Tests for relative performance evaluation based on assumptions derived from proxy statement disclosures

The importance of the long term performance in contracts is studied by Vogel and Lobo (2002) who study the impact of adoption of long-term performance plans on financial analysts’ long-term forecasts, Lam and Chng (2006) who look at the value implications of stock option grants and Lin et al. (2010) who look at warrants and corporate performance.

Although Arya et al. (2005) frame their discussion in the context of divisions within a firm, the concept would also apply to choosing standardized measures at the corporate level to facilitate comparisons with other firms.

Gibbons and Murphy (1990) find support only with a stock returns performance metric, whereas Antle and Smith (1986) find some support with the return on assets performance metric.

Consequently, our goal is to test for RPE based on how firms actually report to use RPE as opposed to how RPE should be used predicated on theoretical studies (e.g., Janakiraman et al. 1992).

We perform sensitivity analyses to assess whether excluding stock options is important to the detection of RPE.

Janakiraman et al. (1992) view a negative weight on peer group performance as a “weak” implication of the RPE hypothesis. “Strong form” tests refer to tests that check for a specific negative relation among the coefficients on firm-own and peer group performance. Hence, negative weight on peer group performance is generally viewed as a minimum requirement for finding support for RPE.

The coefficient of peer group performance should be equal to the coefficient of firm performance times the ratio of the standard deviation of firm performance to the standard deviation of peer group performance times the correlation between firm and peer group performance times negative one if peer group performance is completely filtered out.

Antle and Smith (1986) find some support for the use of RPE among 16 of 39 firms in the chemical, aerospace, and electronics industries during 1947 to 1977 when examining the return on assets performance metric, but generally do not find support for RPE when using stock returns as a performance metric. Albuquerque (2005) finds stronger support for RPE when using peer groups that incorporate both size and industry, rather than simply defining peers as firms in the same industry.

The papers referred to in this section that use weak form tests and strong form tests can be characterized as searching for RPE under the assumption that all firms are equally likely to use RPE in compensation contracts (e.g., Gibbons and Murphy 1990; and Janakiraman et al. 1992). A few other studies attempt to identify environments or subsets of firms where RPE is most likely to be observed. Using similar types of empirical tests, Garvey and Milbourn (2003) and Aggarwal and Samwick (1999b) find evidence consistent with RPE use in selected environments. Joh (1999) finds evidence of RPE among Japanese firms. In these three studies, propositions regarding which environments are most likely to be conducive to RPE employment replace the assumption that all firms are equally likely to employ RPE.

Furthermore, Maher (1987, p. 296) notes that it is difficult to interpret the outcome of research seeking to detect RPE when such research does not take into account how companies make explicit use of RPE.

The reason that some firms were deleted is that financial statements were not filed for firms that were taken over via a pooling of interests.

The RPE associated plans identified in this paper were in place for at least one year.

This rate is similar to Bannister and Newman’s (2003) finding of 28% using a sample based on the 1992 Fortune 250.

We believe that return on assets and return on capital are measured similarly enough that a separate presentation is not warranted. Further, we note that the combined return on assets measure is only the fourth most commonly used RPE measure.

The peer group we refer to is the one the firm uses to assess relative performance as provided in the compensation committee report. This usually is not the peer group used in the proxy performance graph, although it is possible that the two peer groups can be the same (Byrd et al. 1998).

Later in the paper, we explore whether using the level variable ROE (i.e., 1998 ROE) rather than CHROE, influences our empirical findings.

For example, assume that the means of Firm A’s and Firm B’s peer group distributions are equal and that Firm A’s own performance exceeds that of Firm B. Also assume that the standard deviation of Firm A’s peer group is much greater than the standard deviation of firm B’s peer group, so that Firm A’s performance lies at the 60th percentile of its peer group distribution while Firm B’s performance lies at the 90th percentile of its peer group. In a cross-sectional model predicated on two separate performance variables, one for the firm and one for the peer group, Firm A would be expected to pay greater compensation than would Firm B. However, if firms compensate on the basis of where performance falls in the peer distribution, Firm A would actually pay either equal or less compensation than would Firm B, not more. In our model, where firm and peer group distribution are examined jointly, equal compensation would be paid. This would match with actual compensation practices if Firm A and Firm B actually compensate on the basis of which half of the peer group distribution firm performance falls into. If Firm A and Firm B use a more refined compensation system where compensation decisions are made on the basis of which quartile a firm’s performance falls into, then Firm A would actually pay less compensation than Firm B, whereas our model would have both firms paying equal compensation, and the model based on magnitude differences would have Firm A pay more than Firm B. While our model performs better than the model based on magnitude differences, we acknowledge that our structure may not adequately capture the compensation model used by all firms.

We perform these tests because empirical evidence of RPE will not automatically follow when a firm states that it uses RPE. One reason is that RPE may be used in a manner that would not be detectable using standard regression methods. Second, RPE is generally applied to only one or two compensation plans, it remains to be demonstrated that RPE can be detected when using total compensation excluding stock option grants as a compensation measure. Third, sometimes compensation plans state that factors in addition to peer group performance influence the payout and those other factors may dominate RPE.

Significant coefficients could indicate that the performance measures actually used by these firms are highly correlated with RET or ROE.

White's (1980) direct test for heteroscedasticity indicates that the results of the regressions in Table 4, as well as the results for the regressions reported later in this section, are not influenced by heteroscedasticity.

We also ran the model including CEO age and firm size (log of total assets) as control variables. Age (Gibbons and Murphy 1990) and firm size (Rajgopal et al. 2006) have been used in examinations of the relation between pay and performance. The results of this model are quantitatively and qualitatively similar to those reported in Table 6.