Does it pay to ‘Be Like Mike’? Aspiratonal peer firms and relative performance evaluation

Key empirical papers on the relation between CEO compensation and aggregate performance of a competitive peer group are Antle and Smith (1986), Jensen and Murphy (1990), Barro and Barro (1990) Aggarwal and Samwick (1999b), Garvey and Milbourn (2003), and Albuquerque (2009), among others.

The lack of strong empirical support for the predictions from standard RPE theory has long troubled researchers, and recent studies continue to cast doubt on the framework’s sufficiency. For example, Jenter and Kanaan (2015) provide evidence that firms do not filter out common shocks when making firing decisions. In addition, Ma et al. (2017) provide empirical evidence inconsistent with firms optimally selecting peers solely on the exogenous covariance, which is predicted by standard RPE theory. They suggest that firms’ seemingly suboptimal choices result from boards not “getting it right.” However, the results of our study suggest that boards may still get it right, while the standard RPE model with competitive peers is not sufficient to capture the multi-dimensional nature of RPE.

The phrase “Be Like Mike” in the title of our study refers to a Gatorade commercial that originally aired in 1992 and again in 2015. In the commercial, footage of Michael Jordan playing basketball, juxtaposed with video of young kids imitating his moves, was used as a backdrop to the lyrics of a song: “Sometimes I dream that he is me. / You’ve got to see that’s how I dream to be. / I dream I move, I dream I groove. / Like Mike. If I could be like Mike.” The commercial embodies the spirit of our study and the results that we document.

To focus on the key assumptions and resulting empirical predictions of our model, we relegate some discussion and detailed derivations to Appendix A of this paper or refer the reader to Hemmer (2017) wherever appropriate.

A drawback of relying on the simple binomial version of Hemmer (2017) to provide intuition here is that \(p_{\tau }{<}\frac {1}{1+a}\) is required for peers being aspirational as here defined. This is, however, entirely a feature of the discrete binomial version of the model and does not carry over to the continuous Brownian version, from which the actual predictions are derived, obtained in the limit when the length of subperiods approach zero.

This is a fundamental divergence from the opposite assumption underlying the RPE literature. For a firm’s exposure to a common shock to be independent of an agent’s operational choices, the model structure has to be one of “effort-plus-noise” as, for example, in the standard LEN representation that much of the RPE literature relies on. Unfortunately, “effort-plus-noise” models do not lend themselves to the analysis of optimal contracts. (See the introduction of Jewitt (1988) for an excellent discussion of why.) Rather, the LEN RPE contract is the “best” contract available among the class of contracts that are linear in aggregate own and peer performance, and the key “common-shock filtering” result obtained from the LEN model results from restricting the analysis to focus on this exogenously specified, non-optimal contractual form. In fact, without the exogenous restriction to linear contracts, the so-called Mirrlees non-existence result applies, and all firms can attain approximate first-best solutions without the use of RPE. By contrast, the results we are testing in this paper are obtained as properties of optimal RPE contracts derived in an economic environment where optimal RPE contracts actually exist.

While the simple binomial structure we rely on is too restrictive to allow for the sign of γ_p to differ across peers, it is straightforward to generate parameter values in which the magnitude of γ_p is quite different from the perspective of the focal versus the peer firm. Moreover, in the full Brownian model from Hemmer (2017) that emerges as the limiting case of the discrete multinomial models, γ_p can differ across peers in both sign and magnitude for appropriately selected parameter values.

Because there can be multiple such equilibria, we cannot speak to what the agents’ actions look like in equilibrium, nor is this the focus of our study. Rather, we focus on identifying the equilibrium use of peer performance in the contracts that implement a given set of equilibrium actions.

This is significant because the empirical RPE literature that relies on the restriction that the exposure to economy-wide shocks is independent of the operational decisions firms make has largely been unable to establish β_π being different from zero. Hemmer (2017) demonstrates that, when the exposure to the common risk is not independent of managerial decisions (as it cannot be in this setting for the reason discussed above), it is not optimal to simply filter out the common component in the optimal RPE contract.

Many of the components of total compensation are awarded based on contemporaneous performance, and these components conform most closely to the constructs in our model. If long-term incentive payouts depend on actions taken in prior years, this component introduces noise to our measure of total compensation because actions only affect contemporaneous firm performance (based on the structure of our model). However, such timing-driven measurement error in our total compensation proxy should only affect the dependent variable, rather than the independent variables, and therefore should not bias our coefficient estimates.

For example, consider the manager of a passive market index fund that is focused on tracking the underlying market index. He or she is provided incentives to take actions (e.g., periodically rebalancing the portfolio to accommodate changes in the index) that increase the correlation of the fund’s performance with that of the market index, which primarily represents the systematic component. By contrast, a hedge fund manager may focus on emulating an aspirational peer fund manager in an effort to increase abnormal, risk-adjusted return performance, which primarily represents an idiosyncratic component. Thus the informativeness of distinguishing between these two return components crucially hinges on the ability to identify the specific nature of the strategies of the aspirational peer that the firms is attempting to emulate. While interesting, it is outside the scope of our study, so again we leave it for future research.

All test statistics are based on standard errors clustered by firm.

Similar results are obtained for all specifications using value-weighted peer performance measures instead of equal-weighted measures. Specifically, Tables 3 and 4 are replicated using value-weighted performance measures. Results are reported in Tables OA.1 and OA.2, respectively, of the online appendix.

As discussed in Section 2.2, a two-way relation would arise in the framework of Hemmer (2017) if, for example, firms find it optimal to adopt each others’ best practices or learn from each others’ successes. In addition, a one-way relation could arise if the focal firm benefits from imitating the peer’s technology or business model, whereas either (i) the peer is indifferent between imitating or differentiating from the focal firm, or (ii) the peer prefers to innovate to distinguish itself from the focal firm but declines to specify non-aspirational peers for the reasons discussed in Section 2.2. By contrast, traditional RPE models preclude one-way relations by prescribing that firms treat each other symmetrically based on the sign of their covariance when filtering common shocks from compensation.

Representative RPE studies based on industry peer groups include those by Antle and Smith (1986), Gibbons and Murphy (1990), Jensen and Murphy (1990), Barro and Barro (1990), Janakiraman et al. (1992), Aggarwal and Samwick (1999a, b), Garvey and Milbourn (2006), Rajgopal et al. (2006), and Albuquerque (2009; 2014).

The variable perfcorr_jt,k is defined as the performance correlation between firm j and its peer firm k in fiscal year t. We scale by the total number of j’s peer firms in t (N_jt) to increase comparability across firm-year observations.