Industry Homogeneity and Performance Impact on Relative Pay Performance in Executive Compensation

Transcription

1 Industry Homogeneity and Performance Impact on Relative Pay Performance in Executive Compensation Wendell Licon Doctoral Candidate: The University of Texas at Austin Visiting Instructor: The University of Oklahoma November 25, 2002 Earlier drafts of the paper were titled Relative Pay Performance and How it Interacts with Retention Issues in Executive Compensation. I wish to thank the JC Penney Foundation at the University of Oklahoma for research assistance. I would also like to thank Bob Parrino, John Martin, Jay Hartzell, Jayanthi Sunder, and Bob Shealor for their helpful comments. Contact: wlicon@ou.edu

2 Abstract This paper contributes to the existing literature by investigating whether the relationship between executive compensation and firm performance relative to a firm s peers is affected by the degree of industry homogeneity where a firm competes. Since industry homogeneity can be perceived as both a measure of the accuracy of the relative performance signal that is received as well as a measure of the availability of industry specific human capital, the paper also investigates the effect of a firm s relative performance on the relative pay performance relationship within the sample. It documents that the ex ante relative performance compensation mechanism, as measured by a firm s stated intention in its proxy statements, is positively related to the level of industry homogeneity but negatively related to industry performance within an homogeneity level. It also finds that executives in firm s that perform well, relative to a peer benchmark, have greater relative performance involvement, ex poste in their compensation structure. However, industry homogeneity appears to have a slight impact on the ex poste relationship that is the opposite found in the ex ante relationship. The results are consistent with the explanation that a source of variation in relative performance based compensation is related to executive retention concerns. For example, the demand for specialized management expertise may lead firms to avoid or be attracted to relative performance-based compensation schemes either ex ante or ex poste when retention issues might be more central to a firm s future than current performance issues. 2

3 I. Introduction Although Executive Compensation has been studied to a great extent, much of the relationship between compensation and an executive s performance, or the executive team s performance is still unknown. That part of the compensation relationship to which academic work has yet to understand well is how a firm incorporates its performance relative to its peers into the compensation equation. If firms have access to the required information to tie an executive s performance to her compensation in a more exact and measurable way, then firms choosing not to do so leaves our understanding of the compensation-performance relationship at a loss. In fact, Holmstrom s (1979) result that the inclusion of an informative signal into the compensation rule is a Pareto improvement over not including the signal questions the plausibility that firms are not utilizing relative performance evaluation (RPE) in their compensation relationships with their executive officers. Gibbons and Murphy (1990) suggest that utilizing RPE delivers benefits by filtering out events that may reward or punish workers for circumstances beyond their control. And yet evidence suggests that firms do not incorporate RPE in a systematic manner. For instance, Sibson (1991) finds that only 6% of industrial firms explicitly use RPE in their bonus contracts. Do firms choose to ignore these optimizing benefits or are there factors that may prevent firms from improving their payperformance relationships through RPE? For instance, Aggarwal and Samwick (1999b) and Kedia (1999) offer such evidence suggesting that the lack of RPE can be partially explained by the strategic interactions of firms in their product markets. 3

4 The purpose of this study is to document the variation in the use of RPE in executive compensation that may be attributable to the level of industry homogeneity that a firm principally competes. Industry homogeneity provides a measure concerning the accuracy of an RPE signal s but may also proxy for the level of retention concerns that a firm must entertain when establishing its executive compensation mix and level of pay. Those retention concerns, when combined with the practical limitations in compensation may have an effect on a firm s ability to institute RPE with the executive compensation mechanism. Four empirical results are found. First, more homogenous firms are more likely to commit to an ex ante RPE compensation mechanism than less homogenous firms. Second, low performing firms competing in more homogenous industries are more likely to commit to an ex ante RPE pay relation than their higher performing counterparts. Third, higher performing firms exhibit a greater propensity to compensate executives by involving RPE when examined in the post-performance period. Finally, the payperformance relationship exhibits evidence of breaking down in lower performing firms. Those firms tend to have an ex poste pay relationship that is at times negatively related to firm-performance while positively related to an industry index performance measure. The results suggest that the prevalence of RPE in executive compensation may be related to a firm s ability to garner a clear industry performance signal but they also suggest that a firm s ability to incorporate RPE into compensation is a function of a firm s ability to deviate from target compensation levels without aggravating retention concerns. 4

5 The remainder of the paper is organized as follows. Section II discusses relevant literature concerning relative performance and retention. Section III briefly describes some compensation delivery choices and mechanisms and puts forth the paper hypotheses. Section IV describes the data. Section V presents the empirical evidence, and Section VI the conclusions. II. Literature Relative Performance The RPE solution for helping to optimize the principal-agent problem is justified by Holmstrom (1982) in that filtering for risk not in control of the executive will be beneficial to the executive because she will be insured from that risk. In a prior paper, Holmstrom (1979) showed that an optimal contract must incorporate the information contained in an informative signal. This suggests that if the compensation mechanism ignores information that is unique to the relative performance of the firm, then the contract may be inefficient. Therefore, even an imperfect RPE signal may help to improve the principal-agent relationship, absent other complicating factors. Not withstanding the practical complications introduced by the accounting regulations associated with explicitly filtering industry and market shocks, it seems unlikely that firms would ignore such potential benefits to RPE and would find creative ways to introduce an informative signal to their pay-performance relation. Aggarwal and Samwick (1999a) find that executives of companies that have low variances have higher 5

6 pay performance sensitivity. This may suggest that compensation committees are aware of the adverse consequences of high-powered incentives and utilize them when the cost of those consequences is low. It seems plausible that the board would also handle other adverse compensation consequences constructively. Early work focused on measuring RPE and documenting its anemicly perceived effect on compensation. Gibbons and Murphy (1990) found evidence of RPE but also found it difficult to distinguish whether firms benchmark to their industry or some other broad market measure. Their work is also supported by a number of proxy statements that mention that the market for executive talent extends well beyond their respective proxy required competitive industrial group. Antle and Smith (1986) study 39 firms over a long period and find evidence that compensation is more closely linked with relative return on assets than absolute return on assets. While their results were encouraging for RPE in accounting returns, the results were consistent with only partial filtering on market related returns. They warn that RPE may be detrimental to the firm when managers have the ability to change the industry where a firm competes. Dye (1992) finds that the benefits of RPE are lessened when the number of projects available is small but RPE is effective when there are a large number of choices as well as when there are no decisions concerning project selection. Murphy (1999) finds that RPE is increasingly popular with cyclical industries and utilities where it is replacing budget based measurement. He suggests that firms in homogenous industries find the cost of the RPE signal less expensive that those in less homogenous industries. While the lessened cost of the signal makes it cheaper and more efficient to evaluate a firm s executive within an homogenous 6

7 group, it also makes it easier to evaluate other executive candidates within an industry in order for a firm to determine the size and quality of the executive replacement pool from where the firm is most likely to recruit a replacement. This additional information might have an effect on a firm s compensation policy if retention factors impact compensation policy. Along similar reasoning Johnson and Tian (2000a) suggest that the reservation utility level for an executive becomes more accurately determined with a large number of firms in an industry. This may also suggest that in more homogenous industries there is less ambiguity concerning market level pay. Aggarwal and Samwick (1999b) furthered our understanding of industry cross-sectional variation in RPE use. Their results suggest that the nature of competition will affect the maximizing incentives given to managers. They find that for firms in industries where the competition is Bertrand or strategically based, compensation should exhibit a positive instead of negative (or index filtering) sensitivity to rival firm performance. That sensitivity should be increasing in competition. Their empirical evidence supports the hypothesis in a test of manufacturing firms. They also find scant evidence of relative performance in their sample covering Garvey and Milbourn (2001) suggest that older, more diversified, executives do not need the benefits of RPE filtering for industry shocks while younger executives will benefit from RPE. Their empirical findings agree with their assertion. 7

8 Retention Jensen and Meckling (1976) note that firms with a high probability of bankruptcy must pay higher salaries in order to induce executives to take the risk of working for these firms. This higher salary can be viewed as an insurance premium to compensate riskaverse executives. From a retention perspective it can also be perceived as increasing the switching costs associated with an executive seeking another position outside the firm. Gilson and Vestsuypens (1993) find in a study of distressed firms that for newly appointed CEOs, inside replacements are generally paid 35% less than those they replace, while outside replacements are generally paid about 36% more than the those they replace. Although they suggest that outsider replacements may have specialized expertise that may account for this difference, it may also be telling of the level of switching costs that must be overcome in order to entice a strong manager to leave her current employer. It may also have the added benefit of establishing a new high level of switching cost at her new employer that will be needed for retention purposes due to the troubled firm s already high turnover rate. Core and Guay (2001) find evidence consistent with firms using options to attract and retain certain type of employees as well as for incentive purposes while Mehran and Yermack (1999) find a negative relationship between option compensation and CEO turnover which is also consistent with firms using compensation for retention purposes. Firms generally justify resetting executive options that have fallen far out-of-the-money with both incentive and retention arguments. Brenner, Sundaram, and Yermack (2000) mention that firms argue for re-setting strike prices on out-of-the-money options for 8

9 under performing firms. The underlying argument is that these firms believe they have executives that are currently still valuable to the firm as an ongoing entity. They also find that the value to the executive from re-setting the options does not substitute for other compensation and that executives with larger pay packages are more likely to have their options reset. If higher compensation is indicative of the quality of the executive, the implication is that the likelihood of resetting for just those executives should be greater than for lesser compensated executives. Chance, Kumar and Todd (2000) also have similar findings. Achyra, John and Sundaram (2000) show that some re-setting is almost always optimal but the advantages of re-setting decrease as the costs of replacing incumbent managers decrease. If we interpret option re-pricing as a compensation contract renegotiation that is a costly action, as well as a costly signal to shareholders, then it is evident that those costs will be borne by the firm if replacements costs are at a sufficiently high level. It is then clear that a commitment to a specific compensation contract may be a function of a firm s cost to replace potentially lost executive talent since such a commitment raises the possibility of the need for renegotiation. Parrino (1997) finds evidence suggesting that poor performing CEOs are easier to identify and replace when they are working in homogenous industries where poor performers are easier to identify. An extension of that hypothesis is that industry homogeneity is a proxy for the abundance of industry specific human capital that may have an effect on a firm s ability to replace a departed executive. If that is true, firms in more homogenous industries may have lessened retention concerns, holding other variables constant, due to lower replacement costs. Whether industry homogeneity 9

10 affects the compensation mechanism through a clearer signal argument, a replacement cost argument, or some other manifestation is ultimately an empirical question. In short, firms may be able to place a stricter pay-for-performance contract that introduces the possibility of a missed compensation target on the executive when that executive is working in a more homogenous industry. Then we would expect RPE, which fits that description, to be found to a greater extent in more homogenous industries if those firms are looking to optimize their pay-performance relationships. Industry homogeneity may also have a negative effect on a firm s ability to incorporate RPE. Fee and Hadlock (2001) find that executive talent raids by outside firms are more likely if the target executive s firm has outperformed an industry benchmark. Since we know that a firm competing in a more homogenous industry will generate a clearer signal concerning that performance and that signal will be available to industry outsiders as well as insiders, then we would expect the top industry performers, controlling for homogeneity, to be more sensitive to increases in the market wage for specific executive talent compared to the weaker performing firms. This increased sensitivity might hinder a firms ability to commit to compensation RPE ex ante and yet increase the need for the firm to ensure a relatively high ex poste wage to the top firm executives. Himmelberg and Hubbard (2000) find evidence suggesting that the supply of qualified CEOs available to manage large corporations is relatively inelastic. They also theorize that an executive pecking order is created so that top talent ends up working for larger firms where their efforts are put to better use (as well as being better compensated for 10

11 their selection). Empirically, they find less compensation based RPE in larger firms than in smaller firms. Their evidence is consistent with a firm being sensitive to the possibility that their top talent might voluntarily leave the organization and that sensitivity is accounted for through increased compensation. If their results concerning a pecking order are extended beyond firm size to performance levels, then we would expect better performing firms to compensate their executives in a manner consistent with their industry level performance in order to address this top talent retention sensitivity. III. Compensation Tools and Hypothesis Most studies to date have looked at the relationship between compensation and RPE, relative to a constructed benchmark. That relationship will be the ex poste result of both explicit RPE where the firm has made a public commitment and implicit RPE where a firm exercises its option to include RPE in its compensation levels after the performance period. An ex ante commitment will subject a firm to a greater possibility of compensation errors than the ex poste method due to the firm reducing its action possibility set. I begin by examining a firm that publicly commits to an RPE compensation mechanism. Theory says that if RPE provides a valuable signal to a firm concerning the performance of its management team then integrating RPE should help improve the pay-performance relation of the firm. If a firm operates in an homogenous industry where a more statistically meaningful signal of a firm s industry ranked performance level exists, then 11

12 we would expect a greater degree of RPE to be manifested in a firm s pay-performance relation if frictions do not significantly disrupt the meaningful signal theory. An alternative hypothesis is that if industry homogeneity is also a measure of the abundance of industry specific executive talent, then we would also expect that firms competing in more homogenous industries to be less sensitized to the chance of an executive departure, due to compensation factors, and therefore be more willing to commit to an RPE compensation scheme ex ante. As briefly reviewed in Appendix A, the major effect of the accounting and compensation practice frictions are to blunt the ability of firms to finely tune their compensation payouts via an RPE formula. The choice with respect to equity-based compensation is essentially to have RPE calculations impact the size of an initial grant or to possibly accelerate the vesting restrictions on the grant. The income statement expense required when modifying (or indexing the strike price of) an already granted option can be prohibitively expensive and is therefore generally considered inferior to pre-grant calculations or vesting accelerated alternatives. While cash-based programs do not suffer from this friction, they do have a greater problem. The cash flow associated with making a payout a material portion of an executive s annual compensation could be both economically and income statement prohibitively expensive. This makes it difficult to incorporate an economically significant, finely tuned RPE mechanism into the process and may be one of the material causes for limited systematic RPE in compensation payouts. 12

13 Descriptive Model I now proceed with a two-period descriptive model that incorporates a mild yet realistic RPE impact where the initial period is used to calibrate the second period RPE mechanism. The story begins with a trial period where a manager is compensated with a fixed salary, a, plus a share, b, of the profits, π, that the manager created for the firm during the performance period 1. (1) Compensati on0 = a0 + b0π 0 Profits that the manager creates by his own efforts for the firm are composed of her unknown ability, α, plus an effort, e, and a random component, ε. (2) π t = α t + et + ε t 2 where ε ~ N(0, ) is a stochastic component and e is a level of effort that the firm σ ε requires and has designed the contract to generate such an effort level 2. Note also that 2 first period ability is distributed α ~ N ( µ α, σ α ) and evolves over time according to α + t = α t 1 υt where t υ and ε t are uncorrelated and independent of each other 3. 1 I assume that the firm is able to differentiate profit creation among managers. 2 Both Gibbons and Murphy (1992) and Hartzell (1998) have shown that a given level of effort can be elicited from the manager in a similar setting. 3 This is the same general setup and notation used in Hartzell (1998). 13

14 The initial period payout occurs at t=0 when a new compensation contract is agreed upon whereby a firm-wide relative performance variable (or sharing rule modification variable), d, is introduced. The relative performance variable is a function of the firmwide performance compared to the remaining firms in the industry 4. An increased firm level relative performance generates a value of d that is greater than or equal to a lower level of relative performance. For the case where a firm ignores its industry comparable performance in the compensation mechanism, d will always be equal to one. (3) Compensati on1 = a1 + d1b1π 1 Although d 0 is not explicitly in equation (3) it is important in calibrating d 1 since (4) d 1 = d where ~ N (0, σ ) so that E ( d ) = d is a change in the firm s sharing rule modification variable that was caused by a shock to the firm s industry measured performance which might be attributed to executive personnel changes or other factors not related to the performance of the executive of interest. I assume that the shock to a firm s industry relative performance does not affect the profits that an individual executive will produce for the firm. Only the sharing rule 4 Conceptually, one can think of this variable as a function of the industry performance percentile of a firm within an industry. The function should have reasonable upper and lower bounds to preclude unrealistic payouts to or collections from an executive. 14

15 concerning the executive s individual profits will be altered by this shock. It is independent and uncorrelated with ε 1 and α 1. A target level of compensation can be calibrated using d 0 which is the value of the sharing rule modification variable that would have made the initial period payout equal to the payout that the firm would have made had it instituted RPE ex poste. It is easy to see that if a firm chooses to wait until after each period to set the RPE variable for the prior period payout, it greatly increases its ability to deliver the target payout. However, in that case the firm loses some of its ability to align the firm s industry related performance goals with those of its shareholders during the performance period. The target level of compensation for the executive can be calibrated using d 0 with the knowledge that some firm level shock will occur that could negatively or positively disrupt the executive s expected compensation level. The executive and the firm are equally aware of the shock when it occurs and revise their expectations for compensation at t= ½ which can be interpreted as occurring an instant after t=0. The time line is as follows: t= -1 t=0 t= ½ t=1 I I I I Calibration Initial contract Firm level RPE contract period paid. shock occurs. payout occurs. begins. RPE contract Executive leaves/stays. agreed. 15

16 I assume that executives are initially on a firm team that is equal to their own capabilities and efforts. That is, a good manager is on a good (industry performance measure) team. At t =1/2 a significant shock to the team performance level payout, 1, occurs that can produce a scenario where a good executive is now paired with a poor team level payout which causes the good executive s expected level of compensation to decrease relative to her initial target compensation established at t= 0. A large negative shift in expected compensation will encourage the executive to seek employment outside of the firm if that shortfall is large relative to her employment switching costs. With respect to a large positive shock for an initially poorly performing team, an executive could be considered a poor performer on a now high level performance payout team. In that case, the executive might be over compensated (in expectation) relative to the profits that she is expected to generate for the firm. If this over-compensation is large relative to the firm s executive replacement costs, then the executive will be terminated. It is apparent that calibration is central to the firm retaining the executive. In fact, in this setting, over-compensation or under-compensation can generate an executive departure. RPE becomes difficult to commit to, in this setting, if the variance of the shock is large or if executive replacement costs are large. In addition, if a firm is already an industry performance leader, a positive shock may not increase the payout significantly whereas a negative shock will have a good chance of decreasing the executive payout which in turn increases the possibility of a voluntary departure. A poor performing firm has the opposite situation. A negative shock will not hurt their industry ranking very much whereas a positive shock can increase the payouts dramatically. In that case the chance of over compensating an executive exists, but at least the decision to keep an executive remains in the hands of the 16

17 firm rather than with the executive. Since a lower performing firm may be partially shielded from a compensation calibration error, it should be more willing to commit to RPE ex ante than their better performing industry competitors. This does not preclude the need for the better performing competitors to take up their ex poste option to recalibrate their compensation levels with their industry competitors. In summary, this analysis predicts that poorer performing firms have greater compensation freedom to commit to RPE ex ante while their better performing counterparts may not have the freedom to commit to such a mechanism and yet must settle up ex poste in a manner that is reflective of the firm s industry performance. Two aspects of the descriptive model should be addressed. The non-linearity of the profit sharing function in the payout equation is not a prediction of Holmstrom (1982). Given the earlier mentioned preferred methods (vesting and grant values for equity based incentives) of incorporating RPE into compensation, the introduction of the sharing rule slope modification seems more realistic compared to introducing a separate but additive RPE performance sharing percentage rule. The latter would in practice introduce income statement expenses that firms have historically labored to avoid 5. The second aspect is important in that I assume that contract renegotiation does not take place if a public commitment to RPE is made ex ante. That assumption is based upon there being a significant cost to board member reputations in renegotiating a publicly committed contract. Repricing options for executive teams is an example of a contract renegotiation 5 For example, if an option is granted under conditions where vesting is not certain at some predetermined point in time, then the accounting for that option prescribes an expense which is open ended until that vesting date occurs. A term such as dπ added to equation (1) suggests uncertainty with respect to the existence of a grant rather than a timing or predetermined modification as in the case of equation (3). 17

18 and I refer to the backlash in the popular press from that as a significant cost borne by board member reputations as a reason for needing extreme circumstances to renegotiate a contract at the equivalent of t=1/2. The hypotheses that follow from above are as follows: Hypothesis I: Firms competing in more homogenous industries should be more inclined to incorporate relative performance than firms in less homogenous industries. This hypothesis is a test concerning firm use of a more accurate performance signal as well as a test concerning the replacement costs, and therefore retention concerns for a firm. Hypothesis II: Weaker performing firms, compared to their industry competitors should be more inclined to commit to an RPE mechanism than their stronger industry performing counterparts. This hypothesis is a test of performance in the use of RPE which is intended to proxy for retention concerns. Hypothesis III: Higher performing firms should exhibit more ex poste compensation RPE than lower performing firms. Confirmation of this hypothesis is not necessarily reflective of a firm s ex ante incentive calibration but may be part of the ex poste labor market settling-up process to compensate high performing executives at a higher level than their weaker performing industry peers. 18

19 IV. Data The paper proceeds with two separate empirical analyses: 1) An ex ante analysis concerning the propensity for a firm to commit to an RPE compensation mechanism, and 2) An ex poste analysis to determine if there is cross-sectional variation among firms that compensate their executives in a manner consistent with RPE. Two different, yet not mutually exclusive, data sets are utilized. Ex Ante Test Data Data was collected from company proxies on file with the SEC for all firms in the S&P 500 (as defined in the ExecuComp) for the Standard and Poor s defined year ending Data was collected and summarized for each major category of performance related compensation 6. If a firm noted that it was or will be using a relative performance based measure in either its grant calculations (pre or post-grant),vesting conditions (pure performance based or accelerated performance) or post-grant payouts then the firm was deemed to have used an RPE measure for that portion of its compensation package. If a firm did not mention RPE in its Compensation Committee Report on Executive Compensation but evidence was found that it did utilize RPE through review of the company s compensation table and footnotes, then that also qualified for RPE in this study. Data was collected for 462 firms in this manner. 6 These include Bonus, Options, Restricted Stock/Units, Performance Units, Phantom Units, and Long-term Performance Based Cash Plans. 19

20 As a proxy for industry homogeneity, mean partial correlation calculations are taken from Parrino (1997), p Parrino calculates mean partial correlation figures by regressing the monthly returns for each stock in the CRSP database from 1970 through 1988 against an equally weighted return index for each 2-digit SIC code and the equally weighted market return. The partial correlation coefficient for each stock with the 2-digit industry return is then averaged across a random selection maximum of 50 firms (minimum 35 observations) to arrive at the mean partial correlation (MPC) for each 2-digit SIC code. Following the procedure in Aggarwal and Samwick (1999b) I then calculate the sample cumulative distribution function for each 2-digit MPC to create F(M). The largest MPC will then have an F(M) value of 1.00 while the lowest will have a value of Market return and price information is sourced from CRSP while compensation and accounting data is from S&P Compustat and ExecuComp. When the data was combined with the necessary industry homogeneity data, the sample was further reduced to 403 firms and then further reduced to 359 firms for the probit analysis when the accounting, compensation and market return data were required. From the sample of 403 firms, Panel A of Table I shows that 207 of the 403 firms have 2- digit SIC codes which classifies them as manufacturing firms (20-39) while the remaining 196 firms are non-manufacturing. The sample consists of 38 different 2-digit SIC codes with 16 of those represented by manufacturing firms and the remaining 22 being non-manufacturing codes. Panel B suggests that although the sample consists of the largest firms in the United States, the sample is skewed with respect to market value 20

21 and revenues. The median (average) total common stock value and revenue for the firms is $8.6 billion ($25.5 billion) and $5.2 billion ($11.4 billion), respectfully. The mean and median industry homogeneity percentiles are both similar and close to.5 which suggests a reasonably symmetric distribution of firms in the total sample. The fewest number of firms in an industry which was used to generate industry relative performance calculations was 16 firms while the largest number was 1, 014. Greater than 50% of the firms had 0% of sales as research and development expenses and the median (mean) dividend yield was 1.12% (1.59%). The median (mean) percentage of property and plant as a percentage of book value is 26% (29%). Panel C details the frequency of RPE among the firms. If we consider all forms of RPE based compensation which are benchmarked according to any external accounting or market measure, then 134 firms are considered to have used RPE in any of their compensation delivery classifications. If we limit that benchmark to bonus compensation then only 53 firms utilized RPE while 99 utilized RPE for any long-term compensation component. Firms that utilized any RPE method for any long-term compensation component except options were 93 while 69 firms used RPE in their long-term cash related plans. The simple frequency of RPE suggests that the most common compensation tool used was non-option related long-term compensation such as restricted stock/units and long-term cash based plans. When we confine RPE to external industry related market measures, the frequency in all compensation choices is reduced to 94 firms. Only 23 firms elected to have RPE 21

22 influence their yearly bonus plans while 79 utilized RPE for any long-term compensation. Of those 79, 76 firms utilized RPE in their non-option related long-term choices and 57 firms used RPE in their long-term cash based plans. Once again, the predominant RPE tool was restricted stock/units and long-term cash based plans. Ex Poste Test Data The tests utilize the same methodology of Aggarwal and Samwick (1999b) and as such utilize a similar data set and time period as well as more recent data. Compensation data from Standard and Poor s ExecuComp database is combined with data from Compustat and with Herfindahl data from the Commerce Department s Census of Manufacturers. One again industry homogeneity calculations are taken from Parrino (1997), p. 189 and converted to sample cumulative distribution function value for each 2-digit MPC to create F(M). I do a similar calculation for the Herfindahl indices taken from the Commerce Departments s Census of Manufacturers survey at the 4-digit SIC-code level to calculate F(H). F(H) is included in the early regressions to ensure that it is not highly correlated with F(M). Although not reported in the tables, the Pearson Correlation Coefficient for F(M) and F(H) is for the sample. The Herfindahl data is available for 4-digit manufacturing SIC codes 2000 through 3999 only and therefore cannot be used in testing the non-manufacturing data. The ExecuComp data set includes compensation information for the Top 5 (one of which must be the CEO) compensated executives of a firm, ranked by salary plus bonus. The sample includes firm data for firms in the S&P 500, S&P Midcap 400, and S&P 22

23 SmallCap 600. The data include firms for the 1993 through 2000 years as defined by Compustat. Table IV includes summary data for 1995 and 2000 in December 1995 dollars. While the compensation in all categories has increased by a noticeable amount, the greatest increases were in the option portion of long-term compensation. CEO median (mean) total compensation increased from $1.307 ($2.257) million to $2.327 ($6.363) million for an increase of about 78% (182%) during the period. Short-term compensation increased from $.76 million ($1.052 million) to $.862 million ($1.352 million) or about 13% (29%) during the period while long-term compensation increased about 147% (316%) during that period. Of that component, median (mean) option grants increased by about 237% (389%). As a percent of total compensation long-term compensation increased from about 36.7% (36.8%) of the total package in 1995 to about 55.8% (51.8%). While this is consistent with greater pay-performance sensitivity for the sample, it is also consistent with firms endeavoring to increase the target level of pay for executives with a minimal income statement expense. For non-ceos the sample shows a similar trend. Median (mean) total compensation increased about 72% (158%) during the period while the short-term component increased 23% (40%) reflecting greater cash compensation percentage increases for non-ceos than for CEOs. The long-term component increased by 188% (287%) during the period that was primarily composed of 277% (377%) increases in the size of option grants. The greatest percentage increases came from long-term compensation components. Long-term compensation was 30.5% (32.1%) of total compensation in 1995 and increased to 49.8% (48.1%) in

24 Executive compensation increased at a rate much greater than inflation during the period. The simple statistics suggest that firms shifted a greater percentage of total compensation to long-term (performance related) components. Of particular interest is the size of the percentage increases in the CEO and non-ceo samples. It appears that non-ceo compensation increases were similar to CEO increases during the period. If the large increases in compensation during the period were indicative of the need for high-powered talent, the simple statistics suggest that sub-ceo executive talent demand was also high. An additional item is of particular interest. In all sub-samples, greater than 75% of the observations did not grant restricted stock/units. As discussed in the Appendix A, while restricted stock/units will be expensed in all cases and should therefore be RPE neutral, they are a compensation delivery choice for a minority of the firms in the sample. I focus on total compensation, or TDC1 as defined by ExecuComp and Total Current Compensation defined as total salary and bonus. This is a departure from Aggarwal and Samwick (1999b) who study total short-term compensation which also includes shortterm other compensation. The market return data is taken from the Center for Research in Security Prices (CRSP) database. 24

25 V. Empirical Results Ex Ante Tests of Hypothesis I and II Do firms vary in their decision to publicly commit to using RPE? Hypothesis I suggests that more homogenous firms should be more willing to commit to an RPE mechanism. As mentioned earlier, this is a joint hypothesis between a clearer signal and an indication that retention costs affect the ability of a firm to commit to RPE. If retention concerns are a major concern when designing compensation structures, then there should be some variation in firm RPE commitment by a retention proxy such as performance relative to an industry benchmark. If industry ranked performance is a proxy for retention concerns, then lower performing firms should exhibit a greater propensity to publicly commit to an RPE compensation mechanism than higher performing firms as stated in Hypothesis II. In order to test the relation of a firm s willingness to commit, I utilize a probit model where the likelihood of committing to an RPE mechanism is a function of: 1. 1 or 2 year return percentile for a firm within its 2-digit SIC code 2. F(M) or the cumulative density function associated with mean partial correlation for the 2-digit SIC code that is a firm s primary line of business or 2-yr industry return percentile X F(M) for a firm 4. Sales 5. R&D as percent of Sales 6. Number of Board meetings 7. Market/Book Value 8. Dividend Yield for the firm 9. Property and Plant as a percent of Book Value I divide commitment to RPE measures into five broad categories: 25

26 Accounting Relative Measures where a firm s financial statement measures are compared to an industry composite to calculate a type of compensation, Industry Relative Measures where a firm s market derived returns are compared to the market returns of a similar industry composite for a compensation calculation, Broad Market Relative Measures where a firm s market derived returns are compared to the market returns of some broad market measure or some static market level for a compensation calculation, Industry and Broad Market Measures where the market derived returns definition is expanded to include either industry relative or broad market measures, All Relative Measures where any of the above relative measures are counted toward RPE. In order to further classify the data, compensation components were classified according to categories where RPE was found. They include the following: Bonus All Long-term Compensation All Long-term Compensation excluding options Long-term Compensation that is cash based such as performance, phantom and long-term cash incentive plans All Compensation combined. 26

27 The probit relation utilizing 1-year percentile returns are not shown. Those results exhibit a weak relation to the likelihood of committing to an RPE compensation mechanism that is related to a firm s industry percentile performance. Because committing to an RPE measure may be perceived by compensation committees as a strategic tool, it is possible that boards consider industry rank over a period greater than one year. If a board needs to change the motivating factors for executives it seems reasonable that they would make their decision based upon long-run competitive trends rather than a single year s experience. In order to investigate that possibility, I test the same specification utilizing two-year industry rankings. The results for rankings beyond two years and up to five were also very mild. Table II reports the results using two-year rankings. The industry relative measures in columns (1)-(5) are supportive of Hypothesis I and mixed for Hypothesis II. F(M) is positive and highly significant when All Compensation, All LT Compensation and LT Compensation Without Options are the compensation categories, while LT Cash Compensation is positive and significant at the 10% level. The F(M)-2 yr interaction terms are negative and highly significant in two of the five compensation categories, while the All Compensation and LT Compensation Without Options are negative and significant at the 10% level. Higher performance-higher Homogeneity industry firms tend to be less likely to commit to an RPE choice while greater homogeneity alone appears to increase the likelihood that an RPE mechanism will be irrevocably put in place. Note that the 2-yr percentile coefficients in columns (3) and (4) are marginally significant and positive. This weakly suggests that higher ranked firms, absent their 27

28 homogeneity factors, are more likely to commit to an RPE method when that method involves two of the LT compensation definitions. Such a conclusion does not support Hypothesis II and is investigated in an additional specification. The relation with standard deviation suggests that firms with higher levels of risk tend not commit to an RPE method. If firm volatility impacts the likelihood of achieving a target level of compensation then volatility might also negatively impact a firm s ability to retain an executive within an RPE compensation framework. This finding also supports Aggarwal and Samwick s (1999a) result that executive pay performance sensitivity is decreasing in the variance of the firm. The broad market results (not shown) are generally statistically insignificant and the accounting relative results (also not shown) are not significant in any of the variables except a negative significant term for some of the standard deviation coefficients. The All Relative Measures RPE classification is consistent with the results of the Industry Relative Measure classification. The F(M) coefficients in columns (6), (8), (9) and (10) are positive and significant (not at conventional levels for column (10)). The F(M) interaction terms are negative and significant at the 10% level for columns (8) and (9). The results for the 2-yr percentile effects are insignificant in all cases. The results in Table II suggest that the high industry homogeneity firms are more likely to commit to an RPE mechanism, usually through a long-term compensation choice. They also weakly support Hypothesis II in that higher performing, more homogenous 28

29 firms are less likely to commit to RPE than low performing, more homogenous firms. If higher performance ranked firms do have greater retention concerns, as measured by both the cost of replacing a lost executive and the likelihood that an executive will leave the firm voluntarily, then the results support the alternative in Hypothesis I that firms with lower retention concerns will be more likely to publicly announce their intention to utilize an RPE method. In order to investigate the positive relation that Table II found concerning return percentile having a positive effect on the likelihood of commitment, a slightly different specification is used. Table III shows the results where the 2-year Industry %ile Return is interacted with a dummy variable which is equal to one if the firm performed in the top 50% of all firms in its industry and zero otherwise. The Top 50% dummy variable also enters the specification by itself and interacted with the 2-year Industry %ile return X Homogeneity %ile interaction term. The results from this alternative specification further support Hypothesis I and II. The 2-year Industry %ile Return X Top 50%ile dummy term is insignificant in all cases with both measures of RPE. This suggests that the positive inclination of better performing firms in Table II, absent their level of homogeneity, to impose an RPE mechanism is related to the weaker performing firms in the sample and not the upper 50%ile where there is an apparent aversion. In fact, it suggests that the better performing firms in the lower 50%ile, independent of their industry homogeneity, are more likely to commit while the same cannot be said of the better performing firms in the top 50%ile. 29

30 The results from the probit models support Hypothesis I and II which suggests evidence of retention concerns in the compensation relation. Ex Poste Tests of Hypothesis III The Ex Poste empirical tests generally follow the specification of Aggarwal and Samwick (1999b). I test RPE and how it varies due to changes in levels of performance and homogeneity. Homogeneity may serve as a proxy for the clarity of the performance signal received by the firma as well as a proxy for the level of industry specific human capital available as a replacement for a lost executive. In the case of the latter effect, I argue that a large amount of industry specific human capital available will lower retention concerns and this should be reflected in a firm s propensity to utilize RPE. The initial specification is as follows: (4) w ijt = η + η π jt + η π 2 r jt + η F(H 3 j ) π η8f(m j) + η 9 CEO it + ψ + 00 t= 94 0 jt t ε ijt + η F(H 4 j ) π r jt + η 5 F(H ) + η F(M j 6 j ) π 0 jt + η F(M 7 j ) π r jt + where, w ijt = compensation in thousands (total or total bonus plus salary) received by executive i at firm j in year t. 0 π jt = the total 1995-level dollar, in millions, adjusted return to all shareholders of firm j during period t-1 to t. 30

31 r π jt = the total 1995-level dollar, in millions, adjusted return to investing an amount equal to the total common stock value of firm j at time t-1 in an index composed of value weighted firms in the same 2-digit industry as firm j during period t-1 to t. The index is rebalanced monthly. F(H j ) = the cumulative density function associated with the Herfindahl index for the 4-digit SIC code that is firm j s primary line of business. F( M j ) = the cumulative density function associated with mean partial correlation for the 2-digit SIC code that is firm j s primary line of business. ψ t = a dummy variable to indicate whether the compensation was received in CEO it = a dummy variable indicating whether executive i is the CEO during period t. Equation 4 is also adjusted for factors that might be omitted such as sales, research and development as a percentage of sales, market-to-book value, monthly standard deviation of returns for the firm preceding 60 months prior to time t, an indicator for whether the executive is a member of the board of directors, the number of board meetings, and property plant and equipment as a percentage of book value. Prior research has indicated that these factors might impact compensation decisions 7. An additional specification investigated to check for changes in compensation: (5) w ijt = η + η π jt + η π 2 r jt + η F(H η8f(m j) + η 9 CEO it + ψ t= 94 j ) π 0 jt t ε ijt + η F(H 4 j ) π r jt + η 5 F(H ) + η F(M j 6 j ) π 0 jt + η F(M 7 j ) π r jt + 7 See Murphy (1998) 31

32 Equation (5) is also adjusted for the additional variables mentioned above. If industry homogeneity has an effect on firm use of RPE, then η 7 < 0 in equations (4) and (5). Also of interest is η 8 in equations (4) and (5) where if homogeneity helps to hold down replacement costs, then F(M) should have a negative effect on the level of compensation. However, such an homogeneity effect could also reflect lesser-skilled commodity type industries rather than labor market factors alone. While Aggarwal and Samwick (1999) utilized median regressions in the majority of their paper, OLS with standard errors robust to intra-company covariation is used in this analysis where the time series data is pooled. Hypothesis III states that high quality, or higher performing firms utilize RPE in a different manner than weaker performing firms. Therefore the skewed compensation data which is evident in Table IV may contain meaningful information that OLS will not de-emphasize. The results for manufacturing firms are presented in Table V. The specification initially utilizes F(H) in order to make a comparison to Aggarwal and Samwick (1999b). The results in columns (1) and (2) are similar to their findings with the exception that the results are statistically less significant. One departure from their results is in column (1) of Table V where the Herfindahl coefficient F(H) is positive and insignificant and the Herfindahl interaction term with own performance is also positive and insignificant 8. Also, the Table V specification finds the F(H)- index performance interaction term 8 Aggarwal and Samwick (199b) find the own performance interaction term to be positive and significant while the index interaction term is negative and significant. 32