The recent empirical work examining the efficiency of compensation packages using regression analysis is not designed to capture efficiency (Charnes et al., 1989). Regression analysis is designed to determine linear relationships, and devi- ations therefrom. The line of fit is not intended to portray the efficient frontier.
The Data Envelopment Analysis (DEA) model used in this study is designed to evaluate relative efficiency. DEA determines whether the observation in ques- tion is more or less efficient than other comparable observations. Therefore this model is far more appropriate for integrating the theoretically optimal solutions with an evaluation of practice (Banker, 1989; Banker et al., 1989; Callen, 1991).
In its simplest form, the principal/agent compensation problem is to maximize the principal’s residual claim subject to the agent’s compensation and the agent’s minimum welfare which is determined by the market for managers. The prin- cipal in the problem is assumed to be risk neutral and the agent is assumed to 1
2 3 4 5 6 7 8 9 1011 11 12 13 14 15 16 17 18 19 2011 21 22 23 24 25 26 27 28 29 3011 31 32 33 34 35 36 37 38 39 40
Efficient CEO Compensation: A Data Envelopment Analysis Approach 193
be both risk and effort averse. When the agent’s action is observable, the first best solution to the optimal compensation contract is the contract that solves the following optimization problem.
Maximize:
Principal’s welfare (firm performance) Subject to:
The agent’s minimum welfare (market for managers)
The first best solution to this problem is a fixed salary (Holmstrom, 1979).
However, the agent is usually not the owner and the agent’s action is not gener- ally observable. Therefore constraints must be added to address accurate reporting, agent’s risk aversion, effort aversion and the agent’s performance incentives. The second best solution is a solution to the following problem:
Maximize:
Principal’s welfare (firm performance) Subject to:
The agent’s minimum welfare (market for managers) Agent’s reporting incentive
Agent’s effort aversion Agent’s performance incentive Agent’s risk aversion
Compensation packages have evolved to address these problems. These pack- ages generally consist of a fixed salary, a bonus, and some combination of restricted stock, stock options and long term incentive plans (Sloan, 1993;
Bloedorn, 1994). However, the optimal composition of the compensation pack- ages is not known. Performance measures cover both accounting and market measures to address reporting requirements and managers’ risk aversion.
Salary is generally based on past performance. The bonus portion of the com- pensation generally compensates the manager for annual accounting based per- formance measures and is assumed to motivate managers towards short term performance goals. The long term portion of the compensation plan generally con- sists of some combination of restricted stock, long term incentive plans and stock options. These plans are generally stock based and payable over a number of years and are assumed to motivate managers towards long term performance goals.
The large number of different types of compensation schemes exist because of the lack of any truly objective performance measure, and the information asymmetry between owners and managers. Earnings and returns are both composite measures of a firm’s performance. However they are both subject to noise from accounting principles, reporting practices, market and industry driven 1
2 3 4 5 6 7 8 9 1011 11 12 13 14 15 16 17 18 19 2011 21 22 23 24 25 26 27 28 29 3011 31 32 33 34 35 36 37 38 39 40
194 ELIZABETH T. COLE AND JOANNE P. HEALY
factors. Additionally while earnings and returns are closely associated over the long run, the association is much smaller in the short run (Easton, Harris &
Ohlson, 1992).
Stock returns are driven by both firm performance and industry and market wide factors such as actions of competitors and suppliers, regulatory actions, etc. (Lambert, 1993; Janakiraman, Lambert & Larcker, 1992). Earnings filter out many of the industry and market shocks that make it difficult to accurately evaluate the manager’s action, however earnings may be inaccurate due to both manipulation and the estimated nature of many accrual accounting numbers (Lambert, 1993).
The principal’s tradeoff between using accounting based and stock based performance measures is expected to be based on the relative informativeness of these measures (Banker & Datar, 1989; Bushman & Indjejikian, 1993; Kim
& Suh, 1993; Sloan, 1993) and manager’s risk aversion (Lambert, 1993;
Lambert, Larker & Verecchia, 1991). In order to most accurately capture current practice we use both accounting and market based performance measures in the DEA evaluation.
The DEA model reduces the multiple outputs and inputs into a virtual output and input where the ratio of virtual outputs to virtual inputs for the most effi- cient units are rated as one, and the less efficient units are rated as less then one. The definitions from the preceding paragraphs lead to the following model:
Maximize:␣1RETi+ ␣2EPSi
Minimize: ò1SALARYi+ ò2BONUSi + ò3LTCompi Subject to:
␣1RETj + ␣2EPSj + ò1SALARYj+ ò2BONUSj + ò3LTCompj≤ 0 Where:
RET= Stock Return
EPS = Change in Earnings Per Share
SALARY = The fixed portion of the CEO’s compensation
BONUS = The short term bonus portion of the CEO’s compensation LTComp = Long term incentive pay outs, stock options or restricted stock
␣ = the performance measure weight 1
2 3 4 5 6 7 8 9 1011 11 12 13 14 15 16 17 18 19 2011 21 22 23 24 25 26 27 28 29 3011 31 32 33 34 35 36 37 38 39 40
Efficient CEO Compensation: A Data Envelopment Analysis Approach 195
ò = the weight given by the agent to compensation i = the firm in question
j = 1 to n firms in sample where j≠i n= number of firms in the population
For the DEA evaluation all variables are measured as averages from 1993 to 1995.2Salary, Bonus, and LTComp are measured in dollars. RET is measured as the firm’s annual stock return. EPS is measured as the change in earnings per share deflated by lagged stock price to remove the effects of size. All variables are scaled to fall between 1 and 99 to facilitate the DEA evaluation.
The ␣s represent the weights that are given to each of the performance measures by the principal. The weights are determined in part by the presumed effort aversion of the manager and the presumed ability of the different compo- nents of compensation to reduce this effort aversion. The òs represent manager’s risk aversion to each of the components of compensation.
Theoretical optimization models have attempted to determine the optimal values of ␣ and ò. However, without a greater understanding of agent’s risk and effort aversion, it is not possible to determine the optimal values. The DEA model does not require that the functional form (the ␣s, òs) is specified. DEA extracts the functional form individually for each firm in the sample which makes that firm appear in the most favorable light, and allows variation between firms in both owner’s and manager’s preferences.
The final constraints are formed by all other firms’ performance and compen- sation levels which proxies for the market for managers. These constraints force the model to evaluate firms in the context of feasible performance levels given the performance levels achieved by other firms. Indicator variables representing industry affiliation further constrains the model to evaluate firms within industry groupings (Banker & Morey, 1989).