, other variables may also be important in shaping top director pay. A particularly important candidate variable is firm size (see Cosh, 1975; Rosen, 1990; Gregg et al., 1993; Conyon and Leech, 1994; Conyon and Gregg, 1994). In terms of the notation introduced in Section 2.1., this size term can be represented as an element of z in the compensation contract, w(x, z). Early studies of executive pay determination argued that company size, which was assumed to act as a proxy variable reflecting a managerial preference for the growth of the enterprise, should also enter the director pay equation (see Cosh, 1975).3 If firm size does reflect the ability of managers to determine their own pay, then managerial theories predict that director compensation and firm size are positively correlated. However, there is another important force at work here. Company size may be empirically important in shaping executive pay because the costs of incorrect decision-making are greater 122 ANDREW BENITO AND MARTIN J. CONYON in larger firms than in smaller ones. In this way, executive pay may be positively related to firm size because bigger firms require better mangers (see Rosen, 1990). Again, we test for this effect in the sample of companies. 2.4. T HE GOVERNANCE OF DIRECTORS ’ COMPENSATION Agency considerations/theory also suggest that internal corporate governance arrangements are important in shaping executive pay contracts. The company board is uniquely bound up with the issue of director pay-setting. As Jensen notes (1993, p. 862): “The job of the board is to hire, fire, and compensate the CEO, and to provide high level counsel”. The usual place where top pay is actually set in the UK is via a sub-committee of the main board termed the remuneration or compensation committee. The absence of a remuneration committee suggests an opportunity for senior executives to award themselves pay raises which are not congruent with shareholder interests (e.g. Williamson, 1985). In essence, then, an agency cost arises. Companies without remuneration or compensation committees may not have the appropriate mechanisms to determine whether CEOs undertook the desired actions and then appropriately reward the CEO. Remuneration committees potentially have an important monitoring and arbiter function in the setting of top pay. They can produce incremental information about CEOs’ actions. Our expectation is that companies that have adopted remuneration committees will have lower executive compensation if such committees act as a check on the CEOs self-awarding pay increases.4 An integral feature to ensure that the board operates effectively in representing shareholder interests is the position of the chairman. The Cadbury (1992) committee report recommended that there should be a clear division of responsibilities at the head of the company. To many commentators this was interpreted as meaning that the roles of chairman and CEO should not be combined. We examine the effect on executive pay of separating these different functions in the empirical work. The process by which members of the compensation committee are chosen will clearly impact their ability to accurately assess the appropriate compensation package for the CEO. In companies that do not have nominating committees, the effectiveness of the remuneration committee in setting pay may be diminished. This may arise if the selection of the compensation committee membership is subject to the patronage of the CEO. In such instances, committee members may owe their positions to a particular CEO and consequently may make partial decisions. We therefore include in our pay models a nomination committee variable to proxy the selection of remuneration committee members.5 THE GOVERNANCE OF DIRECTORS’ PAY 123 3. Modelling Procedure and Data Considerations 3.1. M ODELLING PROCEDURE Above we argued that company performance, relative performance and governance innovations matter in the determination of directors’ compensation. We specify and estimate the following equation: log(Salary)it = α + β1 StockReti,t−1 + β2 RelPerfi,t−1 + β3 log(S)i,t−1 + β4 RemComi,t−1 + β5 NomComi,t−1 + β6 CEO-Spliti,t−1 + ξt + εit (1) where Salary is cash salary and bonus of the highest paid director (see discussion below) in company i and at year t, StockRet is shareholder returns; RelPerf is relative performance evaluation (average shareholder return in the stock exchange industry); S is total company sales. The variables RemCom, NomCom and CEO-Split are corporate governance indicators for remuneration and nomination committee existence and separation of the role of the CEO and chair respectively (defined fully in the appendix). The model includes time effects, ξt , to filter macroeconomic shocks from the estimating equation.6 The disturbance term is given as: εit = µi + νit (2) where µi captures unobserved company effects that are static and νit specifies the remainder disturbance. A priori, the parameters to be estimated are signed as: β1 > 0; β2 < 0; β3 > 0; β4 < 0; β5 < 0; β6 < 0. Equation 1 characterises a company level panel data model.7 The primary feature of any such panel is the longitudinal organisation of the data. We have data on 211 cross section units (UK stock-market companies) over a number of time periods (1984 to 1994).8 Below we present estimates based on panel data techniques, random and fixed effects methods. We do this since the different methods may give different estimates (e.g. Hausman, 1978) given the models’ different treatment of the error structure, εit (see Baltagi, 1995). We discuss Hausman tests to discriminate between the two procedures. The random effects model provides more efficient estimates under the null of the Hausman test. However, if this null is rejected, then the random effects estimates are inconsistent and the consistent fixed effects estimates are to be preferred. Our econometric strategy, then, stresses the importance of idiosyncratic company effects. They are included to purge from the estimating equation any unobserved time invariant company factors that may contaminate the estimation of the pay-for-performance link. An assumption contained in equation 1 and common to most UK research on top pay determination, is that there is no time series heterogeneity in the payfor-performance coefficient, β1 . However, Gregg et al. (1993) present evidence on the pay-for-performance link for a sample of companies between 1983–91. They 124 ANDREW BENITO AND MARTIN J. CONYON found that there was a significant relationship between top pay and performance between 1983 and 1988, but that the relationship breaks down after 1988. In the light of this, we allow for separate per-period performance effects in our modelling strategy.9 Consider some issues in data construction. This paper defines the pay variable as the direct emoluments of the highest paid director. This includes salary, bonus and benefits but excludes long-term elements of compensation such as share options and equity holdings (see Conyon, Gregg and Machin, 1995). Accounting reporting requirements in the UK, however, make it difficult to collect information to appropriately value such options (e.g. via the Black Scholes method) which explains the omission of this component from the compensation variable. Our results should be seen in the light of this. Conyon, Gregg and Machin (1995) consider whether the performance term should be a market-based measure (such as total shareholder return or shareholder wealth) or an accounting based term (such as accounting rates of return or earnings per share). Since principal-agent mechanisms stress the return to shareholders, we use a market-based measure reflecting share price appreciation and dividend yield (i.e. total shareholder return). Also, director compensation and company performance may be contemporaneously determined. Hence, in our estimating equation the performance variable pre-dates the director compensation term. Our empirical model, then, tests whether directors are subsequently rewarded for good corporate performance and penalised for poor performance (see Conyon and Leech, 1994; Gregg, Machin and Szymanski, 1993). 3.2. DATA CONSIDERATIONS The data for this study are derived from three sources. The directors’ compensation data (namely the combined salary, bonus and benefits of the highest paid director) comes from the Hemmington Scott corporate information database. This has the advantage that it is consistent over time in its construction of the pay measure. This contrasts with the information from Datastream which during 1993 begins to include pension payments. Second, we used the Datastream bank of company accounts. From this source we obtained data on total company sales and shareholder return. Shareholder return was calculated as the annual change in the log of the return index. The return index itself is the mean calculated from daily information over the year. We view this as being more informative concerning corporate performance than simply taking the difference in end-point values. The corporate governance variables were obtained from a proprietary survey of UK companies conducted during the period November 1994–March 1995. The survey includes time-varying information concerning the date of adoption of remuneration and nomination committees and at what date the company separated the roles of CEO and chairman. After merging the available compensation and other company variables into the governance data set, for companies with at least four continuous 125 THE GOVERNANCE OF DIRECTORS’ PAY Table I. Summary statistics Year N Salary (£000) StockRet Sales (£000) RemCom NomCom CEO-Split 1985 138 157.349 (182.477) 0.240 (0.275) 1344.616 (3339.863) 0.333 0.065 0.326 1986 144 163.176 (142.985) 0.291 (0.321) 1460.651 (3478.785) 0.382 0.063 0.354 1987 154 188.483 (165.252) 0.412 (0.278) 1353.671 (3070.546) 0.403 0.071 0.396 1988 164 210.233 (170.099) −0.014 (0.192) 1353.341 (2999.178) 0.439 0.073 0.427 1989 173 226.117 (178.713) 0.139 (0.220) 1414.960 (3285.230) 0.468 0.098 0.462 1990 183 218.324 (161.344) −0.120 (0.268) 1339.089 (3121.607) 0.530 0.120 0.508 1991 203 223.975 (177.873) 0.058 (0.383) 1165.686 (2845.099) 0.655 0.163 0.576 1992 211 234.115 (161.227) 0.046 (0.385) 1152.729 (2814.940) 0.782 0.251 0.611 1993 210 244.885 (160.797) 0.268 (0.307) 1202.312 (2992.363) 0.948 0.462 0.689 1994 118 275.962 (171.304) 0.105 (0.267) 1145.395 (2130.098) 0.992 0.517 0.729 Total 1698 215.926 (169.917) 0.132 (0.336) 1286.012 (3024.059) 0.605 0.191 0.516 Variable definitions: N = number of observations; Salary = highest director direct compensation; StockRet = shareholder return; RemCom = remuneration committee; NomCom = nominating committee; CEO-Split = separation of CEO and chair; standard deviations in parentheses, where applicable. Data sources: Salary, StockRet, Sales from Datastream international and Hemmington Scott; RemCom, NomCom, CEO-Split from authors’ governance survey. observations, the resulting unbalanced panel data set consisted of a maximum of 211 companies.10 The data are described in Table I. Mean highest director compensation has increased from £157,349 in 1985 to £275,962 in 1994 for this (unbalanced) sample of firms. This represents an average annual increase in (mean) top directors’ remuneration of about 75.4% in real terms. 126 ANDREW BENITO AND MARTIN J. CONYON The data on pay-setting mechanisms indicate a marked movement to the principles outlined in the Cadbury committee report. In 1986, 38% of companies in this sample had a remuneration committee. By 1994, the figure had risen to 99%. A similar time series profile is observed when considering the separation of the posts of CEO and chairman. Here, approximately 35% of companies separated the role in 1986 compared with approximately 73% in 1994. Note that most of the changes in these two variables have occurred in the period since 1990: an era characterised by greater awareness of corporate governance issues in the UK. The adoption of a nominations committee has been far less marked and may have implications for top pay setting. In 1985, only 7% of this sample of companies had a nominations committee which had grown to 52% by 1994. Although the growth rate is large, this development occurred from a low base. In the absence of a formal committee for the process of appointment of directors, including nonexecutives, a greater influence of the chief executive may be assumed. Overall, there has been a steady rise in the adoption of remuneration committees and the separation of the posts of CEO and chairman since the mid-1980s. On the other hand, companies have only been likely to adopt a nomination committee since the early 1990s. 4. Econometric Results Our primary econometric results for the pay-for-performance relation are contained in Table II and Figure 1. These examine the pay-for-performance relation and the way that it has changed over time. In particular, Table II examines the broad relationship between compensation and performance for a large sample of Datastream stock market companies. The effects of our governance variables on executive pay are contained in Tables III and IV. In Table III, we evaluate the effect of governance mechanisms on directors’ pay. In Table IV, we expand the model and focus on the interaction effects between shareholder returns and corporate governance variables. 4.1. PAY- FOR - PERFORMANCE RESULTS In Table II, we report our results for a sample of 1093 companies between 1985 and 1994. These examine the statistical correlation between directors’ cash compensation and the stock-market performance of the company. Random and fixed effects are presented for completeness although, in this case, the fixed effects are statistically preferred (Hausman test χ 2 (12) = 55.66). The sample size is considerably larger than that used in previous UK research. Typically, such studies report compensation-performance estimates based on only a few hundred companies (see Conyon, Gregg and Machin, 1995). The results here indicate that after controlling for company-specific effects, company size and macro-economic shocks, directors’ cash compensation is positively related to pre-dated shareholder returns. The esti- 127 THE GOVERNANCE OF DIRECTORS’ PAY Table II. Directors’ compensation, performance and size, 1985–1994. Dependent variable = log(Salary) Random effects StockRet (t − 1) RelPerf (t − 1) log(S) (t − 1) Time effects Hausman test R2 within R2 between R2 overall Observations Companies 0.0762 (0.0193) −0.0642 (0.0476) 0.2492 (0.0056) Yes χ 2 (12) = 55.66 [p = 0.000] 0.101 0.601 0.396 8843 1093 Fixed effects 0.0671 (0.0201) −0.0394 (0.0492) 0.1977 (0.0141) Yes 0.102 0.601 0.393 8843 1093 1. Variable definitions: see notes to Table I. 2. The Hausman test considers the restrictions implied by the random effects model against the fixed effects model. Individual effects are for the company. 3. The regressions contain a constant. 4. Standard errors in parentheses. Figure 1. Sensitivity of directors’ pay to shareholder returns, 1985–1994. Fixed effects estimates. 128 ANDREW BENITO AND MARTIN J. CONYON Table III. Directors’ compensation and corporate governance. Estimates based on a sample of 211 UK stock-market companies, 1985–1994. Dependent variable = log(Salary) Random effects Fixed effects Random effects Fixed effects StockRet (t − 1) 0.1060 (0.0263) 0.1033 (0.0265) 0.1052 (0.0263) 0.1021 (0.0265) RelPerf (t − 1) 0.0358 (0.0595) 0.0688 (0.0599) 0.0356 (0.0594) 0.0679 (0.0599) log(S) (t − 1) 0.2162 (0.0099) 0.1810 (0.0155) 0.2169 (0.0097) 0.1811 (0.0154) RemCom (t − 1) 0.0095 (0.0246) 0.0063 (0.0263) NomCom (t − 1) 0.0067 (0.0319) 0.0119 (0.0342) CEO-Split (t − 1) 0.0180 (0.0268) 0.0471 (0.0309) Time effects Hausman test R2 within R2 between R2 overall Observations Companies Yes χ 2 (14) = 45.89 [p = 0.000] 0.415 0.600 0.577 1555 211 Yes 0.417 0.584 0.561 1555 211 Yes χ 2 (11) = 39.41 [p = 0.000] 0.414 0.603 0.579 1555 211 Yes 0.416 0.592 0.568 1555 211 As for Table II. mated performance effect is approximately 0.07. From the fixed effects model in Table II, a 10% increase in shareholder returns predicts an approximate £1852 increase in directors’ compensation (evaluated at average 1994 pay). On the other hand, the company size effect is quantitatively stronger than the performance term: the estimated elasticity is approximately 0.20. This has lead some authors (e.g. Gregg et al., 1993) to suggest that directors have an incentive to pursue company growth strategies at the potential expense of shareholder wealth maximisation. The results contained in this paper, based on a large sample of 1093 companies, appear broadly consistent with other UK evidence (see Conyon, Gregg and Machin, 1995). The relative performance evaluation term, captured by average industry shareholder return, attracts a negative sign but is statistically insignificant. One potential reason for this result is the problem of identifying the relevant companies for comparison purposes. We have implicitly constrained the comparison companies to be those which are in the same stock exchange industry as firm i.11 However, 129 THE GOVERNANCE OF DIRECTORS’ PAY Table IV. Directors’ compensation and governance interaction effects. Estimates based on a sample of 211 UK stock-market companies, 1985–1994. Dependent variable = log(Salary) Random effects StockRet (t − 1) RelPerf (t − 1) log(S) (t − 1) RemCom (t − 1) NomCom (t − 1) CEO-Split (t − 1) StockRet∗RemCom (t − 1) StockRet∗NomCom (t − 1) StockRet∗CEO-Split (t − 1) Time effects Hausman test R2 within R2 between R2 overall Observations Companies 0.0692 (0.0382) 0.0412 (0.0595) 0.2158 (0.0099) 0.0088 (0.0252) 0.0289 (0.0335) 0.0075 (0.0272) 0.0310 (0.0459) −0.1387 (0.0683) 0.0700 (0.0432) Yes χ 2 (17) = 90.35 [p = 0.000] 0.419 0.598 0.576 1555 211 Fixed effects 0.0540 (0.0384) 0.0741 (0.0598) 0.1803 (0.0155) 0.0039 (0.0269) 0.0345 (0.0358) 0.0357 (0.0313) 0.0485 (0.0460) −0.1412 (0.0683) 0.0768 (0.0432) Yes 0.421 0.580 0.560 1555 211 As for Table II. in practice companies may use a different set of companies (e.g. the FTSE 100 companies) which can explain this observed effect.12 We now consider whether there is any time-series variation in the payperformance link. Gregg et al. (1993) find evidence of a declining relationship between directors’ pay and shareholder returns between 1983 and 1991 based on a sample of 288 large UK companies. We estimated our executive pay equation by allowing the coefficient estimates on the performance terms to be different for each year (as described in note 9). In Figure 1, we plot these coefficient estimates on the performance terms against time. We use our large sample of 1093 companies over the 1985–94 period. The pattern reveals that the estimated performance effect on directors’ pay tends to increase over the period.13 Indeed, one can reject the hypothesis that the coefficients are equal to each other (F(9,775) = 2.85). Our results contrast with Gregg et al. (1993) who found that the pay-for-performance correlation declined between 1983 and 1991. Our findings, at a minimum, call into question the supposed de-coupling of executive pay and corporate performance in the UK over time. The time series heterogeneity of the pay-performance term is clearly an important feature of the data. The differences between our results and those 130 ANDREW BENITO AND MARTIN J. CONYON provided by Gregg et al. may be attributable to a number of factors. Importantly, as noted above, our results are based on a larger sample of companies and over a longer time period. Indeed, the Gregg et al. study is based on very large companies whereas our sample includes relatively small companies also drawn from the population of UK listed companies. In terms of econometric methods employed, there are similarities between the two studies: both our research and that produced by Gregg et al. use panel data econometric methods. Our results also accord to a greater extent with recent US research. Murphy (1998) finds that the elasticity of CEO cash compensation with respect to shareholder return in the Standard and Poors 500 companies has increased in each decade since 1970. For instance, his research reveals that the US cash compensation–performance elasticity was 0.09 in the 1970s, 0.21 in the 1980s and 0.26 between 1990 and 1996. Overall, our panel data results indicate that it is possible to isolate significant size and corporate performance effects on directors’ compensation. This result is established for a large sample of 1093 Datastream companies between 1985 and 1994. The effect of relative performance on pay is less clear. There also appears to be time series heterogeneity in the pay-for-performance relation. In particular, the correlation appears to grow stronger over time in our data set. 4.2. D IRECTORS ’ PAY AND BOARDROOM GOVERNANCE Having considered the relationship between pay and performance, and also the time series heterogeneity in this relationship, we now turn to an assessment of the relationship between directors’ pay and corporate governance variables. In particular, we assess the direct effects of remuneration and nomination committees’ variables on pay as well as the effect of splitting the roles of CEO and chairman. This is then followed by an evaluation of the indirect effects of these variables by focusing on the interaction between the variables and the performance term. In Table III we examine the impact of governance mechanisms on compensation outcomes. Again, for completeness, we present random and fixed effects estimates. Remuneration committee adoption and the company previously separating the posts of CEO and chairman have no significant statistical impact on directors’ compensation. The nomination committee variable also attracts an insignificant coefficient. A formal test that the governance variables are jointly insignificant cannot be rejected (in the fixed effects specification, F(3,1330) = 0.93; p-value = 0.4242). Do our results imply that much of the boardroom governance re-organisation of the early 1990s has been largely cosmetic? We do not believe so for (at least) three reasons. First, the impact of the governance variables may be contained within the company fixed effects. Netting out these idiosyncratic factors (i.e. ignoring between group variation) makes it difficult to isolate an overall average governance effect. Second, we have focused only on one economic variable: the direct compensation of the highest paid director. The average effect of governance THE GOVERNANCE OF DIRECTORS’ PAY 131 innovations on other economic variables (e.g. productivity and Tobin’s q) may be quite different. Third, the boardroom governance innovations of this decade have resulted in greater information becoming available to shareholders and policy-makers alike. Generally, such increased information is to be welcomed.14 In addition to the direct effect of corporate governance variables on directors’ compensation we might expect there to be an indirect effect. For instance, companies that have adopted a remuneration committee may have a stronger pay-performance link. This general idea is captured in the following equation: log(Salary)it = α + β1 StockReti,t−1 + β2 RelPerfi,t−1 + β3 log(S)i,t−1 + β4 RemComi,t−1 + β5 NomComi,t−1 + β6 CEO-Spliti,t−1 + β7 StockRet ∗ RemComi,t−1 + β8 StockRet ∗ NomComi,t−1 +β9 StockRet ∗ CEO-Spliti,t−1 + ξt + εit (3) Estimates of this equation are presented in Table IV. Although there is still little evidence of a direct effect of governance variables on top pay, the interaction terms are at the margin of significance [χ 2 (3) = 7.48; p-value = 0.058]. The positive signs on the interactions with remuneration committee adoption and separation of the posts of chief executive and Chairman may be contrasted with a significantly negative coefficient on the interaction between shareholder return and the nominations committee variable. However, when the (linear and interacted) shareholder return variable is dated contemporaneously, the coefficient on the interaction with remuneration committee existence becomes statistically significant with a point estimate (standard error) of 0.107 (0.045). In this case, the multiplicative term between shareholder return and chief executive/Chairman separation is insignificant with a point estimate (standard error) of 0.047 (0.043). Thus, there is some modest evidence suggesting that the performance effect upon pay may be stronger in companies that have adopted a remuneration committee. This suggestion may be contrasted with the paper by Main and Johnston (1993). They find that the structure of pay, measured as the proportion of options in total pay, is not influenced by the reported existence of a remuneration committee. Main and Johnston (1993) expected that the proportion of pay taken in the form of options would vary positively with remuneration committee existence so as to align management and shareholder interests. Overall, the results pertaining to the effects of boardroom governance variables on directors’ cash compensation are not overwhelming. In the estimated models that considered the direct effects of nomination and remuneration committee and splitting the roles of CEO and Chairman, little effect on directors’ pay was isolated. This is perhaps due to the deep nature of this variable. Future work needs to consider more precisely the dynamics of these pay-setting mechanisms. When we interacted the governance indicator variables with the stockholder return variable there was some modest evidence that these governance variables play a role in shaping directors’ pay outcomes. 132 ANDREW BENITO AND MARTIN J. CONYON 5. Concluding Remarks In this paper, we have examined the determination of directors’ remuneration in UK companies. The study has been motivated by the perception in some of the executive compensation literature (and in the public policy debate) that pay is inadequately tied to company performance, and a suggested recent de-coupling of pay from performance. Further, there has been substantial change in corporate governance practices in UK companies and the effects of such changes are not currently well understood. Hence, we have focused upon the determination of executive pay in terms of corporate performance and internal governance structures. Differing data sets were employed tailored to analysing each set of questions in detail. First, employing a large unbalanced panel data set of 1093 companies we obtain results which corroborate certain findings of previous studies: that pay is linked to corporate performance but its effect is dominated by the company size variable. No significant evidence in support of relative performance evaluations was found. In contrast to some other UK work, we find that the pay-for-performance estimate becomes quantitatively larger over our sample period. This similar effect has been observed using United States data also. Second, access to unique survey data on internal boardroom control systems enabled an analysis of the effect of corporate governance innovations upon senior executive pay over the period 1985–94. Panel data results were presented based on a sub-sample of 211 companies. There was no evidence that adoption of either a remuneration or nominations committee or of the separation of the positions of CEO and Chairman have an influence upon pay awards. There was some modest evidence that adoption of a remuneration committee may lead to a greater sensitivity of pay to performance. The analysis based on fixed effects does not preclude other approaches to executive pay determination (e.g. human capital or tournament theories). We were able to isolate evidence of a cross sectional relationship between pay and governance structure. However, the use of a fixed effects strategy reported in the paper rendered these cross sectional correlations insignificant. The effect of internal boardroom control systems are instead likely to be related to the unique nature of individual companies.15 The result is perhaps consistent with the notion that governance may be a deep variable in the system. Moreover, this urges a cautionary tale in terms of the policy recommendations put forward by the Cadbury Committee and other organisations which have looked for general structural solutions to the perceived problems in governance practices of UK companies. Acknowledgements We would like to thank Severin Borenstein, Brian Main, Luis Gomez-Mejia, Simon Peck, Peter Pope and Mike Waterson for valuable discussions during the preparation of this paper. We also thank two anonymous referees and Massimo Warglien 133 THE GOVERNANCE OF DIRECTORS’ PAY for comments and guidance on our paper. Funding from the Economic and Social Research Council (award R000237246) is gratefully acknowledged. Appendix Table A1. Data definition and description Salary: Highest-paid director emoluments (Source: Hemmington Scott) divided by retail price index (Source: CSO; Base year 1994) (£000). This includes salary, bonus and benefits but excludes pensions, share options and equity. S: Company Sales; source Datastream; variable 104/retail price index (Source: CSO) (£000). StockRet: Total Shareholder return; source Datastream. This is the change in the return index. The return index represents the total value of holding a notional stock. The holding is deemed to return a daily dividend, which is used to purchase new units of the stock at the current price. The gross dividend is used. RelPerf: Relative performance; Average shareholder return of the industry in year t; classified by Datastream industry code. RemCom: Remuneration committee; dummy variable indicating presence (=1) or not (=0) of remuneration committee in year t (Source: Own Survey). NomCom: Nomination committee; dummy variable indicating presence (=1) or not (=0) of nominations committee in year t (Source: Own Survey). CEO-Split: Separation of chairman and CEO; dummy variable indicating separation (=1) or combination (=0) of most senior positions in year t (Source: Own Survey). Table A2. Sample size: The panel for which we have corporate governance data is unbalanced with a minimum of three records per company. The data set consists of the following number of records per company Periods in the sample 3 4 5 6 7 8 9 10 Total Number of companies 15 17 8 11 11 8 72 69 211 134 ANDREW BENITO AND MARTIN J. CONYON Notes 1 Some of the recent UK research on executive compensation includes Conyon (1997), Main et al. (1996), Watson et al. (1994), Conyon and Gregg (1994), Conyon and Leech (1994), Gregg et al. (1993) and Main (1991). The more voluminous work from the United States includes Abowd (1990), Jensen and Murphy (1990), Gibbons and Murphy (1992), Murphy (1985), Rose and Shepard (1994), Winfrey (1994), Hall and Liebman (1998). This literature is comprehensively reviewed by Murphy (1998) and Hallock and Murphy (1999). 2 Note also that Holmstrom (1979) shows that any signal of the individual action is of value if it possesses an association with the observed payoff. 3 For instance, Cosh (1975, p. 77) comments that “Since these theories (i.e. managerial theories) mainly assume that the executive is primarily interested in increasing the size of the company we would expect executive remuneration to be related to company size”. 4 But the function of the remuneration committee is to decide on the correct pay structure. It is conceivable that a remuneration committee may want to increase the pay of the CEO (see Main and Johnston, 1993). 5 Nickell (1995, chapters 2 and 3) discusses whether non-executive directors, who typically make up the majority membership of the remuneration committees in the UK, will be effective in carrying out their monitoring function. He concludes (p. 56) they will not. 6 Operationally, we include a set of time dummies to achieve this. 7 An appealing alternative would be to estimate a general human capital earnings function using director level data and then including information on CEO job tenure, experience, and education. Such data is not consistently available for UK directors. 8 So, in our panel data set i = 1, . . . , N (where maximum N equals 211) and t = 1, . . . , T (where maximum T equals 11). 9 In effect, we estimate equation 1 but allow a separate effect for each year: log(Salary) = α + it βt1 StockRetit−1 + β2 RelPerfit−1 + β3 log(S)it−1 + β4 RemComit−1 + β5 NomComit−1 + β6 Sepit−1 + ξt + εit . We test whether βt1 = β1 . 10 The number of records per company is detailed in the appendix. Also, more detailed definitions on all the variables used in this study are provided there. 11 These stock exchange industries are measured at approximately the two digit SIC level of aggregation. 12 Alternatively, any positive association may be attributable to the use of industry based compensation surveys or more informal comparisons with firms in the same industry in the determination of directors’ pay. 13 The estimated coefficients (standard errors) on the shareholder return variable for each year are: 1985: −0.11 (0.35); 1986: −0.07 (0.06); 1987: −0.14 (0.06); 1988: 0.09 (0.08); 1989: 0.14 (0.07); 1990: 0.08 (0.05); 1991: 0.09 (0.04); 1992: 0.10 (0.04); 1993: 0.15 (0.05); 1994: 0.18 (0.08). A test of the hypothesis that these effects are jointly equal to one another rejects the null: F(9,7775) = 2.85; p-value = 0.002. 14 The recent Hampel Committee report, cited in Coopers and Lybrand (1997), however, expresses concern that there may be excessive information disclosed rendering unclear the salient information regarding executive pay. 15 As a partial consideration of this possibility, re-estimation of our basic model (equation 1) as a pooled regression reveals a set of coefficients (standard errors) on the governance variables of: remuneration committee, 0.018 (0.025); nominating committee, −0.037 (0.032) and separating the pools of CEO and chair, −0.068 (0.022). A null-hypothesis that the coefficients are jointly zero is rejected at the 5% level: F(3,1540) = 3.63. THE GOVERNANCE OF DIRECTORS’ PAY 135 References Abowd, J.: 1990, “Does Performance-Based Managerial Compensation Affect Corporate Performance”, Industrial and Labor Relations Review 43: 52–73. Baltagi, B.: 1995, Econometric Analysis of Panel Data (New York: John Wiley & Sons). Cadbury, Sir Adrian: 1992, Committee on the Financial Aspects of Corporate Governance (London). Conyon, M. J.: 1997, “Corporate Governance and Executive Compensation”, International Journal of Industrial Organization 15: 493–510. Conyon, M. J.: 1994, “Corporate Governance Changes in UK Companies Between 1988 and 1993”, Corporate Governance: An International Review 2: 97–109. Conyon, M. J., P. Gregg and S. Machin: 1995, “Taking Care of Business: Executive Compensation in the UK”, Economic Journal 105: 704–715. Conyon, M. J. and P. Gregg: 1994, “Pay at the Top: A Study of the Sensitivity of Top Director Remuneration to Company Specific Shocks”, National Institute Economic Review 3: 83–92. Conyon, M. J. and D. Leech: 1994, “Top Pay, Company Performance and Corporate Governance”, Oxford Bulletin of Economics and Statistics 56: 229–247. Conyon, M. J. and S. I. Peck: 1998, “Recent Developments in UK Corporate Governance”, in T. Buxton, P. Chapman and P. Temple (eds.), Britain’s Economic Performance (London: Routledge). Coopers and Lybrand: 1997, From Greenbury to Hampel: A Mid-Term Report on Developments in Directors’ Remuneration (London). Cosh, A.: 1975, “The Remuneration of Chief Executives in the United Kingdom”, Economic Journal 85: 75–95. Gibbons, R. and K. J. Murphy: 1990, “Relative Performance Evaluation for Chief Executive Officers”, Industrial and Labor Relations Review 43: 30–51. Gibbons, R. and K. J. Murphy: 1992, “Optimal Incentive Contracts in the Presence of Career Concerns”, Journal of Political Economy 100: 468–505. Hall, B. and J. Liebman: 1998, “Are CEOs Really Paid Like Bureaucrats?”, Quarterly Journal of Economics 113: 653–691. Hallock, K. and K. Murphy: 1999, The Economics of Executive Compensation (Northampton, MA, USA: Edward Elgar). Gregg, P., S. Machin and S. Szymanski: 1993, “The Disappearing Relationship Between Directors’ Pay and Corporate Performance”, British Journal of Industrial Relations 31: 1–10. Hausman, J.: 1978, “Specification Tests in Econometrics”, Econometrica 46: 1251–1271. Hart, O.: 1995, “Corporate Governance: Some Theory and Implications”, Economic Journal 105: 678–689. Holmstrom, B.: 1979, “Moral Hazard and Observability”, Bell Journal of Economics 10: 74–91. Holmstrom, B. and P. Milgrom: 1987, “Aggregation and Linearity in the Provision of Intertemporal Incentives”, Econometrica 55: 303–328. Jensen, M. and K. Murphy: 1990, “Performance Pay and Top Management Incentives”, Journal of Political Economy 98: 225–264. Jensen, M. C.: 1993, “The Modern Industrial Revolution, Exit, and the Failure of Internal Control Mechanisms”, Journal of Finance XLVIII(3): 831–880. Main, B. G. M.: 1991, “Top Executive Pay and Performance”, Managerial and Decision Economics 12: 219–229. Main, B. G. M., A. Bruce and T. Buck: 1996, “Total Board Remuneration and Company Performance”, Economic Journal 106: 1627–1644. Main, B. G. M. and J. Johnston: 1993, “Remuneration Committee and Corporate Governance”, Accounting and Business Research 23: 351–362. Murphy, K. J.: 1998, “Executive Compensation”, in Orley Ashenfelter and David Card (eds.), Handbook of Labor Economics, Vol. 3 (North Holland), forthcoming. 136 ANDREW BENITO AND MARTIN J. CONYON Murphy, K. J.: 1985, “Corporate Performance and Managerial Remuneration: An Empirical Analysis’, Journal of Accounting and Economics 7: 11–42. Nickell, S.: 1995, The Performance of Companies (Oxford: Basil Blackwell). Rose, N. and A. Shepard: 1994, “Firm Diversification and CEO Compensation: Managerial Ability or Executive Entrenchment?”, NBER Working Paper 4723. Rosen, S.: 1990, “Contracts and the Market for Executives’, NBER Working Paper 3542. Watson, R., D. Storey, P. Wynarczyk, K. Keasey and H. Short: 1994, “The Remuneration of NonOwner Managers in Small and Medium Sized UK Enterprises”, Journal of Management Studies 31: 553–568. Williamson, O. E.: 1985, The Economic Institutions of Capitalism: Firms, Markets, Relational Contacting (New York: Free Press). Winfrey, F.: 1994, “The Control of CEO Compensation by Institutional Investors: An Empirical Study of Agency Theory in Large US Industrial Corporations”, Corporate Governance: An International Review 2: 211–224. Address for correspondence: Martin J. Conyon, Warwick Business School, University of Warwick, Coventry CV4 7AL, UK Phone: 44(0) 1203 522960; Fax: 44(0) 1203 523779; E-mail: Martin.conyon@warwick.ac.uk